2 * Copyright (c) 2009, 2010, 2011, 2012 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-provider.h"
26 #include "byte-order.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
36 #include "mac-learning.h"
37 #include "meta-flow.h"
38 #include "multipath.h"
45 #include "ofp-print.h"
46 #include "ofproto-dpif-sflow.h"
47 #include "poll-loop.h"
49 #include "unaligned.h"
51 #include "vlan-bitmap.h"
54 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
56 COVERAGE_DEFINE(ofproto_dpif_ctlr_action);
57 COVERAGE_DEFINE(ofproto_dpif_expired);
58 COVERAGE_DEFINE(ofproto_dpif_no_packet_in);
59 COVERAGE_DEFINE(ofproto_dpif_xlate);
60 COVERAGE_DEFINE(facet_changed_rule);
61 COVERAGE_DEFINE(facet_invalidated);
62 COVERAGE_DEFINE(facet_revalidate);
63 COVERAGE_DEFINE(facet_unexpected);
65 /* Maximum depth of flow table recursion (due to resubmit actions) in a
66 * flow translation. */
67 #define MAX_RESUBMIT_RECURSION 32
69 /* Number of implemented OpenFlow tables. */
70 enum { N_TABLES = 255 };
71 BUILD_ASSERT_DECL(N_TABLES >= 1 && N_TABLES <= 255);
81 * - Do include packets and bytes from facets that have been deleted or
82 * whose own statistics have been folded into the rule.
84 * - Do include packets and bytes sent "by hand" that were accounted to
85 * the rule without any facet being involved (this is a rare corner
86 * case in rule_execute()).
88 * - Do not include packet or bytes that can be obtained from any facet's
89 * packet_count or byte_count member or that can be obtained from the
90 * datapath by, e.g., dpif_flow_get() for any subfacet.
92 uint64_t packet_count; /* Number of packets received. */
93 uint64_t byte_count; /* Number of bytes received. */
95 tag_type tag; /* Caches rule_calculate_tag() result. */
97 struct list facets; /* List of "struct facet"s. */
100 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
102 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
105 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
106 const struct flow *, uint8_t table);
108 static void flow_push_stats(struct rule_dpif *, const struct flow *,
109 uint64_t packets, uint64_t bytes,
112 static uint32_t rule_calculate_tag(const struct flow *,
113 const struct flow_wildcards *,
115 static void rule_invalidate(const struct rule_dpif *);
117 #define MAX_MIRRORS 32
118 typedef uint32_t mirror_mask_t;
119 #define MIRROR_MASK_C(X) UINT32_C(X)
120 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
122 struct ofproto_dpif *ofproto; /* Owning ofproto. */
123 size_t idx; /* In ofproto's "mirrors" array. */
124 void *aux; /* Key supplied by ofproto's client. */
125 char *name; /* Identifier for log messages. */
127 /* Selection criteria. */
128 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
129 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
130 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
132 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
133 struct ofbundle *out; /* Output port or NULL. */
134 int out_vlan; /* Output VLAN or -1. */
135 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
138 int64_t packet_count; /* Number of packets sent. */
139 int64_t byte_count; /* Number of bytes sent. */
142 static void mirror_destroy(struct ofmirror *);
143 static void update_mirror_stats(struct ofproto_dpif *ofproto,
144 mirror_mask_t mirrors,
145 uint64_t packets, uint64_t bytes);
148 struct ofproto_dpif *ofproto; /* Owning ofproto. */
149 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
150 void *aux; /* Key supplied by ofproto's client. */
151 char *name; /* Identifier for log messages. */
154 struct list ports; /* Contains "struct ofport"s. */
155 enum port_vlan_mode vlan_mode; /* VLAN mode */
156 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
157 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
158 * NULL if all VLANs are trunked. */
159 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
160 struct bond *bond; /* Nonnull iff more than one port. */
161 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
164 bool floodable; /* True if no port has OFPPC_NO_FLOOD set. */
166 /* Port mirroring info. */
167 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
168 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
169 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
172 static void bundle_remove(struct ofport *);
173 static void bundle_update(struct ofbundle *);
174 static void bundle_destroy(struct ofbundle *);
175 static void bundle_del_port(struct ofport_dpif *);
176 static void bundle_run(struct ofbundle *);
177 static void bundle_wait(struct ofbundle *);
178 static struct ofbundle *lookup_input_bundle(struct ofproto_dpif *,
179 uint16_t in_port, bool warn);
181 /* A controller may use OFPP_NONE as the ingress port to indicate that
182 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
183 * when an input bundle is needed for validation (e.g., mirroring or
184 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
185 * any 'port' structs, so care must be taken when dealing with it. */
186 static struct ofbundle ofpp_none_bundle = {
188 .vlan_mode = PORT_VLAN_TRUNK
191 static void stp_run(struct ofproto_dpif *ofproto);
192 static void stp_wait(struct ofproto_dpif *ofproto);
193 static int set_stp_port(struct ofport *,
194 const struct ofproto_port_stp_settings *);
196 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
198 struct action_xlate_ctx {
199 /* action_xlate_ctx_init() initializes these members. */
202 struct ofproto_dpif *ofproto;
204 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
205 * this flow when actions change header fields. */
208 /* The packet corresponding to 'flow', or a null pointer if we are
209 * revalidating without a packet to refer to. */
210 const struct ofpbuf *packet;
212 /* Should OFPP_NORMAL MAC learning and NXAST_LEARN actions execute? We
213 * want to execute them if we are actually processing a packet, or if we
214 * are accounting for packets that the datapath has processed, but not if
215 * we are just revalidating. */
218 /* The rule that we are currently translating, or NULL. */
219 struct rule_dpif *rule;
221 /* Union of the set of TCP flags seen so far in this flow. (Used only by
222 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
226 /* If nonnull, called just before executing a resubmit action. In
227 * addition, disables logging of traces when the recursion depth is
230 * This is normally null so the client has to set it manually after
231 * calling action_xlate_ctx_init(). */
232 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *);
234 /* xlate_actions() initializes and uses these members. The client might want
235 * to look at them after it returns. */
237 struct ofpbuf *odp_actions; /* Datapath actions. */
238 tag_type tags; /* Tags associated with actions. */
239 bool may_set_up_flow; /* True ordinarily; false if the actions must
240 * be reassessed for every packet. */
241 bool has_learn; /* Actions include NXAST_LEARN? */
242 bool has_normal; /* Actions output to OFPP_NORMAL? */
243 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
244 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
245 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
247 /* xlate_actions() initializes and uses these members, but the client has no
248 * reason to look at them. */
250 int recurse; /* Recursion level, via xlate_table_action. */
251 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
252 struct flow base_flow; /* Flow at the last commit. */
253 uint32_t orig_skb_priority; /* Priority when packet arrived. */
254 uint8_t table_id; /* OpenFlow table ID where flow was found. */
255 uint32_t sflow_n_outputs; /* Number of output ports. */
256 uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
257 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
258 bool exit; /* No further actions should be processed. */
261 static void action_xlate_ctx_init(struct action_xlate_ctx *,
262 struct ofproto_dpif *, const struct flow *,
263 ovs_be16 initial_tci, struct rule_dpif *,
264 uint8_t tcp_flags, const struct ofpbuf *);
265 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
266 const union ofp_action *in, size_t n_in);
268 /* An exact-match instantiation of an OpenFlow flow.
270 * A facet associates a "struct flow", which represents the Open vSwitch
271 * userspace idea of an exact-match flow, with one or more subfacets. Each
272 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
273 * the facet. When the kernel module (or other dpif implementation) and Open
274 * vSwitch userspace agree on the definition of a flow key, there is exactly
275 * one subfacet per facet. If the dpif implementation supports more-specific
276 * flow matching than userspace, however, a facet can have more than one
277 * subfacet, each of which corresponds to some distinction in flow that
278 * userspace simply doesn't understand.
280 * Flow expiration works in terms of subfacets, so a facet must have at least
281 * one subfacet or it will never expire, leaking memory. */
284 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
285 struct list list_node; /* In owning rule's 'facets' list. */
286 struct rule_dpif *rule; /* Owning rule. */
289 struct list subfacets;
290 long long int used; /* Time last used; time created if not used. */
297 * - Do include packets and bytes sent "by hand", e.g. with
300 * - Do include packets and bytes that were obtained from the datapath
301 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
302 * DPIF_FP_ZERO_STATS).
304 * - Do not include packets or bytes that can be obtained from the
305 * datapath for any existing subfacet.
307 uint64_t packet_count; /* Number of packets received. */
308 uint64_t byte_count; /* Number of bytes received. */
310 /* Resubmit statistics. */
311 uint64_t prev_packet_count; /* Number of packets from last stats push. */
312 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
313 long long int prev_used; /* Used time from last stats push. */
316 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
317 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
318 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
320 /* Properties of datapath actions.
322 * Every subfacet has its own actions because actions can differ slightly
323 * between splintered and non-splintered subfacets due to the VLAN tag
324 * being initially different (present vs. absent). All of them have these
325 * properties in common so we just store one copy of them here. */
326 bool may_install; /* Reassess actions for every packet? */
327 bool has_learn; /* Actions include NXAST_LEARN? */
328 bool has_normal; /* Actions output to OFPP_NORMAL? */
329 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
330 tag_type tags; /* Tags that would require revalidation. */
331 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
334 static struct facet *facet_create(struct rule_dpif *, const struct flow *);
335 static void facet_remove(struct facet *);
336 static void facet_free(struct facet *);
338 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
339 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
340 const struct flow *);
341 static bool facet_revalidate(struct facet *);
342 static bool facet_check_consistency(struct facet *);
344 static void facet_flush_stats(struct facet *);
346 static void facet_update_time(struct facet *, long long int used);
347 static void facet_reset_counters(struct facet *);
348 static void facet_push_stats(struct facet *);
349 static void facet_account(struct facet *);
351 static bool facet_is_controller_flow(struct facet *);
353 /* A dpif flow and actions associated with a facet.
355 * See also the large comment on struct facet. */
358 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
359 struct list list_node; /* In struct facet's 'facets' list. */
360 struct facet *facet; /* Owning facet. */
364 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
365 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
366 * regenerate the ODP flow key from ->facet->flow. */
367 enum odp_key_fitness key_fitness;
371 long long int used; /* Time last used; time created if not used. */
373 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
374 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
378 * These should be essentially identical for every subfacet in a facet, but
379 * may differ in trivial ways due to VLAN splinters. */
380 size_t actions_len; /* Number of bytes in actions[]. */
381 struct nlattr *actions; /* Datapath actions. */
383 bool installed; /* Installed in datapath? */
385 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
386 * splinters can cause it to differ. This value should be removed when
387 * the VLAN splinters feature is no longer needed. */
388 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
391 static struct subfacet *subfacet_create(struct facet *, enum odp_key_fitness,
392 const struct nlattr *key,
393 size_t key_len, ovs_be16 initial_tci);
394 static struct subfacet *subfacet_find(struct ofproto_dpif *,
395 const struct nlattr *key, size_t key_len);
396 static void subfacet_destroy(struct subfacet *);
397 static void subfacet_destroy__(struct subfacet *);
398 static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *,
400 static void subfacet_reset_dp_stats(struct subfacet *,
401 struct dpif_flow_stats *);
402 static void subfacet_update_time(struct subfacet *, long long int used);
403 static void subfacet_update_stats(struct subfacet *,
404 const struct dpif_flow_stats *);
405 static void subfacet_make_actions(struct subfacet *,
406 const struct ofpbuf *packet);
407 static int subfacet_install(struct subfacet *,
408 const struct nlattr *actions, size_t actions_len,
409 struct dpif_flow_stats *);
410 static void subfacet_uninstall(struct subfacet *);
416 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
417 struct list bundle_node; /* In struct ofbundle's "ports" list. */
418 struct cfm *cfm; /* Connectivity Fault Management, if any. */
419 tag_type tag; /* Tag associated with this port. */
420 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
421 bool may_enable; /* May be enabled in bonds. */
424 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
425 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
426 long long int stp_state_entered;
428 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
430 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
432 * This is deprecated. It is only for compatibility with broken device
433 * drivers in old versions of Linux that do not properly support VLANs when
434 * VLAN devices are not used. When broken device drivers are no longer in
435 * widespread use, we will delete these interfaces. */
436 uint16_t realdev_ofp_port;
440 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
441 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
442 * traffic egressing the 'ofport' with that priority should be marked with. */
443 struct priority_to_dscp {
444 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
445 uint32_t priority; /* Priority of this queue (see struct flow). */
447 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
450 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
452 * This is deprecated. It is only for compatibility with broken device drivers
453 * in old versions of Linux that do not properly support VLANs when VLAN
454 * devices are not used. When broken device drivers are no longer in
455 * widespread use, we will delete these interfaces. */
456 struct vlan_splinter {
457 struct hmap_node realdev_vid_node;
458 struct hmap_node vlandev_node;
459 uint16_t realdev_ofp_port;
460 uint16_t vlandev_ofp_port;
464 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
465 uint32_t realdev, ovs_be16 vlan_tci);
466 static uint16_t vsp_vlandev_to_realdev(const struct ofproto_dpif *,
467 uint16_t vlandev, int *vid);
468 static void vsp_remove(struct ofport_dpif *);
469 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
471 static struct ofport_dpif *
472 ofport_dpif_cast(const struct ofport *ofport)
474 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
475 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
478 static void port_run(struct ofport_dpif *);
479 static void port_wait(struct ofport_dpif *);
480 static int set_cfm(struct ofport *, const struct cfm_settings *);
481 static void ofport_clear_priorities(struct ofport_dpif *);
483 struct dpif_completion {
484 struct list list_node;
485 struct ofoperation *op;
488 /* Extra information about a classifier table.
489 * Currently used just for optimized flow revalidation. */
491 /* If either of these is nonnull, then this table has a form that allows
492 * flows to be tagged to avoid revalidating most flows for the most common
493 * kinds of flow table changes. */
494 struct cls_table *catchall_table; /* Table that wildcards all fields. */
495 struct cls_table *other_table; /* Table with any other wildcard set. */
496 uint32_t basis; /* Keeps each table's tags separate. */
499 struct ofproto_dpif {
500 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
509 struct netflow *netflow;
510 struct dpif_sflow *sflow;
511 struct hmap bundles; /* Contains "struct ofbundle"s. */
512 struct mac_learning *ml;
513 struct ofmirror *mirrors[MAX_MIRRORS];
514 bool has_bonded_bundles;
517 struct timer next_expiration;
521 struct hmap subfacets;
524 struct table_dpif tables[N_TABLES];
525 bool need_revalidate;
526 struct tag_set revalidate_set;
528 /* Support for debugging async flow mods. */
529 struct list completions;
531 bool has_bundle_action; /* True when the first bundle action appears. */
532 struct netdev_stats stats; /* To account packets generated and consumed in
537 long long int stp_last_tick;
539 /* VLAN splinters. */
540 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
541 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
544 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
545 * for debugging the asynchronous flow_mod implementation.) */
548 /* All existing ofproto_dpif instances, indexed by ->up.name. */
549 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
551 static void ofproto_dpif_unixctl_init(void);
553 static struct ofproto_dpif *
554 ofproto_dpif_cast(const struct ofproto *ofproto)
556 assert(ofproto->ofproto_class == &ofproto_dpif_class);
557 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
560 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
562 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
564 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
565 const struct ofpbuf *, ovs_be16 initial_tci,
568 /* Packet processing. */
569 static void update_learning_table(struct ofproto_dpif *,
570 const struct flow *, int vlan,
573 #define FLOW_MISS_MAX_BATCH 50
574 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
576 /* Flow expiration. */
577 static int expire(struct ofproto_dpif *);
580 static void send_netflow_active_timeouts(struct ofproto_dpif *);
583 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
585 compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions,
586 const struct flow *, uint32_t odp_port);
587 static void add_mirror_actions(struct action_xlate_ctx *ctx,
588 const struct flow *flow);
589 /* Global variables. */
590 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
592 /* Factory functions. */
595 enumerate_types(struct sset *types)
597 dp_enumerate_types(types);
601 enumerate_names(const char *type, struct sset *names)
603 return dp_enumerate_names(type, names);
607 del(const char *type, const char *name)
612 error = dpif_open(name, type, &dpif);
614 error = dpif_delete(dpif);
620 /* Basic life-cycle. */
622 static struct ofproto *
625 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
630 dealloc(struct ofproto *ofproto_)
632 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
637 construct(struct ofproto *ofproto_)
639 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
640 const char *name = ofproto->up.name;
644 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
646 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
650 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
651 ofproto->n_matches = 0;
653 dpif_flow_flush(ofproto->dpif);
654 dpif_recv_purge(ofproto->dpif);
656 error = dpif_recv_set(ofproto->dpif, true);
658 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
659 dpif_close(ofproto->dpif);
663 ofproto->netflow = NULL;
664 ofproto->sflow = NULL;
666 hmap_init(&ofproto->bundles);
667 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
668 for (i = 0; i < MAX_MIRRORS; i++) {
669 ofproto->mirrors[i] = NULL;
671 ofproto->has_bonded_bundles = false;
673 timer_set_duration(&ofproto->next_expiration, 1000);
675 hmap_init(&ofproto->facets);
676 hmap_init(&ofproto->subfacets);
678 for (i = 0; i < N_TABLES; i++) {
679 struct table_dpif *table = &ofproto->tables[i];
681 table->catchall_table = NULL;
682 table->other_table = NULL;
683 table->basis = random_uint32();
685 ofproto->need_revalidate = false;
686 tag_set_init(&ofproto->revalidate_set);
688 list_init(&ofproto->completions);
690 ofproto_dpif_unixctl_init();
692 ofproto->has_bundle_action = false;
694 hmap_init(&ofproto->vlandev_map);
695 hmap_init(&ofproto->realdev_vid_map);
697 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
698 hash_string(ofproto->up.name, 0));
699 memset(&ofproto->stats, 0, sizeof ofproto->stats);
701 ofproto_init_tables(ofproto_, N_TABLES);
707 complete_operations(struct ofproto_dpif *ofproto)
709 struct dpif_completion *c, *next;
711 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
712 ofoperation_complete(c->op, 0);
713 list_remove(&c->list_node);
719 destruct(struct ofproto *ofproto_)
721 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
722 struct rule_dpif *rule, *next_rule;
723 struct oftable *table;
726 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
727 complete_operations(ofproto);
729 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
730 struct cls_cursor cursor;
732 cls_cursor_init(&cursor, &table->cls, NULL);
733 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
734 ofproto_rule_destroy(&rule->up);
738 for (i = 0; i < MAX_MIRRORS; i++) {
739 mirror_destroy(ofproto->mirrors[i]);
742 netflow_destroy(ofproto->netflow);
743 dpif_sflow_destroy(ofproto->sflow);
744 hmap_destroy(&ofproto->bundles);
745 mac_learning_destroy(ofproto->ml);
747 hmap_destroy(&ofproto->facets);
748 hmap_destroy(&ofproto->subfacets);
750 hmap_destroy(&ofproto->vlandev_map);
751 hmap_destroy(&ofproto->realdev_vid_map);
753 dpif_close(ofproto->dpif);
757 run_fast(struct ofproto *ofproto_)
759 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
762 /* Handle one or more batches of upcalls, until there's nothing left to do
763 * or until we do a fixed total amount of work.
