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 void xlate_actions(struct action_xlate_ctx *,
268 const union ofp_action *in, size_t n_in,
269 struct ofpbuf *odp_actions);
270 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
271 const union ofp_action *in,
274 /* A dpif flow and actions associated with a facet.
276 * See also the large comment on struct facet. */
279 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
280 struct list list_node; /* In struct facet's 'facets' list. */
281 struct facet *facet; /* Owning facet. */
285 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
286 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
287 * regenerate the ODP flow key from ->facet->flow. */
288 enum odp_key_fitness key_fitness;
292 long long int used; /* Time last used; time created if not used. */
294 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
295 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
299 * These should be essentially identical for every subfacet in a facet, but
300 * may differ in trivial ways due to VLAN splinters. */
301 size_t actions_len; /* Number of bytes in actions[]. */
302 struct nlattr *actions; /* Datapath actions. */
304 bool installed; /* Installed in datapath? */
306 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
307 * splinters can cause it to differ. This value should be removed when
308 * the VLAN splinters feature is no longer needed. */
309 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
312 static struct subfacet *subfacet_create(struct facet *, enum odp_key_fitness,
313 const struct nlattr *key,
314 size_t key_len, ovs_be16 initial_tci);
315 static struct subfacet *subfacet_find(struct ofproto_dpif *,
316 const struct nlattr *key, size_t key_len);
317 static void subfacet_destroy(struct subfacet *);
318 static void subfacet_destroy__(struct subfacet *);
319 static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *,
321 static void subfacet_reset_dp_stats(struct subfacet *,
322 struct dpif_flow_stats *);
323 static void subfacet_update_time(struct subfacet *, long long int used);
324 static void subfacet_update_stats(struct subfacet *,
325 const struct dpif_flow_stats *);
326 static void subfacet_make_actions(struct subfacet *,
327 const struct ofpbuf *packet);
328 static int subfacet_install(struct subfacet *,
329 const struct nlattr *actions, size_t actions_len,
330 struct dpif_flow_stats *);
331 static void subfacet_uninstall(struct subfacet *);
333 /* An exact-match instantiation of an OpenFlow flow.
335 * A facet associates a "struct flow", which represents the Open vSwitch
336 * userspace idea of an exact-match flow, with one or more subfacets. Each
337 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
338 * the facet. When the kernel module (or other dpif implementation) and Open
339 * vSwitch userspace agree on the definition of a flow key, there is exactly
340 * one subfacet per facet. If the dpif implementation supports more-specific
341 * flow matching than userspace, however, a facet can have more than one
342 * subfacet, each of which corresponds to some distinction in flow that
343 * userspace simply doesn't understand.
345 * Flow expiration works in terms of subfacets, so a facet must have at least
346 * one subfacet or it will never expire, leaking memory. */
349 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
350 struct list list_node; /* In owning rule's 'facets' list. */
351 struct rule_dpif *rule; /* Owning rule. */
354 struct list subfacets;
355 long long int used; /* Time last used; time created if not used. */
362 * - Do include packets and bytes sent "by hand", e.g. with
365 * - Do include packets and bytes that were obtained from the datapath
366 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
367 * DPIF_FP_ZERO_STATS).
369 * - Do not include packets or bytes that can be obtained from the
370 * datapath for any existing subfacet.
372 uint64_t packet_count; /* Number of packets received. */
373 uint64_t byte_count; /* Number of bytes received. */
375 /* Resubmit statistics. */
376 uint64_t prev_packet_count; /* Number of packets from last stats push. */
377 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
378 long long int prev_used; /* Used time from last stats push. */
381 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
382 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
383 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
385 /* Properties of datapath actions.
387 * Every subfacet has its own actions because actions can differ slightly
388 * between splintered and non-splintered subfacets due to the VLAN tag
389 * being initially different (present vs. absent). All of them have these
390 * properties in common so we just store one copy of them here. */
391 bool may_install; /* Reassess actions for every packet? */
392 bool has_learn; /* Actions include NXAST_LEARN? */
393 bool has_normal; /* Actions output to OFPP_NORMAL? */
394 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
395 tag_type tags; /* Tags that would require revalidation. */
396 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
398 /* Storage for a single subfacet, to reduce malloc() time and space
399 * overhead. (A facet always has at least one subfacet and in the common
400 * case has exactly one subfacet.) */
401 struct subfacet one_subfacet;
404 static struct facet *facet_create(struct rule_dpif *,
405 const struct flow *, uint32_t hash);
406 static void facet_remove(struct facet *);
407 static void facet_free(struct facet *);
409 static struct facet *facet_find(struct ofproto_dpif *,
410 const struct flow *, uint32_t hash);
411 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
412 const struct flow *, uint32_t hash);
413 static bool facet_revalidate(struct facet *);
414 static bool facet_check_consistency(struct facet *);
416 static void facet_flush_stats(struct facet *);
418 static void facet_update_time(struct facet *, long long int used);
419 static void facet_reset_counters(struct facet *);
420 static void facet_push_stats(struct facet *);
421 static void facet_learn(struct facet *);
422 static void facet_account(struct facet *);
424 static bool facet_is_controller_flow(struct facet *);
430 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
431 struct list bundle_node; /* In struct ofbundle's "ports" list. */
432 struct cfm *cfm; /* Connectivity Fault Management, if any. */
433 tag_type tag; /* Tag associated with this port. */
434 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
435 bool may_enable; /* May be enabled in bonds. */
436 long long int carrier_seq; /* Carrier status changes. */
439 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
440 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
441 long long int stp_state_entered;
443 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
445 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
447 * This is deprecated. It is only for compatibility with broken device
448 * drivers in old versions of Linux that do not properly support VLANs when
449 * VLAN devices are not used. When broken device drivers are no longer in
450 * widespread use, we will delete these interfaces. */
451 uint16_t realdev_ofp_port;
455 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
456 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
457 * traffic egressing the 'ofport' with that priority should be marked with. */
458 struct priority_to_dscp {
459 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
460 uint32_t priority; /* Priority of this queue (see struct flow). */
462 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
465 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
467 * This is deprecated. It is only for compatibility with broken device drivers
468 * in old versions of Linux that do not properly support VLANs when VLAN
469 * devices are not used. When broken device drivers are no longer in
470 * widespread use, we will delete these interfaces. */
471 struct vlan_splinter {
472 struct hmap_node realdev_vid_node;
473 struct hmap_node vlandev_node;
474 uint16_t realdev_ofp_port;
475 uint16_t vlandev_ofp_port;
479 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
480 uint32_t realdev, ovs_be16 vlan_tci);
481 static uint16_t vsp_vlandev_to_realdev(const struct ofproto_dpif *,
482 uint16_t vlandev, int *vid);
483 static void vsp_remove(struct ofport_dpif *);
484 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
486 static struct ofport_dpif *
487 ofport_dpif_cast(const struct ofport *ofport)
489 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
490 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
493 static void port_run(struct ofport_dpif *);
494 static void port_wait(struct ofport_dpif *);
495 static int set_cfm(struct ofport *, const struct cfm_settings *);
496 static void ofport_clear_priorities(struct ofport_dpif *);
498 struct dpif_completion {
499 struct list list_node;
500 struct ofoperation *op;
503 /* Extra information about a classifier table.
504 * Currently used just for optimized flow revalidation. */
506 /* If either of these is nonnull, then this table has a form that allows
507 * flows to be tagged to avoid revalidating most flows for the most common
508 * kinds of flow table changes. */
509 struct cls_table *catchall_table; /* Table that wildcards all fields. */
510 struct cls_table *other_table; /* Table with any other wildcard set. */
511 uint32_t basis; /* Keeps each table's tags separate. */
514 struct ofproto_dpif {
515 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
524 struct netflow *netflow;
525 struct dpif_sflow *sflow;
526 struct hmap bundles; /* Contains "struct ofbundle"s. */
527 struct mac_learning *ml;
528 struct ofmirror *mirrors[MAX_MIRRORS];
529 bool has_bonded_bundles;
532 struct timer next_expiration;
536 struct hmap subfacets;
539 struct table_dpif tables[N_TABLES];
540 bool need_revalidate;
541 struct tag_set revalidate_set;
543 /* Support for debugging async flow mods. */
544 struct list completions;
546 bool has_bundle_action; /* True when the first bundle action appears. */
547 struct netdev_stats stats; /* To account packets generated and consumed in
552 long long int stp_last_tick;
554 /* VLAN splinters. */
555 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
556 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
559 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
560 * for debugging the asynchronous flow_mod implementation.) */
563 /* All existing ofproto_dpif instances, indexed by ->up.name. */
564 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
566 static void ofproto_dpif_unixctl_init(void);
568 static struct ofproto_dpif *
569 ofproto_dpif_cast(const struct ofproto *ofproto)
571 assert(ofproto->ofproto_class == &ofproto_dpif_class);
572 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
575 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
577 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
579 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
580 const struct ofpbuf *, ovs_be16 initial_tci,
583 /* Packet processing. */
584 static void update_learning_table(struct ofproto_dpif *,
585 const struct flow *, int vlan,
588 #define FLOW_MISS_MAX_BATCH 50
589 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
591 /* Flow expiration. */
592 static int expire(struct ofproto_dpif *);
595 static void send_netflow_active_timeouts(struct ofproto_dpif *);
598 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
600 compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions,
601 const struct flow *, uint32_t odp_port);
602 static void add_mirror_actions(struct action_xlate_ctx *ctx,
603 const struct flow *flow);
604 /* Global variables. */
605 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
607 /* Factory functions. */
610 enumerate_types(struct sset *types)
612 dp_enumerate_types(types);
616 enumerate_names(const char *type, struct sset *names)
618 return dp_enumerate_names(type, names);
622 del(const char *type, const char *name)
627 error = dpif_open(name, type, &dpif);
629 error = dpif_delete(dpif);
635 /* Basic life-cycle. */
637 static struct ofproto *
640 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
645 dealloc(struct ofproto *ofproto_)
647 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
652 construct(struct ofproto *ofproto_)
654 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
655 const char *name = ofproto->up.name;
659 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
661 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
665 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
666 ofproto->n_matches = 0;
668 dpif_flow_flush(ofproto->dpif);
669 dpif_recv_purge(ofproto->dpif);
671 error = dpif_recv_set(ofproto->dpif, true);
673 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
674 dpif_close(ofproto->dpif);
678 ofproto->netflow = NULL;
679 ofproto->sflow = NULL;
681 hmap_init(&ofproto->bundles);
682 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
683 for (i = 0; i < MAX_MIRRORS; i++) {
684 ofproto->mirrors[i] = NULL;
686 ofproto->has_bonded_bundles = false;
688 timer_set_duration(&ofproto->next_expiration, 1000);
690 hmap_init(&ofproto->facets);
691 hmap_init(&ofproto->subfacets);
693 for (i = 0; i < N_TABLES; i++) {
694 struct table_dpif *table = &ofproto->tables[i];
696 table->catchall_table = NULL;
697 table->other_table = NULL;
698 table->basis = random_uint32();
700 ofproto->need_revalidate = false;
701 tag_set_init(&ofproto->revalidate_set);
703 list_init(&ofproto->completions);
705 ofproto_dpif_unixctl_init();
707 ofproto->has_bundle_action = false;
709 hmap_init(&ofproto->vlandev_map);
710 hmap_init(&ofproto->realdev_vid_map);
712 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
713 hash_string(ofproto->up.name, 0));
714 memset(&ofproto->stats, 0, sizeof ofproto->stats);
716 ofproto_init_tables(ofproto_, N_TABLES);
722 complete_operations(struct ofproto_dpif *ofproto)
724 struct dpif_completion *c, *next;
726 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
727 ofoperation_complete(c->op, 0);
728 list_remove(&c->list_node);
734 destruct(struct ofproto *ofproto_)
736 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
737 struct rule_dpif *rule, *next_rule;
738 struct oftable *table;
741 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
742 complete_operations(ofproto);
744 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
745 struct cls_cursor cursor;
747 cls_cursor_init(&cursor, &table->cls, NULL);
748 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
749 ofproto_rule_destroy(&rule->up);
753 for (i = 0; i < MAX_MIRRORS; i++) {
754 mirror_destroy(ofproto->mirrors[i]);
757 netflow_destroy(ofproto->netflow);
758 dpif_sflow_destroy(ofproto->sflow);
759 hmap_destroy(&ofproto->bundles);
760 mac_learning_destroy(ofproto->ml);
762 hmap_destroy(&ofproto->facets);
763 hmap_destroy(&ofproto->subfacets);
765 hmap_destroy(&ofproto->vlandev_map);
766 hmap_destroy(&ofproto->realdev_vid_map);
768 dpif_close(ofproto->dpif);
772 run_fast(struct ofproto *ofproto_)
774 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
777 /* Handle one or more batches of upcalls, until there's nothing left to do
778 * or until we do a fixed total amount of work.
