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 tag_type rule_calculate_tag(const struct flow *,
113 const struct flow_wildcards *,
115 static void rule_invalidate(const struct rule_dpif *);
117 #define MAX_MIRRORS 32
118 typedef uint32_t mirror_mask_t;
119 #define MIRROR_MASK_C(X) UINT32_C(X)
120 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
122 struct ofproto_dpif *ofproto; /* Owning ofproto. */
123 size_t idx; /* In ofproto's "mirrors" array. */
124 void *aux; /* Key supplied by ofproto's client. */
125 char *name; /* Identifier for log messages. */
127 /* Selection criteria. */
128 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
129 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
130 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
132 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
133 struct ofbundle *out; /* Output port or NULL. */
134 int out_vlan; /* Output VLAN or -1. */
135 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
138 int64_t packet_count; /* Number of packets sent. */
139 int64_t byte_count; /* Number of bytes sent. */
142 static void mirror_destroy(struct ofmirror *);
143 static void update_mirror_stats(struct ofproto_dpif *ofproto,
144 mirror_mask_t mirrors,
145 uint64_t packets, uint64_t bytes);
148 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
149 struct ofproto_dpif *ofproto; /* Owning ofproto. */
150 void *aux; /* Key supplied by ofproto's client. */
151 char *name; /* Identifier for log messages. */
154 struct list ports; /* Contains "struct ofport"s. */
155 enum port_vlan_mode vlan_mode; /* VLAN mode */
156 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
157 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
158 * NULL if all VLANs are trunked. */
159 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
160 struct bond *bond; /* Nonnull iff more than one port. */
161 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
164 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
166 /* Port mirroring info. */
167 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
168 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
169 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
172 static void bundle_remove(struct ofport *);
173 static void bundle_update(struct ofbundle *);
174 static void bundle_destroy(struct ofbundle *);
175 static void bundle_del_port(struct ofport_dpif *);
176 static void bundle_run(struct ofbundle *);
177 static void bundle_wait(struct ofbundle *);
178 static struct ofbundle *lookup_input_bundle(struct ofproto_dpif *,
179 uint16_t in_port, bool warn);
181 /* A controller may use OFPP_NONE as the ingress port to indicate that
182 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
183 * when an input bundle is needed for validation (e.g., mirroring or
184 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
185 * any 'port' structs, so care must be taken when dealing with it. */
186 static struct ofbundle ofpp_none_bundle = {
188 .vlan_mode = PORT_VLAN_TRUNK
191 static void stp_run(struct ofproto_dpif *ofproto);
192 static void stp_wait(struct ofproto_dpif *ofproto);
193 static int set_stp_port(struct ofport *,
194 const struct ofproto_port_stp_settings *);
196 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
198 struct action_xlate_ctx {
199 /* action_xlate_ctx_init() initializes these members. */
202 struct ofproto_dpif *ofproto;
204 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
205 * this flow when actions change header fields. */
208 /* The packet corresponding to 'flow', or a null pointer if we are
209 * revalidating without a packet to refer to. */
210 const struct ofpbuf *packet;
212 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
213 * actions update the flow table?
215 * We want to update these tables if we are actually processing a packet,
216 * or if we are accounting for packets that the datapath has processed, but
217 * not if we are just revalidating. */
220 /* The rule that we are currently translating, or NULL. */
221 struct rule_dpif *rule;
223 /* Union of the set of TCP flags seen so far in this flow. (Used only by
224 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
228 /* If nonnull, called just before executing a resubmit action. In
229 * addition, disables logging of traces when the recursion depth is
232 * This is normally null so the client has to set it manually after
233 * calling action_xlate_ctx_init(). */
234 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *);
236 /* xlate_actions() initializes and uses these members. The client might want
237 * to look at them after it returns. */
239 struct ofpbuf *odp_actions; /* Datapath actions. */
240 tag_type tags; /* Tags associated with actions. */
241 bool may_set_up_flow; /* True ordinarily; false if the actions must
242 * be reassessed for every packet. */
243 bool has_learn; /* Actions include NXAST_LEARN? */
244 bool has_normal; /* Actions output to OFPP_NORMAL? */
245 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
246 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
247 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
249 /* xlate_actions() initializes and uses these members, but the client has no
250 * reason to look at them. */
252 int recurse; /* Recursion level, via xlate_table_action. */
253 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
254 struct flow base_flow; /* Flow at the last commit. */
255 uint32_t orig_skb_priority; /* Priority when packet arrived. */
256 uint8_t table_id; /* OpenFlow table ID where flow was found. */
257 uint32_t sflow_n_outputs; /* Number of output ports. */
258 uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
259 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
260 bool exit; /* No further actions should be processed. */
263 static void action_xlate_ctx_init(struct action_xlate_ctx *,
264 struct ofproto_dpif *, const struct flow *,
265 ovs_be16 initial_tci, struct rule_dpif *,
266 uint8_t tcp_flags, const struct ofpbuf *);
267 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
268 const union ofp_action *in, size_t n_in);
270 /* An exact-match instantiation of an OpenFlow flow.
272 * A facet associates a "struct flow", which represents the Open vSwitch
273 * userspace idea of an exact-match flow, with one or more subfacets. Each
274 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
275 * the facet. When the kernel module (or other dpif implementation) and Open
276 * vSwitch userspace agree on the definition of a flow key, there is exactly
277 * one subfacet per facet. If the dpif implementation supports more-specific
278 * flow matching than userspace, however, a facet can have more than one
279 * subfacet, each of which corresponds to some distinction in flow that
280 * userspace simply doesn't understand.
282 * Flow expiration works in terms of subfacets, so a facet must have at least
283 * one subfacet or it will never expire, leaking memory. */
286 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
287 struct list list_node; /* In owning rule's 'facets' list. */
288 struct rule_dpif *rule; /* Owning rule. */
291 struct list subfacets;
292 long long int used; /* Time last used; time created if not used. */
299 * - Do include packets and bytes sent "by hand", e.g. with
302 * - Do include packets and bytes that were obtained from the datapath
303 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
304 * DPIF_FP_ZERO_STATS).
306 * - Do not include packets or bytes that can be obtained from the
307 * datapath for any existing subfacet.
309 uint64_t packet_count; /* Number of packets received. */
310 uint64_t byte_count; /* Number of bytes received. */
312 /* Resubmit statistics. */
313 uint64_t prev_packet_count; /* Number of packets from last stats push. */
314 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
315 long long int prev_used; /* Used time from last stats push. */
318 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
319 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
320 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
322 /* Properties of datapath actions.
324 * Every subfacet has its own actions because actions can differ slightly
325 * between splintered and non-splintered subfacets due to the VLAN tag
326 * being initially different (present vs. absent). All of them have these
327 * properties in common so we just store one copy of them here. */
328 bool may_install; /* Reassess actions for every packet? */
329 bool has_learn; /* Actions include NXAST_LEARN? */
330 bool has_normal; /* Actions output to OFPP_NORMAL? */
331 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
332 tag_type tags; /* Tags that would require revalidation. */
333 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
336 static struct facet *facet_create(struct rule_dpif *, const struct flow *);
337 static void facet_remove(struct facet *);
338 static void facet_free(struct facet *);
340 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
341 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
342 const struct flow *);
343 static bool facet_revalidate(struct facet *);
344 static bool facet_check_consistency(struct facet *);
346 static void facet_flush_stats(struct facet *);
348 static void facet_update_time(struct facet *, long long int used);
349 static void facet_reset_counters(struct facet *);
350 static void facet_push_stats(struct facet *);
351 static void facet_learn(struct facet *);
352 static void facet_account(struct facet *);
354 static bool facet_is_controller_flow(struct facet *);
356 /* A dpif flow and actions associated with a facet.
358 * See also the large comment on struct facet. */
361 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
362 struct list list_node; /* In struct facet's 'facets' list. */
363 struct facet *facet; /* Owning facet. */
367 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
368 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
369 * regenerate the ODP flow key from ->facet->flow. */
370 enum odp_key_fitness key_fitness;
374 long long int used; /* Time last used; time created if not used. */
376 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
377 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
381 * These should be essentially identical for every subfacet in a facet, but
382 * may differ in trivial ways due to VLAN splinters. */
383 size_t actions_len; /* Number of bytes in actions[]. */
384 struct nlattr *actions; /* Datapath actions. */
386 bool installed; /* Installed in datapath? */
388 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
389 * splinters can cause it to differ. This value should be removed when
390 * the VLAN splinters feature is no longer needed. */
391 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
394 static struct subfacet *subfacet_create(struct facet *, enum odp_key_fitness,
395 const struct nlattr *key,
396 size_t key_len, ovs_be16 initial_tci);
397 static struct subfacet *subfacet_find(struct ofproto_dpif *,
398 const struct nlattr *key, size_t key_len);
399 static void subfacet_destroy(struct subfacet *);
400 static void subfacet_destroy__(struct subfacet *);
401 static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *,
403 static void subfacet_reset_dp_stats(struct subfacet *,
404 struct dpif_flow_stats *);
405 static void subfacet_update_time(struct subfacet *, long long int used);
406 static void subfacet_update_stats(struct subfacet *,
407 const struct dpif_flow_stats *);
408 static void subfacet_make_actions(struct subfacet *,
409 const struct ofpbuf *packet);
410 static int subfacet_install(struct subfacet *,
411 const struct nlattr *actions, size_t actions_len,
412 struct dpif_flow_stats *);
413 static void subfacet_uninstall(struct subfacet *);
419 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
420 struct list bundle_node; /* In struct ofbundle's "ports" list. */
421 struct cfm *cfm; /* Connectivity Fault Management, if any. */
422 tag_type tag; /* Tag associated with this port. */
423 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
424 bool may_enable; /* May be enabled in bonds. */
425 long long int carrier_seq; /* Carrier status changes. */
428 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
429 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
430 long long int stp_state_entered;
432 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
434 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
436 * This is deprecated. It is only for compatibility with broken device
437 * drivers in old versions of Linux that do not properly support VLANs when
438 * VLAN devices are not used. When broken device drivers are no longer in
439 * widespread use, we will delete these interfaces. */
440 uint16_t realdev_ofp_port;
444 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
445 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
446 * traffic egressing the 'ofport' with that priority should be marked with. */
447 struct priority_to_dscp {
448 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
449 uint32_t priority; /* Priority of this queue (see struct flow). */
451 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
454 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
456 * This is deprecated. It is only for compatibility with broken device drivers
457 * in old versions of Linux that do not properly support VLANs when VLAN
458 * devices are not used. When broken device drivers are no longer in
459 * widespread use, we will delete these interfaces. */
460 struct vlan_splinter {
461 struct hmap_node realdev_vid_node;
462 struct hmap_node vlandev_node;
463 uint16_t realdev_ofp_port;
464 uint16_t vlandev_ofp_port;
468 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
469 uint32_t realdev, ovs_be16 vlan_tci);
470 static uint16_t vsp_vlandev_to_realdev(const struct ofproto_dpif *,
471 uint16_t vlandev, int *vid);
472 static void vsp_remove(struct ofport_dpif *);
473 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
475 static struct ofport_dpif *
476 ofport_dpif_cast(const struct ofport *ofport)
478 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
479 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
482 static void port_run(struct ofport_dpif *);
483 static void port_wait(struct ofport_dpif *);
484 static int set_cfm(struct ofport *, const struct cfm_settings *);
485 static void ofport_clear_priorities(struct ofport_dpif *);
487 struct dpif_completion {
488 struct list list_node;
489 struct ofoperation *op;
492 /* Extra information about a classifier table.
493 * Currently used just for optimized flow revalidation. */
495 /* If either of these is nonnull, then this table has a form that allows
496 * flows to be tagged to avoid revalidating most flows for the most common
497 * kinds of flow table changes. */
498 struct cls_table *catchall_table; /* Table that wildcards all fields. */
499 struct cls_table *other_table; /* Table with any other wildcard set. */
500 uint32_t basis; /* Keeps each table's tags separate. */
503 struct ofproto_dpif {
504 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
513 struct netflow *netflow;
514 struct dpif_sflow *sflow;
515 struct hmap bundles; /* Contains "struct ofbundle"s. */
516 struct mac_learning *ml;
517 struct ofmirror *mirrors[MAX_MIRRORS];
518 bool has_bonded_bundles;
521 struct timer next_expiration;
525 struct hmap subfacets;
528 struct table_dpif tables[N_TABLES];
529 bool need_revalidate;
530 struct tag_set revalidate_set;
532 /* Support for debugging async flow mods. */
533 struct list completions;
535 bool has_bundle_action; /* True when the first bundle action appears. */
536 struct netdev_stats stats; /* To account packets generated and consumed in
541 long long int stp_last_tick;
543 /* VLAN splinters. */
544 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
545 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
548 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
549 * for debugging the asynchronous flow_mod implementation.) */
552 /* All existing ofproto_dpif instances, indexed by ->up.name. */
553 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
555 static void ofproto_dpif_unixctl_init(void);
557 static struct ofproto_dpif *
558 ofproto_dpif_cast(const struct ofproto *ofproto)
560 assert(ofproto->ofproto_class == &ofproto_dpif_class);
561 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
564 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
566 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
568 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
569 const struct ofpbuf *, ovs_be16 initial_tci,
572 /* Packet processing. */
573 static void update_learning_table(struct ofproto_dpif *,
574 const struct flow *, int vlan,
577 #define FLOW_MISS_MAX_BATCH 50
578 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
580 /* Flow expiration. */
581 static int expire(struct ofproto_dpif *);
584 static void send_netflow_active_timeouts(struct ofproto_dpif *);
587 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
589 compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions,
590 const struct flow *, uint32_t odp_port);
591 static void add_mirror_actions(struct action_xlate_ctx *ctx,
592 const struct flow *flow);
593 /* Global variables. */
594 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
596 /* Factory functions. */
599 enumerate_types(struct sset *types)
601 dp_enumerate_types(types);
605 enumerate_names(const char *type, struct sset *names)
607 return dp_enumerate_names(type, names);
611 del(const char *type, const char *name)
616 error = dpif_open(name, type, &dpif);
618 error = dpif_delete(dpif);
624 /* Basic life-cycle. */
626 static struct ofproto *
629 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
634 dealloc(struct ofproto *ofproto_)
636 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
641 construct(struct ofproto *ofproto_)
643 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
644 const char *name = ofproto->up.name;
648 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
650 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
654 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
655 ofproto->n_matches = 0;
657 dpif_flow_flush(ofproto->dpif);
658 dpif_recv_purge(ofproto->dpif);
660 error = dpif_recv_set(ofproto->dpif, true);
662 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
663 dpif_close(ofproto->dpif);
667 ofproto->netflow = NULL;
668 ofproto->sflow = NULL;
670 hmap_init(&ofproto->bundles);
671 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
672 for (i = 0; i < MAX_MIRRORS; i++) {
673 ofproto->mirrors[i] = NULL;
675 ofproto->has_bonded_bundles = false;
677 timer_set_duration(&ofproto->next_expiration, 1000);
679 hmap_init(&ofproto->facets);
680 hmap_init(&ofproto->subfacets);
682 for (i = 0; i < N_TABLES; i++) {
683 struct table_dpif *table = &ofproto->tables[i];
685 table->catchall_table = NULL;
686 table->other_table = NULL;
687 table->basis = random_uint32();
689 ofproto->need_revalidate = false;
690 tag_set_init(&ofproto->revalidate_set);
692 list_init(&ofproto->completions);
694 ofproto_dpif_unixctl_init();
696 ofproto->has_bundle_action = false;
698 hmap_init(&ofproto->vlandev_map);
699 hmap_init(&ofproto->realdev_vid_map);
701 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
702 hash_string(ofproto->up.name, 0));
703 memset(&ofproto->stats, 0, sizeof ofproto->stats);
705 ofproto_init_tables(ofproto_, N_TABLES);
711 complete_operations(struct ofproto_dpif *ofproto)
713 struct dpif_completion *c, *next;
715 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
716 ofoperation_complete(c->op, 0);
717 list_remove(&c->list_node);
723 destruct(struct ofproto *ofproto_)
725 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
726 struct rule_dpif *rule, *next_rule;
727 struct oftable *table;
730 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
731 complete_operations(ofproto);
733 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
734 struct cls_cursor cursor;
736 cls_cursor_init(&cursor, &table->cls, NULL);
737 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
738 ofproto_rule_destroy(&rule->up);
742 for (i = 0; i < MAX_MIRRORS; i++) {
743 mirror_destroy(ofproto->mirrors[i]);
746 netflow_destroy(ofproto->netflow);
747 dpif_sflow_destroy(ofproto->sflow);
748 hmap_destroy(&ofproto->bundles);
749 mac_learning_destroy(ofproto->ml);
751 hmap_destroy(&ofproto->facets);
752 hmap_destroy(&ofproto->subfacets);
754 hmap_destroy(&ofproto->vlandev_map);
755 hmap_destroy(&ofproto->realdev_vid_map);
757 dpif_close(ofproto->dpif);
761 run_fast(struct ofproto *ofproto_)
763 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
766 /* Handle one or more batches of upcalls, until there's nothing left to do
767 * or until we do a fixed total amount of work.