765 * We do work in batches because it can be much cheaper to set up a number
766 * of flows and fire off their patches all at once. We do multiple batches
767 * because in some cases handling a packet can cause another packet to be
768 * queued almost immediately as part of the return flow. Both
769 * optimizations can make major improvements on some benchmarks and
770 * presumably for real traffic as well. */
772 while (work < FLOW_MISS_MAX_BATCH) {
773 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
783 run(struct ofproto *ofproto_)
785 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
786 struct ofport_dpif *ofport;
787 struct ofbundle *bundle;
791 complete_operations(ofproto);
793 dpif_run(ofproto->dpif);
795 error = run_fast(ofproto_);
800 if (timer_expired(&ofproto->next_expiration)) {
801 int delay = expire(ofproto);
802 timer_set_duration(&ofproto->next_expiration, delay);
805 if (ofproto->netflow) {
806 if (netflow_run(ofproto->netflow)) {
807 send_netflow_active_timeouts(ofproto);
810 if (ofproto->sflow) {
811 dpif_sflow_run(ofproto->sflow);
814 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
817 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
822 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
824 /* Now revalidate if there's anything to do. */
825 if (ofproto->need_revalidate
826 || !tag_set_is_empty(&ofproto->revalidate_set)) {
827 struct tag_set revalidate_set = ofproto->revalidate_set;
828 bool revalidate_all = ofproto->need_revalidate;
829 struct facet *facet, *next;
831 /* Clear the revalidation flags. */
832 tag_set_init(&ofproto->revalidate_set);
833 ofproto->need_revalidate = false;
835 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
837 || tag_set_intersects(&revalidate_set, facet->tags)) {
838 facet_revalidate(facet);
843 /* Check the consistency of a random facet, to aid debugging. */
844 if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) {
847 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
848 struct facet, hmap_node);
849 if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) {
850 if (!facet_check_consistency(facet)) {
851 ofproto->need_revalidate = true;
860 wait(struct ofproto *ofproto_)
862 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
863 struct ofport_dpif *ofport;
864 struct ofbundle *bundle;
866 if (!clogged && !list_is_empty(&ofproto->completions)) {
867 poll_immediate_wake();
870 dpif_wait(ofproto->dpif);
871 dpif_recv_wait(ofproto->dpif);
872 if (ofproto->sflow) {
873 dpif_sflow_wait(ofproto->sflow);
875 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
876 poll_immediate_wake();
878 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
881 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
884 if (ofproto->netflow) {
885 netflow_wait(ofproto->netflow);
887 mac_learning_wait(ofproto->ml);
889 if (ofproto->need_revalidate) {
890 /* Shouldn't happen, but if it does just go around again. */
891 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
892 poll_immediate_wake();
894 timer_wait(&ofproto->next_expiration);
899 flush(struct ofproto *ofproto_)
901 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
902 struct facet *facet, *next_facet;
904 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
905 /* Mark the facet as not installed so that facet_remove() doesn't
906 * bother trying to uninstall it. There is no point in uninstalling it
907 * individually since we are about to blow away all the facets with
908 * dpif_flow_flush(). */
909 struct subfacet *subfacet;
911 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
912 subfacet->installed = false;
913 subfacet->dp_packet_count = 0;
914 subfacet->dp_byte_count = 0;
918 dpif_flow_flush(ofproto->dpif);
922 get_features(struct ofproto *ofproto_ OVS_UNUSED,
923 bool *arp_match_ip, uint32_t *actions)
925 *arp_match_ip = true;
926 *actions = ((1u << OFPAT_OUTPUT) |
927 (1u << OFPAT_SET_VLAN_VID) |
928 (1u << OFPAT_SET_VLAN_PCP) |
929 (1u << OFPAT_STRIP_VLAN) |
930 (1u << OFPAT_SET_DL_SRC) |
931 (1u << OFPAT_SET_DL_DST) |
932 (1u << OFPAT_SET_NW_SRC) |
933 (1u << OFPAT_SET_NW_DST) |
934 (1u << OFPAT_SET_NW_TOS) |
935 (1u << OFPAT_SET_TP_SRC) |
936 (1u << OFPAT_SET_TP_DST) |
937 (1u << OFPAT_ENQUEUE));
941 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
943 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
944 struct dpif_dp_stats s;
946 strcpy(ots->name, "classifier");
948 dpif_get_dp_stats(ofproto->dpif, &s);
949 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
950 put_32aligned_be64(&ots->matched_count,
951 htonll(s.n_hit + ofproto->n_matches));
954 static struct ofport *
957 struct ofport_dpif *port = xmalloc(sizeof *port);
962 port_dealloc(struct ofport *port_)
964 struct ofport_dpif *port = ofport_dpif_cast(port_);
969 port_construct(struct ofport *port_)
971 struct ofport_dpif *port = ofport_dpif_cast(port_);
972 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
974 ofproto->need_revalidate = true;
975 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
978 port->tag = tag_create_random();
979 port->may_enable = true;
980 port->stp_port = NULL;
981 port->stp_state = STP_DISABLED;
982 hmap_init(&port->priorities);
983 port->realdev_ofp_port = 0;
984 port->vlandev_vid = 0;
986 if (ofproto->sflow) {
987 dpif_sflow_add_port(ofproto->sflow, port_);
994 port_destruct(struct ofport *port_)
996 struct ofport_dpif *port = ofport_dpif_cast(port_);
997 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
999 ofproto->need_revalidate = true;
1000 bundle_remove(port_);
1001 set_cfm(port_, NULL);
1002 if (ofproto->sflow) {
1003 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1006 ofport_clear_priorities(port);
1007 hmap_destroy(&port->priorities);
1011 port_modified(struct ofport *port_)
1013 struct ofport_dpif *port = ofport_dpif_cast(port_);
1015 if (port->bundle && port->bundle->bond) {
1016 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1021 port_reconfigured(struct ofport *port_, ovs_be32 old_config)
1023 struct ofport_dpif *port = ofport_dpif_cast(port_);
1024 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1025 ovs_be32 changed = old_config ^ port->up.opp.config;
1027 if (changed & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
1028 OFPPC_NO_FWD | OFPPC_NO_FLOOD)) {
1029 ofproto->need_revalidate = true;
1031 if (changed & htonl(OFPPC_NO_FLOOD) && port->bundle) {
1032 bundle_update(port->bundle);
1038 set_sflow(struct ofproto *ofproto_,
1039 const struct ofproto_sflow_options *sflow_options)
1041 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1042 struct dpif_sflow *ds = ofproto->sflow;
1044 if (sflow_options) {
1046 struct ofport_dpif *ofport;
1048 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1049 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1050 dpif_sflow_add_port(ds, &ofport->up);
1052 ofproto->need_revalidate = true;
1054 dpif_sflow_set_options(ds, sflow_options);
1057 dpif_sflow_destroy(ds);
1058 ofproto->need_revalidate = true;
1059 ofproto->sflow = NULL;
1066 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1068 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1075 struct ofproto_dpif *ofproto;
1077 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1078 ofproto->need_revalidate = true;
1079 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1082 if (cfm_configure(ofport->cfm, s)) {
1088 cfm_destroy(ofport->cfm);
1094 get_cfm_fault(const struct ofport *ofport_)
1096 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1098 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1102 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1105 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1108 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1115 /* Spanning Tree. */
1118 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1120 struct ofproto_dpif *ofproto = ofproto_;
1121 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1122 struct ofport_dpif *ofport;
1124 ofport = stp_port_get_aux(sp);
1126 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1127 ofproto->up.name, port_num);
1129 struct eth_header *eth = pkt->l2;
1131 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1132 if (eth_addr_is_zero(eth->eth_src)) {
1133 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1134 "with unknown MAC", ofproto->up.name, port_num);
1136 send_packet(ofport, pkt);
1142 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1144 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1146 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1148 /* Only revalidate flows if the configuration changed. */
1149 if (!s != !ofproto->stp) {
1150 ofproto->need_revalidate = true;
1154 if (!ofproto->stp) {
1155 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1156 send_bpdu_cb, ofproto);
1157 ofproto->stp_last_tick = time_msec();
1160 stp_set_bridge_id(ofproto->stp, s->system_id);
1161 stp_set_bridge_priority(ofproto->stp, s->priority);
1162 stp_set_hello_time(ofproto->stp, s->hello_time);
1163 stp_set_max_age(ofproto->stp, s->max_age);
1164 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1166 struct ofport *ofport;
1168 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1169 set_stp_port(ofport, NULL);
1172 stp_destroy(ofproto->stp);
1173 ofproto->stp = NULL;
1180 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1182 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1186 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1187 s->designated_root = stp_get_designated_root(ofproto->stp);
1188 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1197 update_stp_port_state(struct ofport_dpif *ofport)
1199 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1200 enum stp_state state;
1202 /* Figure out new state. */
1203 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1207 if (ofport->stp_state != state) {
1211 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1212 netdev_get_name(ofport->up.netdev),
1213 stp_state_name(ofport->stp_state),
1214 stp_state_name(state));
1215 if (stp_learn_in_state(ofport->stp_state)
1216 != stp_learn_in_state(state)) {
1217 /* xxx Learning action flows should also be flushed. */
1218 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1220 fwd_change = stp_forward_in_state(ofport->stp_state)
1221 != stp_forward_in_state(state);
1223 ofproto->need_revalidate = true;
1224 ofport->stp_state = state;
1225 ofport->stp_state_entered = time_msec();
1227 if (fwd_change && ofport->bundle) {
1228 bundle_update(ofport->bundle);
1231 /* Update the STP state bits in the OpenFlow port description. */
1232 of_state = (ofport->up.opp.state & htonl(~OFPPS_STP_MASK))
1233 | htonl(state == STP_LISTENING ? OFPPS_STP_LISTEN
1234 : state == STP_LEARNING ? OFPPS_STP_LEARN
1235 : state == STP_FORWARDING ? OFPPS_STP_FORWARD
1236 : state == STP_BLOCKING ? OFPPS_STP_BLOCK
1238 ofproto_port_set_state(&ofport->up, of_state);
1242 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1243 * caller is responsible for assigning STP port numbers and ensuring
1244 * there are no duplicates. */
1246 set_stp_port(struct ofport *ofport_,
1247 const struct ofproto_port_stp_settings *s)
1249 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1250 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1251 struct stp_port *sp = ofport->stp_port;
1253 if (!s || !s->enable) {
1255 ofport->stp_port = NULL;
1256 stp_port_disable(sp);
1257 update_stp_port_state(ofport);
1260 } else if (sp && stp_port_no(sp) != s->port_num
1261 && ofport == stp_port_get_aux(sp)) {
1262 /* The port-id changed, so disable the old one if it's not
1263 * already in use by another port. */
1264 stp_port_disable(sp);
1267 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1268 stp_port_enable(sp);
1270 stp_port_set_aux(sp, ofport);
1271 stp_port_set_priority(sp, s->priority);
1272 stp_port_set_path_cost(sp, s->path_cost);
1274 update_stp_port_state(ofport);
1280 get_stp_port_status(struct ofport *ofport_,
1281 struct ofproto_port_stp_status *s)
1283 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1284 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1285 struct stp_port *sp = ofport->stp_port;
1287 if (!ofproto->stp || !sp) {
1293 s->port_id = stp_port_get_id(sp);
1294 s->state = stp_port_get_state(sp);
1295 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1296 s->role = stp_port_get_role(sp);
1297 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1303 stp_run(struct ofproto_dpif *ofproto)
1306 long long int now = time_msec();
1307 long long int elapsed = now - ofproto->stp_last_tick;
1308 struct stp_port *sp;
1311 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1312 ofproto->stp_last_tick = now;
1314 while (stp_get_changed_port(ofproto->stp, &sp)) {
1315 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1318 update_stp_port_state(ofport);
1322 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1323 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1329 stp_wait(struct ofproto_dpif *ofproto)
1332 poll_timer_wait(1000);
1336 /* Returns true if STP should process 'flow'. */
1338 stp_should_process_flow(const struct flow *flow)
1340 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1344 stp_process_packet(const struct ofport_dpif *ofport,
1345 const struct ofpbuf *packet)
1347 struct ofpbuf payload = *packet;
1348 struct eth_header *eth = payload.data;
1349 struct stp_port *sp = ofport->stp_port;
1351 /* Sink packets on ports that have STP disabled when the bridge has
1353 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1357 /* Trim off padding on payload. */
1358 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1359 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1362 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1363 stp_received_bpdu(sp, payload.data, payload.size);
1367 static struct priority_to_dscp *
1368 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1370 struct priority_to_dscp *pdscp;
1373 hash = hash_int(priority, 0);
1374 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1375 if (pdscp->priority == priority) {
1383 ofport_clear_priorities(struct ofport_dpif *ofport)
1385 struct priority_to_dscp *pdscp, *next;
1387 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1388 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1394 set_queues(struct ofport *ofport_,
1395 const struct ofproto_port_queue *qdscp_list,
1398 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1399 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1400 struct hmap new = HMAP_INITIALIZER(&new);
1403 for (i = 0; i < n_qdscp; i++) {
1404 struct priority_to_dscp *pdscp;
1408 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1409 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1414 pdscp = get_priority(ofport, priority);
1416 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1418 pdscp = xmalloc(sizeof *pdscp);
1419 pdscp->priority = priority;
1421 ofproto->need_revalidate = true;
1424 if (pdscp->dscp != dscp) {
1426 ofproto->need_revalidate = true;
1429 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1432 if (!hmap_is_empty(&ofport->priorities)) {
1433 ofport_clear_priorities(ofport);
1434 ofproto->need_revalidate = true;
1437 hmap_swap(&new, &ofport->priorities);
1445 /* Expires all MAC learning entries associated with 'bundle' and forces its
1446 * ofproto to revalidate every flow.
