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 /* An exact-match instantiation of an OpenFlow flow.
276 * A facet associates a "struct flow", which represents the Open vSwitch
277 * userspace idea of an exact-match flow, with one or more subfacets. Each
278 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
279 * the facet. When the kernel module (or other dpif implementation) and Open
280 * vSwitch userspace agree on the definition of a flow key, there is exactly
281 * one subfacet per facet. If the dpif implementation supports more-specific
282 * flow matching than userspace, however, a facet can have more than one
283 * subfacet, each of which corresponds to some distinction in flow that
284 * userspace simply doesn't understand.
286 * Flow expiration works in terms of subfacets, so a facet must have at least
287 * one subfacet or it will never expire, leaking memory. */
290 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
291 struct list list_node; /* In owning rule's 'facets' list. */
292 struct rule_dpif *rule; /* Owning rule. */
295 struct list subfacets;
296 long long int used; /* Time last used; time created if not used. */
303 * - Do include packets and bytes sent "by hand", e.g. with
306 * - Do include packets and bytes that were obtained from the datapath
307 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
308 * DPIF_FP_ZERO_STATS).
310 * - Do not include packets or bytes that can be obtained from the
311 * datapath for any existing subfacet.
313 uint64_t packet_count; /* Number of packets received. */
314 uint64_t byte_count; /* Number of bytes received. */
316 /* Resubmit statistics. */
317 uint64_t prev_packet_count; /* Number of packets from last stats push. */
318 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
319 long long int prev_used; /* Used time from last stats push. */
322 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
323 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
324 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
326 /* Properties of datapath actions.
328 * Every subfacet has its own actions because actions can differ slightly
329 * between splintered and non-splintered subfacets due to the VLAN tag
330 * being initially different (present vs. absent). All of them have these
331 * properties in common so we just store one copy of them here. */
332 bool may_install; /* Reassess actions for every packet? */
333 bool has_learn; /* Actions include NXAST_LEARN? */
334 bool has_normal; /* Actions output to OFPP_NORMAL? */
335 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
336 tag_type tags; /* Tags that would require revalidation. */
337 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
340 static struct facet *facet_create(struct rule_dpif *, const struct flow *);
341 static void facet_remove(struct facet *);
342 static void facet_free(struct facet *);
344 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
345 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
346 const struct flow *);
347 static bool facet_revalidate(struct facet *);
348 static bool facet_check_consistency(struct facet *);
350 static void facet_flush_stats(struct facet *);
352 static void facet_update_time(struct facet *, long long int used);
353 static void facet_reset_counters(struct facet *);
354 static void facet_push_stats(struct facet *);
355 static void facet_learn(struct facet *);
356 static void facet_account(struct facet *);
358 static bool facet_is_controller_flow(struct facet *);
360 /* A dpif flow and actions associated with a facet.
362 * See also the large comment on struct facet. */
365 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
366 struct list list_node; /* In struct facet's 'facets' list. */
367 struct facet *facet; /* Owning facet. */
371 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
372 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
373 * regenerate the ODP flow key from ->facet->flow. */
374 enum odp_key_fitness key_fitness;
378 long long int used; /* Time last used; time created if not used. */
380 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
381 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
385 * These should be essentially identical for every subfacet in a facet, but
386 * may differ in trivial ways due to VLAN splinters. */
387 size_t actions_len; /* Number of bytes in actions[]. */
388 struct nlattr *actions; /* Datapath actions. */
390 bool installed; /* Installed in datapath? */
392 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
393 * splinters can cause it to differ. This value should be removed when
394 * the VLAN splinters feature is no longer needed. */
395 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
398 static struct subfacet *subfacet_create(struct facet *, enum odp_key_fitness,
399 const struct nlattr *key,
400 size_t key_len, ovs_be16 initial_tci);
401 static struct subfacet *subfacet_find(struct ofproto_dpif *,
402 const struct nlattr *key, size_t key_len);
403 static void subfacet_destroy(struct subfacet *);
404 static void subfacet_destroy__(struct subfacet *);
405 static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *,
407 static void subfacet_reset_dp_stats(struct subfacet *,
408 struct dpif_flow_stats *);
409 static void subfacet_update_time(struct subfacet *, long long int used);
410 static void subfacet_update_stats(struct subfacet *,
411 const struct dpif_flow_stats *);
412 static void subfacet_make_actions(struct subfacet *,
413 const struct ofpbuf *packet);
414 static int subfacet_install(struct subfacet *,
415 const struct nlattr *actions, size_t actions_len,
416 struct dpif_flow_stats *);
417 static void subfacet_uninstall(struct subfacet *);
423 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
424 struct list bundle_node; /* In struct ofbundle's "ports" list. */
425 struct cfm *cfm; /* Connectivity Fault Management, if any. */
426 tag_type tag; /* Tag associated with this port. */
427 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
428 bool may_enable; /* May be enabled in bonds. */
429 long long int carrier_seq; /* Carrier status changes. */
432 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
433 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
434 long long int stp_state_entered;
436 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
438 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
440 * This is deprecated. It is only for compatibility with broken device
441 * drivers in old versions of Linux that do not properly support VLANs when
442 * VLAN devices are not used. When broken device drivers are no longer in
443 * widespread use, we will delete these interfaces. */
444 uint16_t realdev_ofp_port;
448 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
449 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
450 * traffic egressing the 'ofport' with that priority should be marked with. */
451 struct priority_to_dscp {
452 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
453 uint32_t priority; /* Priority of this queue (see struct flow). */
455 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
458 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
460 * This is deprecated. It is only for compatibility with broken device drivers
461 * in old versions of Linux that do not properly support VLANs when VLAN
462 * devices are not used. When broken device drivers are no longer in
463 * widespread use, we will delete these interfaces. */
464 struct vlan_splinter {
465 struct hmap_node realdev_vid_node;
466 struct hmap_node vlandev_node;
467 uint16_t realdev_ofp_port;
468 uint16_t vlandev_ofp_port;
472 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
473 uint32_t realdev, ovs_be16 vlan_tci);
474 static uint16_t vsp_vlandev_to_realdev(const struct ofproto_dpif *,
475 uint16_t vlandev, int *vid);
476 static void vsp_remove(struct ofport_dpif *);
477 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
479 static struct ofport_dpif *
480 ofport_dpif_cast(const struct ofport *ofport)
482 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
483 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
486 static void port_run(struct ofport_dpif *);
487 static void port_wait(struct ofport_dpif *);
488 static int set_cfm(struct ofport *, const struct cfm_settings *);
489 static void ofport_clear_priorities(struct ofport_dpif *);
491 struct dpif_completion {
492 struct list list_node;
493 struct ofoperation *op;
496 /* Extra information about a classifier table.