780 * We do work in batches because it can be much cheaper to set up a number
781 * of flows and fire off their patches all at once. We do multiple batches
782 * because in some cases handling a packet can cause another packet to be
783 * queued almost immediately as part of the return flow. Both
784 * optimizations can make major improvements on some benchmarks and
785 * presumably for real traffic as well. */
787 while (work < FLOW_MISS_MAX_BATCH) {
788 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
798 run(struct ofproto *ofproto_)
800 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
801 struct ofport_dpif *ofport;
802 struct ofbundle *bundle;
806 complete_operations(ofproto);
808 dpif_run(ofproto->dpif);
810 error = run_fast(ofproto_);
815 if (timer_expired(&ofproto->next_expiration)) {
816 int delay = expire(ofproto);
817 timer_set_duration(&ofproto->next_expiration, delay);
820 if (ofproto->netflow) {
821 if (netflow_run(ofproto->netflow)) {
822 send_netflow_active_timeouts(ofproto);
825 if (ofproto->sflow) {
826 dpif_sflow_run(ofproto->sflow);
829 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
832 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
837 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
839 /* Now revalidate if there's anything to do. */
840 if (ofproto->need_revalidate
841 || !tag_set_is_empty(&ofproto->revalidate_set)) {
842 struct tag_set revalidate_set = ofproto->revalidate_set;
843 bool revalidate_all = ofproto->need_revalidate;
844 struct facet *facet, *next;
846 /* Clear the revalidation flags. */
847 tag_set_init(&ofproto->revalidate_set);
848 ofproto->need_revalidate = false;
850 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
852 || tag_set_intersects(&revalidate_set, facet->tags)) {
853 facet_revalidate(facet);
858 /* Check the consistency of a random facet, to aid debugging. */
859 if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) {
862 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
863 struct facet, hmap_node);
864 if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) {
865 if (!facet_check_consistency(facet)) {
866 ofproto->need_revalidate = true;
875 wait(struct ofproto *ofproto_)
877 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
878 struct ofport_dpif *ofport;
879 struct ofbundle *bundle;
881 if (!clogged && !list_is_empty(&ofproto->completions)) {
882 poll_immediate_wake();
885 dpif_wait(ofproto->dpif);
886 dpif_recv_wait(ofproto->dpif);
887 if (ofproto->sflow) {
888 dpif_sflow_wait(ofproto->sflow);
890 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
891 poll_immediate_wake();
893 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
896 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
899 if (ofproto->netflow) {
900 netflow_wait(ofproto->netflow);
902 mac_learning_wait(ofproto->ml);
904 if (ofproto->need_revalidate) {
905 /* Shouldn't happen, but if it does just go around again. */
906 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
907 poll_immediate_wake();
909 timer_wait(&ofproto->next_expiration);
914 flush(struct ofproto *ofproto_)
916 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
917 struct facet *facet, *next_facet;
919 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
920 /* Mark the facet as not installed so that facet_remove() doesn't
921 * bother trying to uninstall it. There is no point in uninstalling it
922 * individually since we are about to blow away all the facets with
923 * dpif_flow_flush(). */
924 struct subfacet *subfacet;
926 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
927 subfacet->installed = false;
928 subfacet->dp_packet_count = 0;
929 subfacet->dp_byte_count = 0;
933 dpif_flow_flush(ofproto->dpif);
937 get_features(struct ofproto *ofproto_ OVS_UNUSED,
938 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
940 *arp_match_ip = true;
941 *actions = (OFPUTIL_A_OUTPUT |
942 OFPUTIL_A_SET_VLAN_VID |
943 OFPUTIL_A_SET_VLAN_PCP |
944 OFPUTIL_A_STRIP_VLAN |
945 OFPUTIL_A_SET_DL_SRC |
946 OFPUTIL_A_SET_DL_DST |
947 OFPUTIL_A_SET_NW_SRC |
948 OFPUTIL_A_SET_NW_DST |
949 OFPUTIL_A_SET_NW_TOS |
950 OFPUTIL_A_SET_TP_SRC |
951 OFPUTIL_A_SET_TP_DST |
956 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
958 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
959 struct dpif_dp_stats s;
961 strcpy(ots->name, "classifier");
963 dpif_get_dp_stats(ofproto->dpif, &s);
964 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
965 put_32aligned_be64(&ots->matched_count,
966 htonll(s.n_hit + ofproto->n_matches));
969 static struct ofport *
972 struct ofport_dpif *port = xmalloc(sizeof *port);
977 port_dealloc(struct ofport *port_)
979 struct ofport_dpif *port = ofport_dpif_cast(port_);
984 port_construct(struct ofport *port_)
986 struct ofport_dpif *port = ofport_dpif_cast(port_);
987 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
989 ofproto->need_revalidate = true;
990 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
993 port->tag = tag_create_random();
994 port->may_enable = true;
995 port->stp_port = NULL;
996 port->stp_state = STP_DISABLED;
997 hmap_init(&port->priorities);
998 port->realdev_ofp_port = 0;
999 port->vlandev_vid = 0;
1000 port->carrier_seq = netdev_get_carrier_resets(port->up.netdev);
1002 if (ofproto->sflow) {
1003 dpif_sflow_add_port(ofproto->sflow, port_);
1010 port_destruct(struct ofport *port_)
1012 struct ofport_dpif *port = ofport_dpif_cast(port_);
1013 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1015 ofproto->need_revalidate = true;
1016 bundle_remove(port_);
1017 set_cfm(port_, NULL);
1018 if (ofproto->sflow) {
1019 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1022 ofport_clear_priorities(port);
1023 hmap_destroy(&port->priorities);
1027 port_modified(struct ofport *port_)
1029 struct ofport_dpif *port = ofport_dpif_cast(port_);
1031 if (port->bundle && port->bundle->bond) {
1032 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1037 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1039 struct ofport_dpif *port = ofport_dpif_cast(port_);
1040 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1041 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1043 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1044 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD)) {
1045 ofproto->need_revalidate = true;
1047 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1048 bundle_update(port->bundle);
1054 set_sflow(struct ofproto *ofproto_,
1055 const struct ofproto_sflow_options *sflow_options)
1057 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1058 struct dpif_sflow *ds = ofproto->sflow;
1060 if (sflow_options) {
1062 struct ofport_dpif *ofport;
1064 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1065 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1066 dpif_sflow_add_port(ds, &ofport->up);
1068 ofproto->need_revalidate = true;
1070 dpif_sflow_set_options(ds, sflow_options);
1073 dpif_sflow_destroy(ds);
1074 ofproto->need_revalidate = true;
1075 ofproto->sflow = NULL;
1082 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1084 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1091 struct ofproto_dpif *ofproto;
1093 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1094 ofproto->need_revalidate = true;
1095 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1098 if (cfm_configure(ofport->cfm, s)) {
1104 cfm_destroy(ofport->cfm);
1110 get_cfm_fault(const struct ofport *ofport_)
1112 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1114 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1118 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1121 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1124 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1132 get_cfm_health(const struct ofport *ofport_)
1134 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1136 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1139 /* Spanning Tree. */
1142 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1144 struct ofproto_dpif *ofproto = ofproto_;
1145 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1146 struct ofport_dpif *ofport;
1148 ofport = stp_port_get_aux(sp);
1150 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1151 ofproto->up.name, port_num);
1153 struct eth_header *eth = pkt->l2;
1155 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1156 if (eth_addr_is_zero(eth->eth_src)) {
1157 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1158 "with unknown MAC", ofproto->up.name, port_num);
1160 send_packet(ofport, pkt);
1166 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1168 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1170 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1172 /* Only revalidate flows if the configuration changed. */
1173 if (!s != !ofproto->stp) {
1174 ofproto->need_revalidate = true;
1178 if (!ofproto->stp) {
1179 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1180 send_bpdu_cb, ofproto);
1181 ofproto->stp_last_tick = time_msec();
1184 stp_set_bridge_id(ofproto->stp, s->system_id);
1185 stp_set_bridge_priority(ofproto->stp, s->priority);
1186 stp_set_hello_time(ofproto->stp, s->hello_time);
1187 stp_set_max_age(ofproto->stp, s->max_age);
1188 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1190 struct ofport *ofport;
1192 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1193 set_stp_port(ofport, NULL);
1196 stp_destroy(ofproto->stp);
1197 ofproto->stp = NULL;
1204 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1206 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1210 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1211 s->designated_root = stp_get_designated_root(ofproto->stp);
1212 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1221 update_stp_port_state(struct ofport_dpif *ofport)
1223 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1224 enum stp_state state;
1226 /* Figure out new state. */
1227 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1231 if (ofport->stp_state != state) {
1232 enum ofputil_port_state of_state;
1235 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1236 netdev_get_name(ofport->up.netdev),
1237 stp_state_name(ofport->stp_state),
1238 stp_state_name(state));
1239 if (stp_learn_in_state(ofport->stp_state)
1240 != stp_learn_in_state(state)) {
1241 /* xxx Learning action flows should also be flushed. */
1242 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1244 fwd_change = stp_forward_in_state(ofport->stp_state)
1245 != stp_forward_in_state(state);
1247 ofproto->need_revalidate = true;
1248 ofport->stp_state = state;
1249 ofport->stp_state_entered = time_msec();
1251 if (fwd_change && ofport->bundle) {
1252 bundle_update(ofport->bundle);
1255 /* Update the STP state bits in the OpenFlow port description. */
1256 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1257 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1258 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1259 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1260 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1262 ofproto_port_set_state(&ofport->up, of_state);
1266 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1267 * caller is responsible for assigning STP port numbers and ensuring
1268 * there are no duplicates. */
1270 set_stp_port(struct ofport *ofport_,
1271 const struct ofproto_port_stp_settings *s)
1273 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1274 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1275 struct stp_port *sp = ofport->stp_port;
1277 if (!s || !s->enable) {
1279 ofport->stp_port = NULL;
1280 stp_port_disable(sp);
1281 update_stp_port_state(ofport);
1284 } else if (sp && stp_port_no(sp) != s->port_num
1285 && ofport == stp_port_get_aux(sp)) {
1286 /* The port-id changed, so disable the old one if it's not
1287 * already in use by another port. */
1288 stp_port_disable(sp);
1291 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1292 stp_port_enable(sp);
1294 stp_port_set_aux(sp, ofport);
1295 stp_port_set_priority(sp, s->priority);
1296 stp_port_set_path_cost(sp, s->path_cost);
1298 update_stp_port_state(ofport);
1304 get_stp_port_status(struct ofport *ofport_,
1305 struct ofproto_port_stp_status *s)
1307 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1308 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1309 struct stp_port *sp = ofport->stp_port;
1311 if (!ofproto->stp || !sp) {
1317 s->port_id = stp_port_get_id(sp);
1318 s->state = stp_port_get_state(sp);
1319 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1320 s->role = stp_port_get_role(sp);
1321 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1327 stp_run(struct ofproto_dpif *ofproto)
1330 long long int now = time_msec();
1331 long long int elapsed = now - ofproto->stp_last_tick;
1332 struct stp_port *sp;
1335 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1336 ofproto->stp_last_tick = now;
1338 while (stp_get_changed_port(ofproto->stp, &sp)) {
1339 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1342 update_stp_port_state(ofport);
1346 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1347 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1353 stp_wait(struct ofproto_dpif *ofproto)
1356 poll_timer_wait(1000);
1360 /* Returns true if STP should process 'flow'. */
1362 stp_should_process_flow(const struct flow *flow)
1364 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1368 stp_process_packet(const struct ofport_dpif *ofport,
1369 const struct ofpbuf *packet)
1371 struct ofpbuf payload = *packet;
1372 struct eth_header *eth = payload.data;
1373 struct stp_port *sp = ofport->stp_port;
1375 /* Sink packets on ports that have STP disabled when the bridge has
1377 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1381 /* Trim off padding on payload. */
1382 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1383 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1386 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1387 stp_received_bpdu(sp, payload.data, payload.size);
1391 static struct priority_to_dscp *
1392 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1394 struct priority_to_dscp *pdscp;
1397 hash = hash_int(priority, 0);
1398 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1399 if (pdscp->priority == priority) {
1407 ofport_clear_priorities(struct ofport_dpif *ofport)
1409 struct priority_to_dscp *pdscp, *next;
1411 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1412 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1418 set_queues(struct ofport *ofport_,
1419 const struct ofproto_port_queue *qdscp_list,
1422 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1423 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1424 struct hmap new = HMAP_INITIALIZER(&new);
1427 for (i = 0; i < n_qdscp; i++) {
1428 struct priority_to_dscp *pdscp;
1432 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1433 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1438 pdscp = get_priority(ofport, priority);
1440 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1442 pdscp = xmalloc(sizeof *pdscp);
1443 pdscp->priority = priority;
1445 ofproto->need_revalidate = true;
1448 if (pdscp->dscp != dscp) {
1450 ofproto->need_revalidate = true;
1453 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1456 if (!hmap_is_empty(&ofport->priorities)) {
1457 ofport_clear_priorities(ofport);
1458 ofproto->need_revalidate = true;
1461 hmap_swap(&new, &ofport->priorities);
1469 /* Expires all MAC learning entries associated with 'bundle' and forces its
1470 * ofproto to revalidate every flow.
1472 * Normally MAC learning entries are removed only from the ofproto associated
1473 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1474 * are removed from every ofproto. When patch ports and SLB bonds are in use
1475 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1476 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1477 * with the host from which it migrated. */
1479 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1481 struct ofproto_dpif *ofproto = bundle->ofproto;
1482 struct mac_learning *ml = ofproto->ml;
1483 struct mac_entry *mac, *next_mac;
1485 ofproto->need_revalidate = true;
1486 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1487 if (mac->port.p == bundle) {
1489 struct ofproto_dpif *o;
1491 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1493 struct mac_entry *e;
1495 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1498 tag_set_add(&o->revalidate_set, e->tag);
1499 mac_learning_expire(o->ml, e);
1505 mac_learning_expire(ml, mac);
1510 static struct ofbundle *
1511 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1513 struct ofbundle *bundle;
1515 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1516 &ofproto->bundles) {
1517 if (bundle->aux == aux) {
1524 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1525 * ones that are found to 'bundles'. */
1527 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1528 void **auxes, size_t n_auxes,
1529 struct hmapx *bundles)
1533 hmapx_init(bundles);
1534 for (i = 0; i < n_auxes; i++) {
1535 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1537 hmapx_add(bundles, bundle);
1543 bundle_update(struct ofbundle *bundle)
1545 struct ofport_dpif *port;
1547 bundle->floodable = true;
1548 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1549 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1550 || !stp_forward_in_state(port->stp_state)) {
1551 bundle->floodable = false;
1558 bundle_del_port(struct ofport_dpif *port)
1560 struct ofbundle *bundle = port->bundle;
1562 bundle->ofproto->need_revalidate = true;
1564 list_remove(&port->bundle_node);
1565 port->bundle = NULL;
1568 lacp_slave_unregister(bundle->lacp, port);
1571 bond_slave_unregister(bundle->bond, port);
1574 bundle_update(bundle);
1578 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1579 struct lacp_slave_settings *lacp,
1580 uint32_t bond_stable_id)
1582 struct ofport_dpif *port;
1584 port = get_ofp_port(bundle->ofproto, ofp_port);
1589 if (port->bundle != bundle) {
1590 bundle->ofproto->need_revalidate = true;
1592 bundle_del_port(port);
1595 port->bundle = bundle;
1596 list_push_back(&bundle->ports, &port->bundle_node);
1597 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1598 || !stp_forward_in_state(port->stp_state)) {
1599 bundle->floodable = false;
1603 port->bundle->ofproto->need_revalidate = true;
1604 lacp_slave_register(bundle->lacp, port, lacp);
1607 port->bond_stable_id = bond_stable_id;
1613 bundle_destroy(struct ofbundle *bundle)
1615 struct ofproto_dpif *ofproto;
1616 struct ofport_dpif *port, *next_port;
1623 ofproto = bundle->ofproto;
1624 for (i = 0; i < MAX_MIRRORS; i++) {
1625 struct ofmirror *m = ofproto->mirrors[i];
1627 if (m->out == bundle) {
1629 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1630 || hmapx_find_and_delete(&m->dsts, bundle)) {
1631 ofproto->need_revalidate = true;
1636 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1637 bundle_del_port(port);
1640 bundle_flush_macs(bundle, true);
1641 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1643 free(bundle->trunks);
1644 lacp_destroy(bundle->lacp);
1645 bond_destroy(bundle->bond);
1650 bundle_set(struct ofproto *ofproto_, void *aux,
1651 const struct ofproto_bundle_settings *s)
1653 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1654 bool need_flush = false;
1655 struct ofport_dpif *port;
1656 struct ofbundle *bundle;
1657 unsigned long *trunks;
1663 bundle_destroy(bundle_lookup(ofproto, aux));
1667 assert(s->n_slaves == 1 || s->bond != NULL);
1668 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1670 bundle = bundle_lookup(ofproto, aux);
1672 bundle = xmalloc(sizeof *bundle);
1674 bundle->ofproto = ofproto;
1675 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1676 hash_pointer(aux, 0));
1678 bundle->name = NULL;
1680 list_init(&bundle->ports);
1681 bundle->vlan_mode = PORT_VLAN_TRUNK;
1683 bundle->trunks = NULL;
1684 bundle->use_priority_tags = s->use_priority_tags;
1685 bundle->lacp = NULL;
1686 bundle->bond = NULL;
1688 bundle->floodable = true;
1690 bundle->src_mirrors = 0;
1691 bundle->dst_mirrors = 0;
1692 bundle->mirror_out = 0;
1695 if (!bundle->name || strcmp(s->name, bundle->name)) {
1697 bundle->name = xstrdup(s->name);
1702 if (!bundle->lacp) {
1703 ofproto->need_revalidate = true;
1704 bundle->lacp = lacp_create();
1706 lacp_configure(bundle->lacp, s->lacp);
1708 lacp_destroy(bundle->lacp);
1709 bundle->lacp = NULL;
1712 /* Update set of ports. */
1714 for (i = 0; i < s->n_slaves; i++) {
1715 if (!bundle_add_port(bundle, s->slaves[i],
1716 s->lacp ? &s->lacp_slaves[i] : NULL,
1717 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1721 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1722 struct ofport_dpif *next_port;
1724 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1725 for (i = 0; i < s->n_slaves; i++) {
1726 if (s->slaves[i] == port->up.ofp_port) {
1731 bundle_del_port(port);
1735 assert(list_size(&bundle->ports) <= s->n_slaves);
1737 if (list_is_empty(&bundle->ports)) {
1738 bundle_destroy(bundle);
1742 /* Set VLAN tagging mode */
1743 if (s->vlan_mode != bundle->vlan_mode
1744 || s->use_priority_tags != bundle->use_priority_tags) {
1745 bundle->vlan_mode = s->vlan_mode;
1746 bundle->use_priority_tags = s->use_priority_tags;
1751 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1752 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1754 if (vlan != bundle->vlan) {
1755 bundle->vlan = vlan;
1759 /* Get trunked VLANs. */
1760 switch (s->vlan_mode) {
1761 case PORT_VLAN_ACCESS:
1765 case PORT_VLAN_TRUNK:
1766 trunks = (unsigned long *) s->trunks;
1769 case PORT_VLAN_NATIVE_UNTAGGED:
1770 case PORT_VLAN_NATIVE_TAGGED:
1771 if (vlan != 0 && (!s->trunks
1772 || !bitmap_is_set(s->trunks, vlan)
1773 || bitmap_is_set(s->trunks, 0))) {
1774 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1776 trunks = bitmap_clone(s->trunks, 4096);
1778 trunks = bitmap_allocate1(4096);
1780 bitmap_set1(trunks, vlan);
1781 bitmap_set0(trunks, 0);
1783 trunks = (unsigned long *) s->trunks;
1790 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1791 free(bundle->trunks);
1792 if (trunks == s->trunks) {
1793 bundle->trunks = vlan_bitmap_clone(trunks);
1795 bundle->trunks = trunks;
1800 if (trunks != s->trunks) {