769 * We do work in batches because it can be much cheaper to set up a number
770 * of flows and fire off their patches all at once. We do multiple batches
771 * because in some cases handling a packet can cause another packet to be
772 * queued almost immediately as part of the return flow. Both
773 * optimizations can make major improvements on some benchmarks and
774 * presumably for real traffic as well. */
776 while (work < FLOW_MISS_MAX_BATCH) {
777 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
787 run(struct ofproto *ofproto_)
789 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
790 struct ofport_dpif *ofport;
791 struct ofbundle *bundle;
795 complete_operations(ofproto);
797 dpif_run(ofproto->dpif);
799 error = run_fast(ofproto_);
804 if (timer_expired(&ofproto->next_expiration)) {
805 int delay = expire(ofproto);
806 timer_set_duration(&ofproto->next_expiration, delay);
809 if (ofproto->netflow) {
810 if (netflow_run(ofproto->netflow)) {
811 send_netflow_active_timeouts(ofproto);
814 if (ofproto->sflow) {
815 dpif_sflow_run(ofproto->sflow);
818 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
821 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
826 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
828 /* Now revalidate if there's anything to do. */
829 if (ofproto->need_revalidate
830 || !tag_set_is_empty(&ofproto->revalidate_set)) {
831 struct tag_set revalidate_set = ofproto->revalidate_set;
832 bool revalidate_all = ofproto->need_revalidate;
833 struct facet *facet, *next;
835 /* Clear the revalidation flags. */
836 tag_set_init(&ofproto->revalidate_set);
837 ofproto->need_revalidate = false;
839 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
841 || tag_set_intersects(&revalidate_set, facet->tags)) {
842 facet_revalidate(facet);
847 /* Check the consistency of a random facet, to aid debugging. */
848 if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) {
851 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
852 struct facet, hmap_node);
853 if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) {
854 if (!facet_check_consistency(facet)) {
855 ofproto->need_revalidate = true;
864 wait(struct ofproto *ofproto_)
866 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
867 struct ofport_dpif *ofport;
868 struct ofbundle *bundle;
870 if (!clogged && !list_is_empty(&ofproto->completions)) {
871 poll_immediate_wake();
874 dpif_wait(ofproto->dpif);
875 dpif_recv_wait(ofproto->dpif);
876 if (ofproto->sflow) {
877 dpif_sflow_wait(ofproto->sflow);
879 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
880 poll_immediate_wake();
882 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
885 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
888 if (ofproto->netflow) {
889 netflow_wait(ofproto->netflow);
891 mac_learning_wait(ofproto->ml);
893 if (ofproto->need_revalidate) {
894 /* Shouldn't happen, but if it does just go around again. */
895 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
896 poll_immediate_wake();
898 timer_wait(&ofproto->next_expiration);
903 flush(struct ofproto *ofproto_)
905 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
906 struct facet *facet, *next_facet;
908 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
909 /* Mark the facet as not installed so that facet_remove() doesn't
910 * bother trying to uninstall it. There is no point in uninstalling it
911 * individually since we are about to blow away all the facets with
912 * dpif_flow_flush(). */
913 struct subfacet *subfacet;
915 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
916 subfacet->installed = false;
917 subfacet->dp_packet_count = 0;
918 subfacet->dp_byte_count = 0;
922 dpif_flow_flush(ofproto->dpif);
926 get_features(struct ofproto *ofproto_ OVS_UNUSED,
927 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
929 *arp_match_ip = true;
930 *actions = (OFPUTIL_A_OUTPUT |
931 OFPUTIL_A_SET_VLAN_VID |
932 OFPUTIL_A_SET_VLAN_PCP |
933 OFPUTIL_A_STRIP_VLAN |
934 OFPUTIL_A_SET_DL_SRC |
935 OFPUTIL_A_SET_DL_DST |
936 OFPUTIL_A_SET_NW_SRC |
937 OFPUTIL_A_SET_NW_DST |
938 OFPUTIL_A_SET_NW_TOS |
939 OFPUTIL_A_SET_TP_SRC |
940 OFPUTIL_A_SET_TP_DST |
945 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
947 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
948 struct dpif_dp_stats s;
950 strcpy(ots->name, "classifier");
952 dpif_get_dp_stats(ofproto->dpif, &s);
953 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
954 put_32aligned_be64(&ots->matched_count,
955 htonll(s.n_hit + ofproto->n_matches));
958 static struct ofport *
961 struct ofport_dpif *port = xmalloc(sizeof *port);
966 port_dealloc(struct ofport *port_)
968 struct ofport_dpif *port = ofport_dpif_cast(port_);
973 port_construct(struct ofport *port_)
975 struct ofport_dpif *port = ofport_dpif_cast(port_);
976 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
978 ofproto->need_revalidate = true;
979 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
982 port->tag = tag_create_random();
983 port->may_enable = true;
984 port->stp_port = NULL;
985 port->stp_state = STP_DISABLED;
986 hmap_init(&port->priorities);
987 port->realdev_ofp_port = 0;
988 port->vlandev_vid = 0;
989 port->carrier_seq = netdev_get_carrier_resets(port->up.netdev);
991 if (ofproto->sflow) {
992 dpif_sflow_add_port(ofproto->sflow, port_);
999 port_destruct(struct ofport *port_)
1001 struct ofport_dpif *port = ofport_dpif_cast(port_);
1002 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1004 ofproto->need_revalidate = true;
1005 bundle_remove(port_);
1006 set_cfm(port_, NULL);
1007 if (ofproto->sflow) {
1008 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1011 ofport_clear_priorities(port);
1012 hmap_destroy(&port->priorities);
1016 port_modified(struct ofport *port_)
1018 struct ofport_dpif *port = ofport_dpif_cast(port_);
1020 if (port->bundle && port->bundle->bond) {
1021 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1026 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1028 struct ofport_dpif *port = ofport_dpif_cast(port_);
1029 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1030 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1032 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1033 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD)) {
1034 ofproto->need_revalidate = true;
1036 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1037 bundle_update(port->bundle);
1043 set_sflow(struct ofproto *ofproto_,
1044 const struct ofproto_sflow_options *sflow_options)
1046 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1047 struct dpif_sflow *ds = ofproto->sflow;
1049 if (sflow_options) {
1051 struct ofport_dpif *ofport;
1053 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1054 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1055 dpif_sflow_add_port(ds, &ofport->up);
1057 ofproto->need_revalidate = true;
1059 dpif_sflow_set_options(ds, sflow_options);
1062 dpif_sflow_destroy(ds);
1063 ofproto->need_revalidate = true;
1064 ofproto->sflow = NULL;
1071 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1073 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1080 struct ofproto_dpif *ofproto;
1082 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1083 ofproto->need_revalidate = true;
1084 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1087 if (cfm_configure(ofport->cfm, s)) {
1093 cfm_destroy(ofport->cfm);
1099 get_cfm_fault(const struct ofport *ofport_)
1101 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1103 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1107 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1110 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1113 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1121 get_cfm_health(const struct ofport *ofport_)
1123 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1125 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1128 /* Spanning Tree. */
1131 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1133 struct ofproto_dpif *ofproto = ofproto_;
1134 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1135 struct ofport_dpif *ofport;
1137 ofport = stp_port_get_aux(sp);
1139 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1140 ofproto->up.name, port_num);
1142 struct eth_header *eth = pkt->l2;
1144 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1145 if (eth_addr_is_zero(eth->eth_src)) {
1146 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1147 "with unknown MAC", ofproto->up.name, port_num);
1149 send_packet(ofport, pkt);
1155 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1157 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1159 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1161 /* Only revalidate flows if the configuration changed. */
1162 if (!s != !ofproto->stp) {
1163 ofproto->need_revalidate = true;
1167 if (!ofproto->stp) {
1168 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1169 send_bpdu_cb, ofproto);
1170 ofproto->stp_last_tick = time_msec();
1173 stp_set_bridge_id(ofproto->stp, s->system_id);
1174 stp_set_bridge_priority(ofproto->stp, s->priority);
1175 stp_set_hello_time(ofproto->stp, s->hello_time);
1176 stp_set_max_age(ofproto->stp, s->max_age);
1177 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1179 struct ofport *ofport;
1181 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1182 set_stp_port(ofport, NULL);
1185 stp_destroy(ofproto->stp);
1186 ofproto->stp = NULL;
1193 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1195 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1199 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1200 s->designated_root = stp_get_designated_root(ofproto->stp);
1201 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1210 update_stp_port_state(struct ofport_dpif *ofport)
1212 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1213 enum stp_state state;
1215 /* Figure out new state. */
1216 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1220 if (ofport->stp_state != state) {
1221 enum ofputil_port_state of_state;
1224 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1225 netdev_get_name(ofport->up.netdev),
1226 stp_state_name(ofport->stp_state),
1227 stp_state_name(state));
1228 if (stp_learn_in_state(ofport->stp_state)
1229 != stp_learn_in_state(state)) {
1230 /* xxx Learning action flows should also be flushed. */
1231 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1233 fwd_change = stp_forward_in_state(ofport->stp_state)
1234 != stp_forward_in_state(state);
1236 ofproto->need_revalidate = true;
1237 ofport->stp_state = state;
1238 ofport->stp_state_entered = time_msec();
1240 if (fwd_change && ofport->bundle) {
1241 bundle_update(ofport->bundle);
1244 /* Update the STP state bits in the OpenFlow port description. */
1245 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1246 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1247 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1248 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1249 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1251 ofproto_port_set_state(&ofport->up, of_state);
1255 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1256 * caller is responsible for assigning STP port numbers and ensuring
1257 * there are no duplicates. */
1259 set_stp_port(struct ofport *ofport_,
1260 const struct ofproto_port_stp_settings *s)
1262 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1263 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1264 struct stp_port *sp = ofport->stp_port;
1266 if (!s || !s->enable) {
1268 ofport->stp_port = NULL;
1269 stp_port_disable(sp);
1270 update_stp_port_state(ofport);
1273 } else if (sp && stp_port_no(sp) != s->port_num
1274 && ofport == stp_port_get_aux(sp)) {
1275 /* The port-id changed, so disable the old one if it's not
1276 * already in use by another port. */
1277 stp_port_disable(sp);
1280 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1281 stp_port_enable(sp);
1283 stp_port_set_aux(sp, ofport);
1284 stp_port_set_priority(sp, s->priority);
1285 stp_port_set_path_cost(sp, s->path_cost);
1287 update_stp_port_state(ofport);
1293 get_stp_port_status(struct ofport *ofport_,
1294 struct ofproto_port_stp_status *s)
1296 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1297 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1298 struct stp_port *sp = ofport->stp_port;
1300 if (!ofproto->stp || !sp) {
1306 s->port_id = stp_port_get_id(sp);
1307 s->state = stp_port_get_state(sp);
1308 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1309 s->role = stp_port_get_role(sp);
1310 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1316 stp_run(struct ofproto_dpif *ofproto)
1319 long long int now = time_msec();
1320 long long int elapsed = now - ofproto->stp_last_tick;
1321 struct stp_port *sp;
1324 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1325 ofproto->stp_last_tick = now;
1327 while (stp_get_changed_port(ofproto->stp, &sp)) {
1328 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1331 update_stp_port_state(ofport);
1335 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1336 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1342 stp_wait(struct ofproto_dpif *ofproto)
1345 poll_timer_wait(1000);
1349 /* Returns true if STP should process 'flow'. */
1351 stp_should_process_flow(const struct flow *flow)
1353 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1357 stp_process_packet(const struct ofport_dpif *ofport,
1358 const struct ofpbuf *packet)
1360 struct ofpbuf payload = *packet;
1361 struct eth_header *eth = payload.data;
1362 struct stp_port *sp = ofport->stp_port;
1364 /* Sink packets on ports that have STP disabled when the bridge has
1366 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1370 /* Trim off padding on payload. */
1371 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1372 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1375 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1376 stp_received_bpdu(sp, payload.data, payload.size);
1380 static struct priority_to_dscp *
1381 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1383 struct priority_to_dscp *pdscp;
1386 hash = hash_int(priority, 0);
1387 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1388 if (pdscp->priority == priority) {
1396 ofport_clear_priorities(struct ofport_dpif *ofport)
1398 struct priority_to_dscp *pdscp, *next;
1400 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1401 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1407 set_queues(struct ofport *ofport_,
1408 const struct ofproto_port_queue *qdscp_list,
1411 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1412 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1413 struct hmap new = HMAP_INITIALIZER(&new);
1416 for (i = 0; i < n_qdscp; i++) {
1417 struct priority_to_dscp *pdscp;
1421 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1422 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1427 pdscp = get_priority(ofport, priority);
1429 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1431 pdscp = xmalloc(sizeof *pdscp);
1432 pdscp->priority = priority;
1434 ofproto->need_revalidate = true;
1437 if (pdscp->dscp != dscp) {
1439 ofproto->need_revalidate = true;
1442 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1445 if (!hmap_is_empty(&ofport->priorities)) {
1446 ofport_clear_priorities(ofport);
1447 ofproto->need_revalidate = true;
1450 hmap_swap(&new, &ofport->priorities);
1458 /* Expires all MAC learning entries associated with 'bundle' and forces its
1459 * ofproto to revalidate every flow.
1461 * Normally MAC learning entries are removed only from the ofproto associated
1462 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1463 * are removed from every ofproto. When patch ports and SLB bonds are in use
1464 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1465 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1466 * with the host from which it migrated. */
1468 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1470 struct ofproto_dpif *ofproto = bundle->ofproto;
1471 struct mac_learning *ml = ofproto->ml;
1472 struct mac_entry *mac, *next_mac;
1474 ofproto->need_revalidate = true;
1475 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1476 if (mac->port.p == bundle) {
1478 struct ofproto_dpif *o;
1480 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1482 struct mac_entry *e;
1484 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1487 tag_set_add(&o->revalidate_set, e->tag);
1488 mac_learning_expire(o->ml, e);
1494 mac_learning_expire(ml, mac);
1499 static struct ofbundle *
1500 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1502 struct ofbundle *bundle;
1504 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1505 &ofproto->bundles) {
1506 if (bundle->aux == aux) {
1513 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1514 * ones that are found to 'bundles'. */
1516 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1517 void **auxes, size_t n_auxes,
1518 struct hmapx *bundles)
1522 hmapx_init(bundles);
1523 for (i = 0; i < n_auxes; i++) {
1524 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1526 hmapx_add(bundles, bundle);
1532 bundle_update(struct ofbundle *bundle)
1534 struct ofport_dpif *port;
1536 bundle->floodable = true;
1537 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1538 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1539 || !stp_forward_in_state(port->stp_state)) {
1540 bundle->floodable = false;
1547 bundle_del_port(struct ofport_dpif *port)
1549 struct ofbundle *bundle = port->bundle;
1551 bundle->ofproto->need_revalidate = true;
1553 list_remove(&port->bundle_node);
1554 port->bundle = NULL;
1557 lacp_slave_unregister(bundle->lacp, port);
1560 bond_slave_unregister(bundle->bond, port);
1563 bundle_update(bundle);
1567 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1568 struct lacp_slave_settings *lacp,
1569 uint32_t bond_stable_id)
1571 struct ofport_dpif *port;
1573 port = get_ofp_port(bundle->ofproto, ofp_port);
1578 if (port->bundle != bundle) {
1579 bundle->ofproto->need_revalidate = true;
1581 bundle_del_port(port);
1584 port->bundle = bundle;
1585 list_push_back(&bundle->ports, &port->bundle_node);
1586 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1587 || !stp_forward_in_state(port->stp_state)) {
1588 bundle->floodable = false;
1592 port->bundle->ofproto->need_revalidate = true;
1593 lacp_slave_register(bundle->lacp, port, lacp);
1596 port->bond_stable_id = bond_stable_id;
1602 bundle_destroy(struct ofbundle *bundle)
1604 struct ofproto_dpif *ofproto;
1605 struct ofport_dpif *port, *next_port;
1612 ofproto = bundle->ofproto;
1613 for (i = 0; i < MAX_MIRRORS; i++) {
1614 struct ofmirror *m = ofproto->mirrors[i];
1616 if (m->out == bundle) {
1618 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1619 || hmapx_find_and_delete(&m->dsts, bundle)) {
1620 ofproto->need_revalidate = true;
1625 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1626 bundle_del_port(port);
1629 bundle_flush_macs(bundle, true);
1630 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1632 free(bundle->trunks);
1633 lacp_destroy(bundle->lacp);
1634 bond_destroy(bundle->bond);
1639 bundle_set(struct ofproto *ofproto_, void *aux,
1640 const struct ofproto_bundle_settings *s)
1642 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1643 bool need_flush = false;
1644 struct ofport_dpif *port;
1645 struct ofbundle *bundle;
1646 unsigned long *trunks;
1652 bundle_destroy(bundle_lookup(ofproto, aux));
1656 assert(s->n_slaves == 1 || s->bond != NULL);
1657 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1659 bundle = bundle_lookup(ofproto, aux);
1661 bundle = xmalloc(sizeof *bundle);
1663 bundle->ofproto = ofproto;
1664 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1665 hash_pointer(aux, 0));
1667 bundle->name = NULL;
1669 list_init(&bundle->ports);
1670 bundle->vlan_mode = PORT_VLAN_TRUNK;
1672 bundle->trunks = NULL;
1673 bundle->use_priority_tags = s->use_priority_tags;
1674 bundle->lacp = NULL;
1675 bundle->bond = NULL;
1677 bundle->floodable = true;
1679 bundle->src_mirrors = 0;
1680 bundle->dst_mirrors = 0;
1681 bundle->mirror_out = 0;
1684 if (!bundle->name || strcmp(s->name, bundle->name)) {
1686 bundle->name = xstrdup(s->name);
1691 if (!bundle->lacp) {
1692 ofproto->need_revalidate = true;
1693 bundle->lacp = lacp_create();
1695 lacp_configure(bundle->lacp, s->lacp);
1697 lacp_destroy(bundle->lacp);
1698 bundle->lacp = NULL;
1701 /* Update set of ports. */
1703 for (i = 0; i < s->n_slaves; i++) {
1704 if (!bundle_add_port(bundle, s->slaves[i],
1705 s->lacp ? &s->lacp_slaves[i] : NULL,
1706 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1710 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1711 struct ofport_dpif *next_port;
1713 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1714 for (i = 0; i < s->n_slaves; i++) {
1715 if (s->slaves[i] == port->up.ofp_port) {
1720 bundle_del_port(port);
1724 assert(list_size(&bundle->ports) <= s->n_slaves);
1726 if (list_is_empty(&bundle->ports)) {
1727 bundle_destroy(bundle);
1731 /* Set VLAN tagging mode */
1732 if (s->vlan_mode != bundle->vlan_mode
1733 || s->use_priority_tags != bundle->use_priority_tags) {
1734 bundle->vlan_mode = s->vlan_mode;