1448 * Normally MAC learning entries are removed only from the ofproto associated
1449 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1450 * are removed from every ofproto. When patch ports and SLB bonds are in use
1451 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1452 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1453 * with the host from which it migrated. */
1455 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1457 struct ofproto_dpif *ofproto = bundle->ofproto;
1458 struct mac_learning *ml = ofproto->ml;
1459 struct mac_entry *mac, *next_mac;
1461 ofproto->need_revalidate = true;
1462 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1463 if (mac->port.p == bundle) {
1465 struct ofproto_dpif *o;
1467 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1469 struct mac_entry *e;
1471 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1474 tag_set_add(&o->revalidate_set, e->tag);
1475 mac_learning_expire(o->ml, e);
1481 mac_learning_expire(ml, mac);
1486 static struct ofbundle *
1487 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1489 struct ofbundle *bundle;
1491 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1492 &ofproto->bundles) {
1493 if (bundle->aux == aux) {
1500 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1501 * ones that are found to 'bundles'. */
1503 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1504 void **auxes, size_t n_auxes,
1505 struct hmapx *bundles)
1509 hmapx_init(bundles);
1510 for (i = 0; i < n_auxes; i++) {
1511 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1513 hmapx_add(bundles, bundle);
1519 bundle_update(struct ofbundle *bundle)
1521 struct ofport_dpif *port;
1523 bundle->floodable = true;
1524 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1525 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
1526 bundle->floodable = false;
1533 bundle_del_port(struct ofport_dpif *port)
1535 struct ofbundle *bundle = port->bundle;
1537 bundle->ofproto->need_revalidate = true;
1539 list_remove(&port->bundle_node);
1540 port->bundle = NULL;
1543 lacp_slave_unregister(bundle->lacp, port);
1546 bond_slave_unregister(bundle->bond, port);
1549 bundle_update(bundle);
1553 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1554 struct lacp_slave_settings *lacp,
1555 uint32_t bond_stable_id)
1557 struct ofport_dpif *port;
1559 port = get_ofp_port(bundle->ofproto, ofp_port);
1564 if (port->bundle != bundle) {
1565 bundle->ofproto->need_revalidate = true;
1567 bundle_del_port(port);
1570 port->bundle = bundle;
1571 list_push_back(&bundle->ports, &port->bundle_node);
1572 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
1573 bundle->floodable = false;
1577 port->bundle->ofproto->need_revalidate = true;
1578 lacp_slave_register(bundle->lacp, port, lacp);
1581 port->bond_stable_id = bond_stable_id;
1587 bundle_destroy(struct ofbundle *bundle)
1589 struct ofproto_dpif *ofproto;
1590 struct ofport_dpif *port, *next_port;
1597 ofproto = bundle->ofproto;
1598 for (i = 0; i < MAX_MIRRORS; i++) {
1599 struct ofmirror *m = ofproto->mirrors[i];
1601 if (m->out == bundle) {
1603 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1604 || hmapx_find_and_delete(&m->dsts, bundle)) {
1605 ofproto->need_revalidate = true;
1610 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1611 bundle_del_port(port);
1614 bundle_flush_macs(bundle, true);
1615 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1617 free(bundle->trunks);
1618 lacp_destroy(bundle->lacp);
1619 bond_destroy(bundle->bond);
1624 bundle_set(struct ofproto *ofproto_, void *aux,
1625 const struct ofproto_bundle_settings *s)
1627 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1628 bool need_flush = false;
1629 struct ofport_dpif *port;
1630 struct ofbundle *bundle;
1631 unsigned long *trunks;
1637 bundle_destroy(bundle_lookup(ofproto, aux));
1641 assert(s->n_slaves == 1 || s->bond != NULL);
1642 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1644 bundle = bundle_lookup(ofproto, aux);
1646 bundle = xmalloc(sizeof *bundle);
1648 bundle->ofproto = ofproto;
1649 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1650 hash_pointer(aux, 0));
1652 bundle->name = NULL;
1654 list_init(&bundle->ports);
1655 bundle->vlan_mode = PORT_VLAN_TRUNK;
1657 bundle->trunks = NULL;
1658 bundle->use_priority_tags = s->use_priority_tags;
1659 bundle->lacp = NULL;
1660 bundle->bond = NULL;
1662 bundle->floodable = true;
1664 bundle->src_mirrors = 0;
1665 bundle->dst_mirrors = 0;
1666 bundle->mirror_out = 0;
1669 if (!bundle->name || strcmp(s->name, bundle->name)) {
1671 bundle->name = xstrdup(s->name);
1676 if (!bundle->lacp) {
1677 ofproto->need_revalidate = true;
1678 bundle->lacp = lacp_create();
1680 lacp_configure(bundle->lacp, s->lacp);
1682 lacp_destroy(bundle->lacp);
1683 bundle->lacp = NULL;
1686 /* Update set of ports. */
1688 for (i = 0; i < s->n_slaves; i++) {
1689 if (!bundle_add_port(bundle, s->slaves[i],
1690 s->lacp ? &s->lacp_slaves[i] : NULL,
1691 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1695 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1696 struct ofport_dpif *next_port;
1698 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1699 for (i = 0; i < s->n_slaves; i++) {
1700 if (s->slaves[i] == port->up.ofp_port) {
1705 bundle_del_port(port);
1709 assert(list_size(&bundle->ports) <= s->n_slaves);
1711 if (list_is_empty(&bundle->ports)) {
1712 bundle_destroy(bundle);
1716 /* Set VLAN tagging mode */
1717 if (s->vlan_mode != bundle->vlan_mode
1718 || s->use_priority_tags != bundle->use_priority_tags) {
1719 bundle->vlan_mode = s->vlan_mode;
1720 bundle->use_priority_tags = s->use_priority_tags;
1725 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1726 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1728 if (vlan != bundle->vlan) {
1729 bundle->vlan = vlan;
1733 /* Get trunked VLANs. */
1734 switch (s->vlan_mode) {
1735 case PORT_VLAN_ACCESS:
1739 case PORT_VLAN_TRUNK:
1740 trunks = (unsigned long *) s->trunks;
1743 case PORT_VLAN_NATIVE_UNTAGGED:
1744 case PORT_VLAN_NATIVE_TAGGED:
1745 if (vlan != 0 && (!s->trunks
1746 || !bitmap_is_set(s->trunks, vlan)
1747 || bitmap_is_set(s->trunks, 0))) {
1748 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1750 trunks = bitmap_clone(s->trunks, 4096);
1752 trunks = bitmap_allocate1(4096);
1754 bitmap_set1(trunks, vlan);
1755 bitmap_set0(trunks, 0);
1757 trunks = (unsigned long *) s->trunks;
1764 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1765 free(bundle->trunks);
1766 if (trunks == s->trunks) {
1767 bundle->trunks = vlan_bitmap_clone(trunks);
1769 bundle->trunks = trunks;
1774 if (trunks != s->trunks) {
1779 if (!list_is_short(&bundle->ports)) {
1780 bundle->ofproto->has_bonded_bundles = true;
1782 if (bond_reconfigure(bundle->bond, s->bond)) {
1783 ofproto->need_revalidate = true;
1786 bundle->bond = bond_create(s->bond);
1787 ofproto->need_revalidate = true;
1790 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1791 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1795 bond_destroy(bundle->bond);
1796 bundle->bond = NULL;
1799 /* If we changed something that would affect MAC learning, un-learn
1800 * everything on this port and force flow revalidation. */
1802 bundle_flush_macs(bundle, false);
1809 bundle_remove(struct ofport *port_)
1811 struct ofport_dpif *port = ofport_dpif_cast(port_);
1812 struct ofbundle *bundle = port->bundle;
1815 bundle_del_port(port);
1816 if (list_is_empty(&bundle->ports)) {
1817 bundle_destroy(bundle);
1818 } else if (list_is_short(&bundle->ports)) {
1819 bond_destroy(bundle->bond);
1820 bundle->bond = NULL;
1826 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
1828 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1829 struct ofport_dpif *port = port_;
1830 uint8_t ea[ETH_ADDR_LEN];
1833 error = netdev_get_etheraddr(port->up.netdev, ea);
1835 struct ofpbuf packet;
1838 ofpbuf_init(&packet, 0);
1839 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1841 memcpy(packet_pdu, pdu, pdu_size);
1843 send_packet(port, &packet);
1844 ofpbuf_uninit(&packet);
1846 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1847 "%s (%s)", port->bundle->name,
1848 netdev_get_name(port->up.netdev), strerror(error));
1853 bundle_send_learning_packets(struct ofbundle *bundle)
1855 struct ofproto_dpif *ofproto = bundle->ofproto;
1856 int error, n_packets, n_errors;
1857 struct mac_entry *e;
1859 error = n_packets = n_errors = 0;
1860 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1861 if (e->port.p != bundle) {
1862 struct ofpbuf *learning_packet;
1863 struct ofport_dpif *port;
1867 /* The assignment to "port" is unnecessary but makes "grep"ing for
1868 * struct ofport_dpif more effective. */
1869 learning_packet = bond_compose_learning_packet(bundle->bond,
1873 ret = send_packet(port, learning_packet);
1874 ofpbuf_delete(learning_packet);
1884 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1885 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1886 "packets, last error was: %s",
1887 bundle->name, n_errors, n_packets, strerror(error));
1889 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1890 bundle->name, n_packets);
1895 bundle_run(struct ofbundle *bundle)
1898 lacp_run(bundle->lacp, send_pdu_cb);
1901 struct ofport_dpif *port;
1903 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1904 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1907 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1908 lacp_status(bundle->lacp));
1909 if (bond_should_send_learning_packets(bundle->bond)) {
1910 bundle_send_learning_packets(bundle);
1916 bundle_wait(struct ofbundle *bundle)
1919 lacp_wait(bundle->lacp);
1922 bond_wait(bundle->bond);
1929 mirror_scan(struct ofproto_dpif *ofproto)
1933 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1934 if (!ofproto->mirrors[idx]) {
1941 static struct ofmirror *
1942 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1946 for (i = 0; i < MAX_MIRRORS; i++) {
1947 struct ofmirror *mirror = ofproto->mirrors[i];
1948 if (mirror && mirror->aux == aux) {
1956 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
1958 mirror_update_dups(struct ofproto_dpif *ofproto)
1962 for (i = 0; i < MAX_MIRRORS; i++) {
1963 struct ofmirror *m = ofproto->mirrors[i];
1966 m->dup_mirrors = MIRROR_MASK_C(1) << i;
1970 for (i = 0; i < MAX_MIRRORS; i++) {
1971 struct ofmirror *m1 = ofproto->mirrors[i];
1978 for (j = i + 1; j < MAX_MIRRORS; j++) {
1979 struct ofmirror *m2 = ofproto->mirrors[j];
1981 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
1982 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
1983 m2->dup_mirrors |= m1->dup_mirrors;
1990 mirror_set(struct ofproto *ofproto_, void *aux,
1991 const struct ofproto_mirror_settings *s)
1993 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1994 mirror_mask_t mirror_bit;
1995 struct ofbundle *bundle;
1996 struct ofmirror *mirror;
1997 struct ofbundle *out;
1998 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
1999 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2002 mirror = mirror_lookup(ofproto, aux);
2004 mirror_destroy(mirror);
2010 idx = mirror_scan(ofproto);
2012 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2014 ofproto->up.name, MAX_MIRRORS, s->name);
2018 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2019 mirror->ofproto = ofproto;
2022 mirror->out_vlan = -1;
2023 mirror->name = NULL;
2026 if (!mirror->name || strcmp(s->name, mirror->name)) {
2028 mirror->name = xstrdup(s->name);
2031 /* Get the new configuration. */
2032 if (s->out_bundle) {
2033 out = bundle_lookup(ofproto, s->out_bundle);
2035 mirror_destroy(mirror);
2041 out_vlan = s->out_vlan;
2043 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2044 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2046 /* If the configuration has not changed, do nothing. */
2047 if (hmapx_equals(&srcs, &mirror->srcs)
2048 && hmapx_equals(&dsts, &mirror->dsts)
2049 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2050 && mirror->out == out
2051 && mirror->out_vlan == out_vlan)
2053 hmapx_destroy(&srcs);
2054 hmapx_destroy(&dsts);
2058 hmapx_swap(&srcs, &mirror->srcs);
2059 hmapx_destroy(&srcs);
2061 hmapx_swap(&dsts, &mirror->dsts);
2062 hmapx_destroy(&dsts);
2064 free(mirror->vlans);
2065 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2068 mirror->out_vlan = out_vlan;
2070 /* Update bundles. */
2071 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2072 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2073 if (hmapx_contains(&mirror->srcs, bundle)) {
2074 bundle->src_mirrors |= mirror_bit;
2076 bundle->src_mirrors &= ~mirror_bit;
2079 if (hmapx_contains(&mirror->dsts, bundle)) {
2080 bundle->dst_mirrors |= mirror_bit;
2082 bundle->dst_mirrors &= ~mirror_bit;
2085 if (mirror->out == bundle) {
2086 bundle->mirror_out |= mirror_bit;
2088 bundle->mirror_out &= ~mirror_bit;
2092 ofproto->need_revalidate = true;
2093 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2094 mirror_update_dups(ofproto);
2100 mirror_destroy(struct ofmirror *mirror)
2102 struct ofproto_dpif *ofproto;
2103 mirror_mask_t mirror_bit;
2104 struct ofbundle *bundle;
2110 ofproto = mirror->ofproto;
2111 ofproto->need_revalidate = true;
2112 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2114 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2115 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2116 bundle->src_mirrors &= ~mirror_bit;
2117 bundle->dst_mirrors &= ~mirror_bit;
2118 bundle->mirror_out &= ~mirror_bit;
2121 hmapx_destroy(&mirror->srcs);
2122 hmapx_destroy(&mirror->dsts);
2123 free(mirror->vlans);
2125 ofproto->mirrors[mirror->idx] = NULL;
2129 mirror_update_dups(ofproto);
2133 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2134 uint64_t *packets, uint64_t *bytes)
2136 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2137 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2140 *packets = *bytes = UINT64_MAX;
2144 *packets = mirror->packet_count;
2145 *bytes = mirror->byte_count;
2151 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2153 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2154 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2155 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2161 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2163 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2164 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2165 return bundle && bundle->mirror_out != 0;
2169 forward_bpdu_changed(struct ofproto *ofproto_)
2171 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2172 /* Revalidate cached flows whenever forward_bpdu option changes. */
2173 ofproto->need_revalidate = true;
2177 set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time)
2179 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2180 mac_learning_set_idle_time(ofproto->ml, idle_time);
2185 static struct ofport_dpif *
2186 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
2188 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2189 return ofport ? ofport_dpif_cast(ofport) : NULL;
2192 static struct ofport_dpif *
2193 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
2195 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2199 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2200 struct dpif_port *dpif_port)
2202 ofproto_port->name = dpif_port->name;
2203 ofproto_port->type = dpif_port->type;
2204 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2208 port_run(struct ofport_dpif *ofport)
2210 bool enable = netdev_get_carrier(ofport->up.netdev);
2213 cfm_run(ofport->cfm);
2215 if (cfm_should_send_ccm(ofport->cfm)) {
2216 struct ofpbuf packet;
2218 ofpbuf_init(&packet, 0);
2219 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr);
2220 send_packet(ofport, &packet);
2221 ofpbuf_uninit(&packet);
2224 enable = enable && !cfm_get_fault(ofport->cfm)
2225 && cfm_get_opup(ofport->cfm);
2228 if (ofport->bundle) {
2229 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2232 if (ofport->may_enable != enable) {
2233 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2235 if (ofproto->has_bundle_action) {
2236 ofproto->need_revalidate = true;
2240 ofport->may_enable = enable;
2244 port_wait(struct ofport_dpif *ofport)
2247 cfm_wait(ofport->cfm);
2252 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2253 struct ofproto_port *ofproto_port)
2255 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2256 struct dpif_port dpif_port;
2259 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2261 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2267 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2269 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2273 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2275 *ofp_portp = odp_port_to_ofp_port(odp_port);
2281 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2283 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2286 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2288 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2290 /* The caller is going to close ofport->up.netdev. If this is a
2291 * bonded port, then the bond is using that netdev, so remove it
2292 * from the bond. The client will need to reconfigure everything
2293 * after deleting ports, so then the slave will get re-added. */
2294 bundle_remove(&ofport->up);
2301 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2303 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2306 error = netdev_get_stats(ofport->up.netdev, stats);
2308 if (!error && ofport->odp_port == OVSP_LOCAL) {
2309 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2311 /* ofproto->stats.tx_packets represents packets that we created
2312 * internally and sent to some port (e.g. packets sent with
2313 * send_packet()). Account for them as if they had come from
2314 * OFPP_LOCAL and got forwarded. */
2316 if (stats->rx_packets != UINT64_MAX) {
2317 stats->rx_packets += ofproto->stats.tx_packets;
2320 if (stats->rx_bytes != UINT64_MAX) {
2321 stats->rx_bytes += ofproto->stats.tx_bytes;
2324 /* ofproto->stats.rx_packets represents packets that were received on
2325 * some port and we processed internally and dropped (e.g. STP).
2326 * Account fro them as if they had been forwarded to OFPP_LOCAL. */
2328 if (stats->tx_packets != UINT64_MAX) {
2329 stats->tx_packets += ofproto->stats.rx_packets;
2332 if (stats->tx_bytes != UINT64_MAX) {
2333 stats->tx_bytes += ofproto->stats.rx_bytes;
2340 /* Account packets for LOCAL port. */
2342 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2343 size_t tx_size, size_t rx_size)
2345 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2348 ofproto->stats.rx_packets++;
2349 ofproto->stats.rx_bytes += rx_size;
2352 ofproto->stats.tx_packets++;
2353 ofproto->stats.tx_bytes += tx_size;
2357 struct port_dump_state {
2358 struct dpif_port_dump dump;
2363 port_dump_start(const struct ofproto *ofproto_, void **statep)
2365 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2366 struct port_dump_state *state;
2368 *statep = state = xmalloc(sizeof *state);
2369 dpif_port_dump_start(&state->dump, ofproto->dpif);
2370 state->done = false;
2375 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2376 struct ofproto_port *port)
2378 struct port_dump_state *state = state_;
2379 struct dpif_port dpif_port;
2381 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2382 ofproto_port_from_dpif_port(port, &dpif_port);
2385 int error = dpif_port_dump_done(&state->dump);
2387 return error ? error : EOF;
2392 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2394 struct port_dump_state *state = state_;
2397 dpif_port_dump_done(&state->dump);
2404 port_poll(const struct ofproto *ofproto_, char **devnamep)
2406 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2407 return dpif_port_poll(ofproto->dpif, devnamep);
2411 port_poll_wait(const struct ofproto *ofproto_)
2413 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2414 dpif_port_poll_wait(ofproto->dpif);
2418 port_is_lacp_current(const struct ofport *ofport_)
2420 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2421 return (ofport->bundle && ofport->bundle->lacp
2422 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2426 /* Upcall handling. */
2428 /* Flow miss batching.
2430 * Some dpifs implement operations faster when you hand them off in a batch.
2431 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2432 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2433 * more packets, plus possibly installing the flow in the dpif.
2435 * So far we only batch the operations that affect flow setup time the most.
2436 * It's possible to batch more than that, but the benefit might be minimal. */
2438 struct hmap_node hmap_node;
2440 enum odp_key_fitness key_fitness;
2441 const struct nlattr *key;
2443 ovs_be16 initial_tci;
2444 struct list packets;
2447 struct flow_miss_op {
2448 struct dpif_op dpif_op;
2449 struct subfacet *subfacet;
2452 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2453 * OpenFlow controller as necessary according to their individual
2454 * configurations. */
2456 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
2457 const struct flow *flow)
2459 struct ofputil_packet_in pin;
2461 pin.packet = packet->data;
2462 pin.packet_len = packet->size;
2463 pin.total_len = packet->size;
2464 pin.reason = OFPR_NO_MATCH;
2469 pin.buffer_id = 0; /* not yet known */
2470 pin.send_len = 0; /* not used for flow table misses */
2472 flow_get_metadata(flow, &pin.fmd);
2474 /* Registers aren't meaningful on a miss. */
2475 memset(pin.fmd.reg_masks, 0, sizeof pin.fmd.reg_masks);
2477 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow);
2481 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2482 const struct ofpbuf *packet)
2484 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2490 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2492 cfm_process_heartbeat(ofport->cfm, packet);
2495 } else if (ofport->bundle && ofport->bundle->lacp
2496 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2498 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2501 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2503 stp_process_packet(ofport, packet);
2510 static struct flow_miss *
2511 flow_miss_create(struct hmap *todo, const struct flow *flow,
2512 enum odp_key_fitness key_fitness,
2513 const struct nlattr *key, size_t key_len,
2514 ovs_be16 initial_tci)
2516 uint32_t hash = flow_hash(flow, 0);
2517 struct flow_miss *miss;
2519 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2520 if (flow_equal(&miss->flow, flow)) {
2525 miss = xmalloc(sizeof *miss);
2526 hmap_insert(todo, &miss->hmap_node, hash);
2528 miss->key_fitness = key_fitness;
2530 miss->key_len = key_len;
2531 miss->initial_tci = initial_tci;
2532 list_init(&miss->packets);
2537 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2538 struct flow_miss_op *ops, size_t *n_ops)
2540 const struct flow *flow = &miss->flow;
2541 struct ofpbuf *packet, *next_packet;
2542 struct subfacet *subfacet;
2543 struct facet *facet;
2545 facet = facet_lookup_valid(ofproto, flow);
2547 struct rule_dpif *rule;
2549 rule = rule_dpif_lookup(ofproto, flow, 0);
2551 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
2552 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
2554 if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) {
2555 COVERAGE_INC(ofproto_dpif_no_packet_in);
2556 /* XXX install 'drop' flow entry */
2560 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
2564 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2565 send_packet_in_miss(ofproto, packet, flow);
2571 facet = facet_create(rule, flow);
2574 subfacet = subfacet_create(facet,
2575 miss->key_fitness, miss->key, miss->key_len,
2578 LIST_FOR_EACH_SAFE (packet, next_packet, list_node, &miss->packets) {
2579 struct dpif_flow_stats stats;
2580 struct flow_miss_op *op;
2581 struct dpif_execute *execute;
2583 ofproto->n_matches++;
2585 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2587 * Extra-special case for fail-open mode.