1805 if (!list_is_short(&bundle->ports)) {
1806 bundle->ofproto->has_bonded_bundles = true;
1808 if (bond_reconfigure(bundle->bond, s->bond)) {
1809 ofproto->need_revalidate = true;
1812 bundle->bond = bond_create(s->bond);
1813 ofproto->need_revalidate = true;
1816 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1817 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1821 bond_destroy(bundle->bond);
1822 bundle->bond = NULL;
1825 /* If we changed something that would affect MAC learning, un-learn
1826 * everything on this port and force flow revalidation. */
1828 bundle_flush_macs(bundle, false);
1835 bundle_remove(struct ofport *port_)
1837 struct ofport_dpif *port = ofport_dpif_cast(port_);
1838 struct ofbundle *bundle = port->bundle;
1841 bundle_del_port(port);
1842 if (list_is_empty(&bundle->ports)) {
1843 bundle_destroy(bundle);
1844 } else if (list_is_short(&bundle->ports)) {
1845 bond_destroy(bundle->bond);
1846 bundle->bond = NULL;
1852 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
1854 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1855 struct ofport_dpif *port = port_;
1856 uint8_t ea[ETH_ADDR_LEN];
1859 error = netdev_get_etheraddr(port->up.netdev, ea);
1861 struct ofpbuf packet;
1864 ofpbuf_init(&packet, 0);
1865 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1867 memcpy(packet_pdu, pdu, pdu_size);
1869 send_packet(port, &packet);
1870 ofpbuf_uninit(&packet);
1872 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1873 "%s (%s)", port->bundle->name,
1874 netdev_get_name(port->up.netdev), strerror(error));
1879 bundle_send_learning_packets(struct ofbundle *bundle)
1881 struct ofproto_dpif *ofproto = bundle->ofproto;
1882 int error, n_packets, n_errors;
1883 struct mac_entry *e;
1885 error = n_packets = n_errors = 0;
1886 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1887 if (e->port.p != bundle) {
1888 struct ofpbuf *learning_packet;
1889 struct ofport_dpif *port;
1893 /* The assignment to "port" is unnecessary but makes "grep"ing for
1894 * struct ofport_dpif more effective. */
1895 learning_packet = bond_compose_learning_packet(bundle->bond,
1899 ret = send_packet(port, learning_packet);
1900 ofpbuf_delete(learning_packet);
1910 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1911 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1912 "packets, last error was: %s",
1913 bundle->name, n_errors, n_packets, strerror(error));
1915 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1916 bundle->name, n_packets);
1921 bundle_run(struct ofbundle *bundle)
1924 lacp_run(bundle->lacp, send_pdu_cb);
1927 struct ofport_dpif *port;
1929 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1930 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1933 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1934 lacp_status(bundle->lacp));
1935 if (bond_should_send_learning_packets(bundle->bond)) {
1936 bundle_send_learning_packets(bundle);
1942 bundle_wait(struct ofbundle *bundle)
1945 lacp_wait(bundle->lacp);
1948 bond_wait(bundle->bond);
1955 mirror_scan(struct ofproto_dpif *ofproto)
1959 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1960 if (!ofproto->mirrors[idx]) {
1967 static struct ofmirror *
1968 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1972 for (i = 0; i < MAX_MIRRORS; i++) {
1973 struct ofmirror *mirror = ofproto->mirrors[i];
1974 if (mirror && mirror->aux == aux) {
1982 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
1984 mirror_update_dups(struct ofproto_dpif *ofproto)
1988 for (i = 0; i < MAX_MIRRORS; i++) {
1989 struct ofmirror *m = ofproto->mirrors[i];
1992 m->dup_mirrors = MIRROR_MASK_C(1) << i;
1996 for (i = 0; i < MAX_MIRRORS; i++) {
1997 struct ofmirror *m1 = ofproto->mirrors[i];
2004 for (j = i + 1; j < MAX_MIRRORS; j++) {
2005 struct ofmirror *m2 = ofproto->mirrors[j];
2007 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2008 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2009 m2->dup_mirrors |= m1->dup_mirrors;
2016 mirror_set(struct ofproto *ofproto_, void *aux,
2017 const struct ofproto_mirror_settings *s)
2019 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2020 mirror_mask_t mirror_bit;
2021 struct ofbundle *bundle;
2022 struct ofmirror *mirror;
2023 struct ofbundle *out;
2024 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2025 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2028 mirror = mirror_lookup(ofproto, aux);
2030 mirror_destroy(mirror);
2036 idx = mirror_scan(ofproto);
2038 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2040 ofproto->up.name, MAX_MIRRORS, s->name);
2044 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2045 mirror->ofproto = ofproto;
2048 mirror->out_vlan = -1;
2049 mirror->name = NULL;
2052 if (!mirror->name || strcmp(s->name, mirror->name)) {
2054 mirror->name = xstrdup(s->name);
2057 /* Get the new configuration. */
2058 if (s->out_bundle) {
2059 out = bundle_lookup(ofproto, s->out_bundle);
2061 mirror_destroy(mirror);
2067 out_vlan = s->out_vlan;
2069 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2070 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2072 /* If the configuration has not changed, do nothing. */
2073 if (hmapx_equals(&srcs, &mirror->srcs)
2074 && hmapx_equals(&dsts, &mirror->dsts)
2075 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2076 && mirror->out == out
2077 && mirror->out_vlan == out_vlan)
2079 hmapx_destroy(&srcs);
2080 hmapx_destroy(&dsts);
2084 hmapx_swap(&srcs, &mirror->srcs);
2085 hmapx_destroy(&srcs);
2087 hmapx_swap(&dsts, &mirror->dsts);
2088 hmapx_destroy(&dsts);
2090 free(mirror->vlans);
2091 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2094 mirror->out_vlan = out_vlan;
2096 /* Update bundles. */
2097 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2098 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2099 if (hmapx_contains(&mirror->srcs, bundle)) {
2100 bundle->src_mirrors |= mirror_bit;
2102 bundle->src_mirrors &= ~mirror_bit;
2105 if (hmapx_contains(&mirror->dsts, bundle)) {
2106 bundle->dst_mirrors |= mirror_bit;
2108 bundle->dst_mirrors &= ~mirror_bit;
2111 if (mirror->out == bundle) {
2112 bundle->mirror_out |= mirror_bit;
2114 bundle->mirror_out &= ~mirror_bit;
2118 ofproto->need_revalidate = true;
2119 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2120 mirror_update_dups(ofproto);
2126 mirror_destroy(struct ofmirror *mirror)
2128 struct ofproto_dpif *ofproto;
2129 mirror_mask_t mirror_bit;
2130 struct ofbundle *bundle;
2136 ofproto = mirror->ofproto;
2137 ofproto->need_revalidate = true;
2138 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2140 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2141 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2142 bundle->src_mirrors &= ~mirror_bit;
2143 bundle->dst_mirrors &= ~mirror_bit;
2144 bundle->mirror_out &= ~mirror_bit;
2147 hmapx_destroy(&mirror->srcs);
2148 hmapx_destroy(&mirror->dsts);
2149 free(mirror->vlans);
2151 ofproto->mirrors[mirror->idx] = NULL;
2155 mirror_update_dups(ofproto);
2159 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2160 uint64_t *packets, uint64_t *bytes)
2162 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2163 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2166 *packets = *bytes = UINT64_MAX;
2170 *packets = mirror->packet_count;
2171 *bytes = mirror->byte_count;
2177 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2179 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2180 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2181 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2187 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2189 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2190 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2191 return bundle && bundle->mirror_out != 0;
2195 forward_bpdu_changed(struct ofproto *ofproto_)
2197 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2198 /* Revalidate cached flows whenever forward_bpdu option changes. */
2199 ofproto->need_revalidate = true;
2203 set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time)
2205 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2206 mac_learning_set_idle_time(ofproto->ml, idle_time);
2211 static struct ofport_dpif *
2212 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
2214 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2215 return ofport ? ofport_dpif_cast(ofport) : NULL;
2218 static struct ofport_dpif *
2219 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
2221 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2225 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2226 struct dpif_port *dpif_port)
2228 ofproto_port->name = dpif_port->name;
2229 ofproto_port->type = dpif_port->type;
2230 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2234 port_run(struct ofport_dpif *ofport)
2236 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2237 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2238 bool enable = netdev_get_carrier(ofport->up.netdev);
2240 ofport->carrier_seq = carrier_seq;
2243 cfm_run(ofport->cfm);
2245 if (cfm_should_send_ccm(ofport->cfm)) {
2246 struct ofpbuf packet;
2248 ofpbuf_init(&packet, 0);
2249 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2250 send_packet(ofport, &packet);
2251 ofpbuf_uninit(&packet);
2254 enable = enable && !cfm_get_fault(ofport->cfm)
2255 && cfm_get_opup(ofport->cfm);
2258 if (ofport->bundle) {
2259 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2260 if (carrier_changed) {
2261 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2265 if (ofport->may_enable != enable) {
2266 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2268 if (ofproto->has_bundle_action) {
2269 ofproto->need_revalidate = true;
2273 ofport->may_enable = enable;
2277 port_wait(struct ofport_dpif *ofport)
2280 cfm_wait(ofport->cfm);
2285 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2286 struct ofproto_port *ofproto_port)
2288 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2289 struct dpif_port dpif_port;
2292 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2294 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2300 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2302 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2306 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2308 *ofp_portp = odp_port_to_ofp_port(odp_port);
2314 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2316 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2319 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2321 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2323 /* The caller is going to close ofport->up.netdev. If this is a
2324 * bonded port, then the bond is using that netdev, so remove it
2325 * from the bond. The client will need to reconfigure everything
2326 * after deleting ports, so then the slave will get re-added. */
2327 bundle_remove(&ofport->up);
2334 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2336 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2339 error = netdev_get_stats(ofport->up.netdev, stats);
2341 if (!error && ofport->odp_port == OVSP_LOCAL) {
2342 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2344 /* ofproto->stats.tx_packets represents packets that we created
2345 * internally and sent to some port (e.g. packets sent with
2346 * send_packet()). Account for them as if they had come from
2347 * OFPP_LOCAL and got forwarded. */
2349 if (stats->rx_packets != UINT64_MAX) {
2350 stats->rx_packets += ofproto->stats.tx_packets;
2353 if (stats->rx_bytes != UINT64_MAX) {
2354 stats->rx_bytes += ofproto->stats.tx_bytes;
2357 /* ofproto->stats.rx_packets represents packets that were received on
2358 * some port and we processed internally and dropped (e.g. STP).
2359 * Account fro them as if they had been forwarded to OFPP_LOCAL. */
2361 if (stats->tx_packets != UINT64_MAX) {
2362 stats->tx_packets += ofproto->stats.rx_packets;
2365 if (stats->tx_bytes != UINT64_MAX) {
2366 stats->tx_bytes += ofproto->stats.rx_bytes;
2373 /* Account packets for LOCAL port. */
2375 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2376 size_t tx_size, size_t rx_size)
2378 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2381 ofproto->stats.rx_packets++;
2382 ofproto->stats.rx_bytes += rx_size;
2385 ofproto->stats.tx_packets++;
2386 ofproto->stats.tx_bytes += tx_size;
2390 struct port_dump_state {
2391 struct dpif_port_dump dump;
2396 port_dump_start(const struct ofproto *ofproto_, void **statep)
2398 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2399 struct port_dump_state *state;
2401 *statep = state = xmalloc(sizeof *state);
2402 dpif_port_dump_start(&state->dump, ofproto->dpif);
2403 state->done = false;
2408 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2409 struct ofproto_port *port)
2411 struct port_dump_state *state = state_;
2412 struct dpif_port dpif_port;
2414 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2415 ofproto_port_from_dpif_port(port, &dpif_port);
2418 int error = dpif_port_dump_done(&state->dump);
2420 return error ? error : EOF;
2425 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2427 struct port_dump_state *state = state_;
2430 dpif_port_dump_done(&state->dump);
2437 port_poll(const struct ofproto *ofproto_, char **devnamep)
2439 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2440 return dpif_port_poll(ofproto->dpif, devnamep);
2444 port_poll_wait(const struct ofproto *ofproto_)
2446 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2447 dpif_port_poll_wait(ofproto->dpif);
2451 port_is_lacp_current(const struct ofport *ofport_)
2453 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2454 return (ofport->bundle && ofport->bundle->lacp
2455 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2459 /* Upcall handling. */
2461 /* Flow miss batching.
2463 * Some dpifs implement operations faster when you hand them off in a batch.
2464 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2465 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2466 * more packets, plus possibly installing the flow in the dpif.
2468 * So far we only batch the operations that affect flow setup time the most.
2469 * It's possible to batch more than that, but the benefit might be minimal. */
2471 struct hmap_node hmap_node;
2473 enum odp_key_fitness key_fitness;
2474 const struct nlattr *key;
2476 ovs_be16 initial_tci;
2477 struct list packets;
2480 struct flow_miss_op {
2481 struct dpif_op dpif_op;
2482 struct subfacet *subfacet;
2485 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2486 * OpenFlow controller as necessary according to their individual
2487 * configurations. */
2489 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
2490 const struct flow *flow)
2492 struct ofputil_packet_in pin;
2494 pin.packet = packet->data;
2495 pin.packet_len = packet->size;
2496 pin.reason = OFPR_NO_MATCH;
2497 pin.controller_id = 0;
2502 pin.send_len = 0; /* not used for flow table misses */
2504 flow_get_metadata(flow, &pin.fmd);
2506 /* Registers aren't meaningful on a miss. */
2507 memset(pin.fmd.reg_masks, 0, sizeof pin.fmd.reg_masks);
2509 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
2513 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2514 const struct ofpbuf *packet)
2516 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2522 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2524 cfm_process_heartbeat(ofport->cfm, packet);
2527 } else if (ofport->bundle && ofport->bundle->lacp
2528 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2530 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2533 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2535 stp_process_packet(ofport, packet);
2542 static struct flow_miss *
2543 flow_miss_create(struct hmap *todo, const struct flow *flow,
2544 enum odp_key_fitness key_fitness,
2545 const struct nlattr *key, size_t key_len,
2546 ovs_be16 initial_tci)
2548 uint32_t hash = flow_hash(flow, 0);
2549 struct flow_miss *miss;
2551 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2552 if (flow_equal(&miss->flow, flow)) {
2557 miss = xmalloc(sizeof *miss);
2558 hmap_insert(todo, &miss->hmap_node, hash);
2560 miss->key_fitness = key_fitness;
2562 miss->key_len = key_len;
2563 miss->initial_tci = initial_tci;
2564 list_init(&miss->packets);
2569 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2570 struct flow_miss_op *ops, size_t *n_ops)
2572 const struct flow *flow = &miss->flow;
2573 struct subfacet *subfacet;
2574 struct ofpbuf *packet;
2575 struct facet *facet;
2578 /* The caller must ensure that miss->hmap_node.hash contains
2579 * flow_hash(miss->flow, 0). */
2580 hash = miss->hmap_node.hash;
2582 facet = facet_lookup_valid(ofproto, flow, hash);
2584 struct rule_dpif *rule;
2586 rule = rule_dpif_lookup(ofproto, flow, 0);
2588 /* Don't send a packet-in if OFPUTIL_PC_NO_PACKET_IN asserted. */
2589 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
2591 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
2592 COVERAGE_INC(ofproto_dpif_no_packet_in);
2593 /* XXX install 'drop' flow entry */
2597 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
2601 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2602 send_packet_in_miss(ofproto, packet, flow);
2608 facet = facet_create(rule, flow, hash);
2611 subfacet = subfacet_create(facet,
2612 miss->key_fitness, miss->key, miss->key_len,
2615 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2616 struct dpif_flow_stats stats;
2617 struct flow_miss_op *op;
2618 struct dpif_execute *execute;
2620 ofproto->n_matches++;
2622 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2624 * Extra-special case for fail-open mode.
2626 * We are in fail-open mode and the packet matched the fail-open
2627 * rule, but we are connected to a controller too. We should send
2628 * the packet up to the controller in the hope that it will try to
2629 * set up a flow and thereby allow us to exit fail-open.
2631 * See the top-level comment in fail-open.c for more information.
2633 send_packet_in_miss(ofproto, packet, flow);
2636 if (!facet->may_install || !subfacet->actions) {
2637 subfacet_make_actions(subfacet, packet);
2640 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2641 subfacet_update_stats(subfacet, &stats);
2643 if (!subfacet->actions_len) {
2644 /* No actions to execute, so skip talking to the dpif. */
2648 if (flow->vlan_tci != subfacet->initial_tci) {
2649 /* This packet was received on a VLAN splinter port. We added
2650 * a VLAN to the packet to make the packet resemble the flow,
2651 * but the actions were composed assuming that the packet
2652 * contained no VLAN. So, we must remove the VLAN header from
2653 * the packet before trying to execute the actions. */
2654 eth_pop_vlan(packet);
2657 op = &ops[(*n_ops)++];
2658 execute = &op->dpif_op.u.execute;
2659 op->subfacet = subfacet;
2660 op->dpif_op.type = DPIF_OP_EXECUTE;
2661 execute->key = miss->key;
2662 execute->key_len = miss->key_len;
2663 execute->actions = (facet->may_install
2665 : xmemdup(subfacet->actions,
2666 subfacet->actions_len));
2667 execute->actions_len = subfacet->actions_len;
2668 execute->packet = packet;
2671 if (facet->may_install && subfacet->key_fitness != ODP_FIT_TOO_LITTLE) {
2672 struct flow_miss_op *op = &ops[(*n_ops)++];
2673 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
2675 op->subfacet = subfacet;
2676 op->dpif_op.type = DPIF_OP_FLOW_PUT;
2677 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2678 put->key = miss->key;
2679 put->key_len = miss->key_len;
2680 put->actions = subfacet->actions;
2681 put->actions_len = subfacet->actions_len;
2686 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
2687 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
2688 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
2689 * what a flow key should contain.
2691 * This function also includes some logic to help make VLAN splinters
2692 * transparent to the rest of the upcall processing logic. In particular, if
2693 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
2694 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
2695 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
2697 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
2698 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
2699 * (This differs from the value returned in flow->vlan_tci only for packets
2700 * received on VLAN splinters.)
2702 static enum odp_key_fitness
2703 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
2704 const struct nlattr *key, size_t key_len,
2705 struct flow *flow, ovs_be16 *initial_tci,
2706 struct ofpbuf *packet)
2708 enum odp_key_fitness fitness;
2712 fitness = odp_flow_key_to_flow(key, key_len, flow);
2713 if (fitness == ODP_FIT_ERROR) {
2716 *initial_tci = flow->vlan_tci;
2718 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
2720 /* Cause the flow to be processed as if it came in on the real device
2721 * with the VLAN device's VLAN ID. */
2722 flow->in_port = realdev;
2723 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
2725 /* Make the packet resemble the flow, so that it gets sent to an
2726 * OpenFlow controller properly, so that it looks correct for
2727 * sFlow, and so that flow_extract() will get the correct vlan_tci
2728 * if it is called on 'packet'.
2730 * The allocated space inside 'packet' probably also contains
2731 * 'key', that is, both 'packet' and 'key' are probably part of a
2732 * struct dpif_upcall (see the large comment on that structure
2733 * definition), so pushing data on 'packet' is in general not a
2734 * good idea since it could overwrite 'key' or free it as a side
2735 * effect. However, it's OK in this special case because we know
2736 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
2737 * will just overwrite the 4-byte "struct nlattr", which is fine
2738 * since we don't need that header anymore. */
2739 eth_push_vlan(packet, flow->vlan_tci);
2742 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
2743 if (fitness == ODP_FIT_PERFECT) {
2744 fitness = ODP_FIT_TOO_MUCH;
2752 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
2755 struct dpif_upcall *upcall;
2756 struct flow_miss *miss, *next_miss;
2757 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
2758 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
2767 /* Construct the to-do list.