1735 bundle->use_priority_tags = s->use_priority_tags;
1740 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1741 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1743 if (vlan != bundle->vlan) {
1744 bundle->vlan = vlan;
1748 /* Get trunked VLANs. */
1749 switch (s->vlan_mode) {
1750 case PORT_VLAN_ACCESS:
1754 case PORT_VLAN_TRUNK:
1755 trunks = (unsigned long *) s->trunks;
1758 case PORT_VLAN_NATIVE_UNTAGGED:
1759 case PORT_VLAN_NATIVE_TAGGED:
1760 if (vlan != 0 && (!s->trunks
1761 || !bitmap_is_set(s->trunks, vlan)
1762 || bitmap_is_set(s->trunks, 0))) {
1763 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1765 trunks = bitmap_clone(s->trunks, 4096);
1767 trunks = bitmap_allocate1(4096);
1769 bitmap_set1(trunks, vlan);
1770 bitmap_set0(trunks, 0);
1772 trunks = (unsigned long *) s->trunks;
1779 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1780 free(bundle->trunks);
1781 if (trunks == s->trunks) {
1782 bundle->trunks = vlan_bitmap_clone(trunks);
1784 bundle->trunks = trunks;
1789 if (trunks != s->trunks) {
1794 if (!list_is_short(&bundle->ports)) {
1795 bundle->ofproto->has_bonded_bundles = true;
1797 if (bond_reconfigure(bundle->bond, s->bond)) {
1798 ofproto->need_revalidate = true;
1801 bundle->bond = bond_create(s->bond);
1802 ofproto->need_revalidate = true;
1805 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1806 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1810 bond_destroy(bundle->bond);
1811 bundle->bond = NULL;
1814 /* If we changed something that would affect MAC learning, un-learn
1815 * everything on this port and force flow revalidation. */
1817 bundle_flush_macs(bundle, false);
1824 bundle_remove(struct ofport *port_)
1826 struct ofport_dpif *port = ofport_dpif_cast(port_);
1827 struct ofbundle *bundle = port->bundle;
1830 bundle_del_port(port);
1831 if (list_is_empty(&bundle->ports)) {
1832 bundle_destroy(bundle);
1833 } else if (list_is_short(&bundle->ports)) {
1834 bond_destroy(bundle->bond);
1835 bundle->bond = NULL;
1841 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
1843 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1844 struct ofport_dpif *port = port_;
1845 uint8_t ea[ETH_ADDR_LEN];
1848 error = netdev_get_etheraddr(port->up.netdev, ea);
1850 struct ofpbuf packet;
1853 ofpbuf_init(&packet, 0);
1854 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1856 memcpy(packet_pdu, pdu, pdu_size);
1858 send_packet(port, &packet);
1859 ofpbuf_uninit(&packet);
1861 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1862 "%s (%s)", port->bundle->name,
1863 netdev_get_name(port->up.netdev), strerror(error));
1868 bundle_send_learning_packets(struct ofbundle *bundle)
1870 struct ofproto_dpif *ofproto = bundle->ofproto;
1871 int error, n_packets, n_errors;
1872 struct mac_entry *e;
1874 error = n_packets = n_errors = 0;
1875 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1876 if (e->port.p != bundle) {
1877 struct ofpbuf *learning_packet;
1878 struct ofport_dpif *port;
1882 /* The assignment to "port" is unnecessary but makes "grep"ing for
1883 * struct ofport_dpif more effective. */
1884 learning_packet = bond_compose_learning_packet(bundle->bond,
1888 ret = send_packet(port, learning_packet);
1889 ofpbuf_delete(learning_packet);
1899 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1900 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1901 "packets, last error was: %s",
1902 bundle->name, n_errors, n_packets, strerror(error));
1904 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1905 bundle->name, n_packets);
1910 bundle_run(struct ofbundle *bundle)
1913 lacp_run(bundle->lacp, send_pdu_cb);
1916 struct ofport_dpif *port;
1918 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1919 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1922 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1923 lacp_status(bundle->lacp));
1924 if (bond_should_send_learning_packets(bundle->bond)) {
1925 bundle_send_learning_packets(bundle);
1931 bundle_wait(struct ofbundle *bundle)
1934 lacp_wait(bundle->lacp);
1937 bond_wait(bundle->bond);
1944 mirror_scan(struct ofproto_dpif *ofproto)
1948 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1949 if (!ofproto->mirrors[idx]) {
1956 static struct ofmirror *
1957 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1961 for (i = 0; i < MAX_MIRRORS; i++) {
1962 struct ofmirror *mirror = ofproto->mirrors[i];
1963 if (mirror && mirror->aux == aux) {
1971 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
1973 mirror_update_dups(struct ofproto_dpif *ofproto)
1977 for (i = 0; i < MAX_MIRRORS; i++) {
1978 struct ofmirror *m = ofproto->mirrors[i];
1981 m->dup_mirrors = MIRROR_MASK_C(1) << i;
1985 for (i = 0; i < MAX_MIRRORS; i++) {
1986 struct ofmirror *m1 = ofproto->mirrors[i];
1993 for (j = i + 1; j < MAX_MIRRORS; j++) {
1994 struct ofmirror *m2 = ofproto->mirrors[j];
1996 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
1997 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
1998 m2->dup_mirrors |= m1->dup_mirrors;
2005 mirror_set(struct ofproto *ofproto_, void *aux,
2006 const struct ofproto_mirror_settings *s)
2008 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2009 mirror_mask_t mirror_bit;
2010 struct ofbundle *bundle;
2011 struct ofmirror *mirror;
2012 struct ofbundle *out;
2013 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2014 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2017 mirror = mirror_lookup(ofproto, aux);
2019 mirror_destroy(mirror);
2025 idx = mirror_scan(ofproto);
2027 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2029 ofproto->up.name, MAX_MIRRORS, s->name);
2033 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2034 mirror->ofproto = ofproto;
2037 mirror->out_vlan = -1;
2038 mirror->name = NULL;
2041 if (!mirror->name || strcmp(s->name, mirror->name)) {
2043 mirror->name = xstrdup(s->name);
2046 /* Get the new configuration. */
2047 if (s->out_bundle) {
2048 out = bundle_lookup(ofproto, s->out_bundle);
2050 mirror_destroy(mirror);
2056 out_vlan = s->out_vlan;
2058 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2059 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2061 /* If the configuration has not changed, do nothing. */
2062 if (hmapx_equals(&srcs, &mirror->srcs)
2063 && hmapx_equals(&dsts, &mirror->dsts)
2064 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2065 && mirror->out == out
2066 && mirror->out_vlan == out_vlan)
2068 hmapx_destroy(&srcs);
2069 hmapx_destroy(&dsts);
2073 hmapx_swap(&srcs, &mirror->srcs);
2074 hmapx_destroy(&srcs);
2076 hmapx_swap(&dsts, &mirror->dsts);
2077 hmapx_destroy(&dsts);
2079 free(mirror->vlans);
2080 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2083 mirror->out_vlan = out_vlan;
2085 /* Update bundles. */
2086 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2087 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2088 if (hmapx_contains(&mirror->srcs, bundle)) {
2089 bundle->src_mirrors |= mirror_bit;
2091 bundle->src_mirrors &= ~mirror_bit;
2094 if (hmapx_contains(&mirror->dsts, bundle)) {
2095 bundle->dst_mirrors |= mirror_bit;
2097 bundle->dst_mirrors &= ~mirror_bit;
2100 if (mirror->out == bundle) {
2101 bundle->mirror_out |= mirror_bit;
2103 bundle->mirror_out &= ~mirror_bit;
2107 ofproto->need_revalidate = true;
2108 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2109 mirror_update_dups(ofproto);
2115 mirror_destroy(struct ofmirror *mirror)
2117 struct ofproto_dpif *ofproto;
2118 mirror_mask_t mirror_bit;
2119 struct ofbundle *bundle;
2125 ofproto = mirror->ofproto;
2126 ofproto->need_revalidate = true;
2127 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2129 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2130 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2131 bundle->src_mirrors &= ~mirror_bit;
2132 bundle->dst_mirrors &= ~mirror_bit;
2133 bundle->mirror_out &= ~mirror_bit;
2136 hmapx_destroy(&mirror->srcs);
2137 hmapx_destroy(&mirror->dsts);
2138 free(mirror->vlans);
2140 ofproto->mirrors[mirror->idx] = NULL;
2144 mirror_update_dups(ofproto);
2148 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2149 uint64_t *packets, uint64_t *bytes)
2151 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2152 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2155 *packets = *bytes = UINT64_MAX;
2159 *packets = mirror->packet_count;
2160 *bytes = mirror->byte_count;
2166 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2168 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2169 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2170 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2176 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2178 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2179 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2180 return bundle && bundle->mirror_out != 0;
2184 forward_bpdu_changed(struct ofproto *ofproto_)
2186 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2187 /* Revalidate cached flows whenever forward_bpdu option changes. */
2188 ofproto->need_revalidate = true;
2192 set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time)
2194 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2195 mac_learning_set_idle_time(ofproto->ml, idle_time);
2200 static struct ofport_dpif *
2201 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
2203 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2204 return ofport ? ofport_dpif_cast(ofport) : NULL;
2207 static struct ofport_dpif *
2208 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
2210 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2214 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2215 struct dpif_port *dpif_port)
2217 ofproto_port->name = dpif_port->name;
2218 ofproto_port->type = dpif_port->type;
2219 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2223 port_run(struct ofport_dpif *ofport)
2225 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2226 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2227 bool enable = netdev_get_carrier(ofport->up.netdev);
2229 ofport->carrier_seq = carrier_seq;
2232 cfm_run(ofport->cfm);
2234 if (cfm_should_send_ccm(ofport->cfm)) {
2235 struct ofpbuf packet;
2237 ofpbuf_init(&packet, 0);
2238 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2239 send_packet(ofport, &packet);
2240 ofpbuf_uninit(&packet);
2243 enable = enable && !cfm_get_fault(ofport->cfm)
2244 && cfm_get_opup(ofport->cfm);
2247 if (ofport->bundle) {
2248 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2249 if (carrier_changed) {
2250 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2254 if (ofport->may_enable != enable) {
2255 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2257 if (ofproto->has_bundle_action) {
2258 ofproto->need_revalidate = true;
2262 ofport->may_enable = enable;
2266 port_wait(struct ofport_dpif *ofport)
2269 cfm_wait(ofport->cfm);
2274 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2275 struct ofproto_port *ofproto_port)
2277 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2278 struct dpif_port dpif_port;
2281 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2283 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2289 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2291 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2295 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2297 *ofp_portp = odp_port_to_ofp_port(odp_port);
2303 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2305 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2308 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2310 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2312 /* The caller is going to close ofport->up.netdev. If this is a
2313 * bonded port, then the bond is using that netdev, so remove it
2314 * from the bond. The client will need to reconfigure everything
2315 * after deleting ports, so then the slave will get re-added. */
2316 bundle_remove(&ofport->up);
2323 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2325 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2328 error = netdev_get_stats(ofport->up.netdev, stats);
2330 if (!error && ofport->odp_port == OVSP_LOCAL) {
2331 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2333 /* ofproto->stats.tx_packets represents packets that we created
2334 * internally and sent to some port (e.g. packets sent with
2335 * send_packet()). Account for them as if they had come from
2336 * OFPP_LOCAL and got forwarded. */
2338 if (stats->rx_packets != UINT64_MAX) {
2339 stats->rx_packets += ofproto->stats.tx_packets;
2342 if (stats->rx_bytes != UINT64_MAX) {
2343 stats->rx_bytes += ofproto->stats.tx_bytes;
2346 /* ofproto->stats.rx_packets represents packets that were received on
2347 * some port and we processed internally and dropped (e.g. STP).
2348 * Account fro them as if they had been forwarded to OFPP_LOCAL. */
2350 if (stats->tx_packets != UINT64_MAX) {
2351 stats->tx_packets += ofproto->stats.rx_packets;
2354 if (stats->tx_bytes != UINT64_MAX) {
2355 stats->tx_bytes += ofproto->stats.rx_bytes;
2362 /* Account packets for LOCAL port. */
2364 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2365 size_t tx_size, size_t rx_size)
2367 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2370 ofproto->stats.rx_packets++;
2371 ofproto->stats.rx_bytes += rx_size;
2374 ofproto->stats.tx_packets++;
2375 ofproto->stats.tx_bytes += tx_size;
2379 struct port_dump_state {
2380 struct dpif_port_dump dump;
2385 port_dump_start(const struct ofproto *ofproto_, void **statep)
2387 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2388 struct port_dump_state *state;
2390 *statep = state = xmalloc(sizeof *state);
2391 dpif_port_dump_start(&state->dump, ofproto->dpif);
2392 state->done = false;
2397 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2398 struct ofproto_port *port)
2400 struct port_dump_state *state = state_;
2401 struct dpif_port dpif_port;
2403 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2404 ofproto_port_from_dpif_port(port, &dpif_port);
2407 int error = dpif_port_dump_done(&state->dump);
2409 return error ? error : EOF;
2414 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2416 struct port_dump_state *state = state_;
2419 dpif_port_dump_done(&state->dump);
2426 port_poll(const struct ofproto *ofproto_, char **devnamep)
2428 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2429 return dpif_port_poll(ofproto->dpif, devnamep);
2433 port_poll_wait(const struct ofproto *ofproto_)
2435 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2436 dpif_port_poll_wait(ofproto->dpif);
2440 port_is_lacp_current(const struct ofport *ofport_)
2442 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2443 return (ofport->bundle && ofport->bundle->lacp
2444 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2448 /* Upcall handling. */
2450 /* Flow miss batching.
2452 * Some dpifs implement operations faster when you hand them off in a batch.
2453 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2454 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2455 * more packets, plus possibly installing the flow in the dpif.
2457 * So far we only batch the operations that affect flow setup time the most.
2458 * It's possible to batch more than that, but the benefit might be minimal. */
2460 struct hmap_node hmap_node;
2462 enum odp_key_fitness key_fitness;
2463 const struct nlattr *key;
2465 ovs_be16 initial_tci;
2466 struct list packets;
2469 struct flow_miss_op {
2470 struct dpif_op dpif_op;
2471 struct subfacet *subfacet;
2474 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2475 * OpenFlow controller as necessary according to their individual
2476 * configurations. */
2478 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
2479 const struct flow *flow)
2481 struct ofputil_packet_in pin;
2483 pin.packet = packet->data;
2484 pin.packet_len = packet->size;
2485 pin.reason = OFPR_NO_MATCH;
2486 pin.controller_id = 0;
2491 pin.send_len = 0; /* not used for flow table misses */
2493 flow_get_metadata(flow, &pin.fmd);
2495 /* Registers aren't meaningful on a miss. */
2496 memset(pin.fmd.reg_masks, 0, sizeof pin.fmd.reg_masks);
2498 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
2502 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2503 const struct ofpbuf *packet)
2505 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2511 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2513 cfm_process_heartbeat(ofport->cfm, packet);
2516 } else if (ofport->bundle && ofport->bundle->lacp
2517 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2519 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2522 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2524 stp_process_packet(ofport, packet);
2531 static struct flow_miss *
2532 flow_miss_create(struct hmap *todo, const struct flow *flow,
2533 enum odp_key_fitness key_fitness,
2534 const struct nlattr *key, size_t key_len,
2535 ovs_be16 initial_tci)
2537 uint32_t hash = flow_hash(flow, 0);
2538 struct flow_miss *miss;
2540 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2541 if (flow_equal(&miss->flow, flow)) {
2546 miss = xmalloc(sizeof *miss);
2547 hmap_insert(todo, &miss->hmap_node, hash);
2549 miss->key_fitness = key_fitness;
2551 miss->key_len = key_len;
2552 miss->initial_tci = initial_tci;
2553 list_init(&miss->packets);
2558 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2559 struct flow_miss_op *ops, size_t *n_ops)
2561 const struct flow *flow = &miss->flow;
2562 struct subfacet *subfacet;
2563 struct ofpbuf *packet;
2564 struct facet *facet;
2566 facet = facet_lookup_valid(ofproto, flow);
2568 struct rule_dpif *rule;
2570 rule = rule_dpif_lookup(ofproto, flow, 0);
2572 /* Don't send a packet-in if OFPUTIL_PC_NO_PACKET_IN asserted. */
2573 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
2575 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
2576 COVERAGE_INC(ofproto_dpif_no_packet_in);
2577 /* XXX install 'drop' flow entry */
2581 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
2585 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2586 send_packet_in_miss(ofproto, packet, flow);
2592 facet = facet_create(rule, flow);
2595 subfacet = subfacet_create(facet,
2596 miss->key_fitness, miss->key, miss->key_len,
2599 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2600 struct dpif_flow_stats stats;
2601 struct flow_miss_op *op;
2602 struct dpif_execute *execute;
2604 ofproto->n_matches++;
2606 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2608 * Extra-special case for fail-open mode.
2610 * We are in fail-open mode and the packet matched the fail-open
2611 * rule, but we are connected to a controller too. We should send
2612 * the packet up to the controller in the hope that it will try to
2613 * set up a flow and thereby allow us to exit fail-open.
2615 * See the top-level comment in fail-open.c for more information.
2617 send_packet_in_miss(ofproto, packet, flow);
2620 if (!facet->may_install || !subfacet->actions) {
2621 subfacet_make_actions(subfacet, packet);
2624 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2625 subfacet_update_stats(subfacet, &stats);
2627 if (!subfacet->actions_len) {
2628 /* No actions to execute, so skip talking to the dpif. */
2632 if (flow->vlan_tci != subfacet->initial_tci) {
2633 /* This packet was received on a VLAN splinter port. We added
2634 * a VLAN to the packet to make the packet resemble the flow,
2635 * but the actions were composed assuming that the packet
2636 * contained no VLAN. So, we must remove the VLAN header from
2637 * the packet before trying to execute the actions. */
2638 eth_pop_vlan(packet);
2641 op = &ops[(*n_ops)++];
2642 execute = &op->dpif_op.u.execute;
2643 op->subfacet = subfacet;
2644 op->dpif_op.type = DPIF_OP_EXECUTE;
2645 execute->key = miss->key;
2646 execute->key_len = miss->key_len;
2647 execute->actions = (facet->may_install
2649 : xmemdup(subfacet->actions,
2650 subfacet->actions_len));
2651 execute->actions_len = subfacet->actions_len;
2652 execute->packet = packet;
2655 if (facet->may_install && subfacet->key_fitness != ODP_FIT_TOO_LITTLE) {
2656 struct flow_miss_op *op = &ops[(*n_ops)++];
2657 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
2659 op->subfacet = subfacet;
2660 op->dpif_op.type = DPIF_OP_FLOW_PUT;
2661 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2662 put->key = miss->key;
2663 put->key_len = miss->key_len;
2664 put->actions = subfacet->actions;
2665 put->actions_len = subfacet->actions_len;
2670 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
2671 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
2672 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
2673 * what a flow key should contain.
2675 * This function also includes some logic to help make VLAN splinters
2676 * transparent to the rest of the upcall processing logic. In particular, if
2677 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
2678 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
2679 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
2681 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
2682 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
2683 * (This differs from the value returned in flow->vlan_tci only for packets
2684 * received on VLAN splinters.)