2589 * We are in fail-open mode and the packet matched the fail-open
2590 * rule, but we are connected to a controller too. We should send
2591 * the packet up to the controller in the hope that it will try to
2592 * set up a flow and thereby allow us to exit fail-open.
2594 * See the top-level comment in fail-open.c for more information.
2596 send_packet_in_miss(ofproto, packet, flow);
2599 if (!facet->may_install || !subfacet->actions) {
2600 subfacet_make_actions(subfacet, packet);
2603 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2604 subfacet_update_stats(subfacet, &stats);
2606 if (!subfacet->actions_len) {
2607 /* No actions to execute, so skip talking to the dpif. */
2611 if (flow->vlan_tci != subfacet->initial_tci) {
2612 /* This packet was received on a VLAN splinter port. We added
2613 * a VLAN to the packet to make the packet resemble the flow,
2614 * but the actions were composed assuming that the packet
2615 * contained no VLAN. So, we must remove the VLAN header from
2616 * the packet before trying to execute the actions. */
2617 eth_pop_vlan(packet);
2620 op = &ops[(*n_ops)++];
2621 execute = &op->dpif_op.u.execute;
2622 op->subfacet = subfacet;
2623 op->dpif_op.type = DPIF_OP_EXECUTE;
2624 execute->key = miss->key;
2625 execute->key_len = miss->key_len;
2626 execute->actions = (facet->may_install
2628 : xmemdup(subfacet->actions,
2629 subfacet->actions_len));
2630 execute->actions_len = subfacet->actions_len;
2631 execute->packet = packet;
2634 if (facet->may_install && subfacet->key_fitness != ODP_FIT_TOO_LITTLE) {
2635 struct flow_miss_op *op = &ops[(*n_ops)++];
2636 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
2638 op->subfacet = subfacet;
2639 op->dpif_op.type = DPIF_OP_FLOW_PUT;
2640 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2641 put->key = miss->key;
2642 put->key_len = miss->key_len;
2643 put->actions = subfacet->actions;
2644 put->actions_len = subfacet->actions_len;
2649 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
2650 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
2651 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
2652 * what a flow key should contain.
2654 * This function also includes some logic to help make VLAN splinters
2655 * transparent to the rest of the upcall processing logic. In particular, if
2656 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
2657 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
2658 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
2660 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
2661 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
2662 * (This differs from the value returned in flow->vlan_tci only for packets
2663 * received on VLAN splinters.)
2665 static enum odp_key_fitness
2666 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
2667 const struct nlattr *key, size_t key_len,
2668 struct flow *flow, ovs_be16 *initial_tci,
2669 struct ofpbuf *packet)
2671 enum odp_key_fitness fitness;
2675 fitness = odp_flow_key_to_flow(key, key_len, flow);
2676 if (fitness == ODP_FIT_ERROR) {
2679 *initial_tci = flow->vlan_tci;
2681 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
2683 /* Cause the flow to be processed as if it came in on the real device
2684 * with the VLAN device's VLAN ID. */
2685 flow->in_port = realdev;
2686 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
2688 /* Make the packet resemble the flow, so that it gets sent to an
2689 * OpenFlow controller properly, so that it looks correct for
2690 * sFlow, and so that flow_extract() will get the correct vlan_tci
2691 * if it is called on 'packet'.
2693 * The allocated space inside 'packet' probably also contains
2694 * 'key', that is, both 'packet' and 'key' are probably part of a
2695 * struct dpif_upcall (see the large comment on that structure
2696 * definition), so pushing data on 'packet' is in general not a
2697 * good idea since it could overwrite 'key' or free it as a side
2698 * effect. However, it's OK in this special case because we know
2699 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
2700 * will just overwrite the 4-byte "struct nlattr", which is fine
2701 * since we don't need that header anymore. */
2702 eth_push_vlan(packet, flow->vlan_tci);
2705 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
2706 if (fitness == ODP_FIT_PERFECT) {
2707 fitness = ODP_FIT_TOO_MUCH;
2715 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
2718 struct dpif_upcall *upcall;
2719 struct flow_miss *miss, *next_miss;
2720 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
2721 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
2730 /* Construct the to-do list.
2732 * This just amounts to extracting the flow from each packet and sticking
2733 * the packets that have the same flow in the same "flow_miss" structure so
2734 * that we can process them together. */
2736 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
2737 enum odp_key_fitness fitness;
2738 struct flow_miss *miss;
2739 ovs_be16 initial_tci;
2742 /* Obtain metadata and check userspace/kernel agreement on flow match,
2743 * then set 'flow''s header pointers. */
2744 fitness = ofproto_dpif_extract_flow_key(ofproto,
2745 upcall->key, upcall->key_len,
2746 &flow, &initial_tci,
2748 if (fitness == ODP_FIT_ERROR) {
2749 ofpbuf_delete(upcall->packet);
2752 flow_extract(upcall->packet, flow.skb_priority, flow.tun_id,
2753 flow.in_port, &flow);
2755 /* Handle 802.1ag, LACP, and STP specially. */
2756 if (process_special(ofproto, &flow, upcall->packet)) {
2757 ofproto_update_local_port_stats(&ofproto->up,
2758 0, upcall->packet->size);
2759 ofpbuf_delete(upcall->packet);
2760 ofproto->n_matches++;
2764 /* Add other packets to a to-do list. */
2765 miss = flow_miss_create(&todo, &flow, fitness,
2766 upcall->key, upcall->key_len, initial_tci);
2767 list_push_back(&miss->packets, &upcall->packet->list_node);
2770 /* Process each element in the to-do list, constructing the set of
2771 * operations to batch. */
2773 HMAP_FOR_EACH (miss, hmap_node, &todo) {
2774 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
2776 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
2778 /* Execute batch. */
2779 for (i = 0; i < n_ops; i++) {
2780 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
2782 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
2784 /* Free memory and update facets. */
2785 for (i = 0; i < n_ops; i++) {
2786 struct flow_miss_op *op = &flow_miss_ops[i];
2787 struct dpif_execute *execute;
2789 switch (op->dpif_op.type) {
2790 case DPIF_OP_EXECUTE:
2791 execute = &op->dpif_op.u.execute;
2792 if (op->subfacet->actions != execute->actions) {
2793 free((struct nlattr *) execute->actions);
2797 case DPIF_OP_FLOW_PUT:
2798 if (!op->dpif_op.error) {
2799 op->subfacet->installed = true;
2804 HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) {
2805 ofpbuf_list_delete(&miss->packets);
2806 hmap_remove(&todo, &miss->hmap_node);
2809 hmap_destroy(&todo);
2813 handle_userspace_upcall(struct ofproto_dpif *ofproto,
2814 struct dpif_upcall *upcall)
2816 struct user_action_cookie cookie;
2817 enum odp_key_fitness fitness;
2818 ovs_be16 initial_tci;
2821 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
2823 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
2824 upcall->key_len, &flow,
2825 &initial_tci, upcall->packet);
2826 if (fitness == ODP_FIT_ERROR) {
2827 ofpbuf_delete(upcall->packet);
2831 if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
2832 if (ofproto->sflow) {
2833 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
2837 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
2839 ofpbuf_delete(upcall->packet);
2843 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
2845 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
2849 assert (max_batch <= FLOW_MISS_MAX_BATCH);
2852 for (i = 0; i < max_batch; i++) {
2853 struct dpif_upcall *upcall = &misses[n_misses];
2856 error = dpif_recv(ofproto->dpif, upcall);
2861 switch (upcall->type) {
2862 case DPIF_UC_ACTION:
2863 handle_userspace_upcall(ofproto, upcall);
2867 /* Handle it later. */
2871 case DPIF_N_UC_TYPES:
2873 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32,
2879 handle_miss_upcalls(ofproto, misses, n_misses);
2884 /* Flow expiration. */
2886 static int subfacet_max_idle(const struct ofproto_dpif *);
2887 static void update_stats(struct ofproto_dpif *);
2888 static void rule_expire(struct rule_dpif *);
2889 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
2891 /* This function is called periodically by run(). Its job is to collect
2892 * updates for the flows that have been installed into the datapath, most
2893 * importantly when they last were used, and then use that information to
2894 * expire flows that have not been used recently.
2896 * Returns the number of milliseconds after which it should be called again. */
2898 expire(struct ofproto_dpif *ofproto)
2900 struct rule_dpif *rule, *next_rule;
2901 struct oftable *table;
2904 /* Update stats for each flow in the datapath. */
2905 update_stats(ofproto);
2907 /* Expire subfacets that have been idle too long. */
2908 dp_max_idle = subfacet_max_idle(ofproto);
2909 expire_subfacets(ofproto, dp_max_idle);
2911 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
2912 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
2913 struct cls_cursor cursor;
2915 cls_cursor_init(&cursor, &table->cls, NULL);
2916 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
2921 /* All outstanding data in existing flows has been accounted, so it's a
2922 * good time to do bond rebalancing. */
2923 if (ofproto->has_bonded_bundles) {
2924 struct ofbundle *bundle;
2926 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2928 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
2933 return MIN(dp_max_idle, 1000);
2936 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
2938 * This function also pushes statistics updates to rules which each facet
2939 * resubmits into. Generally these statistics will be accurate. However, if a
2940 * facet changes the rule it resubmits into at some time in between
2941 * update_stats() runs, it is possible that statistics accrued to the
2942 * old rule will be incorrectly attributed to the new rule. This could be
2943 * avoided by calling update_stats() whenever rules are created or
2944 * deleted. However, the performance impact of making so many calls to the
2945 * datapath do not justify the benefit of having perfectly accurate statistics.
2948 update_stats(struct ofproto_dpif *p)
2950 const struct dpif_flow_stats *stats;
2951 struct dpif_flow_dump dump;
2952 const struct nlattr *key;
2955 dpif_flow_dump_start(&dump, p->dpif);
2956 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
2957 struct subfacet *subfacet;
2959 subfacet = subfacet_find(p, key, key_len);
2960 if (subfacet && subfacet->installed) {
2961 struct facet *facet = subfacet->facet;
2963 if (stats->n_packets >= subfacet->dp_packet_count) {
2964 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
2965 facet->packet_count += extra;
2967 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
2970 if (stats->n_bytes >= subfacet->dp_byte_count) {
2971 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
2973 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
2976 subfacet->dp_packet_count = stats->n_packets;
2977 subfacet->dp_byte_count = stats->n_bytes;
2979 facet->tcp_flags |= stats->tcp_flags;
2981 subfacet_update_time(subfacet, stats->used);
2982 facet_account(facet);
2983 facet_push_stats(facet);
2985 if (!VLOG_DROP_WARN(&rl)) {
2989 odp_flow_key_format(key, key_len, &s);
2990 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
2994 COVERAGE_INC(facet_unexpected);
2995 /* There's a flow in the datapath that we know nothing about, or a
2996 * flow that shouldn't be installed but was anyway. Delete it. */
2997 dpif_flow_del(p->dpif, key, key_len, NULL);
3000 dpif_flow_dump_done(&dump);
3003 /* Calculates and returns the number of milliseconds of idle time after which
3004 * subfacets should expire from the datapath. When a subfacet expires, we fold
3005 * its statistics into its facet, and when a facet's last subfacet expires, we
3006 * fold its statistic into its rule. */
3008 subfacet_max_idle(const struct ofproto_dpif *ofproto)
3011 * Idle time histogram.
3013 * Most of the time a switch has a relatively small number of subfacets.
3014 * When this is the case we might as well keep statistics for all of them
3015 * in userspace and to cache them in the kernel datapath for performance as
3018 * As the number of subfacets increases, the memory required to maintain
3019 * statistics about them in userspace and in the kernel becomes
3020 * significant. However, with a large number of subfacets it is likely
3021 * that only a few of them are "heavy hitters" that consume a large amount
3022 * of bandwidth. At this point, only heavy hitters are worth caching in
3023 * the kernel and maintaining in userspaces; other subfacets we can
3026 * The technique used to compute the idle time is to build a histogram with
3027 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3028 * that is installed in the kernel gets dropped in the appropriate bucket.
3029 * After the histogram has been built, we compute the cutoff so that only
3030 * the most-recently-used 1% of subfacets (but at least
3031 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3032 * the most-recently-used bucket of subfacets is kept, so actually an
3033 * arbitrary number of subfacets can be kept in any given expiration run
3034 * (though the next run will delete most of those unless they receive
3037 * This requires a second pass through the subfacets, in addition to the
3038 * pass made by update_stats(), because the former function never looks at
3039 * uninstallable subfacets.
3041 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3042 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3043 int buckets[N_BUCKETS] = { 0 };
3044 int total, subtotal, bucket;
3045 struct subfacet *subfacet;
3049 total = hmap_count(&ofproto->subfacets);
3050 if (total <= ofproto->up.flow_eviction_threshold) {
3051 return N_BUCKETS * BUCKET_WIDTH;
3054 /* Build histogram. */
3056 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3057 long long int idle = now - subfacet->used;
3058 int bucket = (idle <= 0 ? 0
3059 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3060 : (unsigned int) idle / BUCKET_WIDTH);
3064 /* Find the first bucket whose flows should be expired. */
3065 subtotal = bucket = 0;
3067 subtotal += buckets[bucket++];
3068 } while (bucket < N_BUCKETS &&
3069 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3071 if (VLOG_IS_DBG_ENABLED()) {
3075 ds_put_cstr(&s, "keep");
3076 for (i = 0; i < N_BUCKETS; i++) {
3078 ds_put_cstr(&s, ", drop");
3081 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3084 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3088 return bucket * BUCKET_WIDTH;
3092 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3094 long long int cutoff = time_msec() - dp_max_idle;
3095 struct subfacet *subfacet, *next_subfacet;
3097 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3098 &ofproto->subfacets) {
3099 if (subfacet->used < cutoff) {
3100 subfacet_destroy(subfacet);
3105 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3106 * then delete it entirely. */
3108 rule_expire(struct rule_dpif *rule)
3110 struct facet *facet, *next_facet;
3114 /* Has 'rule' expired? */
3116 if (rule->up.hard_timeout
3117 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3118 reason = OFPRR_HARD_TIMEOUT;
3119 } else if (rule->up.idle_timeout && list_is_empty(&rule->facets)
3120 && now > rule->up.used + rule->up.idle_timeout * 1000) {
3121 reason = OFPRR_IDLE_TIMEOUT;
3126 COVERAGE_INC(ofproto_dpif_expired);
3128 /* Update stats. (This is a no-op if the rule expired due to an idle
3129 * timeout, because that only happens when the rule has no facets left.) */
3130 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3131 facet_remove(facet);
3134 /* Get rid of the rule. */
3135 ofproto_rule_expire(&rule->up, reason);
3140 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3142 * The caller must already have determined that no facet with an identical
3143 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3144 * the ofproto's classifier table.
3146 * The facet will initially have no subfacets. The caller should create (at
3147 * least) one subfacet with subfacet_create(). */
3148 static struct facet *
3149 facet_create(struct rule_dpif *rule, const struct flow *flow)
3151 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3152 struct facet *facet;
3154 facet = xzalloc(sizeof *facet);
3155 facet->used = time_msec();
3156 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
3157 list_push_back(&rule->facets, &facet->list_node);
3159 facet->flow = *flow;
3160 list_init(&facet->subfacets);
3161 netflow_flow_init(&facet->nf_flow);
3162 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3168 facet_free(struct facet *facet)
3173 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3174 * 'packet', which arrived on 'in_port'.
3176 * Takes ownership of 'packet'. */
3178 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3179 const struct nlattr *odp_actions, size_t actions_len,
3180 struct ofpbuf *packet)
3182 struct odputil_keybuf keybuf;
3186 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3187 odp_flow_key_from_flow(&key, flow);
3189 error = dpif_execute(ofproto->dpif, key.data, key.size,
3190 odp_actions, actions_len, packet);
3192 ofpbuf_delete(packet);
3196 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3198 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3199 * rule's statistics, via subfacet_uninstall().
3201 * - Removes 'facet' from its rule and from ofproto->facets.
3204 facet_remove(struct facet *facet)
3206 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3207 struct subfacet *subfacet, *next_subfacet;
3209 assert(!list_is_empty(&facet->subfacets));
3211 /* First uninstall all of the subfacets to get final statistics. */
3212 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3213 subfacet_uninstall(subfacet);
3216 /* Flush the final stats to the rule.
3218 * This might require us to have at least one subfacet around so that we
3219 * can use its actions for accounting in facet_account(), which is why we
3220 * have uninstalled but not yet destroyed the subfacets. */
3221 facet_flush_stats(facet);
3223 /* Now we're really all done so destroy everything. */
3224 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3225 &facet->subfacets) {
3226 subfacet_destroy__(subfacet);
3228 hmap_remove(&ofproto->facets, &facet->hmap_node);
3229 list_remove(&facet->list_node);
3234 facet_account(struct facet *facet)
3236 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3238 struct subfacet *subfacet;
3239 const struct nlattr *a;
3243 if (facet->byte_count <= facet->accounted_bytes) {
3246 n_bytes = facet->byte_count - facet->accounted_bytes;
3247 facet->accounted_bytes = facet->byte_count;
3249 /* Feed information from the active flows back into the learning table to
3250 * ensure that table is always in sync with what is actually flowing
3251 * through the datapath. */
3252 if (facet->has_learn || facet->has_normal
3253 || (facet->has_fin_timeout
3254 && facet->tcp_flags & (TCP_FIN | TCP_RST))) {
3255 struct action_xlate_ctx ctx;
3257 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3258 facet->flow.vlan_tci,
3259 facet->rule, facet->tcp_flags, NULL);
3260 ctx.may_learn = true;
3261 ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions,
3262 facet->rule->up.n_actions));
3265 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3269 /* This loop feeds byte counters to bond_account() for rebalancing to use
3270 * as a basis. We also need to track the actual VLAN on which the packet
3271 * is going to be sent to ensure that it matches the one passed to
3272 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3275 * We use the actions from an arbitrary subfacet because they should all
3276 * be equally valid for our purpose. */
3277 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3278 struct subfacet, list_node);
3279 vlan_tci = facet->flow.vlan_tci;
3280 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3281 subfacet->actions, subfacet->actions_len) {
3282 const struct ovs_action_push_vlan *vlan;
3283 struct ofport_dpif *port;
3285 switch (nl_attr_type(a)) {
3286 case OVS_ACTION_ATTR_OUTPUT:
3287 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3288 if (port && port->bundle && port->bundle->bond) {
3289 bond_account(port->bundle->bond, &facet->flow,
3290 vlan_tci_to_vid(vlan_tci), n_bytes);
3294 case OVS_ACTION_ATTR_POP_VLAN:
3295 vlan_tci = htons(0);
3298 case OVS_ACTION_ATTR_PUSH_VLAN:
3299 vlan = nl_attr_get(a);
3300 vlan_tci = vlan->vlan_tci;
3306 /* Returns true if the only action for 'facet' is to send to the controller.