2769 * This just amounts to extracting the flow from each packet and sticking
2770 * the packets that have the same flow in the same "flow_miss" structure so
2771 * that we can process them together. */
2773 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
2774 enum odp_key_fitness fitness;
2775 struct flow_miss *miss;
2776 ovs_be16 initial_tci;
2779 /* Obtain metadata and check userspace/kernel agreement on flow match,
2780 * then set 'flow''s header pointers. */
2781 fitness = ofproto_dpif_extract_flow_key(ofproto,
2782 upcall->key, upcall->key_len,
2783 &flow, &initial_tci,
2785 if (fitness == ODP_FIT_ERROR) {
2786 ofpbuf_delete(upcall->packet);
2789 flow_extract(upcall->packet, flow.skb_priority, flow.tun_id,
2790 flow.in_port, &flow);
2792 /* Handle 802.1ag, LACP, and STP specially. */
2793 if (process_special(ofproto, &flow, upcall->packet)) {
2794 ofproto_update_local_port_stats(&ofproto->up,
2795 0, upcall->packet->size);
2796 ofpbuf_delete(upcall->packet);
2797 ofproto->n_matches++;
2801 /* Add other packets to a to-do list. */
2802 miss = flow_miss_create(&todo, &flow, fitness,
2803 upcall->key, upcall->key_len, initial_tci);
2804 list_push_back(&miss->packets, &upcall->packet->list_node);
2807 /* Process each element in the to-do list, constructing the set of
2808 * operations to batch. */
2810 HMAP_FOR_EACH (miss, hmap_node, &todo) {
2811 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
2813 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
2815 /* Execute batch. */
2816 for (i = 0; i < n_ops; i++) {
2817 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
2819 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
2821 /* Free memory and update facets. */
2822 for (i = 0; i < n_ops; i++) {
2823 struct flow_miss_op *op = &flow_miss_ops[i];
2824 struct dpif_execute *execute;
2826 switch (op->dpif_op.type) {
2827 case DPIF_OP_EXECUTE:
2828 execute = &op->dpif_op.u.execute;
2829 if (op->subfacet->actions != execute->actions) {
2830 free((struct nlattr *) execute->actions);
2834 case DPIF_OP_FLOW_PUT:
2835 if (!op->dpif_op.error) {
2836 op->subfacet->installed = true;
2840 case DPIF_OP_FLOW_DEL:
2844 HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) {
2845 ofpbuf_list_delete(&miss->packets);
2846 hmap_remove(&todo, &miss->hmap_node);
2849 hmap_destroy(&todo);
2853 handle_userspace_upcall(struct ofproto_dpif *ofproto,
2854 struct dpif_upcall *upcall)
2856 struct user_action_cookie cookie;
2857 enum odp_key_fitness fitness;
2858 ovs_be16 initial_tci;
2861 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
2863 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
2864 upcall->key_len, &flow,
2865 &initial_tci, upcall->packet);
2866 if (fitness == ODP_FIT_ERROR) {
2867 ofpbuf_delete(upcall->packet);
2871 if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
2872 if (ofproto->sflow) {
2873 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
2877 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
2879 ofpbuf_delete(upcall->packet);
2883 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
2885 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
2889 assert (max_batch <= FLOW_MISS_MAX_BATCH);
2892 for (i = 0; i < max_batch; i++) {
2893 struct dpif_upcall *upcall = &misses[n_misses];
2896 error = dpif_recv(ofproto->dpif, upcall);
2901 switch (upcall->type) {
2902 case DPIF_UC_ACTION:
2903 handle_userspace_upcall(ofproto, upcall);
2907 /* Handle it later. */
2911 case DPIF_N_UC_TYPES:
2913 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32,
2919 handle_miss_upcalls(ofproto, misses, n_misses);
2924 /* Flow expiration. */
2926 static int subfacet_max_idle(const struct ofproto_dpif *);
2927 static void update_stats(struct ofproto_dpif *);
2928 static void rule_expire(struct rule_dpif *);
2929 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
2931 /* This function is called periodically by run(). Its job is to collect
2932 * updates for the flows that have been installed into the datapath, most
2933 * importantly when they last were used, and then use that information to
2934 * expire flows that have not been used recently.
2936 * Returns the number of milliseconds after which it should be called again. */
2938 expire(struct ofproto_dpif *ofproto)
2940 struct rule_dpif *rule, *next_rule;
2941 struct oftable *table;
2944 /* Update stats for each flow in the datapath. */
2945 update_stats(ofproto);
2947 /* Expire subfacets that have been idle too long. */
2948 dp_max_idle = subfacet_max_idle(ofproto);
2949 expire_subfacets(ofproto, dp_max_idle);
2951 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
2952 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
2953 struct cls_cursor cursor;
2955 cls_cursor_init(&cursor, &table->cls, NULL);
2956 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
2961 /* All outstanding data in existing flows has been accounted, so it's a
2962 * good time to do bond rebalancing. */
2963 if (ofproto->has_bonded_bundles) {
2964 struct ofbundle *bundle;
2966 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2968 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
2973 return MIN(dp_max_idle, 1000);
2976 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
2978 * This function also pushes statistics updates to rules which each facet
2979 * resubmits into. Generally these statistics will be accurate. However, if a
2980 * facet changes the rule it resubmits into at some time in between
2981 * update_stats() runs, it is possible that statistics accrued to the
2982 * old rule will be incorrectly attributed to the new rule. This could be
2983 * avoided by calling update_stats() whenever rules are created or
2984 * deleted. However, the performance impact of making so many calls to the
2985 * datapath do not justify the benefit of having perfectly accurate statistics.
2988 update_stats(struct ofproto_dpif *p)
2990 const struct dpif_flow_stats *stats;
2991 struct dpif_flow_dump dump;
2992 const struct nlattr *key;
2995 dpif_flow_dump_start(&dump, p->dpif);
2996 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
2997 struct subfacet *subfacet;
2999 subfacet = subfacet_find(p, key, key_len);
3000 if (subfacet && subfacet->installed) {
3001 struct facet *facet = subfacet->facet;
3003 if (stats->n_packets >= subfacet->dp_packet_count) {
3004 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
3005 facet->packet_count += extra;
3007 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3010 if (stats->n_bytes >= subfacet->dp_byte_count) {
3011 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
3013 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3016 subfacet->dp_packet_count = stats->n_packets;
3017 subfacet->dp_byte_count = stats->n_bytes;
3019 facet->tcp_flags |= stats->tcp_flags;
3021 subfacet_update_time(subfacet, stats->used);
3022 if (facet->accounted_bytes < facet->byte_count) {
3024 facet_account(facet);
3025 facet->accounted_bytes = facet->byte_count;
3027 facet_push_stats(facet);
3029 if (!VLOG_DROP_WARN(&rl)) {
3033 odp_flow_key_format(key, key_len, &s);
3034 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
3038 COVERAGE_INC(facet_unexpected);
3039 /* There's a flow in the datapath that we know nothing about, or a
3040 * flow that shouldn't be installed but was anyway. Delete it. */
3041 dpif_flow_del(p->dpif, key, key_len, NULL);
3044 dpif_flow_dump_done(&dump);
3047 /* Calculates and returns the number of milliseconds of idle time after which
3048 * subfacets should expire from the datapath. When a subfacet expires, we fold
3049 * its statistics into its facet, and when a facet's last subfacet expires, we
3050 * fold its statistic into its rule. */
3052 subfacet_max_idle(const struct ofproto_dpif *ofproto)
3055 * Idle time histogram.
3057 * Most of the time a switch has a relatively small number of subfacets.
3058 * When this is the case we might as well keep statistics for all of them
3059 * in userspace and to cache them in the kernel datapath for performance as
3062 * As the number of subfacets increases, the memory required to maintain
3063 * statistics about them in userspace and in the kernel becomes
3064 * significant. However, with a large number of subfacets it is likely
3065 * that only a few of them are "heavy hitters" that consume a large amount
3066 * of bandwidth. At this point, only heavy hitters are worth caching in
3067 * the kernel and maintaining in userspaces; other subfacets we can
3070 * The technique used to compute the idle time is to build a histogram with
3071 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3072 * that is installed in the kernel gets dropped in the appropriate bucket.
3073 * After the histogram has been built, we compute the cutoff so that only
3074 * the most-recently-used 1% of subfacets (but at least
3075 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3076 * the most-recently-used bucket of subfacets is kept, so actually an
3077 * arbitrary number of subfacets can be kept in any given expiration run
3078 * (though the next run will delete most of those unless they receive
3081 * This requires a second pass through the subfacets, in addition to the
3082 * pass made by update_stats(), because the former function never looks at
3083 * uninstallable subfacets.
3085 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3086 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3087 int buckets[N_BUCKETS] = { 0 };
3088 int total, subtotal, bucket;
3089 struct subfacet *subfacet;
3093 total = hmap_count(&ofproto->subfacets);
3094 if (total <= ofproto->up.flow_eviction_threshold) {
3095 return N_BUCKETS * BUCKET_WIDTH;
3098 /* Build histogram. */
3100 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3101 long long int idle = now - subfacet->used;
3102 int bucket = (idle <= 0 ? 0
3103 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3104 : (unsigned int) idle / BUCKET_WIDTH);
3108 /* Find the first bucket whose flows should be expired. */
3109 subtotal = bucket = 0;
3111 subtotal += buckets[bucket++];
3112 } while (bucket < N_BUCKETS &&
3113 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3115 if (VLOG_IS_DBG_ENABLED()) {
3119 ds_put_cstr(&s, "keep");
3120 for (i = 0; i < N_BUCKETS; i++) {
3122 ds_put_cstr(&s, ", drop");
3125 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3128 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3132 return bucket * BUCKET_WIDTH;
3135 enum { EXPIRE_MAX_BATCH = 50 };
3138 expire_batch(struct ofproto_dpif *ofproto, struct subfacet **subfacets, int n)
3140 struct odputil_keybuf keybufs[EXPIRE_MAX_BATCH];
3141 struct dpif_op ops[EXPIRE_MAX_BATCH];
3142 struct dpif_op *opsp[EXPIRE_MAX_BATCH];
3143 struct ofpbuf keys[EXPIRE_MAX_BATCH];
3144 struct dpif_flow_stats stats[EXPIRE_MAX_BATCH];
3147 for (i = 0; i < n; i++) {
3148 ops[i].type = DPIF_OP_FLOW_DEL;
3149 subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]);
3150 ops[i].u.flow_del.key = keys[i].data;
3151 ops[i].u.flow_del.key_len = keys[i].size;
3152 ops[i].u.flow_del.stats = &stats[i];
3156 dpif_operate(ofproto->dpif, opsp, n);
3157 for (i = 0; i < n; i++) {
3158 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
3159 subfacets[i]->installed = false;
3160 subfacet_destroy(subfacets[i]);
3165 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3167 long long int cutoff = time_msec() - dp_max_idle;
3169 struct subfacet *subfacet, *next_subfacet;
3170 struct subfacet *batch[EXPIRE_MAX_BATCH];
3174 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3175 &ofproto->subfacets) {
3176 if (subfacet->used < cutoff) {
3177 if (subfacet->installed) {
3178 batch[n_batch++] = subfacet;
3179 if (n_batch >= EXPIRE_MAX_BATCH) {
3180 expire_batch(ofproto, batch, n_batch);
3184 subfacet_destroy(subfacet);
3190 expire_batch(ofproto, batch, n_batch);
3194 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3195 * then delete it entirely. */
3197 rule_expire(struct rule_dpif *rule)
3199 struct facet *facet, *next_facet;
3203 /* Has 'rule' expired? */
3205 if (rule->up.hard_timeout
3206 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3207 reason = OFPRR_HARD_TIMEOUT;
3208 } else if (rule->up.idle_timeout
3209 && now > rule->up.used + rule->up.idle_timeout * 1000) {
3210 reason = OFPRR_IDLE_TIMEOUT;
3215 COVERAGE_INC(ofproto_dpif_expired);
3217 /* Update stats. (This is a no-op if the rule expired due to an idle
3218 * timeout, because that only happens when the rule has no facets left.) */
3219 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3220 facet_remove(facet);
3223 /* Get rid of the rule. */
3224 ofproto_rule_expire(&rule->up, reason);
3229 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3231 * The caller must already have determined that no facet with an identical
3232 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3233 * the ofproto's classifier table.
3235 * 'hash' must be the return value of flow_hash(flow, 0).
3237 * The facet will initially have no subfacets. The caller should create (at
3238 * least) one subfacet with subfacet_create(). */
3239 static struct facet *
3240 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
3242 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3243 struct facet *facet;
3245 facet = xzalloc(sizeof *facet);
3246 facet->used = time_msec();
3247 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
3248 list_push_back(&rule->facets, &facet->list_node);
3250 facet->flow = *flow;
3251 list_init(&facet->subfacets);
3252 netflow_flow_init(&facet->nf_flow);
3253 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3259 facet_free(struct facet *facet)
3264 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3265 * 'packet', which arrived on 'in_port'.
3267 * Takes ownership of 'packet'. */
3269 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3270 const struct nlattr *odp_actions, size_t actions_len,
3271 struct ofpbuf *packet)
3273 struct odputil_keybuf keybuf;
3277 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3278 odp_flow_key_from_flow(&key, flow);
3280 error = dpif_execute(ofproto->dpif, key.data, key.size,
3281 odp_actions, actions_len, packet);
3283 ofpbuf_delete(packet);
3287 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3289 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3290 * rule's statistics, via subfacet_uninstall().
3292 * - Removes 'facet' from its rule and from ofproto->facets.
3295 facet_remove(struct facet *facet)
3297 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3298 struct subfacet *subfacet, *next_subfacet;
3300 assert(!list_is_empty(&facet->subfacets));
3302 /* First uninstall all of the subfacets to get final statistics. */
3303 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3304 subfacet_uninstall(subfacet);
3307 /* Flush the final stats to the rule.
3309 * This might require us to have at least one subfacet around so that we
3310 * can use its actions for accounting in facet_account(), which is why we
3311 * have uninstalled but not yet destroyed the subfacets. */
3312 facet_flush_stats(facet);
3314 /* Now we're really all done so destroy everything. */
3315 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3316 &facet->subfacets) {
3317 subfacet_destroy__(subfacet);
3319 hmap_remove(&ofproto->facets, &facet->hmap_node);
3320 list_remove(&facet->list_node);
3324 /* Feed information from 'facet' back into the learning table to keep it in
3325 * sync with what is actually flowing through the datapath. */
3327 facet_learn(struct facet *facet)
3329 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3330 struct action_xlate_ctx ctx;
3332 if (!facet->has_learn
3333 && !facet->has_normal
3334 && (!facet->has_fin_timeout
3335 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
3339 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3340 facet->flow.vlan_tci,
3341 facet->rule, facet->tcp_flags, NULL);
3342 ctx.may_learn = true;
3343 xlate_actions_for_side_effects(&ctx, facet->rule->up.actions,
3344 facet->rule->up.n_actions);
3348 facet_account(struct facet *facet)
3350 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3351 struct subfacet *subfacet;
3352 const struct nlattr *a;
3357 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3360 n_bytes = facet->byte_count - facet->accounted_bytes;
3362 /* This loop feeds byte counters to bond_account() for rebalancing to use
3363 * as a basis. We also need to track the actual VLAN on which the packet
3364 * is going to be sent to ensure that it matches the one passed to
3365 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3368 * We use the actions from an arbitrary subfacet because they should all
3369 * be equally valid for our purpose. */
3370 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3371 struct subfacet, list_node);
3372 vlan_tci = facet->flow.vlan_tci;
3373 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3374 subfacet->actions, subfacet->actions_len) {
3375 const struct ovs_action_push_vlan *vlan;
3376 struct ofport_dpif *port;
3378 switch (nl_attr_type(a)) {
3379 case OVS_ACTION_ATTR_OUTPUT:
3380 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3381 if (port && port->bundle && port->bundle->bond) {
3382 bond_account(port->bundle->bond, &facet->flow,
3383 vlan_tci_to_vid(vlan_tci), n_bytes);
3387 case OVS_ACTION_ATTR_POP_VLAN:
3388 vlan_tci = htons(0);
3391 case OVS_ACTION_ATTR_PUSH_VLAN:
3392 vlan = nl_attr_get(a);
3393 vlan_tci = vlan->vlan_tci;
3399 /* Returns true if the only action for 'facet' is to send to the controller.
3400 * (We don't report NetFlow expiration messages for such facets because they
3401 * are just part of the control logic for the network, not real traffic). */
3403 facet_is_controller_flow(struct facet *facet)
3406 && facet->rule->up.n_actions == 1
3407 && action_outputs_to_port(&facet->rule->up.actions[0],
3408 htons(OFPP_CONTROLLER)));
3411 /* Folds all of 'facet''s statistics into its rule. Also updates the
3412 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3413 * 'facet''s statistics in the datapath should have been zeroed and folded into
3414 * its packet and byte counts before this function is called. */
3416 facet_flush_stats(struct facet *facet)
3418 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3419 struct subfacet *subfacet;
3421 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3422 assert(!subfacet->dp_byte_count);
3423 assert(!subfacet->dp_packet_count);
3426 facet_push_stats(facet);
3427 if (facet->accounted_bytes < facet->byte_count) {
3428 facet_account(facet);
3429 facet->accounted_bytes = facet->byte_count;
3432 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3433 struct ofexpired expired;
3434 expired.flow = facet->flow;
3435 expired.packet_count = facet->packet_count;
3436 expired.byte_count = facet->byte_count;
3437 expired.used = facet->used;
3438 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3441 facet->rule->packet_count += facet->packet_count;
3442 facet->rule->byte_count += facet->byte_count;
3444 /* Reset counters to prevent double counting if 'facet' ever gets
3446 facet_reset_counters(facet);
3448 netflow_flow_clear(&facet->nf_flow);
3449 facet->tcp_flags = 0;
3452 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3453 * Returns it if found, otherwise a null pointer.
3455 * 'hash' must be the return value of flow_hash(flow, 0).
3457 * The returned facet might need revalidation; use facet_lookup_valid()
3458 * instead if that is important. */
3459 static struct facet *
3460 facet_find(struct ofproto_dpif *ofproto,
3461 const struct flow *flow, uint32_t hash)
3463 struct facet *facet;
3465 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
3466 if (flow_equal(flow, &facet->flow)) {
3474 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3475 * Returns it if found, otherwise a null pointer.
3477 * 'hash' must be the return value of flow_hash(flow, 0).