2686 static enum odp_key_fitness
2687 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
2688 const struct nlattr *key, size_t key_len,
2689 struct flow *flow, ovs_be16 *initial_tci,
2690 struct ofpbuf *packet)
2692 enum odp_key_fitness fitness;
2696 fitness = odp_flow_key_to_flow(key, key_len, flow);
2697 if (fitness == ODP_FIT_ERROR) {
2700 *initial_tci = flow->vlan_tci;
2702 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
2704 /* Cause the flow to be processed as if it came in on the real device
2705 * with the VLAN device's VLAN ID. */
2706 flow->in_port = realdev;
2707 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
2709 /* Make the packet resemble the flow, so that it gets sent to an
2710 * OpenFlow controller properly, so that it looks correct for
2711 * sFlow, and so that flow_extract() will get the correct vlan_tci
2712 * if it is called on 'packet'.
2714 * The allocated space inside 'packet' probably also contains
2715 * 'key', that is, both 'packet' and 'key' are probably part of a
2716 * struct dpif_upcall (see the large comment on that structure
2717 * definition), so pushing data on 'packet' is in general not a
2718 * good idea since it could overwrite 'key' or free it as a side
2719 * effect. However, it's OK in this special case because we know
2720 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
2721 * will just overwrite the 4-byte "struct nlattr", which is fine
2722 * since we don't need that header anymore. */
2723 eth_push_vlan(packet, flow->vlan_tci);
2726 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
2727 if (fitness == ODP_FIT_PERFECT) {
2728 fitness = ODP_FIT_TOO_MUCH;
2736 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
2739 struct dpif_upcall *upcall;
2740 struct flow_miss *miss, *next_miss;
2741 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
2742 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
2751 /* Construct the to-do list.
2753 * This just amounts to extracting the flow from each packet and sticking
2754 * the packets that have the same flow in the same "flow_miss" structure so
2755 * that we can process them together. */
2757 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
2758 enum odp_key_fitness fitness;
2759 struct flow_miss *miss;
2760 ovs_be16 initial_tci;
2763 /* Obtain metadata and check userspace/kernel agreement on flow match,
2764 * then set 'flow''s header pointers. */
2765 fitness = ofproto_dpif_extract_flow_key(ofproto,
2766 upcall->key, upcall->key_len,
2767 &flow, &initial_tci,
2769 if (fitness == ODP_FIT_ERROR) {
2770 ofpbuf_delete(upcall->packet);
2773 flow_extract(upcall->packet, flow.skb_priority, flow.tun_id,
2774 flow.in_port, &flow);
2776 /* Handle 802.1ag, LACP, and STP specially. */
2777 if (process_special(ofproto, &flow, upcall->packet)) {
2778 ofproto_update_local_port_stats(&ofproto->up,
2779 0, upcall->packet->size);
2780 ofpbuf_delete(upcall->packet);
2781 ofproto->n_matches++;
2785 /* Add other packets to a to-do list. */
2786 miss = flow_miss_create(&todo, &flow, fitness,
2787 upcall->key, upcall->key_len, initial_tci);
2788 list_push_back(&miss->packets, &upcall->packet->list_node);
2791 /* Process each element in the to-do list, constructing the set of
2792 * operations to batch. */
2794 HMAP_FOR_EACH (miss, hmap_node, &todo) {
2795 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
2797 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
2799 /* Execute batch. */
2800 for (i = 0; i < n_ops; i++) {
2801 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
2803 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
2805 /* Free memory and update facets. */
2806 for (i = 0; i < n_ops; i++) {
2807 struct flow_miss_op *op = &flow_miss_ops[i];
2808 struct dpif_execute *execute;
2810 switch (op->dpif_op.type) {
2811 case DPIF_OP_EXECUTE:
2812 execute = &op->dpif_op.u.execute;
2813 if (op->subfacet->actions != execute->actions) {
2814 free((struct nlattr *) execute->actions);
2818 case DPIF_OP_FLOW_PUT:
2819 if (!op->dpif_op.error) {
2820 op->subfacet->installed = true;
2824 case DPIF_OP_FLOW_DEL:
2828 HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) {
2829 ofpbuf_list_delete(&miss->packets);
2830 hmap_remove(&todo, &miss->hmap_node);
2833 hmap_destroy(&todo);
2837 handle_userspace_upcall(struct ofproto_dpif *ofproto,
2838 struct dpif_upcall *upcall)
2840 struct user_action_cookie cookie;
2841 enum odp_key_fitness fitness;
2842 ovs_be16 initial_tci;
2845 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
2847 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
2848 upcall->key_len, &flow,
2849 &initial_tci, upcall->packet);
2850 if (fitness == ODP_FIT_ERROR) {
2851 ofpbuf_delete(upcall->packet);
2855 if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
2856 if (ofproto->sflow) {
2857 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
2861 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
2863 ofpbuf_delete(upcall->packet);
2867 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
2869 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
2873 assert (max_batch <= FLOW_MISS_MAX_BATCH);
2876 for (i = 0; i < max_batch; i++) {
2877 struct dpif_upcall *upcall = &misses[n_misses];
2880 error = dpif_recv(ofproto->dpif, upcall);
2885 switch (upcall->type) {
2886 case DPIF_UC_ACTION:
2887 handle_userspace_upcall(ofproto, upcall);
2891 /* Handle it later. */
2895 case DPIF_N_UC_TYPES:
2897 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32,
2903 handle_miss_upcalls(ofproto, misses, n_misses);
2908 /* Flow expiration. */
2910 static int subfacet_max_idle(const struct ofproto_dpif *);
2911 static void update_stats(struct ofproto_dpif *);
2912 static void rule_expire(struct rule_dpif *);
2913 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
2915 /* This function is called periodically by run(). Its job is to collect
2916 * updates for the flows that have been installed into the datapath, most
2917 * importantly when they last were used, and then use that information to
2918 * expire flows that have not been used recently.
2920 * Returns the number of milliseconds after which it should be called again. */
2922 expire(struct ofproto_dpif *ofproto)
2924 struct rule_dpif *rule, *next_rule;
2925 struct oftable *table;
2928 /* Update stats for each flow in the datapath. */
2929 update_stats(ofproto);
2931 /* Expire subfacets that have been idle too long. */
2932 dp_max_idle = subfacet_max_idle(ofproto);
2933 expire_subfacets(ofproto, dp_max_idle);
2935 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
2936 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
2937 struct cls_cursor cursor;
2939 cls_cursor_init(&cursor, &table->cls, NULL);
2940 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
2945 /* All outstanding data in existing flows has been accounted, so it's a
2946 * good time to do bond rebalancing. */
2947 if (ofproto->has_bonded_bundles) {
2948 struct ofbundle *bundle;
2950 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2952 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
2957 return MIN(dp_max_idle, 1000);
2960 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
2962 * This function also pushes statistics updates to rules which each facet
2963 * resubmits into. Generally these statistics will be accurate. However, if a
2964 * facet changes the rule it resubmits into at some time in between
2965 * update_stats() runs, it is possible that statistics accrued to the
2966 * old rule will be incorrectly attributed to the new rule. This could be
2967 * avoided by calling update_stats() whenever rules are created or
2968 * deleted. However, the performance impact of making so many calls to the
2969 * datapath do not justify the benefit of having perfectly accurate statistics.
2972 update_stats(struct ofproto_dpif *p)
2974 const struct dpif_flow_stats *stats;
2975 struct dpif_flow_dump dump;
2976 const struct nlattr *key;
2979 dpif_flow_dump_start(&dump, p->dpif);
2980 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
2981 struct subfacet *subfacet;
2983 subfacet = subfacet_find(p, key, key_len);
2984 if (subfacet && subfacet->installed) {
2985 struct facet *facet = subfacet->facet;
2987 if (stats->n_packets >= subfacet->dp_packet_count) {
2988 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
2989 facet->packet_count += extra;
2991 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
2994 if (stats->n_bytes >= subfacet->dp_byte_count) {
2995 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
2997 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3000 subfacet->dp_packet_count = stats->n_packets;
3001 subfacet->dp_byte_count = stats->n_bytes;
3003 facet->tcp_flags |= stats->tcp_flags;
3005 subfacet_update_time(subfacet, stats->used);
3006 if (facet->accounted_bytes < facet->byte_count) {
3008 facet_account(facet);
3009 facet->accounted_bytes = facet->byte_count;
3011 facet_push_stats(facet);
3013 if (!VLOG_DROP_WARN(&rl)) {
3017 odp_flow_key_format(key, key_len, &s);
3018 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
3022 COVERAGE_INC(facet_unexpected);
3023 /* There's a flow in the datapath that we know nothing about, or a
3024 * flow that shouldn't be installed but was anyway. Delete it. */
3025 dpif_flow_del(p->dpif, key, key_len, NULL);
3028 dpif_flow_dump_done(&dump);
3031 /* Calculates and returns the number of milliseconds of idle time after which
3032 * subfacets should expire from the datapath. When a subfacet expires, we fold
3033 * its statistics into its facet, and when a facet's last subfacet expires, we
3034 * fold its statistic into its rule. */
3036 subfacet_max_idle(const struct ofproto_dpif *ofproto)
3039 * Idle time histogram.
3041 * Most of the time a switch has a relatively small number of subfacets.
3042 * When this is the case we might as well keep statistics for all of them
3043 * in userspace and to cache them in the kernel datapath for performance as
3046 * As the number of subfacets increases, the memory required to maintain
3047 * statistics about them in userspace and in the kernel becomes
3048 * significant. However, with a large number of subfacets it is likely
3049 * that only a few of them are "heavy hitters" that consume a large amount
3050 * of bandwidth. At this point, only heavy hitters are worth caching in
3051 * the kernel and maintaining in userspaces; other subfacets we can
3054 * The technique used to compute the idle time is to build a histogram with
3055 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3056 * that is installed in the kernel gets dropped in the appropriate bucket.
3057 * After the histogram has been built, we compute the cutoff so that only
3058 * the most-recently-used 1% of subfacets (but at least
3059 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3060 * the most-recently-used bucket of subfacets is kept, so actually an
3061 * arbitrary number of subfacets can be kept in any given expiration run
3062 * (though the next run will delete most of those unless they receive
3065 * This requires a second pass through the subfacets, in addition to the
3066 * pass made by update_stats(), because the former function never looks at
3067 * uninstallable subfacets.
3069 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3070 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3071 int buckets[N_BUCKETS] = { 0 };
3072 int total, subtotal, bucket;
3073 struct subfacet *subfacet;
3077 total = hmap_count(&ofproto->subfacets);
3078 if (total <= ofproto->up.flow_eviction_threshold) {
3079 return N_BUCKETS * BUCKET_WIDTH;
3082 /* Build histogram. */
3084 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3085 long long int idle = now - subfacet->used;
3086 int bucket = (idle <= 0 ? 0
3087 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3088 : (unsigned int) idle / BUCKET_WIDTH);
3092 /* Find the first bucket whose flows should be expired. */
3093 subtotal = bucket = 0;
3095 subtotal += buckets[bucket++];
3096 } while (bucket < N_BUCKETS &&
3097 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3099 if (VLOG_IS_DBG_ENABLED()) {
3103 ds_put_cstr(&s, "keep");
3104 for (i = 0; i < N_BUCKETS; i++) {
3106 ds_put_cstr(&s, ", drop");
3109 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3112 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3116 return bucket * BUCKET_WIDTH;
3119 enum { EXPIRE_MAX_BATCH = 50 };
3122 expire_batch(struct ofproto_dpif *ofproto, struct subfacet **subfacets, int n)
3124 struct odputil_keybuf keybufs[EXPIRE_MAX_BATCH];
3125 struct dpif_op ops[EXPIRE_MAX_BATCH];
3126 struct dpif_op *opsp[EXPIRE_MAX_BATCH];
3127 struct ofpbuf keys[EXPIRE_MAX_BATCH];
3128 struct dpif_flow_stats stats[EXPIRE_MAX_BATCH];
3131 for (i = 0; i < n; i++) {
3132 ops[i].type = DPIF_OP_FLOW_DEL;
3133 subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]);
3134 ops[i].u.flow_del.key = keys[i].data;
3135 ops[i].u.flow_del.key_len = keys[i].size;
3136 ops[i].u.flow_del.stats = &stats[i];
3140 dpif_operate(ofproto->dpif, opsp, n);
3141 for (i = 0; i < n; i++) {
3142 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
3143 subfacets[i]->installed = false;
3144 subfacet_destroy(subfacets[i]);
3149 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3151 long long int cutoff = time_msec() - dp_max_idle;
3153 struct subfacet *subfacet, *next_subfacet;
3154 struct subfacet *batch[EXPIRE_MAX_BATCH];
3158 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3159 &ofproto->subfacets) {
3160 if (subfacet->used < cutoff) {
3161 if (subfacet->installed) {
3162 batch[n_batch++] = subfacet;
3163 if (n_batch >= EXPIRE_MAX_BATCH) {
3164 expire_batch(ofproto, batch, n_batch);
3168 subfacet_destroy(subfacet);
3174 expire_batch(ofproto, batch, n_batch);
3178 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3179 * then delete it entirely. */
3181 rule_expire(struct rule_dpif *rule)
3183 struct facet *facet, *next_facet;
3187 /* Has 'rule' expired? */
3189 if (rule->up.hard_timeout
3190 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3191 reason = OFPRR_HARD_TIMEOUT;
3192 } else if (rule->up.idle_timeout
3193 && now > rule->up.used + rule->up.idle_timeout * 1000) {
3194 reason = OFPRR_IDLE_TIMEOUT;
3199 COVERAGE_INC(ofproto_dpif_expired);
3201 /* Update stats. (This is a no-op if the rule expired due to an idle
3202 * timeout, because that only happens when the rule has no facets left.) */
3203 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3204 facet_remove(facet);
3207 /* Get rid of the rule. */
3208 ofproto_rule_expire(&rule->up, reason);
3213 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3215 * The caller must already have determined that no facet with an identical
3216 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3217 * the ofproto's classifier table.
3219 * The facet will initially have no subfacets. The caller should create (at
3220 * least) one subfacet with subfacet_create(). */
3221 static struct facet *
3222 facet_create(struct rule_dpif *rule, const struct flow *flow)
3224 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3225 struct facet *facet;
3227 facet = xzalloc(sizeof *facet);
3228 facet->used = time_msec();
3229 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
3230 list_push_back(&rule->facets, &facet->list_node);
3232 facet->flow = *flow;
3233 list_init(&facet->subfacets);
3234 netflow_flow_init(&facet->nf_flow);
3235 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3241 facet_free(struct facet *facet)
3246 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3247 * 'packet', which arrived on 'in_port'.
3249 * Takes ownership of 'packet'. */
3251 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3252 const struct nlattr *odp_actions, size_t actions_len,
3253 struct ofpbuf *packet)
3255 struct odputil_keybuf keybuf;
3259 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3260 odp_flow_key_from_flow(&key, flow);
3262 error = dpif_execute(ofproto->dpif, key.data, key.size,
3263 odp_actions, actions_len, packet);
3265 ofpbuf_delete(packet);
3269 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3271 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3272 * rule's statistics, via subfacet_uninstall().
3274 * - Removes 'facet' from its rule and from ofproto->facets.
3277 facet_remove(struct facet *facet)
3279 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3280 struct subfacet *subfacet, *next_subfacet;
3282 assert(!list_is_empty(&facet->subfacets));
3284 /* First uninstall all of the subfacets to get final statistics. */
3285 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3286 subfacet_uninstall(subfacet);
3289 /* Flush the final stats to the rule.
3291 * This might require us to have at least one subfacet around so that we
3292 * can use its actions for accounting in facet_account(), which is why we
3293 * have uninstalled but not yet destroyed the subfacets. */
3294 facet_flush_stats(facet);
3296 /* Now we're really all done so destroy everything. */
3297 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3298 &facet->subfacets) {
3299 subfacet_destroy__(subfacet);
3301 hmap_remove(&ofproto->facets, &facet->hmap_node);
3302 list_remove(&facet->list_node);
3306 /* Feed information from 'facet' back into the learning table to keep it in
3307 * sync with what is actually flowing through the datapath. */
3309 facet_learn(struct facet *facet)
3311 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3312 struct action_xlate_ctx ctx;
3314 if (!facet->has_learn
3315 && !facet->has_normal
3316 && (!facet->has_fin_timeout
3317 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
3321 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3322 facet->flow.vlan_tci,
3323 facet->rule, facet->tcp_flags, NULL);
3324 ctx.may_learn = true;
3325 ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions,
3326 facet->rule->up.n_actions));
3330 facet_account(struct facet *facet)
3332 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3333 struct subfacet *subfacet;
3334 const struct nlattr *a;
3339 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3342 n_bytes = facet->byte_count - facet->accounted_bytes;
3344 /* This loop feeds byte counters to bond_account() for rebalancing to use
3345 * as a basis. We also need to track the actual VLAN on which the packet
3346 * is going to be sent to ensure that it matches the one passed to
3347 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3350 * We use the actions from an arbitrary subfacet because they should all
3351 * be equally valid for our purpose. */
3352 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3353 struct subfacet, list_node);
3354 vlan_tci = facet->flow.vlan_tci;
3355 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3356 subfacet->actions, subfacet->actions_len) {
3357 const struct ovs_action_push_vlan *vlan;
3358 struct ofport_dpif *port;
3360 switch (nl_attr_type(a)) {
3361 case OVS_ACTION_ATTR_OUTPUT:
3362 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3363 if (port && port->bundle && port->bundle->bond) {
3364 bond_account(port->bundle->bond, &facet->flow,
3365 vlan_tci_to_vid(vlan_tci), n_bytes);
3369 case OVS_ACTION_ATTR_POP_VLAN:
3370 vlan_tci = htons(0);
3373 case OVS_ACTION_ATTR_PUSH_VLAN:
3374 vlan = nl_attr_get(a);
3375 vlan_tci = vlan->vlan_tci;
3381 /* Returns true if the only action for 'facet' is to send to the controller.