3307 * (We don't report NetFlow expiration messages for such facets because they
3308 * are just part of the control logic for the network, not real traffic). */
3310 facet_is_controller_flow(struct facet *facet)
3313 && facet->rule->up.n_actions == 1
3314 && action_outputs_to_port(&facet->rule->up.actions[0],
3315 htons(OFPP_CONTROLLER)));
3318 /* Folds all of 'facet''s statistics into its rule. Also updates the
3319 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3320 * 'facet''s statistics in the datapath should have been zeroed and folded into
3321 * its packet and byte counts before this function is called. */
3323 facet_flush_stats(struct facet *facet)
3325 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3326 struct subfacet *subfacet;
3328 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3329 assert(!subfacet->dp_byte_count);
3330 assert(!subfacet->dp_packet_count);
3333 facet_push_stats(facet);
3334 facet_account(facet);
3336 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3337 struct ofexpired expired;
3338 expired.flow = facet->flow;
3339 expired.packet_count = facet->packet_count;
3340 expired.byte_count = facet->byte_count;
3341 expired.used = facet->used;
3342 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3345 facet->rule->packet_count += facet->packet_count;
3346 facet->rule->byte_count += facet->byte_count;
3348 /* Reset counters to prevent double counting if 'facet' ever gets
3350 facet_reset_counters(facet);
3352 netflow_flow_clear(&facet->nf_flow);
3353 facet->tcp_flags = 0;
3356 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3357 * Returns it if found, otherwise a null pointer.
3359 * The returned facet might need revalidation; use facet_lookup_valid()
3360 * instead if that is important. */
3361 static struct facet *
3362 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
3364 struct facet *facet;
3366 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
3368 if (flow_equal(flow, &facet->flow)) {
3376 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3377 * Returns it if found, otherwise a null pointer.
3379 * The returned facet is guaranteed to be valid. */
3380 static struct facet *
3381 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
3383 struct facet *facet = facet_find(ofproto, flow);
3385 /* The facet we found might not be valid, since we could be in need of
3386 * revalidation. If it is not valid, don't return it. */
3388 && (ofproto->need_revalidate
3389 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))
3390 && !facet_revalidate(facet)) {
3391 COVERAGE_INC(facet_invalidated);
3399 facet_check_consistency(struct facet *facet)
3401 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3403 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3405 struct rule_dpif *rule;
3406 struct subfacet *subfacet;
3407 bool may_log = false;
3410 /* Check the rule for consistency. */
3411 rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3413 if (!VLOG_DROP_WARN(&rl)) {
3414 char *s = flow_to_string(&facet->flow);
3415 VLOG_WARN("%s: facet should not exist", s);
3419 } else if (rule != facet->rule) {
3420 may_log = !VLOG_DROP_WARN(&rl);
3426 flow_format(&s, &facet->flow);
3427 ds_put_format(&s, ": facet associated with wrong rule (was "
3428 "table=%"PRIu8",", facet->rule->up.table_id);
3429 cls_rule_format(&facet->rule->up.cr, &s);
3430 ds_put_format(&s, ") (should have been table=%"PRIu8",",
3432 cls_rule_format(&rule->up.cr, &s);
3433 ds_put_char(&s, ')');
3435 VLOG_WARN("%s", ds_cstr(&s));
3442 /* Check the datapath actions for consistency. */
3443 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3444 struct action_xlate_ctx ctx;
3445 struct ofpbuf *odp_actions;
3446 bool actions_changed;
3447 bool should_install;
3449 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3450 subfacet->initial_tci, rule, 0, NULL);
3451 odp_actions = xlate_actions(&ctx, rule->up.actions,
3452 rule->up.n_actions);
3454 should_install = (ctx.may_set_up_flow
3455 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3456 if (!should_install && !subfacet->installed) {
3457 /* The actions for uninstallable flows may vary from one packet to
3458 * the next, so don't compare the actions. */
3462 actions_changed = (subfacet->actions_len != odp_actions->size
3463 || memcmp(subfacet->actions, odp_actions->data,
3464 subfacet->actions_len));
3465 if (should_install != subfacet->installed || actions_changed) {
3467 may_log = !VLOG_DROP_WARN(&rl);
3472 struct odputil_keybuf keybuf;
3477 subfacet_get_key(subfacet, &keybuf, &key);
3478 odp_flow_key_format(key.data, key.size, &s);
3480 ds_put_cstr(&s, ": inconsistency in subfacet");
3481 if (should_install != subfacet->installed) {
3482 enum odp_key_fitness fitness = subfacet->key_fitness;
3484 ds_put_format(&s, " (should%s have been installed)",
3485 should_install ? "" : " not");
3486 ds_put_format(&s, " (may_set_up_flow=%s, fitness=%s)",
3487 ctx.may_set_up_flow ? "true" : "false",
3488 odp_key_fitness_to_string(fitness));
3490 if (actions_changed) {
3491 ds_put_cstr(&s, " (actions were: ");
3492 format_odp_actions(&s, subfacet->actions,
3493 subfacet->actions_len);
3494 ds_put_cstr(&s, ") (correct actions: ");
3495 format_odp_actions(&s, odp_actions->data,
3497 ds_put_char(&s, ')');
3499 ds_put_cstr(&s, " (actions: ");
3500 format_odp_actions(&s, subfacet->actions,
3501 subfacet->actions_len);
3502 ds_put_char(&s, ')');
3504 VLOG_WARN("%s", ds_cstr(&s));
3510 ofpbuf_delete(odp_actions);
3516 /* Re-searches the classifier for 'facet':
3518 * - If the rule found is different from 'facet''s current rule, moves
3519 * 'facet' to the new rule and recompiles its actions.
3521 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3522 * where it is and recompiles its actions anyway.
3524 * - If there is none, destroys 'facet'.
3526 * Returns true if 'facet' still exists, false if it has been destroyed. */
3528 facet_revalidate(struct facet *facet)
3530 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3532 struct nlattr *odp_actions;
3535 struct actions *new_actions;
3537 struct action_xlate_ctx ctx;
3538 struct rule_dpif *new_rule;
3539 struct subfacet *subfacet;
3540 bool actions_changed;
3543 COVERAGE_INC(facet_revalidate);
3545 /* Determine the new rule. */
3546 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3548 /* No new rule, so delete the facet. */
3549 facet_remove(facet);
3553 /* Calculate new datapath actions.
3555 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3556 * emit a NetFlow expiration and, if so, we need to have the old state
3557 * around to properly compose it. */
3559 /* If the datapath actions changed or the installability changed,
3560 * then we need to talk to the datapath. */
3563 memset(&ctx, 0, sizeof ctx);
3564 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3565 struct ofpbuf *odp_actions;
3566 bool should_install;
3568 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3569 subfacet->initial_tci, new_rule, 0, NULL);
3570 odp_actions = xlate_actions(&ctx, new_rule->up.actions,
3571 new_rule->up.n_actions);
3572 actions_changed = (subfacet->actions_len != odp_actions->size
3573 || memcmp(subfacet->actions, odp_actions->data,
3574 subfacet->actions_len));
3576 should_install = (ctx.may_set_up_flow
3577 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3578 if (actions_changed || should_install != subfacet->installed) {
3579 if (should_install) {
3580 struct dpif_flow_stats stats;
3582 subfacet_install(subfacet,
3583 odp_actions->data, odp_actions->size, &stats);
3584 subfacet_update_stats(subfacet, &stats);
3586 subfacet_uninstall(subfacet);
3590 new_actions = xcalloc(list_size(&facet->subfacets),
3591 sizeof *new_actions);
3593 new_actions[i].odp_actions = xmemdup(odp_actions->data,
3595 new_actions[i].actions_len = odp_actions->size;
3598 ofpbuf_delete(odp_actions);
3602 facet_flush_stats(facet);
3605 /* Update 'facet' now that we've taken care of all the old state. */
3606 facet->tags = ctx.tags;
3607 facet->nf_flow.output_iface = ctx.nf_output_iface;
3608 facet->may_install = ctx.may_set_up_flow;
3609 facet->has_learn = ctx.has_learn;
3610 facet->has_normal = ctx.has_normal;
3611 facet->has_fin_timeout = ctx.has_fin_timeout;
3612 facet->mirrors = ctx.mirrors;
3615 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3616 if (new_actions[i].odp_actions) {
3617 free(subfacet->actions);
3618 subfacet->actions = new_actions[i].odp_actions;
3619 subfacet->actions_len = new_actions[i].actions_len;
3625 if (facet->rule != new_rule) {
3626 COVERAGE_INC(facet_changed_rule);
3627 list_remove(&facet->list_node);
3628 list_push_back(&new_rule->facets, &facet->list_node);
3629 facet->rule = new_rule;
3630 facet->used = new_rule->up.created;
3631 facet->prev_used = facet->used;
3637 /* Updates 'facet''s used time. Caller is responsible for calling
3638 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3640 facet_update_time(struct facet *facet, long long int used)
3642 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3643 if (used > facet->used) {
3645 ofproto_rule_update_used(&facet->rule->up, used);
3646 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3651 facet_reset_counters(struct facet *facet)
3653 facet->packet_count = 0;
3654 facet->byte_count = 0;
3655 facet->prev_packet_count = 0;
3656 facet->prev_byte_count = 0;
3657 facet->accounted_bytes = 0;
3661 facet_push_stats(struct facet *facet)
3663 uint64_t new_packets, new_bytes;
3665 assert(facet->packet_count >= facet->prev_packet_count);
3666 assert(facet->byte_count >= facet->prev_byte_count);
3667 assert(facet->used >= facet->prev_used);
3669 new_packets = facet->packet_count - facet->prev_packet_count;
3670 new_bytes = facet->byte_count - facet->prev_byte_count;
3672 if (new_packets || new_bytes || facet->used > facet->prev_used) {
3673 facet->prev_packet_count = facet->packet_count;
3674 facet->prev_byte_count = facet->byte_count;
3675 facet->prev_used = facet->used;
3677 flow_push_stats(facet->rule, &facet->flow,
3678 new_packets, new_bytes, facet->used);
3680 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
3681 facet->mirrors, new_packets, new_bytes);
3685 struct ofproto_push {
3686 struct action_xlate_ctx ctx;
3693 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
3695 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3698 rule->packet_count += push->packets;
3699 rule->byte_count += push->bytes;
3700 ofproto_rule_update_used(&rule->up, push->used);
3704 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3705 * 'rule''s actions and mirrors. */
3707 flow_push_stats(struct rule_dpif *rule,
3708 const struct flow *flow, uint64_t packets, uint64_t bytes,
3711 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3712 struct ofproto_push push;
3714 push.packets = packets;
3718 ofproto_rule_update_used(&rule->up, used);
3720 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, rule,
3722 push.ctx.resubmit_hook = push_resubmit;
3723 ofpbuf_delete(xlate_actions(&push.ctx,
3724 rule->up.actions, rule->up.n_actions));
3729 static struct subfacet *
3730 subfacet_find__(struct ofproto_dpif *ofproto,
3731 const struct nlattr *key, size_t key_len, uint32_t key_hash,
3732 const struct flow *flow)
3734 struct subfacet *subfacet;
3736 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
3737 &ofproto->subfacets) {
3739 ? (subfacet->key_len == key_len
3740 && !memcmp(key, subfacet->key, key_len))
3741 : flow_equal(flow, &subfacet->facet->flow)) {
3749 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
3750 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
3751 * there is one, otherwise creates and returns a new subfacet.
3753 * If the returned subfacet is new, then subfacet->actions will be NULL, in
3754 * which case the caller must populate the actions with
3755 * subfacet_make_actions(). */
3756 static struct subfacet *
3757 subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness,
3758 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
3760 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3761 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3762 struct subfacet *subfacet;
3764 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
3766 if (subfacet->facet == facet) {
3770 /* This shouldn't happen. */
3771 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
3772 subfacet_destroy(subfacet);
3775 subfacet = xzalloc(sizeof *subfacet);
3776 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
3777 list_push_back(&facet->subfacets, &subfacet->list_node);
3778 subfacet->facet = facet;
3779 subfacet->used = time_msec();
3780 subfacet->key_fitness = key_fitness;
3781 if (key_fitness != ODP_FIT_PERFECT) {
3782 subfacet->key = xmemdup(key, key_len);
3783 subfacet->key_len = key_len;
3785 subfacet->installed = false;
3786 subfacet->initial_tci = initial_tci;
3791 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
3792 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
3793 static struct subfacet *
3794 subfacet_find(struct ofproto_dpif *ofproto,
3795 const struct nlattr *key, size_t key_len)
3797 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3798 enum odp_key_fitness fitness;
3801 fitness = odp_flow_key_to_flow(key, key_len, &flow);
3802 if (fitness == ODP_FIT_ERROR) {
3806 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
3809 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
3810 * its facet within 'ofproto', and frees it. */
3812 subfacet_destroy__(struct subfacet *subfacet)
3814 struct facet *facet = subfacet->facet;
3815 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3817 subfacet_uninstall(subfacet);
3818 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
3819 list_remove(&subfacet->list_node);
3820 free(subfacet->key);
3821 free(subfacet->actions);
3825 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
3826 * last remaining subfacet in its facet destroys the facet too. */
3828 subfacet_destroy(struct subfacet *subfacet)
3830 struct facet *facet = subfacet->facet;
3832 if (list_is_singleton(&facet->subfacets)) {
3833 /* facet_remove() needs at least one subfacet (it will remove it). */
3834 facet_remove(facet);
3836 subfacet_destroy__(subfacet);
3840 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
3841 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
3842 * for use as temporary storage. */
3844 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
3847 if (!subfacet->key) {
3848 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
3849 odp_flow_key_from_flow(key, &subfacet->facet->flow);
3851 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
3855 /* Composes the datapath actions for 'subfacet' based on its rule's actions. */
3857 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet)
3859 struct facet *facet = subfacet->facet;
3860 struct rule_dpif *rule = facet->rule;
3861 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3862 struct ofpbuf *odp_actions;
3863 struct action_xlate_ctx ctx;
3865 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
3867 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3868 facet->tags = ctx.tags;
3869 facet->may_install = ctx.may_set_up_flow;
3870 facet->has_learn = ctx.has_learn;
3871 facet->has_normal = ctx.has_normal;
3872 facet->has_fin_timeout = ctx.has_fin_timeout;
3873 facet->nf_flow.output_iface = ctx.nf_output_iface;
3874 facet->mirrors = ctx.mirrors;
3876 if (subfacet->actions_len != odp_actions->size
3877 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
3878 free(subfacet->actions);
3879 subfacet->actions_len = odp_actions->size;
3880 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
3883 ofpbuf_delete(odp_actions);
3886 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
3887 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
3888 * in the datapath will be zeroed and 'stats' will be updated with traffic new
3889 * since 'subfacet' was last updated.