3479 * The returned facet is guaranteed to be valid. */
3480 static struct facet *
3481 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
3484 struct facet *facet = facet_find(ofproto, flow, hash);
3486 /* The facet we found might not be valid, since we could be in need of
3487 * revalidation. If it is not valid, don't return it. */
3489 && (ofproto->need_revalidate
3490 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))
3491 && !facet_revalidate(facet)) {
3492 COVERAGE_INC(facet_invalidated);
3500 facet_check_consistency(struct facet *facet)
3502 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3504 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3506 uint64_t odp_actions_stub[1024 / 8];
3507 struct ofpbuf odp_actions;
3509 struct rule_dpif *rule;
3510 struct subfacet *subfacet;
3511 bool may_log = false;
3514 /* Check the rule for consistency. */
3515 rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3517 if (!VLOG_DROP_WARN(&rl)) {
3518 char *s = flow_to_string(&facet->flow);
3519 VLOG_WARN("%s: facet should not exist", s);
3523 } else if (rule != facet->rule) {
3524 may_log = !VLOG_DROP_WARN(&rl);
3530 flow_format(&s, &facet->flow);
3531 ds_put_format(&s, ": facet associated with wrong rule (was "
3532 "table=%"PRIu8",", facet->rule->up.table_id);
3533 cls_rule_format(&facet->rule->up.cr, &s);
3534 ds_put_format(&s, ") (should have been table=%"PRIu8",",
3536 cls_rule_format(&rule->up.cr, &s);
3537 ds_put_char(&s, ')');
3539 VLOG_WARN("%s", ds_cstr(&s));
3546 /* Check the datapath actions for consistency. */
3547 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
3548 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3549 struct action_xlate_ctx ctx;
3550 bool actions_changed;
3551 bool should_install;
3553 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3554 subfacet->initial_tci, rule, 0, NULL);
3555 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions,
3558 should_install = (ctx.may_set_up_flow
3559 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3560 if (!should_install && !subfacet->installed) {
3561 /* The actions for uninstallable flows may vary from one packet to
3562 * the next, so don't compare the actions. */
3566 actions_changed = (subfacet->actions_len != odp_actions.size
3567 || memcmp(subfacet->actions, odp_actions.data,
3568 subfacet->actions_len));
3569 if (should_install != subfacet->installed || actions_changed) {
3571 may_log = !VLOG_DROP_WARN(&rl);
3576 struct odputil_keybuf keybuf;
3581 subfacet_get_key(subfacet, &keybuf, &key);
3582 odp_flow_key_format(key.data, key.size, &s);
3584 ds_put_cstr(&s, ": inconsistency in subfacet");
3585 if (should_install != subfacet->installed) {
3586 enum odp_key_fitness fitness = subfacet->key_fitness;
3588 ds_put_format(&s, " (should%s have been installed)",
3589 should_install ? "" : " not");
3590 ds_put_format(&s, " (may_set_up_flow=%s, fitness=%s)",
3591 ctx.may_set_up_flow ? "true" : "false",
3592 odp_key_fitness_to_string(fitness));
3594 if (actions_changed) {
3595 ds_put_cstr(&s, " (actions were: ");
3596 format_odp_actions(&s, subfacet->actions,
3597 subfacet->actions_len);
3598 ds_put_cstr(&s, ") (correct actions: ");
3599 format_odp_actions(&s, odp_actions.data, odp_actions.size);
3600 ds_put_char(&s, ')');
3602 ds_put_cstr(&s, " (actions: ");
3603 format_odp_actions(&s, subfacet->actions,
3604 subfacet->actions_len);
3605 ds_put_char(&s, ')');
3607 VLOG_WARN("%s", ds_cstr(&s));
3612 ofpbuf_uninit(&odp_actions);
3617 /* Re-searches the classifier for 'facet':
3619 * - If the rule found is different from 'facet''s current rule, moves
3620 * 'facet' to the new rule and recompiles its actions.
3622 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3623 * where it is and recompiles its actions anyway.
3625 * - If there is none, destroys 'facet'.
3627 * Returns true if 'facet' still exists, false if it has been destroyed. */
3629 facet_revalidate(struct facet *facet)
3631 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3633 struct nlattr *odp_actions;
3636 struct actions *new_actions;
3638 struct action_xlate_ctx ctx;
3639 uint64_t odp_actions_stub[1024 / 8];
3640 struct ofpbuf odp_actions;
3642 struct rule_dpif *new_rule;
3643 struct subfacet *subfacet;
3644 bool actions_changed;
3647 COVERAGE_INC(facet_revalidate);
3649 /* Determine the new rule. */
3650 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3652 /* No new rule, so delete the facet. */
3653 facet_remove(facet);
3657 /* Calculate new datapath actions.
3659 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3660 * emit a NetFlow expiration and, if so, we need to have the old state
3661 * around to properly compose it. */
3663 /* If the datapath actions changed or the installability changed,
3664 * then we need to talk to the datapath. */
3667 memset(&ctx, 0, sizeof ctx);
3668 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
3669 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3670 bool should_install;
3672 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3673 subfacet->initial_tci, new_rule, 0, NULL);
3674 xlate_actions(&ctx, new_rule->up.actions, new_rule->up.n_actions,
3676 actions_changed = (subfacet->actions_len != odp_actions.size
3677 || memcmp(subfacet->actions, odp_actions.data,
3678 subfacet->actions_len));
3680 should_install = (ctx.may_set_up_flow
3681 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3682 if (actions_changed || should_install != subfacet->installed) {
3683 if (should_install) {
3684 struct dpif_flow_stats stats;
3686 subfacet_install(subfacet,
3687 odp_actions.data, odp_actions.size, &stats);
3688 subfacet_update_stats(subfacet, &stats);
3690 subfacet_uninstall(subfacet);
3694 new_actions = xcalloc(list_size(&facet->subfacets),
3695 sizeof *new_actions);
3697 new_actions[i].odp_actions = xmemdup(odp_actions.data,
3699 new_actions[i].actions_len = odp_actions.size;
3704 ofpbuf_uninit(&odp_actions);
3707 facet_flush_stats(facet);
3710 /* Update 'facet' now that we've taken care of all the old state. */
3711 facet->tags = ctx.tags;
3712 facet->nf_flow.output_iface = ctx.nf_output_iface;
3713 facet->may_install = ctx.may_set_up_flow;
3714 facet->has_learn = ctx.has_learn;
3715 facet->has_normal = ctx.has_normal;
3716 facet->has_fin_timeout = ctx.has_fin_timeout;
3717 facet->mirrors = ctx.mirrors;
3720 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3721 if (new_actions[i].odp_actions) {
3722 free(subfacet->actions);
3723 subfacet->actions = new_actions[i].odp_actions;
3724 subfacet->actions_len = new_actions[i].actions_len;
3730 if (facet->rule != new_rule) {
3731 COVERAGE_INC(facet_changed_rule);
3732 list_remove(&facet->list_node);
3733 list_push_back(&new_rule->facets, &facet->list_node);
3734 facet->rule = new_rule;
3735 facet->used = new_rule->up.created;
3736 facet->prev_used = facet->used;
3742 /* Updates 'facet''s used time. Caller is responsible for calling
3743 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3745 facet_update_time(struct facet *facet, long long int used)
3747 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3748 if (used > facet->used) {
3750 ofproto_rule_update_used(&facet->rule->up, used);
3751 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3756 facet_reset_counters(struct facet *facet)
3758 facet->packet_count = 0;
3759 facet->byte_count = 0;
3760 facet->prev_packet_count = 0;
3761 facet->prev_byte_count = 0;
3762 facet->accounted_bytes = 0;
3766 facet_push_stats(struct facet *facet)
3768 uint64_t new_packets, new_bytes;
3770 assert(facet->packet_count >= facet->prev_packet_count);
3771 assert(facet->byte_count >= facet->prev_byte_count);
3772 assert(facet->used >= facet->prev_used);
3774 new_packets = facet->packet_count - facet->prev_packet_count;
3775 new_bytes = facet->byte_count - facet->prev_byte_count;
3777 if (new_packets || new_bytes || facet->used > facet->prev_used) {
3778 facet->prev_packet_count = facet->packet_count;
3779 facet->prev_byte_count = facet->byte_count;
3780 facet->prev_used = facet->used;
3782 flow_push_stats(facet->rule, &facet->flow,
3783 new_packets, new_bytes, facet->used);
3785 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
3786 facet->mirrors, new_packets, new_bytes);
3790 struct ofproto_push {
3791 struct action_xlate_ctx ctx;
3798 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
3800 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3803 rule->packet_count += push->packets;
3804 rule->byte_count += push->bytes;
3805 ofproto_rule_update_used(&rule->up, push->used);
3809 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3810 * 'rule''s actions and mirrors. */
3812 flow_push_stats(struct rule_dpif *rule,
3813 const struct flow *flow, uint64_t packets, uint64_t bytes,
3816 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3817 struct ofproto_push push;
3819 push.packets = packets;
3823 ofproto_rule_update_used(&rule->up, used);
3825 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, rule,
3827 push.ctx.resubmit_hook = push_resubmit;
3828 xlate_actions_for_side_effects(&push.ctx,
3829 rule->up.actions, rule->up.n_actions);
3834 static struct subfacet *
3835 subfacet_find__(struct ofproto_dpif *ofproto,
3836 const struct nlattr *key, size_t key_len, uint32_t key_hash,
3837 const struct flow *flow)
3839 struct subfacet *subfacet;
3841 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
3842 &ofproto->subfacets) {
3844 ? (subfacet->key_len == key_len
3845 && !memcmp(key, subfacet->key, key_len))
3846 : flow_equal(flow, &subfacet->facet->flow)) {
3854 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
3855 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
3856 * there is one, otherwise creates and returns a new subfacet.
3858 * If the returned subfacet is new, then subfacet->actions will be NULL, in
3859 * which case the caller must populate the actions with
3860 * subfacet_make_actions(). */
3861 static struct subfacet *
3862 subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness,
3863 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
3865 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3866 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3867 struct subfacet *subfacet;
3869 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
3871 if (subfacet->facet == facet) {
3875 /* This shouldn't happen. */
3876 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
3877 subfacet_destroy(subfacet);
3880 subfacet = (list_is_empty(&facet->subfacets)
3881 ? &facet->one_subfacet
3882 : xmalloc(sizeof *subfacet));
3883 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
3884 list_push_back(&facet->subfacets, &subfacet->list_node);
3885 subfacet->facet = facet;
3886 subfacet->key_fitness = key_fitness;
3887 if (key_fitness != ODP_FIT_PERFECT) {
3888 subfacet->key = xmemdup(key, key_len);
3889 subfacet->key_len = key_len;
3891 subfacet->key = NULL;
3892 subfacet->key_len = 0;
3894 subfacet->used = time_msec();
3895 subfacet->dp_packet_count = 0;
3896 subfacet->dp_byte_count = 0;
3897 subfacet->actions_len = 0;
3898 subfacet->actions = NULL;
3899 subfacet->installed = false;
3900 subfacet->initial_tci = initial_tci;
3905 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
3906 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
3907 static struct subfacet *
3908 subfacet_find(struct ofproto_dpif *ofproto,
3909 const struct nlattr *key, size_t key_len)
3911 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3912 enum odp_key_fitness fitness;
3915 fitness = odp_flow_key_to_flow(key, key_len, &flow);
3916 if (fitness == ODP_FIT_ERROR) {
3920 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
3923 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
3924 * its facet within 'ofproto', and frees it. */
3926 subfacet_destroy__(struct subfacet *subfacet)
3928 struct facet *facet = subfacet->facet;
3929 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3931 subfacet_uninstall(subfacet);
3932 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
3933 list_remove(&subfacet->list_node);
3934 free(subfacet->key);
3935 free(subfacet->actions);
3936 if (subfacet != &facet->one_subfacet) {
3941 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
3942 * last remaining subfacet in its facet destroys the facet too. */
3944 subfacet_destroy(struct subfacet *subfacet)
3946 struct facet *facet = subfacet->facet;
3948 if (list_is_singleton(&facet->subfacets)) {
3949 /* facet_remove() needs at least one subfacet (it will remove it). */
3950 facet_remove(facet);
3952 subfacet_destroy__(subfacet);
3956 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
3957 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
3958 * for use as temporary storage. */
3960 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
3963 if (!subfacet->key) {
3964 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
3965 odp_flow_key_from_flow(key, &subfacet->facet->flow);
3967 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
3971 /* Composes the datapath actions for 'subfacet' based on its rule's actions. */
3973 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet)
3975 struct facet *facet = subfacet->facet;
3976 struct rule_dpif *rule = facet->rule;
3977 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3979 struct action_xlate_ctx ctx;
3980 uint64_t odp_actions_stub[1024 / 8];
3981 struct ofpbuf odp_actions;
3983 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
3984 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
3986 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions, &odp_actions);
3987 facet->tags = ctx.tags;
3988 facet->may_install = ctx.may_set_up_flow;
3989 facet->has_learn = ctx.has_learn;
3990 facet->has_normal = ctx.has_normal;
3991 facet->has_fin_timeout = ctx.has_fin_timeout;
3992 facet->nf_flow.output_iface = ctx.nf_output_iface;
3993 facet->mirrors = ctx.mirrors;
3995 if (subfacet->actions_len != odp_actions.size
3996 || memcmp(subfacet->actions, odp_actions.data, odp_actions.size)) {
3997 free(subfacet->actions);
3998 subfacet->actions_len = odp_actions.size;
3999 subfacet->actions = xmemdup(odp_actions.data, odp_actions.size);
4002 ofpbuf_uninit(&odp_actions);
4005 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4006 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4007 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4008 * since 'subfacet' was last updated.
4010 * Returns 0 if successful, otherwise a positive errno value. */
4012 subfacet_install(struct subfacet *subfacet,
4013 const struct nlattr *actions, size_t actions_len,
4014 struct dpif_flow_stats *stats)
4016 struct facet *facet = subfacet->facet;
4017 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4018 struct odputil_keybuf keybuf;
4019 enum dpif_flow_put_flags flags;
4023 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
4025 flags |= DPIF_FP_ZERO_STATS;
4028 subfacet_get_key(subfacet, &keybuf, &key);
4029 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
4030 actions, actions_len, stats);
4033 subfacet_reset_dp_stats(subfacet, stats);
4039 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4041 subfacet_uninstall(struct subfacet *subfacet)
4043 if (subfacet->installed) {
4044 struct rule_dpif *rule = subfacet->facet->rule;
4045 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4046 struct odputil_keybuf keybuf;
4047 struct dpif_flow_stats stats;
4051 subfacet_get_key(subfacet, &keybuf, &key);
4052 error = dpif_flow_del(ofproto->dpif, key.data, key.size, &stats);
4053 subfacet_reset_dp_stats(subfacet, &stats);
4055 subfacet_update_stats(subfacet, &stats);
4057 subfacet->installed = false;
4059 assert(subfacet->dp_packet_count == 0);
4060 assert(subfacet->dp_byte_count == 0);
4064 /* Resets 'subfacet''s datapath statistics counters. This should be called
4065 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4066 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4067 * was reset in the datapath. 'stats' will be modified to include only
4068 * statistics new since 'subfacet' was last updated. */
4070 subfacet_reset_dp_stats(struct subfacet *subfacet,
4071 struct dpif_flow_stats *stats)
4074 && subfacet->dp_packet_count <= stats->n_packets
4075 && subfacet->dp_byte_count <= stats->n_bytes) {
4076 stats->n_packets -= subfacet->dp_packet_count;
4077 stats->n_bytes -= subfacet->dp_byte_count;
4080 subfacet->dp_packet_count = 0;
4081 subfacet->dp_byte_count = 0;
4084 /* Updates 'subfacet''s used time. The caller is responsible for calling
4085 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
4087 subfacet_update_time(struct subfacet *subfacet, long long int used)
4089 if (used > subfacet->used) {
4090 subfacet->used = used;
4091 facet_update_time(subfacet->facet, used);
4095 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4097 * Because of the meaning of a subfacet's counters, it only makes sense to do
4098 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4099 * represents a packet that was sent by hand or if it represents statistics
4100 * that have been cleared out of the datapath. */
4102 subfacet_update_stats(struct subfacet *subfacet,
4103 const struct dpif_flow_stats *stats)
4105 if (stats->n_packets || stats->used > subfacet->used) {
4106 struct facet *facet = subfacet->facet;
4108 subfacet_update_time(subfacet, stats->used);
4109 facet->packet_count += stats->n_packets;
4110 facet->byte_count += stats->n_bytes;
4111 facet->tcp_flags |= stats->tcp_flags;
4112 facet_push_stats(facet);
4113 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
4119 static struct rule_dpif *
4120 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4123 struct cls_rule *cls_rule;
4124 struct classifier *cls;
4126 if (table_id >= N_TABLES) {
4130 cls = &ofproto->up.tables[table_id].cls;
4131 if (flow->nw_frag & FLOW_NW_FRAG_ANY
4132 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4133 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
4134 * are unavailable. */
4135 struct flow ofpc_normal_flow = *flow;
4136 ofpc_normal_flow.tp_src = htons(0);
4137 ofpc_normal_flow.tp_dst = htons(0);
4138 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
4140 cls_rule = classifier_lookup(cls, flow);
4142 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
4146 complete_operation(struct rule_dpif *rule)
4148 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4150 rule_invalidate(rule);
4152 struct dpif_completion *c = xmalloc(sizeof *c);
4153 c->op = rule->up.pending;
4154 list_push_back(&ofproto->completions, &c->list_node);
4156 ofoperation_complete(rule->up.pending, 0);
4160 static struct rule *
4163 struct rule_dpif *rule = xmalloc(sizeof *rule);
4168 rule_dealloc(struct rule *rule_)
4170 struct rule_dpif *rule = rule_dpif_cast(rule_);
4175 rule_construct(struct rule *rule_)
4177 struct rule_dpif *rule = rule_dpif_cast(rule_);
4178 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4179 struct rule_dpif *victim;
4183 error = validate_actions(rule->up.actions, rule->up.n_actions,
4184 &rule->up.cr.flow, ofproto->max_ports);
4189 rule->packet_count = 0;
4190 rule->byte_count = 0;
4192 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
4193 if (victim && !list_is_empty(&victim->facets)) {
4194 struct facet *facet;
4196 rule->facets = victim->facets;
4197 list_moved(&rule->facets);
4198 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4199 /* XXX: We're only clearing our local counters here. It's possible
4200 * that quite a few packets are unaccounted for in the datapath
4201 * statistics. These will be accounted to the new rule instead of
4202 * cleared as required. This could be fixed by clearing out the
4203 * datapath statistics for this facet, but currently it doesn't
4205 facet_reset_counters(facet);
4209 /* Must avoid list_moved() in this case. */
4210 list_init(&rule->facets);
4213 table_id = rule->up.table_id;
4214 rule->tag = (victim ? victim->tag
4216 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
4217 ofproto->tables[table_id].basis));
4219 complete_operation(rule);
4224 rule_destruct(struct rule *rule_)
4226 struct rule_dpif *rule = rule_dpif_cast(rule_);
4227 struct facet *facet, *next_facet;
4229 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4230 facet_revalidate(facet);
4233 complete_operation(rule);
4237 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4239 struct rule_dpif *rule = rule_dpif_cast(rule_);
4240 struct facet *facet;
4242 /* Start from historical data for 'rule' itself that are no longer tracked
4243 * in facets. This counts, for example, facets that have expired. */
4244 *packets = rule->packet_count;
4245 *bytes = rule->byte_count;
4247 /* Add any statistics that are tracked by facets. This includes
4248 * statistical data recently updated by ofproto_update_stats() as well as
4249 * stats for packets that were executed "by hand" via dpif_execute(). */
4250 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4251 *packets += facet->packet_count;
4252 *bytes += facet->byte_count;
4257 rule_execute(struct rule *rule_, const struct flow *flow,
4258 struct ofpbuf *packet)
4260 struct rule_dpif *rule = rule_dpif_cast(rule_);
4261 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4263 size_t size = packet->size;
4265 struct action_xlate_ctx ctx;
4266 uint64_t odp_actions_stub[1024 / 8];
4267 struct ofpbuf odp_actions;
4269 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4270 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
4271 rule, packet_get_tcp_flags(packet, flow), packet);
4272 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions, &odp_actions);
4273 if (execute_odp_actions(ofproto, flow, odp_actions.data,
4274 odp_actions.size, packet)) {
4275 rule->packet_count++;
4276 rule->byte_count += size;
4277 flow_push_stats(rule, flow, 1, size, time_msec());
4279 ofpbuf_uninit(&odp_actions);
4285 rule_modify_actions(struct rule *rule_)
4287 struct rule_dpif *rule = rule_dpif_cast(rule_);
4288 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4291 error = validate_actions(rule->up.actions, rule->up.n_actions,
4292 &rule->up.cr.flow, ofproto->max_ports);
4294 ofoperation_complete(rule->up.pending, error);
4298 complete_operation(rule);
4301 /* Sends 'packet' out 'ofport'.