3382 * (We don't report NetFlow expiration messages for such facets because they
3383 * are just part of the control logic for the network, not real traffic). */
3385 facet_is_controller_flow(struct facet *facet)
3388 && facet->rule->up.n_actions == 1
3389 && action_outputs_to_port(&facet->rule->up.actions[0],
3390 htons(OFPP_CONTROLLER)));
3393 /* Folds all of 'facet''s statistics into its rule. Also updates the
3394 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3395 * 'facet''s statistics in the datapath should have been zeroed and folded into
3396 * its packet and byte counts before this function is called. */
3398 facet_flush_stats(struct facet *facet)
3400 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3401 struct subfacet *subfacet;
3403 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3404 assert(!subfacet->dp_byte_count);
3405 assert(!subfacet->dp_packet_count);
3408 facet_push_stats(facet);
3409 if (facet->accounted_bytes < facet->byte_count) {
3410 facet_account(facet);
3411 facet->accounted_bytes = facet->byte_count;
3414 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3415 struct ofexpired expired;
3416 expired.flow = facet->flow;
3417 expired.packet_count = facet->packet_count;
3418 expired.byte_count = facet->byte_count;
3419 expired.used = facet->used;
3420 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3423 facet->rule->packet_count += facet->packet_count;
3424 facet->rule->byte_count += facet->byte_count;
3426 /* Reset counters to prevent double counting if 'facet' ever gets
3428 facet_reset_counters(facet);
3430 netflow_flow_clear(&facet->nf_flow);
3431 facet->tcp_flags = 0;
3434 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3435 * Returns it if found, otherwise a null pointer.
3437 * The returned facet might need revalidation; use facet_lookup_valid()
3438 * instead if that is important. */
3439 static struct facet *
3440 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
3442 struct facet *facet;
3444 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
3446 if (flow_equal(flow, &facet->flow)) {
3454 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3455 * Returns it if found, otherwise a null pointer.
3457 * The returned facet is guaranteed to be valid. */
3458 static struct facet *
3459 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
3461 struct facet *facet = facet_find(ofproto, flow);
3463 /* The facet we found might not be valid, since we could be in need of
3464 * revalidation. If it is not valid, don't return it. */
3466 && (ofproto->need_revalidate
3467 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))
3468 && !facet_revalidate(facet)) {
3469 COVERAGE_INC(facet_invalidated);
3477 facet_check_consistency(struct facet *facet)
3479 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3481 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3483 struct rule_dpif *rule;
3484 struct subfacet *subfacet;
3485 bool may_log = false;
3488 /* Check the rule for consistency. */
3489 rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3491 if (!VLOG_DROP_WARN(&rl)) {
3492 char *s = flow_to_string(&facet->flow);
3493 VLOG_WARN("%s: facet should not exist", s);
3497 } else if (rule != facet->rule) {
3498 may_log = !VLOG_DROP_WARN(&rl);
3504 flow_format(&s, &facet->flow);
3505 ds_put_format(&s, ": facet associated with wrong rule (was "
3506 "table=%"PRIu8",", facet->rule->up.table_id);
3507 cls_rule_format(&facet->rule->up.cr, &s);
3508 ds_put_format(&s, ") (should have been table=%"PRIu8",",
3510 cls_rule_format(&rule->up.cr, &s);
3511 ds_put_char(&s, ')');
3513 VLOG_WARN("%s", ds_cstr(&s));
3520 /* Check the datapath actions for consistency. */
3521 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3522 struct action_xlate_ctx ctx;
3523 struct ofpbuf *odp_actions;
3524 bool actions_changed;
3525 bool should_install;
3527 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3528 subfacet->initial_tci, rule, 0, NULL);
3529 odp_actions = xlate_actions(&ctx, rule->up.actions,
3530 rule->up.n_actions);
3532 should_install = (ctx.may_set_up_flow
3533 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3534 if (!should_install && !subfacet->installed) {
3535 /* The actions for uninstallable flows may vary from one packet to
3536 * the next, so don't compare the actions. */
3540 actions_changed = (subfacet->actions_len != odp_actions->size
3541 || memcmp(subfacet->actions, odp_actions->data,
3542 subfacet->actions_len));
3543 if (should_install != subfacet->installed || actions_changed) {
3545 may_log = !VLOG_DROP_WARN(&rl);
3550 struct odputil_keybuf keybuf;
3555 subfacet_get_key(subfacet, &keybuf, &key);
3556 odp_flow_key_format(key.data, key.size, &s);
3558 ds_put_cstr(&s, ": inconsistency in subfacet");
3559 if (should_install != subfacet->installed) {
3560 enum odp_key_fitness fitness = subfacet->key_fitness;
3562 ds_put_format(&s, " (should%s have been installed)",
3563 should_install ? "" : " not");
3564 ds_put_format(&s, " (may_set_up_flow=%s, fitness=%s)",
3565 ctx.may_set_up_flow ? "true" : "false",
3566 odp_key_fitness_to_string(fitness));
3568 if (actions_changed) {
3569 ds_put_cstr(&s, " (actions were: ");
3570 format_odp_actions(&s, subfacet->actions,
3571 subfacet->actions_len);
3572 ds_put_cstr(&s, ") (correct actions: ");
3573 format_odp_actions(&s, odp_actions->data,
3575 ds_put_char(&s, ')');
3577 ds_put_cstr(&s, " (actions: ");
3578 format_odp_actions(&s, subfacet->actions,
3579 subfacet->actions_len);
3580 ds_put_char(&s, ')');
3582 VLOG_WARN("%s", ds_cstr(&s));
3588 ofpbuf_delete(odp_actions);
3594 /* Re-searches the classifier for 'facet':
3596 * - If the rule found is different from 'facet''s current rule, moves
3597 * 'facet' to the new rule and recompiles its actions.
3599 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3600 * where it is and recompiles its actions anyway.
3602 * - If there is none, destroys 'facet'.
3604 * Returns true if 'facet' still exists, false if it has been destroyed. */
3606 facet_revalidate(struct facet *facet)
3608 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3610 struct nlattr *odp_actions;
3613 struct actions *new_actions;
3615 struct action_xlate_ctx ctx;
3616 struct rule_dpif *new_rule;
3617 struct subfacet *subfacet;
3618 bool actions_changed;
3621 COVERAGE_INC(facet_revalidate);
3623 /* Determine the new rule. */
3624 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3626 /* No new rule, so delete the facet. */
3627 facet_remove(facet);
3631 /* Calculate new datapath actions.
3633 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3634 * emit a NetFlow expiration and, if so, we need to have the old state
3635 * around to properly compose it. */
3637 /* If the datapath actions changed or the installability changed,
3638 * then we need to talk to the datapath. */
3641 memset(&ctx, 0, sizeof ctx);
3642 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3643 struct ofpbuf *odp_actions;
3644 bool should_install;
3646 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3647 subfacet->initial_tci, new_rule, 0, NULL);
3648 odp_actions = xlate_actions(&ctx, new_rule->up.actions,
3649 new_rule->up.n_actions);
3650 actions_changed = (subfacet->actions_len != odp_actions->size
3651 || memcmp(subfacet->actions, odp_actions->data,
3652 subfacet->actions_len));
3654 should_install = (ctx.may_set_up_flow
3655 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3656 if (actions_changed || should_install != subfacet->installed) {
3657 if (should_install) {
3658 struct dpif_flow_stats stats;
3660 subfacet_install(subfacet,
3661 odp_actions->data, odp_actions->size, &stats);
3662 subfacet_update_stats(subfacet, &stats);
3664 subfacet_uninstall(subfacet);
3668 new_actions = xcalloc(list_size(&facet->subfacets),
3669 sizeof *new_actions);
3671 new_actions[i].odp_actions = xmemdup(odp_actions->data,
3673 new_actions[i].actions_len = odp_actions->size;
3676 ofpbuf_delete(odp_actions);
3680 facet_flush_stats(facet);
3683 /* Update 'facet' now that we've taken care of all the old state. */
3684 facet->tags = ctx.tags;
3685 facet->nf_flow.output_iface = ctx.nf_output_iface;
3686 facet->may_install = ctx.may_set_up_flow;
3687 facet->has_learn = ctx.has_learn;
3688 facet->has_normal = ctx.has_normal;
3689 facet->has_fin_timeout = ctx.has_fin_timeout;
3690 facet->mirrors = ctx.mirrors;
3693 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3694 if (new_actions[i].odp_actions) {
3695 free(subfacet->actions);
3696 subfacet->actions = new_actions[i].odp_actions;
3697 subfacet->actions_len = new_actions[i].actions_len;
3703 if (facet->rule != new_rule) {
3704 COVERAGE_INC(facet_changed_rule);
3705 list_remove(&facet->list_node);
3706 list_push_back(&new_rule->facets, &facet->list_node);
3707 facet->rule = new_rule;
3708 facet->used = new_rule->up.created;
3709 facet->prev_used = facet->used;
3715 /* Updates 'facet''s used time. Caller is responsible for calling
3716 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3718 facet_update_time(struct facet *facet, long long int used)
3720 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3721 if (used > facet->used) {
3723 ofproto_rule_update_used(&facet->rule->up, used);
3724 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3729 facet_reset_counters(struct facet *facet)
3731 facet->packet_count = 0;
3732 facet->byte_count = 0;
3733 facet->prev_packet_count = 0;
3734 facet->prev_byte_count = 0;
3735 facet->accounted_bytes = 0;
3739 facet_push_stats(struct facet *facet)
3741 uint64_t new_packets, new_bytes;
3743 assert(facet->packet_count >= facet->prev_packet_count);
3744 assert(facet->byte_count >= facet->prev_byte_count);
3745 assert(facet->used >= facet->prev_used);
3747 new_packets = facet->packet_count - facet->prev_packet_count;
3748 new_bytes = facet->byte_count - facet->prev_byte_count;
3750 if (new_packets || new_bytes || facet->used > facet->prev_used) {
3751 facet->prev_packet_count = facet->packet_count;
3752 facet->prev_byte_count = facet->byte_count;
3753 facet->prev_used = facet->used;
3755 flow_push_stats(facet->rule, &facet->flow,
3756 new_packets, new_bytes, facet->used);
3758 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
3759 facet->mirrors, new_packets, new_bytes);
3763 struct ofproto_push {
3764 struct action_xlate_ctx ctx;
3771 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
3773 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3776 rule->packet_count += push->packets;
3777 rule->byte_count += push->bytes;
3778 ofproto_rule_update_used(&rule->up, push->used);
3782 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3783 * 'rule''s actions and mirrors. */
3785 flow_push_stats(struct rule_dpif *rule,
3786 const struct flow *flow, uint64_t packets, uint64_t bytes,
3789 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3790 struct ofproto_push push;
3792 push.packets = packets;
3796 ofproto_rule_update_used(&rule->up, used);
3798 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, rule,
3800 push.ctx.resubmit_hook = push_resubmit;
3801 ofpbuf_delete(xlate_actions(&push.ctx,
3802 rule->up.actions, rule->up.n_actions));
3807 static struct subfacet *
3808 subfacet_find__(struct ofproto_dpif *ofproto,
3809 const struct nlattr *key, size_t key_len, uint32_t key_hash,
3810 const struct flow *flow)
3812 struct subfacet *subfacet;
3814 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
3815 &ofproto->subfacets) {
3817 ? (subfacet->key_len == key_len
3818 && !memcmp(key, subfacet->key, key_len))
3819 : flow_equal(flow, &subfacet->facet->flow)) {
3827 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
3828 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
3829 * there is one, otherwise creates and returns a new subfacet.
3831 * If the returned subfacet is new, then subfacet->actions will be NULL, in
3832 * which case the caller must populate the actions with
3833 * subfacet_make_actions(). */
3834 static struct subfacet *
3835 subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness,
3836 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
3838 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3839 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3840 struct subfacet *subfacet;
3842 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
3844 if (subfacet->facet == facet) {
3848 /* This shouldn't happen. */
3849 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
3850 subfacet_destroy(subfacet);
3853 subfacet = xzalloc(sizeof *subfacet);
3854 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
3855 list_push_back(&facet->subfacets, &subfacet->list_node);
3856 subfacet->facet = facet;
3857 subfacet->used = time_msec();
3858 subfacet->key_fitness = key_fitness;
3859 if (key_fitness != ODP_FIT_PERFECT) {
3860 subfacet->key = xmemdup(key, key_len);
3861 subfacet->key_len = key_len;
3863 subfacet->installed = false;
3864 subfacet->initial_tci = initial_tci;
3869 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
3870 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
3871 static struct subfacet *
3872 subfacet_find(struct ofproto_dpif *ofproto,
3873 const struct nlattr *key, size_t key_len)
3875 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3876 enum odp_key_fitness fitness;
3879 fitness = odp_flow_key_to_flow(key, key_len, &flow);
3880 if (fitness == ODP_FIT_ERROR) {
3884 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
3887 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
3888 * its facet within 'ofproto', and frees it. */
3890 subfacet_destroy__(struct subfacet *subfacet)
3892 struct facet *facet = subfacet->facet;
3893 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3895 subfacet_uninstall(subfacet);
3896 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
3897 list_remove(&subfacet->list_node);
3898 free(subfacet->key);
3899 free(subfacet->actions);
3903 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
3904 * last remaining subfacet in its facet destroys the facet too. */
3906 subfacet_destroy(struct subfacet *subfacet)
3908 struct facet *facet = subfacet->facet;
3910 if (list_is_singleton(&facet->subfacets)) {
3911 /* facet_remove() needs at least one subfacet (it will remove it). */
3912 facet_remove(facet);
3914 subfacet_destroy__(subfacet);
3918 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
3919 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
3920 * for use as temporary storage. */
3922 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
3925 if (!subfacet->key) {
3926 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
3927 odp_flow_key_from_flow(key, &subfacet->facet->flow);
3929 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
3933 /* Composes the datapath actions for 'subfacet' based on its rule's actions. */
3935 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet)
3937 struct facet *facet = subfacet->facet;
3938 struct rule_dpif *rule = facet->rule;
3939 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3940 struct ofpbuf *odp_actions;
3941 struct action_xlate_ctx ctx;
3943 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
3945 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3946 facet->tags = ctx.tags;
3947 facet->may_install = ctx.may_set_up_flow;
3948 facet->has_learn = ctx.has_learn;
3949 facet->has_normal = ctx.has_normal;
3950 facet->has_fin_timeout = ctx.has_fin_timeout;
3951 facet->nf_flow.output_iface = ctx.nf_output_iface;
3952 facet->mirrors = ctx.mirrors;
3954 if (subfacet->actions_len != odp_actions->size
3955 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
3956 free(subfacet->actions);
3957 subfacet->actions_len = odp_actions->size;
3958 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
3961 ofpbuf_delete(odp_actions);
3964 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
3965 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
3966 * in the datapath will be zeroed and 'stats' will be updated with traffic new
3967 * since 'subfacet' was last updated.
3969 * Returns 0 if successful, otherwise a positive errno value. */
3971 subfacet_install(struct subfacet *subfacet,
3972 const struct nlattr *actions, size_t actions_len,
3973 struct dpif_flow_stats *stats)
3975 struct facet *facet = subfacet->facet;
3976 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3977 struct odputil_keybuf keybuf;
3978 enum dpif_flow_put_flags flags;
3982 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3984 flags |= DPIF_FP_ZERO_STATS;
3987 subfacet_get_key(subfacet, &keybuf, &key);
3988 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
3989 actions, actions_len, stats);
3992 subfacet_reset_dp_stats(subfacet, stats);
3998 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4000 subfacet_uninstall(struct subfacet *subfacet)
4002 if (subfacet->installed) {
4003 struct rule_dpif *rule = subfacet->facet->rule;
4004 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4005 struct odputil_keybuf keybuf;
4006 struct dpif_flow_stats stats;
4010 subfacet_get_key(subfacet, &keybuf, &key);
4011 error = dpif_flow_del(ofproto->dpif, key.data, key.size, &stats);
4012 subfacet_reset_dp_stats(subfacet, &stats);
4014 subfacet_update_stats(subfacet, &stats);
4016 subfacet->installed = false;
4018 assert(subfacet->dp_packet_count == 0);
4019 assert(subfacet->dp_byte_count == 0);
4023 /* Resets 'subfacet''s datapath statistics counters. This should be called
4024 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4025 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4026 * was reset in the datapath. 'stats' will be modified to include only
4027 * statistics new since 'subfacet' was last updated. */
4029 subfacet_reset_dp_stats(struct subfacet *subfacet,
4030 struct dpif_flow_stats *stats)
4033 && subfacet->dp_packet_count <= stats->n_packets
4034 && subfacet->dp_byte_count <= stats->n_bytes) {
4035 stats->n_packets -= subfacet->dp_packet_count;
4036 stats->n_bytes -= subfacet->dp_byte_count;
4039 subfacet->dp_packet_count = 0;
4040 subfacet->dp_byte_count = 0;
4043 /* Updates 'subfacet''s used time. The caller is responsible for calling
4044 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
4046 subfacet_update_time(struct subfacet *subfacet, long long int used)
4048 if (used > subfacet->used) {
4049 subfacet->used = used;
4050 facet_update_time(subfacet->facet, used);
4054 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4056 * Because of the meaning of a subfacet's counters, it only makes sense to do
4057 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4058 * represents a packet that was sent by hand or if it represents statistics
4059 * that have been cleared out of the datapath. */
4061 subfacet_update_stats(struct subfacet *subfacet,
4062 const struct dpif_flow_stats *stats)
4064 if (stats->n_packets || stats->used > subfacet->used) {
4065 struct facet *facet = subfacet->facet;
4067 subfacet_update_time(subfacet, stats->used);
4068 facet->packet_count += stats->n_packets;
4069 facet->byte_count += stats->n_bytes;
4070 facet->tcp_flags |= stats->tcp_flags;
4071 facet_push_stats(facet);
4072 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
4078 static struct rule_dpif *
4079 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4082 struct cls_rule *cls_rule;
4083 struct classifier *cls;
4085 if (table_id >= N_TABLES) {
4089 cls = &ofproto->up.tables[table_id].cls;
4090 if (flow->nw_frag & FLOW_NW_FRAG_ANY
4091 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4092 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
4093 * are unavailable. */
4094 struct flow ofpc_normal_flow = *flow;
4095 ofpc_normal_flow.tp_src = htons(0);
4096 ofpc_normal_flow.tp_dst = htons(0);
4097 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
4099 cls_rule = classifier_lookup(cls, flow);
4101 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
4105 complete_operation(struct rule_dpif *rule)
4107 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4109 rule_invalidate(rule);
4111 struct dpif_completion *c = xmalloc(sizeof *c);
4112 c->op = rule->up.pending;
4113 list_push_back(&ofproto->completions, &c->list_node);
4115 ofoperation_complete(rule->up.pending, 0);
4119 static struct rule *
4122 struct rule_dpif *rule = xmalloc(sizeof *rule);
4127 rule_dealloc(struct rule *rule_)
4129 struct rule_dpif *rule = rule_dpif_cast(rule_);
4134 rule_construct(struct rule *rule_)
4136 struct rule_dpif *rule = rule_dpif_cast(rule_);
4137 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4138 struct rule_dpif *victim;
4142 error = validate_actions(rule->up.actions, rule->up.n_actions,
4143 &rule->up.cr.flow, ofproto->max_ports);
4148 rule->packet_count = 0;
4149 rule->byte_count = 0;
4151 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
4152 if (victim && !list_is_empty(&victim->facets)) {
4153 struct facet *facet;
4155 rule->facets = victim->facets;
4156 list_moved(&rule->facets);
4157 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4158 /* XXX: We're only clearing our local counters here. It's possible
4159 * that quite a few packets are unaccounted for in the datapath
4160 * statistics. These will be accounted to the new rule instead of
4161 * cleared as required. This could be fixed by clearing out the
4162 * datapath statistics for this facet, but currently it doesn't
4164 facet_reset_counters(facet);
4168 /* Must avoid list_moved() in this case. */
4169 list_init(&rule->facets);
4172 table_id = rule->up.table_id;
4173 rule->tag = (victim ? victim->tag
4175 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
4176 ofproto->tables[table_id].basis));
4178 complete_operation(rule);
4183 rule_destruct(struct rule *rule_)
4185 struct rule_dpif *rule = rule_dpif_cast(rule_);
4186 struct facet *facet, *next_facet;
4188 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4189 facet_revalidate(facet);
4192 complete_operation(rule);
4196 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4198 struct rule_dpif *rule = rule_dpif_cast(rule_);
4199 struct facet *facet;
4201 /* Start from historical data for 'rule' itself that are no longer tracked
4202 * in facets. This counts, for example, facets that have expired. */
4203 *packets = rule->packet_count;
4204 *bytes = rule->byte_count;
4206 /* Add any statistics that are tracked by facets. This includes
4207 * statistical data recently updated by ofproto_update_stats() as well as
4208 * stats for packets that were executed "by hand" via dpif_execute(). */
4209 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4210 *packets += facet->packet_count;
4211 *bytes += facet->byte_count;
4216 rule_execute(struct rule *rule_, const struct flow *flow,
4217 struct ofpbuf *packet)
4219 struct rule_dpif *rule = rule_dpif_cast(rule_);
4220 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4221 struct action_xlate_ctx ctx;
4222 struct ofpbuf *odp_actions;
4225 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
4226 rule, packet_get_tcp_flags(packet, flow), packet);
4227 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
4228 size = packet->size;
4229 if (execute_odp_actions(ofproto, flow, odp_actions->data,
4230 odp_actions->size, packet)) {
4231 rule->packet_count++;
4232 rule->byte_count += size;
4233 flow_push_stats(rule, flow, 1, size, time_msec());
4235 ofpbuf_delete(odp_actions);
4241 rule_modify_actions(struct rule *rule_)
4243 struct rule_dpif *rule = rule_dpif_cast(rule_);
4244 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4247 error = validate_actions(rule->up.actions, rule->up.n_actions,
4248 &rule->up.cr.flow, ofproto->max_ports);
4250 ofoperation_complete(rule->up.pending, error);
4254 complete_operation(rule);
4257 /* Sends 'packet' out 'ofport'.