3891 * Returns 0 if successful, otherwise a positive errno value. */
3893 subfacet_install(struct subfacet *subfacet,
3894 const struct nlattr *actions, size_t actions_len,
3895 struct dpif_flow_stats *stats)
3897 struct facet *facet = subfacet->facet;
3898 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3899 struct odputil_keybuf keybuf;
3900 enum dpif_flow_put_flags flags;
3904 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3906 flags |= DPIF_FP_ZERO_STATS;
3909 subfacet_get_key(subfacet, &keybuf, &key);
3910 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
3911 actions, actions_len, stats);
3914 subfacet_reset_dp_stats(subfacet, stats);
3920 /* If 'subfacet' is installed in the datapath, uninstalls it. */
3922 subfacet_uninstall(struct subfacet *subfacet)
3924 if (subfacet->installed) {
3925 struct rule_dpif *rule = subfacet->facet->rule;
3926 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3927 struct odputil_keybuf keybuf;
3928 struct dpif_flow_stats stats;
3932 subfacet_get_key(subfacet, &keybuf, &key);
3933 error = dpif_flow_del(ofproto->dpif, key.data, key.size, &stats);
3934 subfacet_reset_dp_stats(subfacet, &stats);
3936 subfacet_update_stats(subfacet, &stats);
3938 subfacet->installed = false;
3940 assert(subfacet->dp_packet_count == 0);
3941 assert(subfacet->dp_byte_count == 0);
3945 /* Resets 'subfacet''s datapath statistics counters. This should be called
3946 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
3947 * non-null, it should contain the statistics returned by dpif when 'subfacet'
3948 * was reset in the datapath. 'stats' will be modified to include only
3949 * statistics new since 'subfacet' was last updated. */
3951 subfacet_reset_dp_stats(struct subfacet *subfacet,
3952 struct dpif_flow_stats *stats)
3955 && subfacet->dp_packet_count <= stats->n_packets
3956 && subfacet->dp_byte_count <= stats->n_bytes) {
3957 stats->n_packets -= subfacet->dp_packet_count;
3958 stats->n_bytes -= subfacet->dp_byte_count;
3961 subfacet->dp_packet_count = 0;
3962 subfacet->dp_byte_count = 0;
3965 /* Updates 'subfacet''s used time. The caller is responsible for calling
3966 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
3968 subfacet_update_time(struct subfacet *subfacet, long long int used)
3970 if (used > subfacet->used) {
3971 subfacet->used = used;
3972 facet_update_time(subfacet->facet, used);
3976 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
3978 * Because of the meaning of a subfacet's counters, it only makes sense to do
3979 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
3980 * represents a packet that was sent by hand or if it represents statistics
3981 * that have been cleared out of the datapath. */
3983 subfacet_update_stats(struct subfacet *subfacet,
3984 const struct dpif_flow_stats *stats)
3986 if (stats->n_packets || stats->used > subfacet->used) {
3987 struct facet *facet = subfacet->facet;
3989 subfacet_update_time(subfacet, stats->used);
3990 facet->packet_count += stats->n_packets;
3991 facet->byte_count += stats->n_bytes;
3992 facet->tcp_flags |= stats->tcp_flags;
3993 facet_push_stats(facet);
3994 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
4000 static struct rule_dpif *
4001 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4004 struct cls_rule *cls_rule;
4005 struct classifier *cls;
4007 if (table_id >= N_TABLES) {
4011 cls = &ofproto->up.tables[table_id].cls;
4012 if (flow->nw_frag & FLOW_NW_FRAG_ANY
4013 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4014 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
4015 * are unavailable. */
4016 struct flow ofpc_normal_flow = *flow;
4017 ofpc_normal_flow.tp_src = htons(0);
4018 ofpc_normal_flow.tp_dst = htons(0);
4019 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
4021 cls_rule = classifier_lookup(cls, flow);
4023 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
4027 complete_operation(struct rule_dpif *rule)
4029 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4031 rule_invalidate(rule);
4033 struct dpif_completion *c = xmalloc(sizeof *c);
4034 c->op = rule->up.pending;
4035 list_push_back(&ofproto->completions, &c->list_node);
4037 ofoperation_complete(rule->up.pending, 0);
4041 static struct rule *
4044 struct rule_dpif *rule = xmalloc(sizeof *rule);
4049 rule_dealloc(struct rule *rule_)
4051 struct rule_dpif *rule = rule_dpif_cast(rule_);
4056 rule_construct(struct rule *rule_)
4058 struct rule_dpif *rule = rule_dpif_cast(rule_);
4059 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4060 struct rule_dpif *victim;
4064 error = validate_actions(rule->up.actions, rule->up.n_actions,
4065 &rule->up.cr.flow, ofproto->max_ports);
4070 rule->packet_count = 0;
4071 rule->byte_count = 0;
4073 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
4074 if (victim && !list_is_empty(&victim->facets)) {
4075 struct facet *facet;
4077 rule->facets = victim->facets;
4078 list_moved(&rule->facets);
4079 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4080 /* XXX: We're only clearing our local counters here. It's possible
4081 * that quite a few packets are unaccounted for in the datapath
4082 * statistics. These will be accounted to the new rule instead of
4083 * cleared as required. This could be fixed by clearing out the
4084 * datapath statistics for this facet, but currently it doesn't
4086 facet_reset_counters(facet);
4090 /* Must avoid list_moved() in this case. */
4091 list_init(&rule->facets);
4094 table_id = rule->up.table_id;
4095 rule->tag = (victim ? victim->tag
4097 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
4098 ofproto->tables[table_id].basis));
4100 complete_operation(rule);
4105 rule_destruct(struct rule *rule_)
4107 struct rule_dpif *rule = rule_dpif_cast(rule_);
4108 struct facet *facet, *next_facet;
4110 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4111 facet_revalidate(facet);
4114 complete_operation(rule);
4118 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4120 struct rule_dpif *rule = rule_dpif_cast(rule_);
4121 struct facet *facet;
4123 /* Start from historical data for 'rule' itself that are no longer tracked
4124 * in facets. This counts, for example, facets that have expired. */
4125 *packets = rule->packet_count;
4126 *bytes = rule->byte_count;
4128 /* Add any statistics that are tracked by facets. This includes
4129 * statistical data recently updated by ofproto_update_stats() as well as
4130 * stats for packets that were executed "by hand" via dpif_execute(). */
4131 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4132 *packets += facet->packet_count;
4133 *bytes += facet->byte_count;
4138 rule_execute(struct rule *rule_, const struct flow *flow,
4139 struct ofpbuf *packet)
4141 struct rule_dpif *rule = rule_dpif_cast(rule_);
4142 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4143 struct action_xlate_ctx ctx;
4144 struct ofpbuf *odp_actions;
4147 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
4148 rule, packet_get_tcp_flags(packet, flow), packet);
4149 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
4150 size = packet->size;
4151 if (execute_odp_actions(ofproto, flow, odp_actions->data,
4152 odp_actions->size, packet)) {
4153 rule->packet_count++;
4154 rule->byte_count += size;
4155 flow_push_stats(rule, flow, 1, size, time_msec());
4157 ofpbuf_delete(odp_actions);
4163 rule_modify_actions(struct rule *rule_)
4165 struct rule_dpif *rule = rule_dpif_cast(rule_);
4166 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4169 error = validate_actions(rule->up.actions, rule->up.n_actions,
4170 &rule->up.cr.flow, ofproto->max_ports);
4172 ofoperation_complete(rule->up.pending, error);
4176 complete_operation(rule);
4179 /* Sends 'packet' out 'ofport'.
4180 * May modify 'packet'.
4181 * Returns 0 if successful, otherwise a positive errno value. */
4183 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4185 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4186 struct ofpbuf key, odp_actions;
4187 struct odputil_keybuf keybuf;
4192 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
4193 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4195 if (odp_port != ofport->odp_port) {
4196 eth_pop_vlan(packet);
4197 flow.vlan_tci = htons(0);
4200 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4201 odp_flow_key_from_flow(&key, &flow);
4203 ofpbuf_init(&odp_actions, 32);
4204 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4206 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4207 error = dpif_execute(ofproto->dpif,
4209 odp_actions.data, odp_actions.size,
4211 ofpbuf_uninit(&odp_actions);
4214 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4215 ofproto->up.name, odp_port, strerror(error));
4217 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
4221 /* OpenFlow to datapath action translation. */
4223 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
4224 struct action_xlate_ctx *ctx);
4225 static void xlate_normal(struct action_xlate_ctx *);
4228 put_userspace_action(const struct ofproto_dpif *ofproto,
4229 struct ofpbuf *odp_actions,
4230 const struct flow *flow,
4231 const struct user_action_cookie *cookie)
4235 pid = dpif_port_get_pid(ofproto->dpif,
4236 ofp_port_to_odp_port(flow->in_port));
4238 return odp_put_userspace_action(pid, cookie, odp_actions);
4241 /* Compose SAMPLE action for sFlow. */
4243 compose_sflow_action(const struct ofproto_dpif *ofproto,
4244 struct ofpbuf *odp_actions,
4245 const struct flow *flow,
4248 uint32_t port_ifindex;
4249 uint32_t probability;
4250 struct user_action_cookie cookie;
4251 size_t sample_offset, actions_offset;
4252 int cookie_offset, n_output;
4254 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4258 if (odp_port == OVSP_NONE) {
4262 port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4266 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4268 /* Number of packets out of UINT_MAX to sample. */
4269 probability = dpif_sflow_get_probability(ofproto->sflow);
4270 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4272 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4274 cookie.type = USER_ACTION_COOKIE_SFLOW;
4275 cookie.data = port_ifindex;
4276 cookie.n_output = n_output;
4277 cookie.vlan_tci = 0;
4278 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4280 nl_msg_end_nested(odp_actions, actions_offset);
4281 nl_msg_end_nested(odp_actions, sample_offset);
4282 return cookie_offset;
4285 /* SAMPLE action must be first action in any given list of actions.
4286 * At this point we do not have all information required to build it. So try to
4287 * build sample action as complete as possible. */
4289 add_sflow_action(struct action_xlate_ctx *ctx)
4291 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4293 &ctx->flow, OVSP_NONE);
4294 ctx->sflow_odp_port = 0;
4295 ctx->sflow_n_outputs = 0;
4298 /* Fix SAMPLE action according to data collected while composing ODP actions.
4299 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4300 * USERSPACE action's user-cookie which is required for sflow. */
4302 fix_sflow_action(struct action_xlate_ctx *ctx)
4304 const struct flow *base = &ctx->base_flow;
4305 struct user_action_cookie *cookie;
4307 if (!ctx->user_cookie_offset) {
4311 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4313 assert(cookie != NULL);
4314 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4316 if (ctx->sflow_n_outputs) {
4317 cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow,
4318 ctx->sflow_odp_port);
4320 if (ctx->sflow_n_outputs >= 255) {
4321 cookie->n_output = 255;
4323 cookie->n_output = ctx->sflow_n_outputs;
4325 cookie->vlan_tci = base->vlan_tci;
4329 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4332 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4333 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4334 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4335 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4339 struct priority_to_dscp *pdscp;
4341 if (ofport->up.opp.config & htonl(OFPPC_NO_FWD)
4342 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4346 pdscp = get_priority(ofport, ctx->flow.skb_priority);
4348 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4349 ctx->flow.nw_tos |= pdscp->dscp;
4352 /* We may not have an ofport record for this port, but it doesn't hurt
4353 * to allow forwarding to it anyhow. Maybe such a port will appear
4354 * later and we're pre-populating the flow table. */
4357 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4358 ctx->flow.vlan_tci);
4359 if (out_port != odp_port) {
4360 ctx->flow.vlan_tci = htons(0);
4362 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4363 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4365 ctx->sflow_odp_port = odp_port;
4366 ctx->sflow_n_outputs++;
4367 ctx->nf_output_iface = ofp_port;
4368 ctx->flow.vlan_tci = flow_vlan_tci;
4369 ctx->flow.nw_tos = flow_nw_tos;
4373 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4375 compose_output_action__(ctx, ofp_port, true);
4379 xlate_table_action(struct action_xlate_ctx *ctx,
4380 uint16_t in_port, uint8_t table_id)
4382 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4383 struct ofproto_dpif *ofproto = ctx->ofproto;
4384 struct rule_dpif *rule;
4385 uint16_t old_in_port;
4386 uint8_t old_table_id;
4388 old_table_id = ctx->table_id;
4389 ctx->table_id = table_id;
4391 /* Look up a flow with 'in_port' as the input port. */
4392 old_in_port = ctx->flow.in_port;
4393 ctx->flow.in_port = in_port;
4394 rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
4397 if (table_id > 0 && table_id < N_TABLES) {
4398 struct table_dpif *table = &ofproto->tables[table_id];
4399 if (table->other_table) {
4402 : rule_calculate_tag(&ctx->flow,
4403 &table->other_table->wc,
4408 /* Restore the original input port. Otherwise OFPP_NORMAL and
4409 * OFPP_IN_PORT will have surprising behavior. */
4410 ctx->flow.in_port = old_in_port;
4412 if (ctx->resubmit_hook) {
4413 ctx->resubmit_hook(ctx, rule);
4417 struct rule_dpif *old_rule = ctx->rule;
4421 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
4422 ctx->rule = old_rule;
4426 ctx->table_id = old_table_id;
4428 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4430 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
4431 MAX_RESUBMIT_RECURSION);
4432 ctx->max_resubmit_trigger = true;
4437 xlate_resubmit_table(struct action_xlate_ctx *ctx,
4438 const struct nx_action_resubmit *nar)
4443 in_port = (nar->in_port == htons(OFPP_IN_PORT)
4445 : ntohs(nar->in_port));
4446 table_id = nar->table == 255 ? ctx->table_id : nar->table;
4448 xlate_table_action(ctx, in_port, table_id);
4452 flood_packets(struct action_xlate_ctx *ctx, bool all)
4454 struct ofport_dpif *ofport;
4456 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
4457 uint16_t ofp_port = ofport->up.ofp_port;
4459 if (ofp_port == ctx->flow.in_port) {
4464 compose_output_action__(ctx, ofp_port, false);
4465 } else if (!(ofport->up.opp.config & htonl(OFPPC_NO_FLOOD))) {
4466 compose_output_action(ctx, ofp_port);
4470 ctx->nf_output_iface = NF_OUT_FLOOD;
4474 execute_controller_action(struct action_xlate_ctx *ctx, int len,
4475 enum ofp_packet_in_reason reason)
4477 struct ofputil_packet_in pin;
4478 struct ofpbuf *packet;
4480 ctx->may_set_up_flow = false;
4485 packet = ofpbuf_clone(ctx->packet);
4487 if (packet->l2 && packet->l3) {
4488 struct eth_header *eh;
4490 eth_pop_vlan(packet);
4492 assert(eh->eth_type == ctx->flow.dl_type);
4493 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
4494 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
4496 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
4497 eth_push_vlan(packet, ctx->flow.vlan_tci);
4501 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4502 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
4503 ctx->flow.nw_tos, ctx->flow.nw_ttl);
4507 if (ctx->flow.nw_proto == IPPROTO_TCP) {
4508 packet_set_tcp_port(packet, ctx->flow.tp_src,
4510 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
4511 packet_set_udp_port(packet, ctx->flow.tp_src,
4518 pin.packet = packet->data;
4519 pin.packet_len = packet->size;
4520 pin.reason = reason;
4521 pin.table_id = ctx->table_id;
4522 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
4526 pin.total_len = packet->size;
4527 flow_get_metadata(&ctx->flow, &pin.fmd);
4529 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin, &ctx->flow);
4530 ofpbuf_delete(packet);
4534 compose_dec_ttl(struct action_xlate_ctx *ctx)
4536 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
4537 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
4541 if (ctx->flow.nw_ttl > 1) {
4545 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL);
4547 /* Stop processing for current table. */
4553 xlate_output_action__(struct action_xlate_ctx *ctx,
4554 uint16_t port, uint16_t max_len)
4556 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
4558 ctx->nf_output_iface = NF_OUT_DROP;
4562 compose_output_action(ctx, ctx->flow.in_port);
4565 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
4571 flood_packets(ctx, false);
4574 flood_packets(ctx, true);
4576 case OFPP_CONTROLLER:
4577 execute_controller_action(ctx, max_len, OFPR_ACTION);
4583 if (port != ctx->flow.in_port) {
4584 compose_output_action(ctx, port);
4589 if (prev_nf_output_iface == NF_OUT_FLOOD) {
4590 ctx->nf_output_iface = NF_OUT_FLOOD;
4591 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
4592 ctx->nf_output_iface = prev_nf_output_iface;
4593 } else if (prev_nf_output_iface != NF_OUT_DROP &&
4594 ctx->nf_output_iface != NF_OUT_FLOOD) {
4595 ctx->nf_output_iface = NF_OUT_MULTI;
4600 xlate_output_reg_action(struct action_xlate_ctx *ctx,
4601 const struct nx_action_output_reg *naor)
4603 struct mf_subfield src;
4606 nxm_decode(&src, naor->src, naor->ofs_nbits);
4607 ofp_port = mf_get_subfield(&src, &ctx->flow);
4609 if (ofp_port <= UINT16_MAX) {
4610 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
4615 xlate_output_action(struct action_xlate_ctx *ctx,
4616 const struct ofp_action_output *oao)
4618 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
4622 xlate_enqueue_action(struct action_xlate_ctx *ctx,
4623 const struct ofp_action_enqueue *oae)
4626 uint32_t flow_priority, priority;
4629 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
4632 /* Fall back to ordinary output action. */
4633 xlate_output_action__(ctx, ntohs(oae->port), 0);
4637 /* Figure out datapath output port. */
4638 ofp_port = ntohs(oae->port);
4639 if (ofp_port == OFPP_IN_PORT) {
4640 ofp_port = ctx->flow.in_port;
4641 } else if (ofp_port == ctx->flow.in_port) {
4645 /* Add datapath actions. */
4646 flow_priority = ctx->flow.skb_priority;
4647 ctx->flow.skb_priority = priority;
4648 compose_output_action(ctx, ofp_port);
4649 ctx->flow.skb_priority = flow_priority;
4651 /* Update NetFlow output port. */
4652 if (ctx->nf_output_iface == NF_OUT_DROP) {
4653 ctx->nf_output_iface = ofp_port;
4654 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
4655 ctx->nf_output_iface = NF_OUT_MULTI;
4660 xlate_set_queue_action(struct action_xlate_ctx *ctx,
4661 const struct nx_action_set_queue *nasq)
4666 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
4669 /* Couldn't translate queue to a priority, so ignore. A warning
4670 * has already been logged. */
4674 ctx->flow.skb_priority = priority;
4677 struct xlate_reg_state {
4683 xlate_autopath(struct action_xlate_ctx *ctx,
4684 const struct nx_action_autopath *naa)
4686 uint16_t ofp_port = ntohl(naa->id);
4687 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
4689 if (!port || !port->bundle) {
4690 ofp_port = OFPP_NONE;
4691 } else if (port->bundle->bond) {
4692 /* Autopath does not support VLAN hashing. */
4693 struct ofport_dpif *slave = bond_choose_output_slave(
4694 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
4696 ofp_port = slave->up.ofp_port;
4699 autopath_execute(naa, &ctx->flow, ofp_port);