4302 * May modify 'packet'.
4303 * Returns 0 if successful, otherwise a positive errno value. */
4305 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4307 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4308 struct ofpbuf key, odp_actions;
4309 struct odputil_keybuf keybuf;
4314 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
4315 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4317 if (odp_port != ofport->odp_port) {
4318 eth_pop_vlan(packet);
4319 flow.vlan_tci = htons(0);
4322 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4323 odp_flow_key_from_flow(&key, &flow);
4325 ofpbuf_init(&odp_actions, 32);
4326 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4328 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4329 error = dpif_execute(ofproto->dpif,
4331 odp_actions.data, odp_actions.size,
4333 ofpbuf_uninit(&odp_actions);
4336 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4337 ofproto->up.name, odp_port, strerror(error));
4339 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
4343 /* OpenFlow to datapath action translation. */
4345 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
4346 struct action_xlate_ctx *ctx);
4347 static void xlate_normal(struct action_xlate_ctx *);
4350 put_userspace_action(const struct ofproto_dpif *ofproto,
4351 struct ofpbuf *odp_actions,
4352 const struct flow *flow,
4353 const struct user_action_cookie *cookie)
4357 pid = dpif_port_get_pid(ofproto->dpif,
4358 ofp_port_to_odp_port(flow->in_port));
4360 return odp_put_userspace_action(pid, cookie, odp_actions);
4363 /* Compose SAMPLE action for sFlow. */
4365 compose_sflow_action(const struct ofproto_dpif *ofproto,
4366 struct ofpbuf *odp_actions,
4367 const struct flow *flow,
4370 uint32_t port_ifindex;
4371 uint32_t probability;
4372 struct user_action_cookie cookie;
4373 size_t sample_offset, actions_offset;
4374 int cookie_offset, n_output;
4376 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4380 if (odp_port == OVSP_NONE) {
4384 port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4388 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4390 /* Number of packets out of UINT_MAX to sample. */
4391 probability = dpif_sflow_get_probability(ofproto->sflow);
4392 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4394 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4396 cookie.type = USER_ACTION_COOKIE_SFLOW;
4397 cookie.data = port_ifindex;
4398 cookie.n_output = n_output;
4399 cookie.vlan_tci = 0;
4400 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4402 nl_msg_end_nested(odp_actions, actions_offset);
4403 nl_msg_end_nested(odp_actions, sample_offset);
4404 return cookie_offset;
4407 /* SAMPLE action must be first action in any given list of actions.
4408 * At this point we do not have all information required to build it. So try to
4409 * build sample action as complete as possible. */
4411 add_sflow_action(struct action_xlate_ctx *ctx)
4413 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4415 &ctx->flow, OVSP_NONE);
4416 ctx->sflow_odp_port = 0;
4417 ctx->sflow_n_outputs = 0;
4420 /* Fix SAMPLE action according to data collected while composing ODP actions.
4421 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4422 * USERSPACE action's user-cookie which is required for sflow. */
4424 fix_sflow_action(struct action_xlate_ctx *ctx)
4426 const struct flow *base = &ctx->base_flow;
4427 struct user_action_cookie *cookie;
4429 if (!ctx->user_cookie_offset) {
4433 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4435 assert(cookie != NULL);
4436 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4438 if (ctx->sflow_n_outputs) {
4439 cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow,
4440 ctx->sflow_odp_port);
4442 if (ctx->sflow_n_outputs >= 255) {
4443 cookie->n_output = 255;
4445 cookie->n_output = ctx->sflow_n_outputs;
4447 cookie->vlan_tci = base->vlan_tci;
4451 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4454 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4455 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4456 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4457 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4461 struct priority_to_dscp *pdscp;
4463 if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD
4464 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4468 pdscp = get_priority(ofport, ctx->flow.skb_priority);
4470 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4471 ctx->flow.nw_tos |= pdscp->dscp;
4474 /* We may not have an ofport record for this port, but it doesn't hurt
4475 * to allow forwarding to it anyhow. Maybe such a port will appear
4476 * later and we're pre-populating the flow table. */
4479 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4480 ctx->flow.vlan_tci);
4481 if (out_port != odp_port) {
4482 ctx->flow.vlan_tci = htons(0);
4484 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4485 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4487 ctx->sflow_odp_port = odp_port;
4488 ctx->sflow_n_outputs++;
4489 ctx->nf_output_iface = ofp_port;
4490 ctx->flow.vlan_tci = flow_vlan_tci;
4491 ctx->flow.nw_tos = flow_nw_tos;
4495 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4497 compose_output_action__(ctx, ofp_port, true);
4501 xlate_table_action(struct action_xlate_ctx *ctx,
4502 uint16_t in_port, uint8_t table_id)
4504 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4505 struct ofproto_dpif *ofproto = ctx->ofproto;
4506 struct rule_dpif *rule;
4507 uint16_t old_in_port;
4508 uint8_t old_table_id;
4510 old_table_id = ctx->table_id;
4511 ctx->table_id = table_id;
4513 /* Look up a flow with 'in_port' as the input port. */
4514 old_in_port = ctx->flow.in_port;
4515 ctx->flow.in_port = in_port;
4516 rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
4519 if (table_id > 0 && table_id < N_TABLES) {
4520 struct table_dpif *table = &ofproto->tables[table_id];
4521 if (table->other_table) {
4522 ctx->tags |= (rule && rule->tag
4524 : rule_calculate_tag(&ctx->flow,
4525 &table->other_table->wc,
4530 /* Restore the original input port. Otherwise OFPP_NORMAL and
4531 * OFPP_IN_PORT will have surprising behavior. */
4532 ctx->flow.in_port = old_in_port;
4534 if (ctx->resubmit_hook) {
4535 ctx->resubmit_hook(ctx, rule);
4539 struct rule_dpif *old_rule = ctx->rule;
4543 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
4544 ctx->rule = old_rule;
4548 ctx->table_id = old_table_id;
4550 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4552 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
4553 MAX_RESUBMIT_RECURSION);
4554 ctx->max_resubmit_trigger = true;
4559 xlate_resubmit_table(struct action_xlate_ctx *ctx,
4560 const struct nx_action_resubmit *nar)
4565 in_port = (nar->in_port == htons(OFPP_IN_PORT)
4567 : ntohs(nar->in_port));
4568 table_id = nar->table == 255 ? ctx->table_id : nar->table;
4570 xlate_table_action(ctx, in_port, table_id);
4574 flood_packets(struct action_xlate_ctx *ctx, bool all)
4576 struct ofport_dpif *ofport;
4578 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
4579 uint16_t ofp_port = ofport->up.ofp_port;
4581 if (ofp_port == ctx->flow.in_port) {
4586 compose_output_action__(ctx, ofp_port, false);
4587 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
4588 compose_output_action(ctx, ofp_port);
4592 ctx->nf_output_iface = NF_OUT_FLOOD;
4596 execute_controller_action(struct action_xlate_ctx *ctx, int len,
4597 enum ofp_packet_in_reason reason,
4598 uint16_t controller_id)
4600 struct ofputil_packet_in pin;
4601 struct ofpbuf *packet;
4603 ctx->may_set_up_flow = false;
4608 packet = ofpbuf_clone(ctx->packet);
4610 if (packet->l2 && packet->l3) {
4611 struct eth_header *eh;
4613 eth_pop_vlan(packet);
4616 /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
4617 * LLC frame. Calculating the Ethernet type of these frames is more
4618 * trouble than seems appropriate for a simple assertion. */
4619 assert(ntohs(eh->eth_type) < ETH_TYPE_MIN
4620 || eh->eth_type == ctx->flow.dl_type);
4622 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
4623 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
4625 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
4626 eth_push_vlan(packet, ctx->flow.vlan_tci);
4630 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4631 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
4632 ctx->flow.nw_tos, ctx->flow.nw_ttl);
4636 if (ctx->flow.nw_proto == IPPROTO_TCP) {
4637 packet_set_tcp_port(packet, ctx->flow.tp_src,
4639 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
4640 packet_set_udp_port(packet, ctx->flow.tp_src,
4647 pin.packet = packet->data;
4648 pin.packet_len = packet->size;
4649 pin.reason = reason;
4650 pin.controller_id = controller_id;
4651 pin.table_id = ctx->table_id;
4652 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
4655 flow_get_metadata(&ctx->flow, &pin.fmd);
4657 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
4658 ofpbuf_delete(packet);
4662 compose_dec_ttl(struct action_xlate_ctx *ctx)
4664 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
4665 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
4669 if (ctx->flow.nw_ttl > 1) {
4673 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
4675 /* Stop processing for current table. */
4681 xlate_output_action__(struct action_xlate_ctx *ctx,
4682 uint16_t port, uint16_t max_len)
4684 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
4686 ctx->nf_output_iface = NF_OUT_DROP;
4690 compose_output_action(ctx, ctx->flow.in_port);
4693 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
4699 flood_packets(ctx, false);
4702 flood_packets(ctx, true);
4704 case OFPP_CONTROLLER:
4705 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
4711 if (port != ctx->flow.in_port) {
4712 compose_output_action(ctx, port);
4717 if (prev_nf_output_iface == NF_OUT_FLOOD) {
4718 ctx->nf_output_iface = NF_OUT_FLOOD;
4719 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
4720 ctx->nf_output_iface = prev_nf_output_iface;
4721 } else if (prev_nf_output_iface != NF_OUT_DROP &&
4722 ctx->nf_output_iface != NF_OUT_FLOOD) {
4723 ctx->nf_output_iface = NF_OUT_MULTI;
4728 xlate_output_reg_action(struct action_xlate_ctx *ctx,
4729 const struct nx_action_output_reg *naor)
4731 struct mf_subfield src;
4734 nxm_decode(&src, naor->src, naor->ofs_nbits);
4735 ofp_port = mf_get_subfield(&src, &ctx->flow);
4737 if (ofp_port <= UINT16_MAX) {
4738 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
4743 xlate_output_action(struct action_xlate_ctx *ctx,
4744 const struct ofp_action_output *oao)
4746 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
4750 xlate_enqueue_action(struct action_xlate_ctx *ctx,
4751 const struct ofp_action_enqueue *oae)
4754 uint32_t flow_priority, priority;
4757 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
4760 /* Fall back to ordinary output action. */
4761 xlate_output_action__(ctx, ntohs(oae->port), 0);
4765 /* Figure out datapath output port. */
4766 ofp_port = ntohs(oae->port);
4767 if (ofp_port == OFPP_IN_PORT) {
4768 ofp_port = ctx->flow.in_port;
4769 } else if (ofp_port == ctx->flow.in_port) {
4773 /* Add datapath actions. */
4774 flow_priority = ctx->flow.skb_priority;
4775 ctx->flow.skb_priority = priority;
4776 compose_output_action(ctx, ofp_port);
4777 ctx->flow.skb_priority = flow_priority;
4779 /* Update NetFlow output port. */
4780 if (ctx->nf_output_iface == NF_OUT_DROP) {
4781 ctx->nf_output_iface = ofp_port;
4782 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
4783 ctx->nf_output_iface = NF_OUT_MULTI;
4788 xlate_set_queue_action(struct action_xlate_ctx *ctx,
4789 const struct nx_action_set_queue *nasq)
4794 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
4797 /* Couldn't translate queue to a priority, so ignore. A warning
4798 * has already been logged. */
4802 ctx->flow.skb_priority = priority;
4805 struct xlate_reg_state {
4811 xlate_autopath(struct action_xlate_ctx *ctx,
4812 const struct nx_action_autopath *naa)
4814 uint16_t ofp_port = ntohl(naa->id);
4815 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
4817 if (!port || !port->bundle) {
4818 ofp_port = OFPP_NONE;
4819 } else if (port->bundle->bond) {
4820 /* Autopath does not support VLAN hashing. */
4821 struct ofport_dpif *slave = bond_choose_output_slave(
4822 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
4824 ofp_port = slave->up.ofp_port;
4827 autopath_execute(naa, &ctx->flow, ofp_port);
4831 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
4833 struct ofproto_dpif *ofproto = ofproto_;
4834 struct ofport_dpif *port;
4844 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
4847 port = get_ofp_port(ofproto, ofp_port);
4848 return port ? port->may_enable : false;
4853 xlate_learn_action(struct action_xlate_ctx *ctx,
4854 const struct nx_action_learn *learn)
4856 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
4857 struct ofputil_flow_mod fm;
4860 learn_execute(learn, &ctx->flow, &fm);
4862 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
4863 if (error && !VLOG_DROP_WARN(&rl)) {
4864 VLOG_WARN("learning action failed to modify flow table (%s)",
4865 ofperr_get_name(error));
4871 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
4872 * means "infinite". */
4874 reduce_timeout(uint16_t max, uint16_t *timeout)
4876 if (max && (!*timeout || *timeout > max)) {
4882 xlate_fin_timeout(struct action_xlate_ctx *ctx,
4883 const struct nx_action_fin_timeout *naft)
4885 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
4886 struct rule_dpif *rule = ctx->rule;
4888 reduce_timeout(ntohs(naft->fin_idle_timeout), &rule->up.idle_timeout);
4889 reduce_timeout(ntohs(naft->fin_hard_timeout), &rule->up.hard_timeout);
4894 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
4896 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
4897 ? OFPUTIL_PC_NO_RECV_STP
4898 : OFPUTIL_PC_NO_RECV)) {
4902 /* Only drop packets here if both forwarding and learning are
4903 * disabled. If just learning is enabled, we need to have
4904 * OFPP_NORMAL and the learning action have a look at the packet
4905 * before we can drop it. */
4906 if (!stp_forward_in_state(port->stp_state)
4907 && !stp_learn_in_state(port->stp_state)) {
4915 do_xlate_actions(const union ofp_action *in, size_t n_in,
4916 struct action_xlate_ctx *ctx)
4918 const struct ofport_dpif *port;
4919 const union ofp_action *ia;
4920 bool was_evictable = true;
4923 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
4924 if (port && !may_receive(port, ctx)) {
4925 /* Drop this flow. */
4930 /* Don't let the rule we're working on get evicted underneath us. */
4931 was_evictable = ctx->rule->up.evictable;
4932 ctx->rule->up.evictable = false;
4934 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
4935 const struct ofp_action_dl_addr *oada;
4936 const struct nx_action_resubmit *nar;
4937 const struct nx_action_set_tunnel *nast;
4938 const struct nx_action_set_queue *nasq;
4939 const struct nx_action_multipath *nam;
4940 const struct nx_action_autopath *naa;
4941 const struct nx_action_bundle *nab;
4942 const struct nx_action_output_reg *naor;
4943 const struct nx_action_controller *nac;
4944 enum ofputil_action_code code;
4951 code = ofputil_decode_action_unsafe(ia);
4953 case OFPUTIL_OFPAT10_OUTPUT:
4954 xlate_output_action(ctx, &ia->output);
4957 case OFPUTIL_OFPAT10_SET_VLAN_VID:
4958 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
4959 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
4962 case OFPUTIL_OFPAT10_SET_VLAN_PCP:
4963 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
4964 ctx->flow.vlan_tci |= htons(
4965 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
4968 case OFPUTIL_OFPAT10_STRIP_VLAN:
4969 ctx->flow.vlan_tci = htons(0);
4972 case OFPUTIL_OFPAT10_SET_DL_SRC:
4973 oada = ((struct ofp_action_dl_addr *) ia);
4974 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
4977 case OFPUTIL_OFPAT10_SET_DL_DST:
4978 oada = ((struct ofp_action_dl_addr *) ia);
4979 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
4982 case OFPUTIL_OFPAT10_SET_NW_SRC:
4983 ctx->flow.nw_src = ia->nw_addr.nw_addr;
4986 case OFPUTIL_OFPAT10_SET_NW_DST:
4987 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
4990 case OFPUTIL_OFPAT10_SET_NW_TOS:
4991 /* OpenFlow 1.0 only supports IPv4. */
4992 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4993 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4994 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
4998 case OFPUTIL_OFPAT10_SET_TP_SRC:
4999 ctx->flow.tp_src = ia->tp_port.tp_port;
5002 case OFPUTIL_OFPAT10_SET_TP_DST:
5003 ctx->flow.tp_dst = ia->tp_port.tp_port;
5006 case OFPUTIL_OFPAT10_ENQUEUE:
5007 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
5010 case OFPUTIL_NXAST_RESUBMIT:
5011 nar = (const struct nx_action_resubmit *) ia;
5012 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
5015 case OFPUTIL_NXAST_RESUBMIT_TABLE:
5016 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
5019 case OFPUTIL_NXAST_SET_TUNNEL:
5020 nast = (const struct nx_action_set_tunnel *) ia;
5021 tun_id = htonll(ntohl(nast->tun_id));
5022 ctx->flow.tun_id = tun_id;
5025 case OFPUTIL_NXAST_SET_QUEUE:
5026 nasq = (const struct nx_action_set_queue *) ia;
5027 xlate_set_queue_action(ctx, nasq);
5030 case OFPUTIL_NXAST_POP_QUEUE:
5031 ctx->flow.skb_priority = ctx->orig_skb_priority;
5034 case OFPUTIL_NXAST_REG_MOVE:
5035 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
5039 case OFPUTIL_NXAST_REG_LOAD:
5040 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
5044 case OFPUTIL_NXAST_NOTE:
5045 /* Nothing to do. */
5048 case OFPUTIL_NXAST_SET_TUNNEL64:
5049 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
5050 ctx->flow.tun_id = tun_id;
5053 case OFPUTIL_NXAST_MULTIPATH:
5054 nam = (const struct nx_action_multipath *) ia;
5055 multipath_execute(nam, &ctx->flow);
5058 case OFPUTIL_NXAST_AUTOPATH:
5059 naa = (const struct nx_action_autopath *) ia;