4258 * May modify 'packet'.
4259 * Returns 0 if successful, otherwise a positive errno value. */
4261 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4263 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4264 struct ofpbuf key, odp_actions;
4265 struct odputil_keybuf keybuf;
4270 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
4271 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4273 if (odp_port != ofport->odp_port) {
4274 eth_pop_vlan(packet);
4275 flow.vlan_tci = htons(0);
4278 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4279 odp_flow_key_from_flow(&key, &flow);
4281 ofpbuf_init(&odp_actions, 32);
4282 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4284 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4285 error = dpif_execute(ofproto->dpif,
4287 odp_actions.data, odp_actions.size,
4289 ofpbuf_uninit(&odp_actions);
4292 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4293 ofproto->up.name, odp_port, strerror(error));
4295 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
4299 /* OpenFlow to datapath action translation. */
4301 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
4302 struct action_xlate_ctx *ctx);
4303 static void xlate_normal(struct action_xlate_ctx *);
4306 put_userspace_action(const struct ofproto_dpif *ofproto,
4307 struct ofpbuf *odp_actions,
4308 const struct flow *flow,
4309 const struct user_action_cookie *cookie)
4313 pid = dpif_port_get_pid(ofproto->dpif,
4314 ofp_port_to_odp_port(flow->in_port));
4316 return odp_put_userspace_action(pid, cookie, odp_actions);
4319 /* Compose SAMPLE action for sFlow. */
4321 compose_sflow_action(const struct ofproto_dpif *ofproto,
4322 struct ofpbuf *odp_actions,
4323 const struct flow *flow,
4326 uint32_t port_ifindex;
4327 uint32_t probability;
4328 struct user_action_cookie cookie;
4329 size_t sample_offset, actions_offset;
4330 int cookie_offset, n_output;
4332 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4336 if (odp_port == OVSP_NONE) {
4340 port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4344 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4346 /* Number of packets out of UINT_MAX to sample. */
4347 probability = dpif_sflow_get_probability(ofproto->sflow);
4348 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4350 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4352 cookie.type = USER_ACTION_COOKIE_SFLOW;
4353 cookie.data = port_ifindex;
4354 cookie.n_output = n_output;
4355 cookie.vlan_tci = 0;
4356 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4358 nl_msg_end_nested(odp_actions, actions_offset);
4359 nl_msg_end_nested(odp_actions, sample_offset);
4360 return cookie_offset;
4363 /* SAMPLE action must be first action in any given list of actions.
4364 * At this point we do not have all information required to build it. So try to
4365 * build sample action as complete as possible. */
4367 add_sflow_action(struct action_xlate_ctx *ctx)
4369 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4371 &ctx->flow, OVSP_NONE);
4372 ctx->sflow_odp_port = 0;
4373 ctx->sflow_n_outputs = 0;
4376 /* Fix SAMPLE action according to data collected while composing ODP actions.
4377 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4378 * USERSPACE action's user-cookie which is required for sflow. */
4380 fix_sflow_action(struct action_xlate_ctx *ctx)
4382 const struct flow *base = &ctx->base_flow;
4383 struct user_action_cookie *cookie;
4385 if (!ctx->user_cookie_offset) {
4389 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4391 assert(cookie != NULL);
4392 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4394 if (ctx->sflow_n_outputs) {
4395 cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow,
4396 ctx->sflow_odp_port);
4398 if (ctx->sflow_n_outputs >= 255) {
4399 cookie->n_output = 255;
4401 cookie->n_output = ctx->sflow_n_outputs;
4403 cookie->vlan_tci = base->vlan_tci;
4407 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4410 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4411 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4412 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4413 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4417 struct priority_to_dscp *pdscp;
4419 if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD
4420 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4424 pdscp = get_priority(ofport, ctx->flow.skb_priority);
4426 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4427 ctx->flow.nw_tos |= pdscp->dscp;
4430 /* We may not have an ofport record for this port, but it doesn't hurt
4431 * to allow forwarding to it anyhow. Maybe such a port will appear
4432 * later and we're pre-populating the flow table. */
4435 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4436 ctx->flow.vlan_tci);
4437 if (out_port != odp_port) {
4438 ctx->flow.vlan_tci = htons(0);
4440 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4441 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4443 ctx->sflow_odp_port = odp_port;
4444 ctx->sflow_n_outputs++;
4445 ctx->nf_output_iface = ofp_port;
4446 ctx->flow.vlan_tci = flow_vlan_tci;
4447 ctx->flow.nw_tos = flow_nw_tos;
4451 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4453 compose_output_action__(ctx, ofp_port, true);
4457 xlate_table_action(struct action_xlate_ctx *ctx,
4458 uint16_t in_port, uint8_t table_id)
4460 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4461 struct ofproto_dpif *ofproto = ctx->ofproto;
4462 struct rule_dpif *rule;
4463 uint16_t old_in_port;
4464 uint8_t old_table_id;
4466 old_table_id = ctx->table_id;
4467 ctx->table_id = table_id;
4469 /* Look up a flow with 'in_port' as the input port. */
4470 old_in_port = ctx->flow.in_port;
4471 ctx->flow.in_port = in_port;
4472 rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
4475 if (table_id > 0 && table_id < N_TABLES) {
4476 struct table_dpif *table = &ofproto->tables[table_id];
4477 if (table->other_table) {
4478 ctx->tags |= (rule && rule->tag
4480 : rule_calculate_tag(&ctx->flow,
4481 &table->other_table->wc,
4486 /* Restore the original input port. Otherwise OFPP_NORMAL and
4487 * OFPP_IN_PORT will have surprising behavior. */
4488 ctx->flow.in_port = old_in_port;
4490 if (ctx->resubmit_hook) {
4491 ctx->resubmit_hook(ctx, rule);
4495 struct rule_dpif *old_rule = ctx->rule;
4499 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
4500 ctx->rule = old_rule;
4504 ctx->table_id = old_table_id;
4506 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4508 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
4509 MAX_RESUBMIT_RECURSION);
4510 ctx->max_resubmit_trigger = true;
4515 xlate_resubmit_table(struct action_xlate_ctx *ctx,
4516 const struct nx_action_resubmit *nar)
4521 in_port = (nar->in_port == htons(OFPP_IN_PORT)
4523 : ntohs(nar->in_port));
4524 table_id = nar->table == 255 ? ctx->table_id : nar->table;
4526 xlate_table_action(ctx, in_port, table_id);
4530 flood_packets(struct action_xlate_ctx *ctx, bool all)
4532 struct ofport_dpif *ofport;
4534 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
4535 uint16_t ofp_port = ofport->up.ofp_port;
4537 if (ofp_port == ctx->flow.in_port) {
4542 compose_output_action__(ctx, ofp_port, false);
4543 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
4544 compose_output_action(ctx, ofp_port);
4548 ctx->nf_output_iface = NF_OUT_FLOOD;
4552 execute_controller_action(struct action_xlate_ctx *ctx, int len,
4553 enum ofp_packet_in_reason reason,
4554 uint16_t controller_id)
4556 struct ofputil_packet_in pin;
4557 struct ofpbuf *packet;
4559 ctx->may_set_up_flow = false;
4564 packet = ofpbuf_clone(ctx->packet);
4566 if (packet->l2 && packet->l3) {
4567 struct eth_header *eh;
4569 eth_pop_vlan(packet);
4572 /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
4573 * LLC frame. Calculating the Ethernet type of these frames is more
4574 * trouble than seems appropriate for a simple assertion. */
4575 assert(ntohs(eh->eth_type) < ETH_TYPE_MIN
4576 || eh->eth_type == ctx->flow.dl_type);
4578 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
4579 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
4581 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
4582 eth_push_vlan(packet, ctx->flow.vlan_tci);
4586 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4587 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
4588 ctx->flow.nw_tos, ctx->flow.nw_ttl);
4592 if (ctx->flow.nw_proto == IPPROTO_TCP) {
4593 packet_set_tcp_port(packet, ctx->flow.tp_src,
4595 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
4596 packet_set_udp_port(packet, ctx->flow.tp_src,
4603 pin.packet = packet->data;
4604 pin.packet_len = packet->size;
4605 pin.reason = reason;
4606 pin.controller_id = controller_id;
4607 pin.table_id = ctx->table_id;
4608 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
4611 flow_get_metadata(&ctx->flow, &pin.fmd);
4613 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
4614 ofpbuf_delete(packet);
4618 compose_dec_ttl(struct action_xlate_ctx *ctx)
4620 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
4621 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
4625 if (ctx->flow.nw_ttl > 1) {
4629 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
4631 /* Stop processing for current table. */
4637 xlate_output_action__(struct action_xlate_ctx *ctx,
4638 uint16_t port, uint16_t max_len)
4640 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
4642 ctx->nf_output_iface = NF_OUT_DROP;
4646 compose_output_action(ctx, ctx->flow.in_port);
4649 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
4655 flood_packets(ctx, false);
4658 flood_packets(ctx, true);
4660 case OFPP_CONTROLLER:
4661 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
4667 if (port != ctx->flow.in_port) {
4668 compose_output_action(ctx, port);
4673 if (prev_nf_output_iface == NF_OUT_FLOOD) {
4674 ctx->nf_output_iface = NF_OUT_FLOOD;
4675 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
4676 ctx->nf_output_iface = prev_nf_output_iface;
4677 } else if (prev_nf_output_iface != NF_OUT_DROP &&
4678 ctx->nf_output_iface != NF_OUT_FLOOD) {
4679 ctx->nf_output_iface = NF_OUT_MULTI;
4684 xlate_output_reg_action(struct action_xlate_ctx *ctx,
4685 const struct nx_action_output_reg *naor)
4687 struct mf_subfield src;
4690 nxm_decode(&src, naor->src, naor->ofs_nbits);
4691 ofp_port = mf_get_subfield(&src, &ctx->flow);
4693 if (ofp_port <= UINT16_MAX) {
4694 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
4699 xlate_output_action(struct action_xlate_ctx *ctx,
4700 const struct ofp_action_output *oao)
4702 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
4706 xlate_enqueue_action(struct action_xlate_ctx *ctx,
4707 const struct ofp_action_enqueue *oae)
4710 uint32_t flow_priority, priority;
4713 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
4716 /* Fall back to ordinary output action. */
4717 xlate_output_action__(ctx, ntohs(oae->port), 0);
4721 /* Figure out datapath output port. */
4722 ofp_port = ntohs(oae->port);
4723 if (ofp_port == OFPP_IN_PORT) {
4724 ofp_port = ctx->flow.in_port;
4725 } else if (ofp_port == ctx->flow.in_port) {
4729 /* Add datapath actions. */
4730 flow_priority = ctx->flow.skb_priority;
4731 ctx->flow.skb_priority = priority;
4732 compose_output_action(ctx, ofp_port);
4733 ctx->flow.skb_priority = flow_priority;
4735 /* Update NetFlow output port. */
4736 if (ctx->nf_output_iface == NF_OUT_DROP) {
4737 ctx->nf_output_iface = ofp_port;
4738 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
4739 ctx->nf_output_iface = NF_OUT_MULTI;
4744 xlate_set_queue_action(struct action_xlate_ctx *ctx,
4745 const struct nx_action_set_queue *nasq)
4750 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
4753 /* Couldn't translate queue to a priority, so ignore. A warning
4754 * has already been logged. */
4758 ctx->flow.skb_priority = priority;
4761 struct xlate_reg_state {
4767 xlate_autopath(struct action_xlate_ctx *ctx,
4768 const struct nx_action_autopath *naa)
4770 uint16_t ofp_port = ntohl(naa->id);
4771 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
4773 if (!port || !port->bundle) {
4774 ofp_port = OFPP_NONE;
4775 } else if (port->bundle->bond) {
4776 /* Autopath does not support VLAN hashing. */
4777 struct ofport_dpif *slave = bond_choose_output_slave(
4778 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
4780 ofp_port = slave->up.ofp_port;
4783 autopath_execute(naa, &ctx->flow, ofp_port);
4787 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
4789 struct ofproto_dpif *ofproto = ofproto_;
4790 struct ofport_dpif *port;
4800 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
4803 port = get_ofp_port(ofproto, ofp_port);
4804 return port ? port->may_enable : false;
4809 xlate_learn_action(struct action_xlate_ctx *ctx,
4810 const struct nx_action_learn *learn)
4812 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
4813 struct ofputil_flow_mod fm;
4816 learn_execute(learn, &ctx->flow, &fm);
4818 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
4819 if (error && !VLOG_DROP_WARN(&rl)) {
4820 VLOG_WARN("learning action failed to modify flow table (%s)",
4821 ofperr_get_name(error));
4827 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
4828 * means "infinite". */
4830 reduce_timeout(uint16_t max, uint16_t *timeout)
4832 if (max && (!*timeout || *timeout > max)) {
4838 xlate_fin_timeout(struct action_xlate_ctx *ctx,
4839 const struct nx_action_fin_timeout *naft)
4841 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
4842 struct rule_dpif *rule = ctx->rule;
4844 reduce_timeout(ntohs(naft->fin_idle_timeout), &rule->up.idle_timeout);
4845 reduce_timeout(ntohs(naft->fin_hard_timeout), &rule->up.hard_timeout);
4850 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
4852 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
4853 ? OFPUTIL_PC_NO_RECV_STP
4854 : OFPUTIL_PC_NO_RECV)) {
4858 /* Only drop packets here if both forwarding and learning are
4859 * disabled. If just learning is enabled, we need to have
4860 * OFPP_NORMAL and the learning action have a look at the packet
4861 * before we can drop it. */
4862 if (!stp_forward_in_state(port->stp_state)
4863 && !stp_learn_in_state(port->stp_state)) {
4871 do_xlate_actions(const union ofp_action *in, size_t n_in,
4872 struct action_xlate_ctx *ctx)
4874 const struct ofport_dpif *port;
4875 const union ofp_action *ia;
4876 bool was_evictable = true;
4879 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
4880 if (port && !may_receive(port, ctx)) {
4881 /* Drop this flow. */
4886 /* Don't let the rule we're working on get evicted underneath us. */
4887 was_evictable = ctx->rule->up.evictable;
4888 ctx->rule->up.evictable = false;
4890 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
4891 const struct ofp_action_dl_addr *oada;
4892 const struct nx_action_resubmit *nar;
4893 const struct nx_action_set_tunnel *nast;
4894 const struct nx_action_set_queue *nasq;
4895 const struct nx_action_multipath *nam;
4896 const struct nx_action_autopath *naa;
4897 const struct nx_action_bundle *nab;
4898 const struct nx_action_output_reg *naor;
4899 const struct nx_action_controller *nac;
4900 enum ofputil_action_code code;
4907 code = ofputil_decode_action_unsafe(ia);
4909 case OFPUTIL_OFPAT10_OUTPUT:
4910 xlate_output_action(ctx, &ia->output);
4913 case OFPUTIL_OFPAT10_SET_VLAN_VID:
4914 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
4915 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
4918 case OFPUTIL_OFPAT10_SET_VLAN_PCP:
4919 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
4920 ctx->flow.vlan_tci |= htons(
4921 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
4924 case OFPUTIL_OFPAT10_STRIP_VLAN:
4925 ctx->flow.vlan_tci = htons(0);
4928 case OFPUTIL_OFPAT10_SET_DL_SRC:
4929 oada = ((struct ofp_action_dl_addr *) ia);
4930 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
4933 case OFPUTIL_OFPAT10_SET_DL_DST:
4934 oada = ((struct ofp_action_dl_addr *) ia);
4935 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
4938 case OFPUTIL_OFPAT10_SET_NW_SRC:
4939 ctx->flow.nw_src = ia->nw_addr.nw_addr;
4942 case OFPUTIL_OFPAT10_SET_NW_DST:
4943 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
4946 case OFPUTIL_OFPAT10_SET_NW_TOS:
4947 /* OpenFlow 1.0 only supports IPv4. */
4948 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4949 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4950 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
4954 case OFPUTIL_OFPAT10_SET_TP_SRC:
4955 ctx->flow.tp_src = ia->tp_port.tp_port;
4958 case OFPUTIL_OFPAT10_SET_TP_DST:
4959 ctx->flow.tp_dst = ia->tp_port.tp_port;
4962 case OFPUTIL_OFPAT10_ENQUEUE:
4963 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
4966 case OFPUTIL_NXAST_RESUBMIT:
4967 nar = (const struct nx_action_resubmit *) ia;
4968 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
4971 case OFPUTIL_NXAST_RESUBMIT_TABLE:
4972 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
4975 case OFPUTIL_NXAST_SET_TUNNEL:
4976 nast = (const struct nx_action_set_tunnel *) ia;
4977 tun_id = htonll(ntohl(nast->tun_id));
4978 ctx->flow.tun_id = tun_id;
4981 case OFPUTIL_NXAST_SET_QUEUE:
4982 nasq = (const struct nx_action_set_queue *) ia;
4983 xlate_set_queue_action(ctx, nasq);
4986 case OFPUTIL_NXAST_POP_QUEUE:
4987 ctx->flow.skb_priority = ctx->orig_skb_priority;
4990 case OFPUTIL_NXAST_REG_MOVE:
4991 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
4995 case OFPUTIL_NXAST_REG_LOAD:
4996 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
5000 case OFPUTIL_NXAST_NOTE:
5001 /* Nothing to do. */
5004 case OFPUTIL_NXAST_SET_TUNNEL64:
5005 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
5006 ctx->flow.tun_id = tun_id;
5009 case OFPUTIL_NXAST_MULTIPATH:
5010 nam = (const struct nx_action_multipath *) ia;
5011 multipath_execute(nam, &ctx->flow);
5014 case OFPUTIL_NXAST_AUTOPATH:
5015 naa = (const struct nx_action_autopath *) ia;
5016 xlate_autopath(ctx, naa);
5019 case OFPUTIL_NXAST_BUNDLE:
5020 ctx->ofproto->has_bundle_action = true;
5021 nab = (const struct nx_action_bundle *) ia;
5022 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
5027 case OFPUTIL_NXAST_BUNDLE_LOAD:
5028 ctx->ofproto->has_bundle_action = true;
5029 nab = (const struct nx_action_bundle *) ia;
5030 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
5034 case OFPUTIL_NXAST_OUTPUT_REG:
5035 naor = (const struct nx_action_output_reg *) ia;
5036 xlate_output_reg_action(ctx, naor);
5039 case OFPUTIL_NXAST_LEARN:
5040 ctx->has_learn = true;
5041 if (ctx->may_learn) {
5042 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
5046 case OFPUTIL_NXAST_DEC_TTL:
5047 if (compose_dec_ttl(ctx)) {
5052 case OFPUTIL_NXAST_EXIT:
5056 case OFPUTIL_NXAST_FIN_TIMEOUT:
5057 ctx->has_fin_timeout = true;
5058 xlate_fin_timeout(ctx, (const struct nx_action_fin_timeout *) ia);
5061 case OFPUTIL_NXAST_CONTROLLER:
5062 nac = (const struct nx_action_controller *) ia;
5063 execute_controller_action(ctx, ntohs(nac->max_len), nac->reason,
5064 ntohs(nac->controller_id));
5070 /* We've let OFPP_NORMAL and the learning action look at the packet,
5071 * so drop it now if forwarding is disabled. */
5072 if (port && !stp_forward_in_state(port->stp_state)) {
5073 ofpbuf_clear(ctx->odp_actions);
5074 add_sflow_action(ctx);
5077 ctx->rule->up.evictable = was_evictable;
5082 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
5083 struct ofproto_dpif *ofproto, const struct flow *flow,
5084 ovs_be16 initial_tci, struct rule_dpif *rule,
5085 uint8_t tcp_flags, const struct ofpbuf *packet)
5087 ctx->ofproto = ofproto;
5089 ctx->base_flow = ctx->flow;
5090 ctx->base_flow.tun_id = 0;
5091 ctx->base_flow.vlan_tci = initial_tci;
5093 ctx->packet = packet;
5094 ctx->may_learn = packet != NULL;
5095 ctx->tcp_flags = tcp_flags;
5096 ctx->resubmit_hook = NULL;
5099 static struct ofpbuf *
5100 xlate_actions(struct action_xlate_ctx *ctx,
5101 const union ofp_action *in, size_t n_in)
5103 struct flow orig_flow = ctx->flow;
5105 COVERAGE_INC(ofproto_dpif_xlate);
5107 ctx->odp_actions = ofpbuf_new(512);
5108 ofpbuf_reserve(ctx->odp_actions, NL_A_U32_SIZE);
5110 ctx->may_set_up_flow = true;
5111 ctx->has_learn = false;
5112 ctx->has_normal = false;
5113 ctx->has_fin_timeout = false;
5114 ctx->nf_output_iface = NF_OUT_DROP;
5117 ctx->max_resubmit_trigger = false;
5118 ctx->orig_skb_priority = ctx->flow.skb_priority;
5122 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
5123 switch (ctx->ofproto->up.frag_handling) {
5124 case OFPC_FRAG_NORMAL:
5125 /* We must pretend that transport ports are unavailable. */
5126 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
5127 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
5130 case OFPC_FRAG_DROP:
5131 return ctx->odp_actions;
5133 case OFPC_FRAG_REASM:
5136 case OFPC_FRAG_NX_MATCH:
5137 /* Nothing to do. */
5140 case OFPC_INVALID_TTL_TO_CONTROLLER:
5145 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
5146 ctx->may_set_up_flow = false;
5147 return ctx->odp_actions;
5149 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5150 struct flow original_flow = ctx->flow;
5151 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
5153 add_sflow_action(ctx);
5154 do_xlate_actions(in, n_in, ctx);
5156 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook
5157 && !VLOG_DROP_ERR(&trace_rl)) {
5158 struct ds ds = DS_EMPTY_INITIALIZER;
5160 ofproto_trace(ctx->ofproto, &original_flow, ctx->packet,
5162 VLOG_ERR("Trace triggered by excessive resubmit recursion:\n%s",
5167 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
5168 ctx->odp_actions->data,
5169 ctx->odp_actions->size)) {
5170 ctx->may_set_up_flow = false;
5172 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
5174 compose_output_action(ctx, OFPP_LOCAL);
5177 add_mirror_actions(ctx, &orig_flow);
5178 fix_sflow_action(ctx);
5181 return ctx->odp_actions;
5184 /* OFPP_NORMAL implementation. */
5186 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
5188 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
5189 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
5190 * the bundle on which the packet was received, returns the VLAN to which the
5193 * Both 'vid' and the return value are in the range 0...4095. */
5195 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
5197 switch (in_bundle->vlan_mode) {
5198 case PORT_VLAN_ACCESS:
5199 return in_bundle->vlan;
5202 case PORT_VLAN_TRUNK:
5205 case PORT_VLAN_NATIVE_UNTAGGED:
5206 case PORT_VLAN_NATIVE_TAGGED:
5207 return vid ? vid : in_bundle->vlan;
5214 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
5215 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
5218 * 'vid' should be the VID obtained from the 802.1Q header that was received as
5219 * part of a packet (specify 0 if there was no 802.1Q header), in the range
5222 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
5224 /* Allow any VID on the OFPP_NONE port. */
5225 if (in_bundle == &ofpp_none_bundle) {
5229 switch (in_bundle->vlan_mode) {
5230 case PORT_VLAN_ACCESS:
5233 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5234 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
5235 "packet received on port %s configured as VLAN "
5236 "%"PRIu16" access port",
5237 in_bundle->ofproto->up.name, vid,
5238 in_bundle->name, in_bundle->vlan);
5244 case PORT_VLAN_NATIVE_UNTAGGED:
5245 case PORT_VLAN_NATIVE_TAGGED:
5247 /* Port must always carry its native VLAN. */
5251 case PORT_VLAN_TRUNK:
5252 if (!ofbundle_includes_vlan(in_bundle, vid)) {
5254 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5255 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
5256 "received on port %s not configured for trunking "
5258 in_bundle->ofproto->up.name, vid,
5259 in_bundle->name, vid);
5271 /* Given 'vlan', the VLAN that a packet belongs to, and
5272 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
5273 * that should be included in the 802.1Q header. (If the return value is 0,
5274 * then the 802.1Q header should only be included in the packet if there is a
5277 * Both 'vlan' and the return value are in the range 0...4095. */
5279 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
5281 switch (out_bundle->vlan_mode) {
5282 case PORT_VLAN_ACCESS:
5285 case PORT_VLAN_TRUNK:
5286 case PORT_VLAN_NATIVE_TAGGED:
5289 case PORT_VLAN_NATIVE_UNTAGGED:
5290 return vlan == out_bundle->vlan ? 0 : vlan;
5298 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
5301 struct ofport_dpif *port;
5303 ovs_be16 tci, old_tci;
5305 vid = output_vlan_to_vid(out_bundle, vlan);
5306 if (!out_bundle->bond) {
5307 port = ofbundle_get_a_port(out_bundle);
5309 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
5312 /* No slaves enabled, so drop packet. */
5317 old_tci = ctx->flow.vlan_tci;
5319 if (tci || out_bundle->use_priority_tags) {
5320 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
5322 tci |= htons(VLAN_CFI);
5325 ctx->flow.vlan_tci = tci;
5327 compose_output_action(ctx, port->up.ofp_port);
5328 ctx->flow.vlan_tci = old_tci;
5332 mirror_mask_ffs(mirror_mask_t mask)
5334 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
5339 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
5341 return (bundle->vlan_mode != PORT_VLAN_ACCESS
5342 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
5346 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
5348 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
5351 /* Returns an arbitrary interface within 'bundle'. */
5352 static struct ofport_dpif *
5353 ofbundle_get_a_port(const struct ofbundle *bundle)
5355 return CONTAINER_OF(list_front(&bundle->ports),
5356 struct ofport_dpif, bundle_node);
5360 vlan_is_mirrored(const struct ofmirror *m, int vlan)
5362 return !m->vlans || bitmap_is_set(m->vlans, vlan);
5365 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
5366 * to a VLAN. In general most packets may be mirrored but we want to drop
5367 * protocols that may confuse switches. */
5369 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
5371 /* If you change this function's behavior, please update corresponding
5372 * documentation in vswitch.xml at the same time. */
5373 if (dst[0] != 0x01) {
5374 /* All the currently banned MACs happen to start with 01 currently, so
5375 * this is a quick way to eliminate most of the good ones. */
5377 if (eth_addr_is_reserved(dst)) {
5378 /* Drop STP, IEEE pause frames, and other reserved protocols
5379 * (01-80-c2-00-00-0x). */
5383 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
5385 if ((dst[3] & 0xfe) == 0xcc &&
5386 (dst[4] & 0xfe) == 0xcc &&
5387 (dst[5] & 0xfe) == 0xcc) {
5388 /* Drop the following protocols plus others following the same
5391 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
5392 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
5393 STP Uplink Fast (01-00-0c-cd-cd-cd) */
5397 if (!(dst[3] | dst[4] | dst[5])) {
5398 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
5407 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5409 struct ofproto_dpif *ofproto = ctx->ofproto;
5410 mirror_mask_t mirrors;
5411 struct ofbundle *in_bundle;
5414 const struct nlattr *a;
5417 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5418 ctx->packet != NULL);
5422 mirrors = in_bundle->src_mirrors;
5424 /* Drop frames on bundles reserved for mirroring. */
5425 if (in_bundle->mirror_out) {
5426 if (ctx->packet != NULL) {
5427 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5428 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5429 "%s, which is reserved exclusively for mirroring",
5430 ctx->ofproto->up.name, in_bundle->name);
5436 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5437 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5440 vlan = input_vid_to_vlan(in_bundle, vid);
5442 /* Look at the output ports to check for destination selections. */
5444 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5445 ctx->odp_actions->size) {
5446 enum ovs_action_attr type = nl_attr_type(a);
5447 struct ofport_dpif *ofport;
5449 if (type != OVS_ACTION_ATTR_OUTPUT) {
5453 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5454 if (ofport && ofport->bundle) {
5455 mirrors |= ofport->bundle->dst_mirrors;
5463 /* Restore the original packet before adding the mirror actions. */
5464 ctx->flow = *orig_flow;
5469 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5471 if (!vlan_is_mirrored(m, vlan)) {
5472 mirrors &= mirrors - 1;
5476 mirrors &= ~m->dup_mirrors;
5477 ctx->mirrors |= m->dup_mirrors;
5479 output_normal(ctx, m->out, vlan);
5480 } else if (eth_dst_may_rspan(orig_flow->dl_dst)
5481 && vlan != m->out_vlan) {
5482 struct ofbundle *bundle;
5484 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5485 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5486 && !bundle->mirror_out) {
5487 output_normal(ctx, bundle, m->out_vlan);
5495 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5496 uint64_t packets, uint64_t bytes)
5502 for (; mirrors; mirrors &= mirrors - 1) {
5505 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5508 /* In normal circumstances 'm' will not be NULL. However,
5509 * if mirrors are reconfigured, we can temporarily get out
5510 * of sync in facet_revalidate(). We could "correct" the
5511 * mirror list before reaching here, but doing that would
5512 * not properly account the traffic stats we've currently
5513 * accumulated for previous mirror configuration. */
5517 m->packet_count += packets;
5518 m->byte_count += bytes;
5522 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
5523 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
5524 * indicate this; newer upstream kernels use gratuitous ARP requests. */
5526 is_gratuitous_arp(const struct flow *flow)
5528 return (flow->dl_type == htons(ETH_TYPE_ARP)
5529 && eth_addr_is_broadcast(flow->dl_dst)
5530 && (flow->nw_proto == ARP_OP_REPLY
5531 || (flow->nw_proto == ARP_OP_REQUEST
5532 && flow->nw_src == flow->nw_dst)));
5536 update_learning_table(struct ofproto_dpif *ofproto,
5537 const struct flow *flow, int vlan,
5538 struct ofbundle *in_bundle)
5540 struct mac_entry *mac;
5542 /* Don't learn the OFPP_NONE port. */
5543 if (in_bundle == &ofpp_none_bundle) {
5547 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
5551 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
5552 if (is_gratuitous_arp(flow)) {
5553 /* We don't want to learn from gratuitous ARP packets that are
5554 * reflected back over bond slaves so we lock the learning table. */
5555 if (!in_bundle->bond) {
5556 mac_entry_set_grat_arp_lock(mac);
5557 } else if (mac_entry_is_grat_arp_locked(mac)) {
5562 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
5563 /* The log messages here could actually be useful in debugging,
5564 * so keep the rate limit relatively high. */
5565 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5566 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
5567 "on port %s in VLAN %d",
5568 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
5569 in_bundle->name, vlan);
5571 mac->port.p = in_bundle;
5572 tag_set_add(&ofproto->revalidate_set,
5573 mac_learning_changed(ofproto->ml, mac));
5577 static struct ofbundle *
5578 lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn)
5580 struct ofport_dpif *ofport;
5582 /* Special-case OFPP_NONE, which a controller may use as the ingress
5583 * port for traffic that it is sourcing. */
5584 if (in_port == OFPP_NONE) {
5585 return &ofpp_none_bundle;
5588 /* Find the port and bundle for the received packet. */
5589 ofport = get_ofp_port(ofproto, in_port);
5590 if (ofport && ofport->bundle) {
5591 return ofport->bundle;
5594 /* Odd. A few possible reasons here:
5596 * - We deleted a port but there are still a few packets queued up
5599 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
5600 * we don't know about.
5602 * - The ofproto client didn't configure the port as part of a bundle.
5605 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5607 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
5608 "port %"PRIu16, ofproto->up.name, in_port);
5613 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
5614 * dropped. Returns true if they may be forwarded, false if they should be
5617 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
5618 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
5620 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
5621 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
5622 * checked by input_vid_is_valid().
5624 * May also add tags to '*tags', although the current implementation only does
5625 * so in one special case.
5628 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
5629 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
5631 struct ofbundle *in_bundle = in_port->bundle;
5633 /* Drop frames for reserved multicast addresses
5634 * only if forward_bpdu option is absent. */
5635 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
5639 if (in_bundle->bond) {
5640 struct mac_entry *mac;
5642 switch (bond_check_admissibility(in_bundle->bond, in_port,
5643 flow->dl_dst, tags)) {
5650 case BV_DROP_IF_MOVED:
5651 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
5652 if (mac && mac->port.p != in_bundle &&
5653 (!is_gratuitous_arp(flow)
5654 || mac_entry_is_grat_arp_locked(mac))) {
5665 xlate_normal(struct action_xlate_ctx *ctx)
5667 struct ofport_dpif *in_port;
5668 struct ofbundle *in_bundle;
5669 struct mac_entry *mac;
5673 ctx->has_normal = true;
5675 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
5676 ctx->packet != NULL);
5681 /* We know 'in_port' exists unless it is "ofpp_none_bundle",
5682 * since lookup_input_bundle() succeeded. */
5683 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5685 /* Drop malformed frames. */
5686 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
5687 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
5688 if (ctx->packet != NULL) {
5689 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5690 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
5691 "VLAN tag received on port %s",
5692 ctx->ofproto->up.name, in_bundle->name);
5697 /* Drop frames on bundles reserved for mirroring. */
5698 if (in_bundle->mirror_out) {
5699 if (ctx->packet != NULL) {
5700 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5701 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5702 "%s, which is reserved exclusively for mirroring",
5703 ctx->ofproto->up.name, in_bundle->name);
5709 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
5710 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5713 vlan = input_vid_to_vlan(in_bundle, vid);
5715 /* Check other admissibility requirements. */
5717 !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
5721 /* Learn source MAC. */
5722 if (ctx->may_learn) {
5723 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
5726 /* Determine output bundle. */
5727 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
5730 if (mac->port.p != in_bundle) {
5731 output_normal(ctx, mac->port.p, vlan);
5734 struct ofbundle *bundle;
5736 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
5737 if (bundle != in_bundle
5738 && ofbundle_includes_vlan(bundle, vlan)
5739 && bundle->floodable
5740 && !bundle->mirror_out) {
5741 output_normal(ctx, bundle, vlan);
5744 ctx->nf_output_iface = NF_OUT_FLOOD;
5748 /* Optimized flow revalidation.