4703 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
4705 struct ofproto_dpif *ofproto = ofproto_;
4706 struct ofport_dpif *port;
4716 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
4719 port = get_ofp_port(ofproto, ofp_port);
4720 return port ? port->may_enable : false;
4725 xlate_learn_action(struct action_xlate_ctx *ctx,
4726 const struct nx_action_learn *learn)
4728 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
4729 struct ofputil_flow_mod fm;
4732 learn_execute(learn, &ctx->flow, &fm);
4734 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
4735 if (error && !VLOG_DROP_WARN(&rl)) {
4736 VLOG_WARN("learning action failed to modify flow table (%s)",
4737 ofperr_get_name(error));
4743 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
4744 * means "infinite". */
4746 reduce_timeout(uint16_t max, uint16_t *timeout)
4748 if (max && (!*timeout || *timeout > max)) {
4754 xlate_fin_timeout(struct action_xlate_ctx *ctx,
4755 const struct nx_action_fin_timeout *naft)
4757 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
4758 struct rule_dpif *rule = ctx->rule;
4760 reduce_timeout(ntohs(naft->fin_idle_timeout), &rule->up.idle_timeout);
4761 reduce_timeout(ntohs(naft->fin_hard_timeout), &rule->up.hard_timeout);
4766 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
4768 if (port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
4769 ? htonl(OFPPC_NO_RECV_STP)
4770 : htonl(OFPPC_NO_RECV))) {
4774 /* Only drop packets here if both forwarding and learning are
4775 * disabled. If just learning is enabled, we need to have
4776 * OFPP_NORMAL and the learning action have a look at the packet
4777 * before we can drop it. */
4778 if (!stp_forward_in_state(port->stp_state)
4779 && !stp_learn_in_state(port->stp_state)) {
4787 do_xlate_actions(const union ofp_action *in, size_t n_in,
4788 struct action_xlate_ctx *ctx)
4790 const struct ofport_dpif *port;
4791 const union ofp_action *ia;
4792 bool was_evictable = true;
4795 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
4796 if (port && !may_receive(port, ctx)) {
4797 /* Drop this flow. */
4802 /* Don't let the rule we're working on get evicted underneath us. */
4803 was_evictable = ctx->rule->up.evictable;
4804 ctx->rule->up.evictable = false;
4806 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
4807 const struct ofp_action_dl_addr *oada;
4808 const struct nx_action_resubmit *nar;
4809 const struct nx_action_set_tunnel *nast;
4810 const struct nx_action_set_queue *nasq;
4811 const struct nx_action_multipath *nam;
4812 const struct nx_action_autopath *naa;
4813 const struct nx_action_bundle *nab;
4814 const struct nx_action_output_reg *naor;
4815 enum ofputil_action_code code;
4822 code = ofputil_decode_action_unsafe(ia);
4824 case OFPUTIL_OFPAT_OUTPUT:
4825 xlate_output_action(ctx, &ia->output);
4828 case OFPUTIL_OFPAT_SET_VLAN_VID:
4829 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
4830 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
4833 case OFPUTIL_OFPAT_SET_VLAN_PCP:
4834 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
4835 ctx->flow.vlan_tci |= htons(
4836 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
4839 case OFPUTIL_OFPAT_STRIP_VLAN:
4840 ctx->flow.vlan_tci = htons(0);
4843 case OFPUTIL_OFPAT_SET_DL_SRC:
4844 oada = ((struct ofp_action_dl_addr *) ia);
4845 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
4848 case OFPUTIL_OFPAT_SET_DL_DST:
4849 oada = ((struct ofp_action_dl_addr *) ia);
4850 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
4853 case OFPUTIL_OFPAT_SET_NW_SRC:
4854 ctx->flow.nw_src = ia->nw_addr.nw_addr;
4857 case OFPUTIL_OFPAT_SET_NW_DST:
4858 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
4861 case OFPUTIL_OFPAT_SET_NW_TOS:
4862 /* OpenFlow 1.0 only supports IPv4. */
4863 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4864 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4865 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
4869 case OFPUTIL_OFPAT_SET_TP_SRC:
4870 ctx->flow.tp_src = ia->tp_port.tp_port;
4873 case OFPUTIL_OFPAT_SET_TP_DST:
4874 ctx->flow.tp_dst = ia->tp_port.tp_port;
4877 case OFPUTIL_OFPAT_ENQUEUE:
4878 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
4881 case OFPUTIL_NXAST_RESUBMIT:
4882 nar = (const struct nx_action_resubmit *) ia;
4883 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
4886 case OFPUTIL_NXAST_RESUBMIT_TABLE:
4887 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
4890 case OFPUTIL_NXAST_SET_TUNNEL:
4891 nast = (const struct nx_action_set_tunnel *) ia;
4892 tun_id = htonll(ntohl(nast->tun_id));
4893 ctx->flow.tun_id = tun_id;
4896 case OFPUTIL_NXAST_SET_QUEUE:
4897 nasq = (const struct nx_action_set_queue *) ia;
4898 xlate_set_queue_action(ctx, nasq);
4901 case OFPUTIL_NXAST_POP_QUEUE:
4902 ctx->flow.skb_priority = ctx->orig_skb_priority;
4905 case OFPUTIL_NXAST_REG_MOVE:
4906 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
4910 case OFPUTIL_NXAST_REG_LOAD:
4911 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
4915 case OFPUTIL_NXAST_NOTE:
4916 /* Nothing to do. */
4919 case OFPUTIL_NXAST_SET_TUNNEL64:
4920 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
4921 ctx->flow.tun_id = tun_id;
4924 case OFPUTIL_NXAST_MULTIPATH:
4925 nam = (const struct nx_action_multipath *) ia;
4926 multipath_execute(nam, &ctx->flow);
4929 case OFPUTIL_NXAST_AUTOPATH:
4930 naa = (const struct nx_action_autopath *) ia;
4931 xlate_autopath(ctx, naa);
4934 case OFPUTIL_NXAST_BUNDLE:
4935 ctx->ofproto->has_bundle_action = true;
4936 nab = (const struct nx_action_bundle *) ia;
4937 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
4942 case OFPUTIL_NXAST_BUNDLE_LOAD:
4943 ctx->ofproto->has_bundle_action = true;
4944 nab = (const struct nx_action_bundle *) ia;
4945 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
4949 case OFPUTIL_NXAST_OUTPUT_REG:
4950 naor = (const struct nx_action_output_reg *) ia;
4951 xlate_output_reg_action(ctx, naor);
4954 case OFPUTIL_NXAST_LEARN:
4955 ctx->has_learn = true;
4956 if (ctx->may_learn) {
4957 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
4961 case OFPUTIL_NXAST_DEC_TTL:
4962 if (compose_dec_ttl(ctx)) {
4967 case OFPUTIL_NXAST_EXIT:
4971 case OFPUTIL_NXAST_FIN_TIMEOUT:
4972 ctx->has_fin_timeout = true;
4973 xlate_fin_timeout(ctx, (const struct nx_action_fin_timeout *) ia);
4979 /* We've let OFPP_NORMAL and the learning action look at the packet,
4980 * so drop it now if forwarding is disabled. */
4981 if (port && !stp_forward_in_state(port->stp_state)) {
4982 ofpbuf_clear(ctx->odp_actions);
4983 add_sflow_action(ctx);
4986 ctx->rule->up.evictable = was_evictable;
4991 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
4992 struct ofproto_dpif *ofproto, const struct flow *flow,
4993 ovs_be16 initial_tci, struct rule_dpif *rule,
4994 uint8_t tcp_flags, const struct ofpbuf *packet)
4996 ctx->ofproto = ofproto;
4998 ctx->base_flow = ctx->flow;
4999 ctx->base_flow.tun_id = 0;
5000 ctx->base_flow.vlan_tci = initial_tci;
5002 ctx->packet = packet;
5003 ctx->may_learn = packet != NULL;
5004 ctx->tcp_flags = tcp_flags;
5005 ctx->resubmit_hook = NULL;
5008 static struct ofpbuf *
5009 xlate_actions(struct action_xlate_ctx *ctx,
5010 const union ofp_action *in, size_t n_in)
5012 struct flow orig_flow = ctx->flow;
5014 COVERAGE_INC(ofproto_dpif_xlate);
5016 ctx->odp_actions = ofpbuf_new(512);
5017 ofpbuf_reserve(ctx->odp_actions, NL_A_U32_SIZE);
5019 ctx->may_set_up_flow = true;
5020 ctx->has_learn = false;
5021 ctx->has_normal = false;
5022 ctx->has_fin_timeout = false;
5023 ctx->nf_output_iface = NF_OUT_DROP;
5026 ctx->max_resubmit_trigger = false;
5027 ctx->orig_skb_priority = ctx->flow.skb_priority;
5031 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
5032 switch (ctx->ofproto->up.frag_handling) {
5033 case OFPC_FRAG_NORMAL:
5034 /* We must pretend that transport ports are unavailable. */
5035 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
5036 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
5039 case OFPC_FRAG_DROP:
5040 return ctx->odp_actions;
5042 case OFPC_FRAG_REASM:
5045 case OFPC_FRAG_NX_MATCH:
5046 /* Nothing to do. */
5049 case OFPC_INVALID_TTL_TO_CONTROLLER:
5054 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
5055 ctx->may_set_up_flow = false;
5056 return ctx->odp_actions;
5058 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5059 struct flow original_flow = ctx->flow;
5060 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
5062 add_sflow_action(ctx);
5063 do_xlate_actions(in, n_in, ctx);
5065 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook
5066 && !VLOG_DROP_ERR(&trace_rl)) {
5067 struct ds ds = DS_EMPTY_INITIALIZER;
5069 ofproto_trace(ctx->ofproto, &original_flow, ctx->packet,
5071 VLOG_ERR("Trace triggered by excessive resubmit recursion:\n%s",
5076 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
5077 ctx->odp_actions->data,
5078 ctx->odp_actions->size)) {
5079 ctx->may_set_up_flow = false;
5081 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
5083 compose_output_action(ctx, OFPP_LOCAL);
5086 add_mirror_actions(ctx, &orig_flow);
5087 fix_sflow_action(ctx);
5090 return ctx->odp_actions;
5093 /* OFPP_NORMAL implementation. */
5095 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
5097 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
5098 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
5099 * the bundle on which the packet was received, returns the VLAN to which the
5102 * Both 'vid' and the return value are in the range 0...4095. */
5104 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
5106 switch (in_bundle->vlan_mode) {
5107 case PORT_VLAN_ACCESS:
5108 return in_bundle->vlan;
5111 case PORT_VLAN_TRUNK:
5114 case PORT_VLAN_NATIVE_UNTAGGED:
5115 case PORT_VLAN_NATIVE_TAGGED:
5116 return vid ? vid : in_bundle->vlan;
5123 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
5124 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
5127 * 'vid' should be the VID obtained from the 802.1Q header that was received as
5128 * part of a packet (specify 0 if there was no 802.1Q header), in the range
5131 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
5133 /* Allow any VID on the OFPP_NONE port. */
5134 if (in_bundle == &ofpp_none_bundle) {
5138 switch (in_bundle->vlan_mode) {
5139 case PORT_VLAN_ACCESS:
5142 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5143 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
5144 "packet received on port %s configured as VLAN "
5145 "%"PRIu16" access port",
5146 in_bundle->ofproto->up.name, vid,
5147 in_bundle->name, in_bundle->vlan);
5153 case PORT_VLAN_NATIVE_UNTAGGED:
5154 case PORT_VLAN_NATIVE_TAGGED:
5156 /* Port must always carry its native VLAN. */
5160 case PORT_VLAN_TRUNK:
5161 if (!ofbundle_includes_vlan(in_bundle, vid)) {
5163 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5164 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
5165 "received on port %s not configured for trunking "
5167 in_bundle->ofproto->up.name, vid,
5168 in_bundle->name, vid);
5180 /* Given 'vlan', the VLAN that a packet belongs to, and
5181 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
5182 * that should be included in the 802.1Q header. (If the return value is 0,
5183 * then the 802.1Q header should only be included in the packet if there is a
5186 * Both 'vlan' and the return value are in the range 0...4095. */
5188 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
5190 switch (out_bundle->vlan_mode) {
5191 case PORT_VLAN_ACCESS:
5194 case PORT_VLAN_TRUNK:
5195 case PORT_VLAN_NATIVE_TAGGED:
5198 case PORT_VLAN_NATIVE_UNTAGGED:
5199 return vlan == out_bundle->vlan ? 0 : vlan;
5207 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
5210 struct ofport_dpif *port;
5212 ovs_be16 tci, old_tci;
5214 vid = output_vlan_to_vid(out_bundle, vlan);
5215 if (!out_bundle->bond) {
5216 port = ofbundle_get_a_port(out_bundle);
5218 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
5221 /* No slaves enabled, so drop packet. */
5226 old_tci = ctx->flow.vlan_tci;
5228 if (tci || out_bundle->use_priority_tags) {
5229 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
5231 tci |= htons(VLAN_CFI);
5234 ctx->flow.vlan_tci = tci;
5236 compose_output_action(ctx, port->up.ofp_port);
5237 ctx->flow.vlan_tci = old_tci;
5241 mirror_mask_ffs(mirror_mask_t mask)
5243 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
5248 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
5250 return (bundle->vlan_mode != PORT_VLAN_ACCESS
5251 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
5255 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
5257 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
5260 /* Returns an arbitrary interface within 'bundle'. */
5261 static struct ofport_dpif *
5262 ofbundle_get_a_port(const struct ofbundle *bundle)
5264 return CONTAINER_OF(list_front(&bundle->ports),
5265 struct ofport_dpif, bundle_node);
5269 vlan_is_mirrored(const struct ofmirror *m, int vlan)
5271 return !m->vlans || bitmap_is_set(m->vlans, vlan);
5274 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
5275 * to a VLAN. In general most packets may be mirrored but we want to drop
5276 * protocols that may confuse switches. */
5278 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
5280 /* If you change this function's behavior, please update corresponding
5281 * documentation in vswitch.xml at the same time. */
5282 if (dst[0] != 0x01) {
5283 /* All the currently banned MACs happen to start with 01 currently, so
5284 * this is a quick way to eliminate most of the good ones. */
5286 if (eth_addr_is_reserved(dst)) {
5287 /* Drop STP, IEEE pause frames, and other reserved protocols
5288 * (01-80-c2-00-00-0x). */
5292 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
5294 if ((dst[3] & 0xfe) == 0xcc &&
5295 (dst[4] & 0xfe) == 0xcc &&
5296 (dst[5] & 0xfe) == 0xcc) {
5297 /* Drop the following protocols plus others following the same
5300 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
5301 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
5302 STP Uplink Fast (01-00-0c-cd-cd-cd) */
5306 if (!(dst[3] | dst[4] | dst[5])) {
5307 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
5316 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5318 struct ofproto_dpif *ofproto = ctx->ofproto;
5319 mirror_mask_t mirrors;
5320 struct ofbundle *in_bundle;
5323 const struct nlattr *a;
5326 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5327 ctx->packet != NULL);
5331 mirrors = in_bundle->src_mirrors;
5333 /* Drop frames on bundles reserved for mirroring. */
5334 if (in_bundle->mirror_out) {
5335 if (ctx->packet != NULL) {
5336 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5337 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5338 "%s, which is reserved exclusively for mirroring",
5339 ctx->ofproto->up.name, in_bundle->name);
5345 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5346 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5349 vlan = input_vid_to_vlan(in_bundle, vid);
5351 /* Look at the output ports to check for destination selections. */
5353 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5354 ctx->odp_actions->size) {
5355 enum ovs_action_attr type = nl_attr_type(a);
5356 struct ofport_dpif *ofport;
5358 if (type != OVS_ACTION_ATTR_OUTPUT) {
5362 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5363 if (ofport && ofport->bundle) {
5364 mirrors |= ofport->bundle->dst_mirrors;
5372 /* Restore the original packet before adding the mirror actions. */
5373 ctx->flow = *orig_flow;
5378 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5380 if (!vlan_is_mirrored(m, vlan)) {
5381 mirrors &= mirrors - 1;
5385 mirrors &= ~m->dup_mirrors;
5386 ctx->mirrors |= m->dup_mirrors;
5388 output_normal(ctx, m->out, vlan);
5389 } else if (eth_dst_may_rspan(orig_flow->dl_dst)
5390 && vlan != m->out_vlan) {
5391 struct ofbundle *bundle;
5393 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5394 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5395 && !bundle->mirror_out) {
5396 output_normal(ctx, bundle, m->out_vlan);
5404 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5405 uint64_t packets, uint64_t bytes)
5411 for (; mirrors; mirrors &= mirrors - 1) {
5414 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5417 /* In normal circumstances 'm' will not be NULL. However,
5418 * if mirrors are reconfigured, we can temporarily get out
5419 * of sync in facet_revalidate(). We could "correct" the
5420 * mirror list before reaching here, but doing that would
5421 * not properly account the traffic stats we've currently
5422 * accumulated for previous mirror configuration. */
5426 m->packet_count += packets;
5427 m->byte_count += bytes;
5431 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
5432 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
5433 * indicate this; newer upstream kernels use gratuitous ARP requests. */
5435 is_gratuitous_arp(const struct flow *flow)
5437 return (flow->dl_type == htons(ETH_TYPE_ARP)
5438 && eth_addr_is_broadcast(flow->dl_dst)
5439 && (flow->nw_proto == ARP_OP_REPLY
5440 || (flow->nw_proto == ARP_OP_REQUEST
5441 && flow->nw_src == flow->nw_dst)));
5445 update_learning_table(struct ofproto_dpif *ofproto,
5446 const struct flow *flow, int vlan,
5447 struct ofbundle *in_bundle)
5449 struct mac_entry *mac;
5451 /* Don't learn the OFPP_NONE port. */
5452 if (in_bundle == &ofpp_none_bundle) {
5456 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
5460 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
5461 if (is_gratuitous_arp(flow)) {
5462 /* We don't want to learn from gratuitous ARP packets that are
5463 * reflected back over bond slaves so we lock the learning table. */
5464 if (!in_bundle->bond) {
5465 mac_entry_set_grat_arp_lock(mac);
5466 } else if (mac_entry_is_grat_arp_locked(mac)) {
5471 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
5472 /* The log messages here could actually be useful in debugging,
5473 * so keep the rate limit relatively high. */
5474 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5475 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
5476 "on port %s in VLAN %d",
5477 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
5478 in_bundle->name, vlan);
5480 mac->port.p = in_bundle;
5481 tag_set_add(&ofproto->revalidate_set,
5482 mac_learning_changed(ofproto->ml, mac));
5486 static struct ofbundle *
5487 lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn)
5489 struct ofport_dpif *ofport;
5491 /* Special-case OFPP_NONE, which a controller may use as the ingress
5492 * port for traffic that it is sourcing. */
5493 if (in_port == OFPP_NONE) {
5494 return &ofpp_none_bundle;
5497 /* Find the port and bundle for the received packet. */
5498 ofport = get_ofp_port(ofproto, in_port);
5499 if (ofport && ofport->bundle) {
5500 return ofport->bundle;
5503 /* Odd. A few possible reasons here:
5505 * - We deleted a port but there are still a few packets queued up
5508 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
5509 * we don't know about.
5511 * - The ofproto client didn't configure the port as part of a bundle.
5514 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5516 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
5517 "port %"PRIu16, ofproto->up.name, in_port);
5522 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
5523 * dropped. Returns true if they may be forwarded, false if they should be
5526 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
5527 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
5529 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
5530 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
5531 * checked by input_vid_is_valid().
5533 * May also add tags to '*tags', although the current implementation only does
5534 * so in one special case.