5060 xlate_autopath(ctx, naa);
5063 case OFPUTIL_NXAST_BUNDLE:
5064 ctx->ofproto->has_bundle_action = true;
5065 nab = (const struct nx_action_bundle *) ia;
5066 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
5071 case OFPUTIL_NXAST_BUNDLE_LOAD:
5072 ctx->ofproto->has_bundle_action = true;
5073 nab = (const struct nx_action_bundle *) ia;
5074 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
5078 case OFPUTIL_NXAST_OUTPUT_REG:
5079 naor = (const struct nx_action_output_reg *) ia;
5080 xlate_output_reg_action(ctx, naor);
5083 case OFPUTIL_NXAST_LEARN:
5084 ctx->has_learn = true;
5085 if (ctx->may_learn) {
5086 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
5090 case OFPUTIL_NXAST_DEC_TTL:
5091 if (compose_dec_ttl(ctx)) {
5096 case OFPUTIL_NXAST_EXIT:
5100 case OFPUTIL_NXAST_FIN_TIMEOUT:
5101 ctx->has_fin_timeout = true;
5102 xlate_fin_timeout(ctx, (const struct nx_action_fin_timeout *) ia);
5105 case OFPUTIL_NXAST_CONTROLLER:
5106 nac = (const struct nx_action_controller *) ia;
5107 execute_controller_action(ctx, ntohs(nac->max_len), nac->reason,
5108 ntohs(nac->controller_id));
5114 /* We've let OFPP_NORMAL and the learning action look at the packet,
5115 * so drop it now if forwarding is disabled. */
5116 if (port && !stp_forward_in_state(port->stp_state)) {
5117 ofpbuf_clear(ctx->odp_actions);
5118 add_sflow_action(ctx);
5121 ctx->rule->up.evictable = was_evictable;
5126 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
5127 struct ofproto_dpif *ofproto, const struct flow *flow,
5128 ovs_be16 initial_tci, struct rule_dpif *rule,
5129 uint8_t tcp_flags, const struct ofpbuf *packet)
5131 ctx->ofproto = ofproto;
5133 ctx->base_flow = ctx->flow;
5134 ctx->base_flow.tun_id = 0;
5135 ctx->base_flow.vlan_tci = initial_tci;
5137 ctx->packet = packet;
5138 ctx->may_learn = packet != NULL;
5139 ctx->tcp_flags = tcp_flags;
5140 ctx->resubmit_hook = NULL;
5143 /* Translates the 'n_in' "union ofp_action"s in 'in' into datapath actions in
5144 * 'odp_actions', using 'ctx'. */
5146 xlate_actions(struct action_xlate_ctx *ctx,
5147 const union ofp_action *in, size_t n_in,
5148 struct ofpbuf *odp_actions)
5150 struct flow orig_flow = ctx->flow;
5152 COVERAGE_INC(ofproto_dpif_xlate);
5154 ofpbuf_clear(odp_actions);
5155 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
5157 ctx->odp_actions = odp_actions;
5159 ctx->may_set_up_flow = true;
5160 ctx->has_learn = false;
5161 ctx->has_normal = false;
5162 ctx->has_fin_timeout = false;
5163 ctx->nf_output_iface = NF_OUT_DROP;
5166 ctx->max_resubmit_trigger = false;
5167 ctx->orig_skb_priority = ctx->flow.skb_priority;
5171 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
5172 switch (ctx->ofproto->up.frag_handling) {
5173 case OFPC_FRAG_NORMAL:
5174 /* We must pretend that transport ports are unavailable. */
5175 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
5176 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
5179 case OFPC_FRAG_DROP:
5182 case OFPC_FRAG_REASM:
5185 case OFPC_FRAG_NX_MATCH:
5186 /* Nothing to do. */
5189 case OFPC_INVALID_TTL_TO_CONTROLLER:
5194 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
5195 ctx->may_set_up_flow = false;
5197 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5198 struct flow original_flow = ctx->flow;
5199 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
5201 add_sflow_action(ctx);
5202 do_xlate_actions(in, n_in, ctx);
5204 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook
5205 && !VLOG_DROP_ERR(&trace_rl)) {
5206 struct ds ds = DS_EMPTY_INITIALIZER;
5208 ofproto_trace(ctx->ofproto, &original_flow, ctx->packet,
5210 VLOG_ERR("Trace triggered by excessive resubmit recursion:\n%s",
5215 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
5216 ctx->odp_actions->data,
5217 ctx->odp_actions->size)) {
5218 ctx->may_set_up_flow = false;
5220 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
5222 compose_output_action(ctx, OFPP_LOCAL);
5225 add_mirror_actions(ctx, &orig_flow);
5226 fix_sflow_action(ctx);
5230 /* Translates the 'n_in' "union ofp_action"s in 'in' into datapath actions,
5231 * using 'ctx', and discards the datapath actions. */
5233 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
5234 const union ofp_action *in, size_t n_in)
5236 uint64_t odp_actions_stub[1024 / 8];
5237 struct ofpbuf odp_actions;
5239 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5240 xlate_actions(ctx, in, n_in, &odp_actions);
5241 ofpbuf_uninit(&odp_actions);
5244 /* OFPP_NORMAL implementation. */
5246 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
5248 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
5249 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
5250 * the bundle on which the packet was received, returns the VLAN to which the
5253 * Both 'vid' and the return value are in the range 0...4095. */
5255 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
5257 switch (in_bundle->vlan_mode) {
5258 case PORT_VLAN_ACCESS:
5259 return in_bundle->vlan;
5262 case PORT_VLAN_TRUNK:
5265 case PORT_VLAN_NATIVE_UNTAGGED:
5266 case PORT_VLAN_NATIVE_TAGGED:
5267 return vid ? vid : in_bundle->vlan;
5274 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
5275 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
5278 * 'vid' should be the VID obtained from the 802.1Q header that was received as
5279 * part of a packet (specify 0 if there was no 802.1Q header), in the range
5282 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
5284 /* Allow any VID on the OFPP_NONE port. */
5285 if (in_bundle == &ofpp_none_bundle) {
5289 switch (in_bundle->vlan_mode) {
5290 case PORT_VLAN_ACCESS:
5293 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5294 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
5295 "packet received on port %s configured as VLAN "
5296 "%"PRIu16" access port",
5297 in_bundle->ofproto->up.name, vid,
5298 in_bundle->name, in_bundle->vlan);
5304 case PORT_VLAN_NATIVE_UNTAGGED:
5305 case PORT_VLAN_NATIVE_TAGGED:
5307 /* Port must always carry its native VLAN. */
5311 case PORT_VLAN_TRUNK:
5312 if (!ofbundle_includes_vlan(in_bundle, vid)) {
5314 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5315 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
5316 "received on port %s not configured for trunking "
5318 in_bundle->ofproto->up.name, vid,
5319 in_bundle->name, vid);
5331 /* Given 'vlan', the VLAN that a packet belongs to, and
5332 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
5333 * that should be included in the 802.1Q header. (If the return value is 0,
5334 * then the 802.1Q header should only be included in the packet if there is a
5337 * Both 'vlan' and the return value are in the range 0...4095. */
5339 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
5341 switch (out_bundle->vlan_mode) {
5342 case PORT_VLAN_ACCESS:
5345 case PORT_VLAN_TRUNK:
5346 case PORT_VLAN_NATIVE_TAGGED:
5349 case PORT_VLAN_NATIVE_UNTAGGED:
5350 return vlan == out_bundle->vlan ? 0 : vlan;
5358 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
5361 struct ofport_dpif *port;
5363 ovs_be16 tci, old_tci;
5365 vid = output_vlan_to_vid(out_bundle, vlan);
5366 if (!out_bundle->bond) {
5367 port = ofbundle_get_a_port(out_bundle);
5369 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
5372 /* No slaves enabled, so drop packet. */
5377 old_tci = ctx->flow.vlan_tci;
5379 if (tci || out_bundle->use_priority_tags) {
5380 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
5382 tci |= htons(VLAN_CFI);
5385 ctx->flow.vlan_tci = tci;
5387 compose_output_action(ctx, port->up.ofp_port);
5388 ctx->flow.vlan_tci = old_tci;
5392 mirror_mask_ffs(mirror_mask_t mask)
5394 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
5399 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
5401 return (bundle->vlan_mode != PORT_VLAN_ACCESS
5402 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
5406 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
5408 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
5411 /* Returns an arbitrary interface within 'bundle'. */
5412 static struct ofport_dpif *
5413 ofbundle_get_a_port(const struct ofbundle *bundle)
5415 return CONTAINER_OF(list_front(&bundle->ports),
5416 struct ofport_dpif, bundle_node);
5420 vlan_is_mirrored(const struct ofmirror *m, int vlan)
5422 return !m->vlans || bitmap_is_set(m->vlans, vlan);
5425 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
5426 * to a VLAN. In general most packets may be mirrored but we want to drop
5427 * protocols that may confuse switches. */
5429 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
5431 /* If you change this function's behavior, please update corresponding
5432 * documentation in vswitch.xml at the same time. */
5433 if (dst[0] != 0x01) {
5434 /* All the currently banned MACs happen to start with 01 currently, so
5435 * this is a quick way to eliminate most of the good ones. */
5437 if (eth_addr_is_reserved(dst)) {
5438 /* Drop STP, IEEE pause frames, and other reserved protocols
5439 * (01-80-c2-00-00-0x). */
5443 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
5445 if ((dst[3] & 0xfe) == 0xcc &&
5446 (dst[4] & 0xfe) == 0xcc &&
5447 (dst[5] & 0xfe) == 0xcc) {
5448 /* Drop the following protocols plus others following the same
5451 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
5452 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
5453 STP Uplink Fast (01-00-0c-cd-cd-cd) */
5457 if (!(dst[3] | dst[4] | dst[5])) {
5458 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
5467 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5469 struct ofproto_dpif *ofproto = ctx->ofproto;
5470 mirror_mask_t mirrors;
5471 struct ofbundle *in_bundle;
5474 const struct nlattr *a;
5477 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5478 ctx->packet != NULL);
5482 mirrors = in_bundle->src_mirrors;
5484 /* Drop frames on bundles reserved for mirroring. */
5485 if (in_bundle->mirror_out) {
5486 if (ctx->packet != NULL) {
5487 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5488 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5489 "%s, which is reserved exclusively for mirroring",
5490 ctx->ofproto->up.name, in_bundle->name);
5496 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5497 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5500 vlan = input_vid_to_vlan(in_bundle, vid);
5502 /* Look at the output ports to check for destination selections. */
5504 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5505 ctx->odp_actions->size) {
5506 enum ovs_action_attr type = nl_attr_type(a);
5507 struct ofport_dpif *ofport;
5509 if (type != OVS_ACTION_ATTR_OUTPUT) {
5513 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5514 if (ofport && ofport->bundle) {
5515 mirrors |= ofport->bundle->dst_mirrors;
5523 /* Restore the original packet before adding the mirror actions. */
5524 ctx->flow = *orig_flow;
5529 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5531 if (!vlan_is_mirrored(m, vlan)) {
5532 mirrors &= mirrors - 1;
5536 mirrors &= ~m->dup_mirrors;
5537 ctx->mirrors |= m->dup_mirrors;
5539 output_normal(ctx, m->out, vlan);
5540 } else if (eth_dst_may_rspan(orig_flow->dl_dst)
5541 && vlan != m->out_vlan) {
5542 struct ofbundle *bundle;
5544 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5545 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5546 && !bundle->mirror_out) {
5547 output_normal(ctx, bundle, m->out_vlan);
5555 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5556 uint64_t packets, uint64_t bytes)
5562 for (; mirrors; mirrors &= mirrors - 1) {
5565 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5568 /* In normal circumstances 'm' will not be NULL. However,
5569 * if mirrors are reconfigured, we can temporarily get out
5570 * of sync in facet_revalidate(). We could "correct" the
5571 * mirror list before reaching here, but doing that would
5572 * not properly account the traffic stats we've currently
5573 * accumulated for previous mirror configuration. */
5577 m->packet_count += packets;
5578 m->byte_count += bytes;
5582 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
5583 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
5584 * indicate this; newer upstream kernels use gratuitous ARP requests. */
5586 is_gratuitous_arp(const struct flow *flow)
5588 return (flow->dl_type == htons(ETH_TYPE_ARP)
5589 && eth_addr_is_broadcast(flow->dl_dst)
5590 && (flow->nw_proto == ARP_OP_REPLY
5591 || (flow->nw_proto == ARP_OP_REQUEST
5592 && flow->nw_src == flow->nw_dst)));
5596 update_learning_table(struct ofproto_dpif *ofproto,
5597 const struct flow *flow, int vlan,
5598 struct ofbundle *in_bundle)
5600 struct mac_entry *mac;
5602 /* Don't learn the OFPP_NONE port. */
5603 if (in_bundle == &ofpp_none_bundle) {
5607 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
5611 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
5612 if (is_gratuitous_arp(flow)) {
5613 /* We don't want to learn from gratuitous ARP packets that are
5614 * reflected back over bond slaves so we lock the learning table. */
5615 if (!in_bundle->bond) {
5616 mac_entry_set_grat_arp_lock(mac);
5617 } else if (mac_entry_is_grat_arp_locked(mac)) {
5622 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
5623 /* The log messages here could actually be useful in debugging,
5624 * so keep the rate limit relatively high. */
5625 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5626 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
5627 "on port %s in VLAN %d",
5628 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
5629 in_bundle->name, vlan);
5631 mac->port.p = in_bundle;
5632 tag_set_add(&ofproto->revalidate_set,
5633 mac_learning_changed(ofproto->ml, mac));
5637 static struct ofbundle *
5638 lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn)
5640 struct ofport_dpif *ofport;
5642 /* Special-case OFPP_NONE, which a controller may use as the ingress
5643 * port for traffic that it is sourcing. */
5644 if (in_port == OFPP_NONE) {
5645 return &ofpp_none_bundle;
5648 /* Find the port and bundle for the received packet. */
5649 ofport = get_ofp_port(ofproto, in_port);
5650 if (ofport && ofport->bundle) {
5651 return ofport->bundle;
5654 /* Odd. A few possible reasons here:
5656 * - We deleted a port but there are still a few packets queued up
5659 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
5660 * we don't know about.
5662 * - The ofproto client didn't configure the port as part of a bundle.
5665 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5667 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
5668 "port %"PRIu16, ofproto->up.name, in_port);
5673 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
5674 * dropped. Returns true if they may be forwarded, false if they should be
5677 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
5678 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
5680 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
5681 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
5682 * checked by input_vid_is_valid().
5684 * May also add tags to '*tags', although the current implementation only does
5685 * so in one special case.
5688 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
5689 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
5691 struct ofbundle *in_bundle = in_port->bundle;
5693 /* Drop frames for reserved multicast addresses
5694 * only if forward_bpdu option is absent. */
5695 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
5699 if (in_bundle->bond) {
5700 struct mac_entry *mac;
5702 switch (bond_check_admissibility(in_bundle->bond, in_port,
5703 flow->dl_dst, tags)) {
5710 case BV_DROP_IF_MOVED:
5711 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
5712 if (mac && mac->port.p != in_bundle &&
5713 (!is_gratuitous_arp(flow)
5714 || mac_entry_is_grat_arp_locked(mac))) {
5725 xlate_normal(struct action_xlate_ctx *ctx)
5727 struct ofport_dpif *in_port;
5728 struct ofbundle *in_bundle;
5729 struct mac_entry *mac;
5733 ctx->has_normal = true;
5735 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
5736 ctx->packet != NULL);
5741 /* We know 'in_port' exists unless it is "ofpp_none_bundle",
5742 * since lookup_input_bundle() succeeded. */
5743 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5745 /* Drop malformed frames. */
5746 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
5747 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
5748 if (ctx->packet != NULL) {
5749 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5750 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
5751 "VLAN tag received on port %s",
5752 ctx->ofproto->up.name, in_bundle->name);
5757 /* Drop frames on bundles reserved for mirroring. */
5758 if (in_bundle->mirror_out) {
5759 if (ctx->packet != NULL) {
5760 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5761 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5762 "%s, which is reserved exclusively for mirroring",
5763 ctx->ofproto->up.name, in_bundle->name);
5769 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
5770 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5773 vlan = input_vid_to_vlan(in_bundle, vid);
5775 /* Check other admissibility requirements. */
5777 !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
5781 /* Learn source MAC. */
5782 if (ctx->may_learn) {
5783 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
5786 /* Determine output bundle. */
5787 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
5790 if (mac->port.p != in_bundle) {
5791 output_normal(ctx, mac->port.p, vlan);
5794 struct ofbundle *bundle;
5796 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
5797 if (bundle != in_bundle
5798 && ofbundle_includes_vlan(bundle, vlan)
5799 && bundle->floodable
5800 && !bundle->mirror_out) {
5801 output_normal(ctx, bundle, vlan);
5804 ctx->nf_output_iface = NF_OUT_FLOOD;
5808 /* Optimized flow revalidation.