5750 * It's a difficult problem, in general, to tell which facets need to have
5751 * their actions recalculated whenever the OpenFlow flow table changes. We
5752 * don't try to solve that general problem: for most kinds of OpenFlow flow
5753 * table changes, we recalculate the actions for every facet. This is
5754 * relatively expensive, but it's good enough if the OpenFlow flow table
5755 * doesn't change very often.
5757 * However, we can expect one particular kind of OpenFlow flow table change to
5758 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5759 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5760 * table, we add a special case that applies to flow tables in which every rule
5761 * has the same form (that is, the same wildcards), except that the table is
5762 * also allowed to have a single "catch-all" flow that matches all packets. We
5763 * optimize this case by tagging all of the facets that resubmit into the table
5764 * and invalidating the same tag whenever a flow changes in that table. The
5765 * end result is that we revalidate just the facets that need it (and sometimes
5766 * a few more, but not all of the facets or even all of the facets that
5767 * resubmit to the table modified by MAC learning). */
5769 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
5770 * into an OpenFlow table with the given 'basis'. */
5772 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
5775 if (flow_wildcards_is_catchall(wc)) {
5778 struct flow tag_flow = *flow;
5779 flow_zero_wildcards(&tag_flow, wc);
5780 return tag_create_deterministic(flow_hash(&tag_flow, secret));
5784 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5785 * taggability of that table.
5787 * This function must be called after *each* change to a flow table. If you
5788 * skip calling it on some changes then the pointer comparisons at the end can
5789 * be invalid if you get unlucky. For example, if a flow removal causes a
5790 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5791 * different wildcards to be created with the same address, then this function
5792 * will incorrectly skip revalidation. */
5794 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5796 struct table_dpif *table = &ofproto->tables[table_id];
5797 const struct oftable *oftable = &ofproto->up.tables[table_id];
5798 struct cls_table *catchall, *other;
5799 struct cls_table *t;
5801 catchall = other = NULL;
5803 switch (hmap_count(&oftable->cls.tables)) {
5805 /* We could tag this OpenFlow table but it would make the logic a
5806 * little harder and it's a corner case that doesn't seem worth it
5812 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
5813 if (cls_table_is_catchall(t)) {
5815 } else if (!other) {
5818 /* Indicate that we can't tag this by setting both tables to
5819 * NULL. (We know that 'catchall' is already NULL.) */
5826 /* Can't tag this table. */
5830 if (table->catchall_table != catchall || table->other_table != other) {
5831 table->catchall_table = catchall;
5832 table->other_table = other;
5833 ofproto->need_revalidate = true;
5837 /* Given 'rule' that has changed in some way (either it is a rule being
5838 * inserted, a rule being deleted, or a rule whose actions are being
5839 * modified), marks facets for revalidation to ensure that packets will be
5840 * forwarded correctly according to the new state of the flow table.
5842 * This function must be called after *each* change to a flow table. See
5843 * the comment on table_update_taggable() for more information. */
5845 rule_invalidate(const struct rule_dpif *rule)
5847 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5849 table_update_taggable(ofproto, rule->up.table_id);
5851 if (!ofproto->need_revalidate) {
5852 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5854 if (table->other_table && rule->tag) {
5855 tag_set_add(&ofproto->revalidate_set, rule->tag);
5857 ofproto->need_revalidate = true;
5863 set_frag_handling(struct ofproto *ofproto_,
5864 enum ofp_config_flags frag_handling)
5866 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5868 if (frag_handling != OFPC_FRAG_REASM) {
5869 ofproto->need_revalidate = true;
5877 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5878 const struct flow *flow,
5879 const union ofp_action *ofp_actions, size_t n_ofp_actions)
5881 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5884 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
5885 return OFPERR_NXBRC_BAD_IN_PORT;
5888 error = validate_actions(ofp_actions, n_ofp_actions, flow,
5889 ofproto->max_ports);
5891 struct odputil_keybuf keybuf;
5892 struct ofpbuf *odp_actions;
5893 struct ofproto_push push;
5896 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5897 odp_flow_key_from_flow(&key, flow);
5899 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, NULL,
5900 packet_get_tcp_flags(packet, flow), packet);
5902 /* Ensure that resubmits in 'ofp_actions' get accounted to their
5903 * matching rules. */
5905 push.bytes = packet->size;
5906 push.used = time_msec();
5907 push.ctx.resubmit_hook = push_resubmit;
5909 odp_actions = xlate_actions(&push.ctx, ofp_actions, n_ofp_actions);
5910 dpif_execute(ofproto->dpif, key.data, key.size,
5911 odp_actions->data, odp_actions->size, packet);
5912 ofpbuf_delete(odp_actions);
5920 set_netflow(struct ofproto *ofproto_,
5921 const struct netflow_options *netflow_options)
5923 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5925 if (netflow_options) {
5926 if (!ofproto->netflow) {
5927 ofproto->netflow = netflow_create();
5929 return netflow_set_options(ofproto->netflow, netflow_options);
5931 netflow_destroy(ofproto->netflow);
5932 ofproto->netflow = NULL;
5938 get_netflow_ids(const struct ofproto *ofproto_,
5939 uint8_t *engine_type, uint8_t *engine_id)
5941 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5943 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
5947 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5949 if (!facet_is_controller_flow(facet) &&
5950 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5951 struct subfacet *subfacet;
5952 struct ofexpired expired;
5954 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5955 if (subfacet->installed) {
5956 struct dpif_flow_stats stats;
5958 subfacet_install(subfacet, subfacet->actions,
5959 subfacet->actions_len, &stats);
5960 subfacet_update_stats(subfacet, &stats);
5964 expired.flow = facet->flow;
5965 expired.packet_count = facet->packet_count;
5966 expired.byte_count = facet->byte_count;
5967 expired.used = facet->used;
5968 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5973 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5975 struct facet *facet;
5977 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5978 send_active_timeout(ofproto, facet);
5982 static struct ofproto_dpif *
5983 ofproto_dpif_lookup(const char *name)
5985 struct ofproto_dpif *ofproto;
5987 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5988 hash_string(name, 0), &all_ofproto_dpifs) {
5989 if (!strcmp(ofproto->up.name, name)) {
5997 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5998 const char *argv[], void *aux OVS_UNUSED)
6000 struct ofproto_dpif *ofproto;
6003 ofproto = ofproto_dpif_lookup(argv[1]);
6005 unixctl_command_reply_error(conn, "no such bridge");
6008 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6010 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6011 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6015 unixctl_command_reply(conn, "table successfully flushed");
6019 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6020 const char *argv[], void *aux OVS_UNUSED)
6022 struct ds ds = DS_EMPTY_INITIALIZER;
6023 const struct ofproto_dpif *ofproto;
6024 const struct mac_entry *e;
6026 ofproto = ofproto_dpif_lookup(argv[1]);
6028 unixctl_command_reply_error(conn, "no such bridge");
6032 ds_put_cstr(&ds, " port VLAN MAC Age\n");
6033 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
6034 struct ofbundle *bundle = e->port.p;
6035 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
6036 ofbundle_get_a_port(bundle)->odp_port,
6037 e->vlan, ETH_ADDR_ARGS(e->mac),
6038 mac_entry_age(ofproto->ml, e));
6040 unixctl_command_reply(conn, ds_cstr(&ds));
6045 struct action_xlate_ctx ctx;
6051 trace_format_rule(struct ds *result, uint8_t table_id, int level,
6052 const struct rule_dpif *rule)
6054 ds_put_char_multiple(result, '\t', level);
6056 ds_put_cstr(result, "No match\n");
6060 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
6061 table_id, ntohll(rule->up.flow_cookie));
6062 cls_rule_format(&rule->up.cr, result);
6063 ds_put_char(result, '\n');
6065 ds_put_char_multiple(result, '\t', level);
6066 ds_put_cstr(result, "OpenFlow ");
6067 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
6068 ds_put_char(result, '\n');
6072 trace_format_flow(struct ds *result, int level, const char *title,
6073 struct trace_ctx *trace)
6075 ds_put_char_multiple(result, '\t', level);
6076 ds_put_format(result, "%s: ", title);
6077 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
6078 ds_put_cstr(result, "unchanged");
6080 flow_format(result, &trace->ctx.flow);
6081 trace->flow = trace->ctx.flow;
6083 ds_put_char(result, '\n');
6087 trace_format_regs(struct ds *result, int level, const char *title,
6088 struct trace_ctx *trace)
6092 ds_put_char_multiple(result, '\t', level);
6093 ds_put_format(result, "%s:", title);
6094 for (i = 0; i < FLOW_N_REGS; i++) {
6095 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
6097 ds_put_char(result, '\n');
6101 trace_format_odp(struct ds *result, int level, const char *title,
6102 struct trace_ctx *trace)
6104 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
6106 ds_put_char_multiple(result, '\t', level);
6107 ds_put_format(result, "%s: ", title);
6108 format_odp_actions(result, odp_actions->data, odp_actions->size);
6109 ds_put_char(result, '\n');
6113 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
6115 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
6116 struct ds *result = trace->result;
6118 ds_put_char(result, '\n');
6119 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
6120 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
6121 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
6122 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
6126 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
6127 void *aux OVS_UNUSED)
6129 const char *dpname = argv[1];
6130 struct ofproto_dpif *ofproto;
6131 struct ofpbuf odp_key;
6132 struct ofpbuf *packet;
6133 ovs_be16 initial_tci;
6139 ofpbuf_init(&odp_key, 0);
6142 ofproto = ofproto_dpif_lookup(dpname);
6144 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
6148 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
6149 /* ofproto/trace dpname flow [-generate] */
6150 const char *flow_s = argv[2];
6151 const char *generate_s = argv[3];
6154 /* Convert string to datapath key. */
6155 ofpbuf_init(&odp_key, 0);
6156 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
6158 unixctl_command_reply_error(conn, "Bad flow syntax");
6162 /* Convert odp_key to flow. */
6163 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
6164 odp_key.size, &flow,
6165 &initial_tci, NULL);
6166 if (error == ODP_FIT_ERROR) {
6167 unixctl_command_reply_error(conn, "Invalid flow");
6171 /* Generate a packet, if requested. */
6173 packet = ofpbuf_new(0);
6174 flow_compose(packet, &flow);
6176 } else if (argc == 6) {
6177 /* ofproto/trace dpname priority tun_id in_port packet */
6178 const char *priority_s = argv[2];
6179 const char *tun_id_s = argv[3];
6180 const char *in_port_s = argv[4];
6181 const char *packet_s = argv[5];
6182 uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s));
6183 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
6184 uint32_t priority = atoi(priority_s);
6187 msg = eth_from_hex(packet_s, &packet);
6189 unixctl_command_reply_error(conn, msg);
6193 ds_put_cstr(&result, "Packet: ");
6194 s = ofp_packet_to_string(packet->data, packet->size);
6195 ds_put_cstr(&result, s);
6198 flow_extract(packet, priority, tun_id, in_port, &flow);
6199 initial_tci = flow.vlan_tci;
6201 unixctl_command_reply_error(conn, "Bad command syntax");
6205 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
6206 unixctl_command_reply(conn, ds_cstr(&result));
6209 ds_destroy(&result);
6210 ofpbuf_delete(packet);
6211 ofpbuf_uninit(&odp_key);
6215 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
6216 const struct ofpbuf *packet, ovs_be16 initial_tci,
6219 struct rule_dpif *rule;
6221 ds_put_cstr(ds, "Flow: ");
6222 flow_format(ds, flow);
6223 ds_put_char(ds, '\n');
6225 rule = rule_dpif_lookup(ofproto, flow, 0);
6226 trace_format_rule(ds, 0, 0, rule);
6228 struct trace_ctx trace;
6229 struct ofpbuf *odp_actions;
6232 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
6235 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
6236 rule, tcp_flags, packet);
6237 trace.ctx.resubmit_hook = trace_resubmit;
6238 odp_actions = xlate_actions(&trace.ctx,
6239 rule->up.actions, rule->up.n_actions);
6241 ds_put_char(ds, '\n');
6242 trace_format_flow(ds, 0, "Final flow", &trace);
6243 ds_put_cstr(ds, "Datapath actions: ");
6244 format_odp_actions(ds, odp_actions->data, odp_actions->size);
6245 ofpbuf_delete(odp_actions);
6247 if (!trace.ctx.may_set_up_flow) {
6249 ds_put_cstr(ds, "\nThis flow is not cachable.");
6251 ds_put_cstr(ds, "\nThe datapath actions are incomplete--"
6252 "for complete actions, please supply a packet.");
6259 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6260 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6263 unixctl_command_reply(conn, NULL);
6267 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6268 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6271 unixctl_command_reply(conn, NULL);
6274 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
6275 * 'reply' describing the results. */
6277 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
6279 struct facet *facet;
6283 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6284 if (!facet_check_consistency(facet)) {
6289 ofproto->need_revalidate = true;
6293 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
6294 ofproto->up.name, errors);
6296 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
6301 ofproto_dpif_self_check(struct unixctl_conn *conn,
6302 int argc, const char *argv[], void *aux OVS_UNUSED)
6304 struct ds reply = DS_EMPTY_INITIALIZER;
6305 struct ofproto_dpif *ofproto;
6308 ofproto = ofproto_dpif_lookup(argv[1]);
6310 unixctl_command_reply_error(conn, "Unknown ofproto (use "
6311 "ofproto/list for help)");
6314 ofproto_dpif_self_check__(ofproto, &reply);
6316 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6317 ofproto_dpif_self_check__(ofproto, &reply);
6321 unixctl_command_reply(conn, ds_cstr(&reply));
6326 ofproto_dpif_unixctl_init(void)
6328 static bool registered;
6334 unixctl_command_register(
6336 "bridge {tun_id in_port packet | odp_flow [-generate]}",
6337 2, 5, ofproto_unixctl_trace, NULL);
6338 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6339 ofproto_unixctl_fdb_flush, NULL);
6340 unixctl_command_register("fdb/show", "bridge", 1, 1,
6341 ofproto_unixctl_fdb_show, NULL);
6342 unixctl_command_register("ofproto/clog", "", 0, 0,
6343 ofproto_dpif_clog, NULL);
6344 unixctl_command_register("ofproto/unclog", "", 0, 0,
6345 ofproto_dpif_unclog, NULL);
6346 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6347 ofproto_dpif_self_check, NULL);
6350 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6352 * This is deprecated. It is only for compatibility with broken device drivers
6353 * in old versions of Linux that do not properly support VLANs when VLAN
6354 * devices are not used. When broken device drivers are no longer in
6355 * widespread use, we will delete these interfaces. */
6358 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
6360 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6361 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6363 if (realdev_ofp_port == ofport->realdev_ofp_port
6364 && vid == ofport->vlandev_vid) {
6368 ofproto->need_revalidate = true;
6370 if (ofport->realdev_ofp_port) {
6373 if (realdev_ofp_port && ofport->bundle) {
6374 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6375 * themselves be part of a bundle. */
6376 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6379 ofport->realdev_ofp_port = realdev_ofp_port;
6380 ofport->vlandev_vid = vid;
6382 if (realdev_ofp_port) {
6383 vsp_add(ofport, realdev_ofp_port, vid);
6390 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
6392 return hash_2words(realdev_ofp_port, vid);
6395 /* Returns the ODP port number of the Linux VLAN device that corresponds to
6396 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
6397 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
6398 * it would return the port number of eth0.9.
6400 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
6401 * function just returns its 'realdev_odp_port' argument. */
6403 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6404 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
6406 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6407 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
6408 int vid = vlan_tci_to_vid(vlan_tci);
6409 const struct vlan_splinter *vsp;
6411 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6412 hash_realdev_vid(realdev_ofp_port, vid),
6413 &ofproto->realdev_vid_map) {
6414 if (vsp->realdev_ofp_port == realdev_ofp_port
6415 && vsp->vid == vid) {
6416 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
6420 return realdev_odp_port;
6423 static struct vlan_splinter *
6424 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
6426 struct vlan_splinter *vsp;
6428 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
6429 &ofproto->vlandev_map) {
6430 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6438 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6439 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6440 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6441 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6442 * eth0 and store 9 in '*vid'.
6444 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6445 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6448 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6449 uint16_t vlandev_ofp_port, int *vid)
6451 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6452 const struct vlan_splinter *vsp;
6454 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6459 return vsp->realdev_ofp_port;
6466 vsp_remove(struct ofport_dpif *port)
6468 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6469 struct vlan_splinter *vsp;
6471 vsp = vlandev_find(ofproto, port->up.ofp_port);
6473 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6474 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6477 port->realdev_ofp_port = 0;
6479 VLOG_ERR("missing vlan device record");
6484 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
6486 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6488 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6489 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6490 == realdev_ofp_port)) {
6491 struct vlan_splinter *vsp;
6493 vsp = xmalloc(sizeof *vsp);
6494 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6495 hash_int(port->up.ofp_port, 0));
6496 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6497 hash_realdev_vid(realdev_ofp_port, vid));
6498 vsp->realdev_ofp_port = realdev_ofp_port;
6499 vsp->vlandev_ofp_port = port->up.ofp_port;
6502 port->realdev_ofp_port = realdev_ofp_port;
6504 VLOG_ERR("duplicate vlan device record");
6508 const struct ofproto_class ofproto_dpif_class = {
6537 port_is_lacp_current,
6538 NULL, /* rule_choose_table */
6545 rule_modify_actions,
6553 get_cfm_remote_mpids,
6558 get_stp_port_status,
6565 is_mirror_output_bundle,
6566 forward_bpdu_changed,