5537 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
5538 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
5540 struct ofbundle *in_bundle = in_port->bundle;
5542 /* Drop frames for reserved multicast addresses
5543 * only if forward_bpdu option is absent. */
5544 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
5548 if (in_bundle->bond) {
5549 struct mac_entry *mac;
5551 switch (bond_check_admissibility(in_bundle->bond, in_port,
5552 flow->dl_dst, tags)) {
5559 case BV_DROP_IF_MOVED:
5560 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
5561 if (mac && mac->port.p != in_bundle &&
5562 (!is_gratuitous_arp(flow)
5563 || mac_entry_is_grat_arp_locked(mac))) {
5574 xlate_normal(struct action_xlate_ctx *ctx)
5576 struct ofport_dpif *in_port;
5577 struct ofbundle *in_bundle;
5578 struct mac_entry *mac;
5582 ctx->has_normal = true;
5584 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
5585 ctx->packet != NULL);
5590 /* We know 'in_port' exists unless it is "ofpp_none_bundle",
5591 * since lookup_input_bundle() succeeded. */
5592 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5594 /* Drop malformed frames. */
5595 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
5596 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
5597 if (ctx->packet != NULL) {
5598 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5599 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
5600 "VLAN tag received on port %s",
5601 ctx->ofproto->up.name, in_bundle->name);
5606 /* Drop frames on bundles reserved for mirroring. */
5607 if (in_bundle->mirror_out) {
5608 if (ctx->packet != NULL) {
5609 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5610 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5611 "%s, which is reserved exclusively for mirroring",
5612 ctx->ofproto->up.name, in_bundle->name);
5618 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
5619 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5622 vlan = input_vid_to_vlan(in_bundle, vid);
5624 /* Check other admissibility requirements. */
5626 !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
5630 /* Learn source MAC. */
5631 if (ctx->may_learn) {
5632 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
5635 /* Determine output bundle. */
5636 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
5639 if (mac->port.p != in_bundle) {
5640 output_normal(ctx, mac->port.p, vlan);
5643 struct ofbundle *bundle;
5645 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
5646 if (bundle != in_bundle
5647 && ofbundle_includes_vlan(bundle, vlan)
5648 && bundle->floodable
5649 && !bundle->mirror_out) {
5650 output_normal(ctx, bundle, vlan);
5653 ctx->nf_output_iface = NF_OUT_FLOOD;
5657 /* Optimized flow revalidation.
5659 * It's a difficult problem, in general, to tell which facets need to have
5660 * their actions recalculated whenever the OpenFlow flow table changes. We
5661 * don't try to solve that general problem: for most kinds of OpenFlow flow
5662 * table changes, we recalculate the actions for every facet. This is
5663 * relatively expensive, but it's good enough if the OpenFlow flow table
5664 * doesn't change very often.
5666 * However, we can expect one particular kind of OpenFlow flow table change to
5667 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5668 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5669 * table, we add a special case that applies to flow tables in which every rule
5670 * has the same form (that is, the same wildcards), except that the table is
5671 * also allowed to have a single "catch-all" flow that matches all packets. We
5672 * optimize this case by tagging all of the facets that resubmit into the table
5673 * and invalidating the same tag whenever a flow changes in that table. The
5674 * end result is that we revalidate just the facets that need it (and sometimes
5675 * a few more, but not all of the facets or even all of the facets that
5676 * resubmit to the table modified by MAC learning). */
5678 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
5679 * into an OpenFlow table with the given 'basis'. */
5681 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
5684 if (flow_wildcards_is_catchall(wc)) {
5687 struct flow tag_flow = *flow;
5688 flow_zero_wildcards(&tag_flow, wc);
5689 return tag_create_deterministic(flow_hash(&tag_flow, secret));
5693 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5694 * taggability of that table.
5696 * This function must be called after *each* change to a flow table. If you
5697 * skip calling it on some changes then the pointer comparisons at the end can
5698 * be invalid if you get unlucky. For example, if a flow removal causes a
5699 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5700 * different wildcards to be created with the same address, then this function
5701 * will incorrectly skip revalidation. */
5703 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5705 struct table_dpif *table = &ofproto->tables[table_id];
5706 const struct oftable *oftable = &ofproto->up.tables[table_id];
5707 struct cls_table *catchall, *other;
5708 struct cls_table *t;
5710 catchall = other = NULL;
5712 switch (hmap_count(&oftable->cls.tables)) {
5714 /* We could tag this OpenFlow table but it would make the logic a
5715 * little harder and it's a corner case that doesn't seem worth it
5721 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
5722 if (cls_table_is_catchall(t)) {
5724 } else if (!other) {
5727 /* Indicate that we can't tag this by setting both tables to
5728 * NULL. (We know that 'catchall' is already NULL.) */
5735 /* Can't tag this table. */
5739 if (table->catchall_table != catchall || table->other_table != other) {
5740 table->catchall_table = catchall;
5741 table->other_table = other;
5742 ofproto->need_revalidate = true;
5746 /* Given 'rule' that has changed in some way (either it is a rule being
5747 * inserted, a rule being deleted, or a rule whose actions are being
5748 * modified), marks facets for revalidation to ensure that packets will be
5749 * forwarded correctly according to the new state of the flow table.
5751 * This function must be called after *each* change to a flow table. See
5752 * the comment on table_update_taggable() for more information. */
5754 rule_invalidate(const struct rule_dpif *rule)
5756 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5758 table_update_taggable(ofproto, rule->up.table_id);
5760 if (!ofproto->need_revalidate) {
5761 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5763 if (table->other_table && rule->tag) {
5764 tag_set_add(&ofproto->revalidate_set, rule->tag);
5766 ofproto->need_revalidate = true;
5772 set_frag_handling(struct ofproto *ofproto_,
5773 enum ofp_config_flags frag_handling)
5775 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5777 if (frag_handling != OFPC_FRAG_REASM) {
5778 ofproto->need_revalidate = true;
5786 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5787 const struct flow *flow,
5788 const union ofp_action *ofp_actions, size_t n_ofp_actions)
5790 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5793 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
5794 return OFPERR_NXBRC_BAD_IN_PORT;
5797 error = validate_actions(ofp_actions, n_ofp_actions, flow,
5798 ofproto->max_ports);
5800 struct odputil_keybuf keybuf;
5801 struct ofpbuf *odp_actions;
5802 struct ofproto_push push;
5805 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5806 odp_flow_key_from_flow(&key, flow);
5808 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, NULL,
5809 packet_get_tcp_flags(packet, flow), packet);
5811 /* Ensure that resubmits in 'ofp_actions' get accounted to their
5812 * matching rules. */
5814 push.bytes = packet->size;
5815 push.used = time_msec();
5816 push.ctx.resubmit_hook = push_resubmit;
5818 odp_actions = xlate_actions(&push.ctx, ofp_actions, n_ofp_actions);
5819 dpif_execute(ofproto->dpif, key.data, key.size,
5820 odp_actions->data, odp_actions->size, packet);
5821 ofpbuf_delete(odp_actions);
5829 set_netflow(struct ofproto *ofproto_,
5830 const struct netflow_options *netflow_options)
5832 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5834 if (netflow_options) {
5835 if (!ofproto->netflow) {
5836 ofproto->netflow = netflow_create();
5838 return netflow_set_options(ofproto->netflow, netflow_options);
5840 netflow_destroy(ofproto->netflow);
5841 ofproto->netflow = NULL;
5847 get_netflow_ids(const struct ofproto *ofproto_,
5848 uint8_t *engine_type, uint8_t *engine_id)
5850 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5852 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
5856 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5858 if (!facet_is_controller_flow(facet) &&
5859 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5860 struct subfacet *subfacet;
5861 struct ofexpired expired;
5863 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5864 if (subfacet->installed) {
5865 struct dpif_flow_stats stats;
5867 subfacet_install(subfacet, subfacet->actions,
5868 subfacet->actions_len, &stats);
5869 subfacet_update_stats(subfacet, &stats);
5873 expired.flow = facet->flow;
5874 expired.packet_count = facet->packet_count;
5875 expired.byte_count = facet->byte_count;
5876 expired.used = facet->used;
5877 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5882 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5884 struct facet *facet;
5886 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5887 send_active_timeout(ofproto, facet);
5891 static struct ofproto_dpif *
5892 ofproto_dpif_lookup(const char *name)
5894 struct ofproto_dpif *ofproto;
5896 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5897 hash_string(name, 0), &all_ofproto_dpifs) {
5898 if (!strcmp(ofproto->up.name, name)) {
5906 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5907 const char *argv[], void *aux OVS_UNUSED)
5909 struct ofproto_dpif *ofproto;
5912 ofproto = ofproto_dpif_lookup(argv[1]);
5914 unixctl_command_reply_error(conn, "no such bridge");
5917 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
5919 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5920 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
5924 unixctl_command_reply(conn, "table successfully flushed");
5928 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5929 const char *argv[], void *aux OVS_UNUSED)
5931 struct ds ds = DS_EMPTY_INITIALIZER;
5932 const struct ofproto_dpif *ofproto;
5933 const struct mac_entry *e;
5935 ofproto = ofproto_dpif_lookup(argv[1]);
5937 unixctl_command_reply_error(conn, "no such bridge");
5941 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5942 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5943 struct ofbundle *bundle = e->port.p;
5944 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
5945 ofbundle_get_a_port(bundle)->odp_port,
5946 e->vlan, ETH_ADDR_ARGS(e->mac),
5947 mac_entry_age(ofproto->ml, e));
5949 unixctl_command_reply(conn, ds_cstr(&ds));
5954 struct action_xlate_ctx ctx;
5960 trace_format_rule(struct ds *result, uint8_t table_id, int level,
5961 const struct rule_dpif *rule)
5963 ds_put_char_multiple(result, '\t', level);
5965 ds_put_cstr(result, "No match\n");
5969 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5970 table_id, ntohll(rule->up.flow_cookie));
5971 cls_rule_format(&rule->up.cr, result);
5972 ds_put_char(result, '\n');
5974 ds_put_char_multiple(result, '\t', level);
5975 ds_put_cstr(result, "OpenFlow ");
5976 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
5977 ds_put_char(result, '\n');
5981 trace_format_flow(struct ds *result, int level, const char *title,
5982 struct trace_ctx *trace)
5984 ds_put_char_multiple(result, '\t', level);
5985 ds_put_format(result, "%s: ", title);
5986 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
5987 ds_put_cstr(result, "unchanged");
5989 flow_format(result, &trace->ctx.flow);
5990 trace->flow = trace->ctx.flow;
5992 ds_put_char(result, '\n');
5996 trace_format_regs(struct ds *result, int level, const char *title,
5997 struct trace_ctx *trace)
6001 ds_put_char_multiple(result, '\t', level);
6002 ds_put_format(result, "%s:", title);
6003 for (i = 0; i < FLOW_N_REGS; i++) {
6004 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
6006 ds_put_char(result, '\n');
6010 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
6012 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
6013 struct ds *result = trace->result;
6015 ds_put_char(result, '\n');
6016 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
6017 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
6018 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
6022 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
6023 void *aux OVS_UNUSED)
6025 const char *dpname = argv[1];
6026 struct ofproto_dpif *ofproto;
6027 struct ofpbuf odp_key;
6028 struct ofpbuf *packet;
6029 ovs_be16 initial_tci;
6035 ofpbuf_init(&odp_key, 0);
6038 ofproto = ofproto_dpif_lookup(dpname);
6040 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
6044 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
6045 /* ofproto/trace dpname flow [-generate] */
6046 const char *flow_s = argv[2];
6047 const char *generate_s = argv[3];
6050 /* Convert string to datapath key. */
6051 ofpbuf_init(&odp_key, 0);
6052 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
6054 unixctl_command_reply_error(conn, "Bad flow syntax");
6058 /* Convert odp_key to flow. */
6059 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
6060 odp_key.size, &flow,
6061 &initial_tci, NULL);
6062 if (error == ODP_FIT_ERROR) {
6063 unixctl_command_reply_error(conn, "Invalid flow");
6067 /* Generate a packet, if requested. */
6069 packet = ofpbuf_new(0);
6070 flow_compose(packet, &flow);
6072 } else if (argc == 6) {
6073 /* ofproto/trace dpname priority tun_id in_port packet */
6074 const char *priority_s = argv[2];
6075 const char *tun_id_s = argv[3];
6076 const char *in_port_s = argv[4];
6077 const char *packet_s = argv[5];
6078 uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s));
6079 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
6080 uint32_t priority = atoi(priority_s);
6083 msg = eth_from_hex(packet_s, &packet);
6085 unixctl_command_reply_error(conn, msg);
6089 ds_put_cstr(&result, "Packet: ");
6090 s = ofp_packet_to_string(packet->data, packet->size);
6091 ds_put_cstr(&result, s);
6094 flow_extract(packet, priority, tun_id, in_port, &flow);
6095 initial_tci = flow.vlan_tci;
6097 unixctl_command_reply_error(conn, "Bad command syntax");
6101 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
6102 unixctl_command_reply(conn, ds_cstr(&result));
6105 ds_destroy(&result);
6106 ofpbuf_delete(packet);
6107 ofpbuf_uninit(&odp_key);
6111 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
6112 const struct ofpbuf *packet, ovs_be16 initial_tci,
6115 struct rule_dpif *rule;
6117 ds_put_cstr(ds, "Flow: ");
6118 flow_format(ds, flow);
6119 ds_put_char(ds, '\n');
6121 rule = rule_dpif_lookup(ofproto, flow, 0);
6122 trace_format_rule(ds, 0, 0, rule);
6124 struct trace_ctx trace;
6125 struct ofpbuf *odp_actions;
6128 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
6131 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
6132 rule, tcp_flags, packet);
6133 trace.ctx.resubmit_hook = trace_resubmit;
6134 odp_actions = xlate_actions(&trace.ctx,
6135 rule->up.actions, rule->up.n_actions);
6137 ds_put_char(ds, '\n');
6138 trace_format_flow(ds, 0, "Final flow", &trace);
6139 ds_put_cstr(ds, "Datapath actions: ");
6140 format_odp_actions(ds, odp_actions->data, odp_actions->size);
6141 ofpbuf_delete(odp_actions);
6143 if (!trace.ctx.may_set_up_flow) {
6145 ds_put_cstr(ds, "\nThis flow is not cachable.");
6147 ds_put_cstr(ds, "\nThe datapath actions are incomplete--"
6148 "for complete actions, please supply a packet.");
6155 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6156 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6159 unixctl_command_reply(conn, NULL);
6163 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6164 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6167 unixctl_command_reply(conn, NULL);
6170 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
6171 * 'reply' describing the results. */
6173 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
6175 struct facet *facet;
6179 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6180 if (!facet_check_consistency(facet)) {
6185 ofproto->need_revalidate = true;
6189 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
6190 ofproto->up.name, errors);
6192 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
6197 ofproto_dpif_self_check(struct unixctl_conn *conn,
6198 int argc, const char *argv[], void *aux OVS_UNUSED)
6200 struct ds reply = DS_EMPTY_INITIALIZER;
6201 struct ofproto_dpif *ofproto;
6204 ofproto = ofproto_dpif_lookup(argv[1]);
6206 unixctl_command_reply_error(conn, "Unknown ofproto (use "
6207 "ofproto/list for help)");
6210 ofproto_dpif_self_check__(ofproto, &reply);
6212 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6213 ofproto_dpif_self_check__(ofproto, &reply);
6217 unixctl_command_reply(conn, ds_cstr(&reply));
6222 ofproto_dpif_unixctl_init(void)
6224 static bool registered;
6230 unixctl_command_register(
6232 "bridge {tun_id in_port packet | odp_flow [-generate]}",
6233 2, 5, ofproto_unixctl_trace, NULL);
6234 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6235 ofproto_unixctl_fdb_flush, NULL);
6236 unixctl_command_register("fdb/show", "bridge", 1, 1,
6237 ofproto_unixctl_fdb_show, NULL);
6238 unixctl_command_register("ofproto/clog", "", 0, 0,
6239 ofproto_dpif_clog, NULL);
6240 unixctl_command_register("ofproto/unclog", "", 0, 0,
6241 ofproto_dpif_unclog, NULL);
6242 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6243 ofproto_dpif_self_check, NULL);
6246 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6248 * This is deprecated. It is only for compatibility with broken device drivers
6249 * in old versions of Linux that do not properly support VLANs when VLAN
6250 * devices are not used. When broken device drivers are no longer in
6251 * widespread use, we will delete these interfaces. */
6254 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
6256 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6257 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6259 if (realdev_ofp_port == ofport->realdev_ofp_port
6260 && vid == ofport->vlandev_vid) {
6264 ofproto->need_revalidate = true;
6266 if (ofport->realdev_ofp_port) {
6269 if (realdev_ofp_port && ofport->bundle) {
6270 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6271 * themselves be part of a bundle. */
6272 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6275 ofport->realdev_ofp_port = realdev_ofp_port;
6276 ofport->vlandev_vid = vid;
6278 if (realdev_ofp_port) {
6279 vsp_add(ofport, realdev_ofp_port, vid);
6286 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
6288 return hash_2words(realdev_ofp_port, vid);
6292 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6293 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
6295 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6296 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
6297 int vid = vlan_tci_to_vid(vlan_tci);
6298 const struct vlan_splinter *vsp;
6300 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6301 hash_realdev_vid(realdev_ofp_port, vid),
6302 &ofproto->realdev_vid_map) {
6303 if (vsp->realdev_ofp_port == realdev_ofp_port
6304 && vsp->vid == vid) {
6305 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
6309 return realdev_odp_port;
6312 static struct vlan_splinter *
6313 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
6315 struct vlan_splinter *vsp;
6317 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
6318 &ofproto->vlandev_map) {
6319 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6328 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6329 uint16_t vlandev_ofp_port, int *vid)
6331 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6332 const struct vlan_splinter *vsp;
6334 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6339 return vsp->realdev_ofp_port;
6346 vsp_remove(struct ofport_dpif *port)
6348 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6349 struct vlan_splinter *vsp;
6351 vsp = vlandev_find(ofproto, port->up.ofp_port);
6353 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6354 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6357 port->realdev_ofp_port = 0;
6359 VLOG_ERR("missing vlan device record");
6364 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
6366 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6368 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6369 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6370 == realdev_ofp_port)) {
6371 struct vlan_splinter *vsp;
6373 vsp = xmalloc(sizeof *vsp);
6374 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6375 hash_int(port->up.ofp_port, 0));
6376 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6377 hash_realdev_vid(realdev_ofp_port, vid));
6378 vsp->realdev_ofp_port = realdev_ofp_port;
6379 vsp->vlandev_ofp_port = port->up.ofp_port;
6382 port->realdev_ofp_port = realdev_ofp_port;
6384 VLOG_ERR("duplicate vlan device record");
6388 const struct ofproto_class ofproto_dpif_class = {
6417 port_is_lacp_current,
6418 NULL, /* rule_choose_table */
6425 rule_modify_actions,
6433 get_cfm_remote_mpids,
6437 get_stp_port_status,
6444 is_mirror_output_bundle,
6445 forward_bpdu_changed,