5810 * It's a difficult problem, in general, to tell which facets need to have
5811 * their actions recalculated whenever the OpenFlow flow table changes. We
5812 * don't try to solve that general problem: for most kinds of OpenFlow flow
5813 * table changes, we recalculate the actions for every facet. This is
5814 * relatively expensive, but it's good enough if the OpenFlow flow table
5815 * doesn't change very often.
5817 * However, we can expect one particular kind of OpenFlow flow table change to
5818 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5819 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5820 * table, we add a special case that applies to flow tables in which every rule
5821 * has the same form (that is, the same wildcards), except that the table is
5822 * also allowed to have a single "catch-all" flow that matches all packets. We
5823 * optimize this case by tagging all of the facets that resubmit into the table
5824 * and invalidating the same tag whenever a flow changes in that table. The
5825 * end result is that we revalidate just the facets that need it (and sometimes
5826 * a few more, but not all of the facets or even all of the facets that
5827 * resubmit to the table modified by MAC learning). */
5829 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
5830 * into an OpenFlow table with the given 'basis'. */
5832 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
5835 if (flow_wildcards_is_catchall(wc)) {
5838 struct flow tag_flow = *flow;
5839 flow_zero_wildcards(&tag_flow, wc);
5840 return tag_create_deterministic(flow_hash(&tag_flow, secret));
5844 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5845 * taggability of that table.
5847 * This function must be called after *each* change to a flow table. If you
5848 * skip calling it on some changes then the pointer comparisons at the end can
5849 * be invalid if you get unlucky. For example, if a flow removal causes a
5850 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5851 * different wildcards to be created with the same address, then this function
5852 * will incorrectly skip revalidation. */
5854 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5856 struct table_dpif *table = &ofproto->tables[table_id];
5857 const struct oftable *oftable = &ofproto->up.tables[table_id];
5858 struct cls_table *catchall, *other;
5859 struct cls_table *t;
5861 catchall = other = NULL;
5863 switch (hmap_count(&oftable->cls.tables)) {
5865 /* We could tag this OpenFlow table but it would make the logic a
5866 * little harder and it's a corner case that doesn't seem worth it
5872 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
5873 if (cls_table_is_catchall(t)) {
5875 } else if (!other) {
5878 /* Indicate that we can't tag this by setting both tables to
5879 * NULL. (We know that 'catchall' is already NULL.) */
5886 /* Can't tag this table. */
5890 if (table->catchall_table != catchall || table->other_table != other) {
5891 table->catchall_table = catchall;
5892 table->other_table = other;
5893 ofproto->need_revalidate = true;
5897 /* Given 'rule' that has changed in some way (either it is a rule being
5898 * inserted, a rule being deleted, or a rule whose actions are being
5899 * modified), marks facets for revalidation to ensure that packets will be
5900 * forwarded correctly according to the new state of the flow table.
5902 * This function must be called after *each* change to a flow table. See
5903 * the comment on table_update_taggable() for more information. */
5905 rule_invalidate(const struct rule_dpif *rule)
5907 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5909 table_update_taggable(ofproto, rule->up.table_id);
5911 if (!ofproto->need_revalidate) {
5912 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5914 if (table->other_table && rule->tag) {
5915 tag_set_add(&ofproto->revalidate_set, rule->tag);
5917 ofproto->need_revalidate = true;
5923 set_frag_handling(struct ofproto *ofproto_,
5924 enum ofp_config_flags frag_handling)
5926 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5928 if (frag_handling != OFPC_FRAG_REASM) {
5929 ofproto->need_revalidate = true;
5937 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5938 const struct flow *flow,
5939 const union ofp_action *ofp_actions, size_t n_ofp_actions)
5941 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5944 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
5945 return OFPERR_NXBRC_BAD_IN_PORT;
5948 error = validate_actions(ofp_actions, n_ofp_actions, flow,
5949 ofproto->max_ports);
5951 struct odputil_keybuf keybuf;
5954 uint64_t odp_actions_stub[1024 / 8];
5955 struct ofpbuf odp_actions;
5956 struct ofproto_push push;
5958 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5959 odp_flow_key_from_flow(&key, flow);
5961 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, NULL,
5962 packet_get_tcp_flags(packet, flow), packet);
5964 /* Ensure that resubmits in 'ofp_actions' get accounted to their
5965 * matching rules. */
5967 push.bytes = packet->size;
5968 push.used = time_msec();
5969 push.ctx.resubmit_hook = push_resubmit;
5971 ofpbuf_use_stub(&odp_actions,
5972 odp_actions_stub, sizeof odp_actions_stub);
5973 xlate_actions(&push.ctx, ofp_actions, n_ofp_actions, &odp_actions);
5974 dpif_execute(ofproto->dpif, key.data, key.size,
5975 odp_actions.data, odp_actions.size, packet);
5976 ofpbuf_uninit(&odp_actions);
5984 set_netflow(struct ofproto *ofproto_,
5985 const struct netflow_options *netflow_options)
5987 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5989 if (netflow_options) {
5990 if (!ofproto->netflow) {
5991 ofproto->netflow = netflow_create();
5993 return netflow_set_options(ofproto->netflow, netflow_options);
5995 netflow_destroy(ofproto->netflow);
5996 ofproto->netflow = NULL;
6002 get_netflow_ids(const struct ofproto *ofproto_,
6003 uint8_t *engine_type, uint8_t *engine_id)
6005 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6007 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
6011 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
6013 if (!facet_is_controller_flow(facet) &&
6014 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
6015 struct subfacet *subfacet;
6016 struct ofexpired expired;
6018 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
6019 if (subfacet->installed) {
6020 struct dpif_flow_stats stats;
6022 subfacet_install(subfacet, subfacet->actions,
6023 subfacet->actions_len, &stats);
6024 subfacet_update_stats(subfacet, &stats);
6028 expired.flow = facet->flow;
6029 expired.packet_count = facet->packet_count;
6030 expired.byte_count = facet->byte_count;
6031 expired.used = facet->used;
6032 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
6037 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
6039 struct facet *facet;
6041 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6042 send_active_timeout(ofproto, facet);
6046 static struct ofproto_dpif *
6047 ofproto_dpif_lookup(const char *name)
6049 struct ofproto_dpif *ofproto;
6051 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
6052 hash_string(name, 0), &all_ofproto_dpifs) {
6053 if (!strcmp(ofproto->up.name, name)) {
6061 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
6062 const char *argv[], void *aux OVS_UNUSED)
6064 struct ofproto_dpif *ofproto;
6067 ofproto = ofproto_dpif_lookup(argv[1]);
6069 unixctl_command_reply_error(conn, "no such bridge");
6072 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6074 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6075 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6079 unixctl_command_reply(conn, "table successfully flushed");
6083 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6084 const char *argv[], void *aux OVS_UNUSED)
6086 struct ds ds = DS_EMPTY_INITIALIZER;
6087 const struct ofproto_dpif *ofproto;
6088 const struct mac_entry *e;
6090 ofproto = ofproto_dpif_lookup(argv[1]);
6092 unixctl_command_reply_error(conn, "no such bridge");
6096 ds_put_cstr(&ds, " port VLAN MAC Age\n");
6097 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
6098 struct ofbundle *bundle = e->port.p;
6099 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
6100 ofbundle_get_a_port(bundle)->odp_port,
6101 e->vlan, ETH_ADDR_ARGS(e->mac),
6102 mac_entry_age(ofproto->ml, e));
6104 unixctl_command_reply(conn, ds_cstr(&ds));
6109 struct action_xlate_ctx ctx;
6115 trace_format_rule(struct ds *result, uint8_t table_id, int level,
6116 const struct rule_dpif *rule)
6118 ds_put_char_multiple(result, '\t', level);
6120 ds_put_cstr(result, "No match\n");
6124 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
6125 table_id, ntohll(rule->up.flow_cookie));
6126 cls_rule_format(&rule->up.cr, result);
6127 ds_put_char(result, '\n');
6129 ds_put_char_multiple(result, '\t', level);
6130 ds_put_cstr(result, "OpenFlow ");
6131 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
6132 ds_put_char(result, '\n');
6136 trace_format_flow(struct ds *result, int level, const char *title,
6137 struct trace_ctx *trace)
6139 ds_put_char_multiple(result, '\t', level);
6140 ds_put_format(result, "%s: ", title);
6141 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
6142 ds_put_cstr(result, "unchanged");
6144 flow_format(result, &trace->ctx.flow);
6145 trace->flow = trace->ctx.flow;
6147 ds_put_char(result, '\n');
6151 trace_format_regs(struct ds *result, int level, const char *title,
6152 struct trace_ctx *trace)
6156 ds_put_char_multiple(result, '\t', level);
6157 ds_put_format(result, "%s:", title);
6158 for (i = 0; i < FLOW_N_REGS; i++) {
6159 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
6161 ds_put_char(result, '\n');
6165 trace_format_odp(struct ds *result, int level, const char *title,
6166 struct trace_ctx *trace)
6168 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
6170 ds_put_char_multiple(result, '\t', level);
6171 ds_put_format(result, "%s: ", title);
6172 format_odp_actions(result, odp_actions->data, odp_actions->size);
6173 ds_put_char(result, '\n');
6177 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
6179 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
6180 struct ds *result = trace->result;
6182 ds_put_char(result, '\n');
6183 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
6184 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
6185 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
6186 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
6190 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
6191 void *aux OVS_UNUSED)
6193 const char *dpname = argv[1];
6194 struct ofproto_dpif *ofproto;
6195 struct ofpbuf odp_key;
6196 struct ofpbuf *packet;
6197 ovs_be16 initial_tci;
6203 ofpbuf_init(&odp_key, 0);
6206 ofproto = ofproto_dpif_lookup(dpname);
6208 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
6212 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
6213 /* ofproto/trace dpname flow [-generate] */
6214 const char *flow_s = argv[2];
6215 const char *generate_s = argv[3];
6218 /* Convert string to datapath key. */
6219 ofpbuf_init(&odp_key, 0);
6220 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
6222 unixctl_command_reply_error(conn, "Bad flow syntax");
6226 /* Convert odp_key to flow. */
6227 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
6228 odp_key.size, &flow,
6229 &initial_tci, NULL);
6230 if (error == ODP_FIT_ERROR) {
6231 unixctl_command_reply_error(conn, "Invalid flow");
6235 /* Generate a packet, if requested. */
6237 packet = ofpbuf_new(0);
6238 flow_compose(packet, &flow);
6240 } else if (argc == 6) {
6241 /* ofproto/trace dpname priority tun_id in_port packet */
6242 const char *priority_s = argv[2];
6243 const char *tun_id_s = argv[3];
6244 const char *in_port_s = argv[4];
6245 const char *packet_s = argv[5];
6246 uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s));
6247 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
6248 uint32_t priority = atoi(priority_s);
6251 msg = eth_from_hex(packet_s, &packet);
6253 unixctl_command_reply_error(conn, msg);
6257 ds_put_cstr(&result, "Packet: ");
6258 s = ofp_packet_to_string(packet->data, packet->size);
6259 ds_put_cstr(&result, s);
6262 flow_extract(packet, priority, tun_id, in_port, &flow);
6263 initial_tci = flow.vlan_tci;
6265 unixctl_command_reply_error(conn, "Bad command syntax");
6269 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
6270 unixctl_command_reply(conn, ds_cstr(&result));
6273 ds_destroy(&result);
6274 ofpbuf_delete(packet);
6275 ofpbuf_uninit(&odp_key);
6279 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
6280 const struct ofpbuf *packet, ovs_be16 initial_tci,
6283 struct rule_dpif *rule;
6285 ds_put_cstr(ds, "Flow: ");
6286 flow_format(ds, flow);
6287 ds_put_char(ds, '\n');
6289 rule = rule_dpif_lookup(ofproto, flow, 0);
6290 trace_format_rule(ds, 0, 0, rule);
6292 uint64_t odp_actions_stub[1024 / 8];
6293 struct ofpbuf odp_actions;
6295 struct trace_ctx trace;
6298 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
6301 ofpbuf_use_stub(&odp_actions,
6302 odp_actions_stub, sizeof odp_actions_stub);
6303 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
6304 rule, tcp_flags, packet);
6305 trace.ctx.resubmit_hook = trace_resubmit;
6306 xlate_actions(&trace.ctx, rule->up.actions, rule->up.n_actions,
6309 ds_put_char(ds, '\n');
6310 trace_format_flow(ds, 0, "Final flow", &trace);
6311 ds_put_cstr(ds, "Datapath actions: ");
6312 format_odp_actions(ds, odp_actions.data, odp_actions.size);
6313 ofpbuf_uninit(&odp_actions);
6315 if (!trace.ctx.may_set_up_flow) {
6317 ds_put_cstr(ds, "\nThis flow is not cachable.");
6319 ds_put_cstr(ds, "\nThe datapath actions are incomplete--"
6320 "for complete actions, please supply a packet.");
6327 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6328 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6331 unixctl_command_reply(conn, NULL);
6335 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6336 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6339 unixctl_command_reply(conn, NULL);
6342 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
6343 * 'reply' describing the results. */
6345 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
6347 struct facet *facet;
6351 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6352 if (!facet_check_consistency(facet)) {
6357 ofproto->need_revalidate = true;
6361 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
6362 ofproto->up.name, errors);
6364 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
6369 ofproto_dpif_self_check(struct unixctl_conn *conn,
6370 int argc, const char *argv[], void *aux OVS_UNUSED)
6372 struct ds reply = DS_EMPTY_INITIALIZER;
6373 struct ofproto_dpif *ofproto;
6376 ofproto = ofproto_dpif_lookup(argv[1]);
6378 unixctl_command_reply_error(conn, "Unknown ofproto (use "
6379 "ofproto/list for help)");
6382 ofproto_dpif_self_check__(ofproto, &reply);
6384 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6385 ofproto_dpif_self_check__(ofproto, &reply);
6389 unixctl_command_reply(conn, ds_cstr(&reply));
6394 ofproto_dpif_unixctl_init(void)
6396 static bool registered;
6402 unixctl_command_register(
6404 "bridge {tun_id in_port packet | odp_flow [-generate]}",
6405 2, 5, ofproto_unixctl_trace, NULL);
6406 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6407 ofproto_unixctl_fdb_flush, NULL);
6408 unixctl_command_register("fdb/show", "bridge", 1, 1,
6409 ofproto_unixctl_fdb_show, NULL);
6410 unixctl_command_register("ofproto/clog", "", 0, 0,
6411 ofproto_dpif_clog, NULL);
6412 unixctl_command_register("ofproto/unclog", "", 0, 0,
6413 ofproto_dpif_unclog, NULL);
6414 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6415 ofproto_dpif_self_check, NULL);
6418 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6420 * This is deprecated. It is only for compatibility with broken device drivers
6421 * in old versions of Linux that do not properly support VLANs when VLAN
6422 * devices are not used. When broken device drivers are no longer in
6423 * widespread use, we will delete these interfaces. */
6426 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
6428 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6429 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6431 if (realdev_ofp_port == ofport->realdev_ofp_port
6432 && vid == ofport->vlandev_vid) {
6436 ofproto->need_revalidate = true;
6438 if (ofport->realdev_ofp_port) {
6441 if (realdev_ofp_port && ofport->bundle) {
6442 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6443 * themselves be part of a bundle. */
6444 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6447 ofport->realdev_ofp_port = realdev_ofp_port;
6448 ofport->vlandev_vid = vid;
6450 if (realdev_ofp_port) {
6451 vsp_add(ofport, realdev_ofp_port, vid);
6458 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
6460 return hash_2words(realdev_ofp_port, vid);
6463 /* Returns the ODP port number of the Linux VLAN device that corresponds to
6464 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
6465 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
6466 * it would return the port number of eth0.9.
6468 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
6469 * function just returns its 'realdev_odp_port' argument. */
6471 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6472 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
6474 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6475 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
6476 int vid = vlan_tci_to_vid(vlan_tci);
6477 const struct vlan_splinter *vsp;
6479 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6480 hash_realdev_vid(realdev_ofp_port, vid),
6481 &ofproto->realdev_vid_map) {
6482 if (vsp->realdev_ofp_port == realdev_ofp_port
6483 && vsp->vid == vid) {
6484 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
6488 return realdev_odp_port;
6491 static struct vlan_splinter *
6492 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
6494 struct vlan_splinter *vsp;
6496 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
6497 &ofproto->vlandev_map) {
6498 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6506 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6507 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6508 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6509 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6510 * eth0 and store 9 in '*vid'.
6512 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6513 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6516 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6517 uint16_t vlandev_ofp_port, int *vid)
6519 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6520 const struct vlan_splinter *vsp;
6522 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6527 return vsp->realdev_ofp_port;
6534 vsp_remove(struct ofport_dpif *port)
6536 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6537 struct vlan_splinter *vsp;
6539 vsp = vlandev_find(ofproto, port->up.ofp_port);
6541 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6542 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6545 port->realdev_ofp_port = 0;
6547 VLOG_ERR("missing vlan device record");
6552 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
6554 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6556 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6557 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6558 == realdev_ofp_port)) {
6559 struct vlan_splinter *vsp;
6561 vsp = xmalloc(sizeof *vsp);
6562 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6563 hash_int(port->up.ofp_port, 0));
6564 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6565 hash_realdev_vid(realdev_ofp_port, vid));
6566 vsp->realdev_ofp_port = realdev_ofp_port;
6567 vsp->vlandev_ofp_port = port->up.ofp_port;
6570 port->realdev_ofp_port = realdev_ofp_port;
6572 VLOG_ERR("duplicate vlan device record");
6576 const struct ofproto_class ofproto_dpif_class = {
6605 port_is_lacp_current,
6606 NULL, /* rule_choose_table */
6613 rule_modify_actions,
6621 get_cfm_remote_mpids,
6626 get_stp_port_status,
6633 is_mirror_output_bundle,
6634 forward_bpdu_changed,