497 * Currently used just for optimized flow revalidation. */
499 /* If either of these is nonnull, then this table has a form that allows
500 * flows to be tagged to avoid revalidating most flows for the most common
501 * kinds of flow table changes. */
502 struct cls_table *catchall_table; /* Table that wildcards all fields. */
503 struct cls_table *other_table; /* Table with any other wildcard set. */
504 uint32_t basis; /* Keeps each table's tags separate. */
507 struct ofproto_dpif {
508 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
517 struct netflow *netflow;
518 struct dpif_sflow *sflow;
519 struct hmap bundles; /* Contains "struct ofbundle"s. */
520 struct mac_learning *ml;
521 struct ofmirror *mirrors[MAX_MIRRORS];
522 bool has_bonded_bundles;
525 struct timer next_expiration;
529 struct hmap subfacets;
532 struct table_dpif tables[N_TABLES];
533 bool need_revalidate;
534 struct tag_set revalidate_set;
536 /* Support for debugging async flow mods. */
537 struct list completions;
539 bool has_bundle_action; /* True when the first bundle action appears. */
540 struct netdev_stats stats; /* To account packets generated and consumed in
545 long long int stp_last_tick;
547 /* VLAN splinters. */
548 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
549 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
552 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
553 * for debugging the asynchronous flow_mod implementation.) */
556 /* All existing ofproto_dpif instances, indexed by ->up.name. */
557 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
559 static void ofproto_dpif_unixctl_init(void);
561 static struct ofproto_dpif *
562 ofproto_dpif_cast(const struct ofproto *ofproto)
564 assert(ofproto->ofproto_class == &ofproto_dpif_class);
565 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
568 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
570 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
572 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
573 const struct ofpbuf *, ovs_be16 initial_tci,
576 /* Packet processing. */
577 static void update_learning_table(struct ofproto_dpif *,
578 const struct flow *, int vlan,
581 #define FLOW_MISS_MAX_BATCH 50
582 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
584 /* Flow expiration. */
585 static int expire(struct ofproto_dpif *);
588 static void send_netflow_active_timeouts(struct ofproto_dpif *);
591 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
593 compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions,
594 const struct flow *, uint32_t odp_port);
595 static void add_mirror_actions(struct action_xlate_ctx *ctx,
596 const struct flow *flow);
597 /* Global variables. */
598 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
600 /* Factory functions. */
603 enumerate_types(struct sset *types)
605 dp_enumerate_types(types);
609 enumerate_names(const char *type, struct sset *names)
611 return dp_enumerate_names(type, names);
615 del(const char *type, const char *name)
620 error = dpif_open(name, type, &dpif);
622 error = dpif_delete(dpif);
628 /* Basic life-cycle. */
630 static struct ofproto *
633 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
638 dealloc(struct ofproto *ofproto_)
640 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
645 construct(struct ofproto *ofproto_)
647 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
648 const char *name = ofproto->up.name;
652 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
654 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
658 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
659 ofproto->n_matches = 0;
661 dpif_flow_flush(ofproto->dpif);
662 dpif_recv_purge(ofproto->dpif);
664 error = dpif_recv_set(ofproto->dpif, true);
666 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
667 dpif_close(ofproto->dpif);
671 ofproto->netflow = NULL;
672 ofproto->sflow = NULL;
674 hmap_init(&ofproto->bundles);
675 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
676 for (i = 0; i < MAX_MIRRORS; i++) {
677 ofproto->mirrors[i] = NULL;
679 ofproto->has_bonded_bundles = false;
681 timer_set_duration(&ofproto->next_expiration, 1000);
683 hmap_init(&ofproto->facets);
684 hmap_init(&ofproto->subfacets);
686 for (i = 0; i < N_TABLES; i++) {
687 struct table_dpif *table = &ofproto->tables[i];
689 table->catchall_table = NULL;
690 table->other_table = NULL;
691 table->basis = random_uint32();
693 ofproto->need_revalidate = false;
694 tag_set_init(&ofproto->revalidate_set);
696 list_init(&ofproto->completions);
698 ofproto_dpif_unixctl_init();
700 ofproto->has_bundle_action = false;
702 hmap_init(&ofproto->vlandev_map);
703 hmap_init(&ofproto->realdev_vid_map);
705 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
706 hash_string(ofproto->up.name, 0));
707 memset(&ofproto->stats, 0, sizeof ofproto->stats);
709 ofproto_init_tables(ofproto_, N_TABLES);
715 complete_operations(struct ofproto_dpif *ofproto)
717 struct dpif_completion *c, *next;
719 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
720 ofoperation_complete(c->op, 0);
721 list_remove(&c->list_node);
727 destruct(struct ofproto *ofproto_)
729 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
730 struct rule_dpif *rule, *next_rule;
731 struct oftable *table;
734 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
735 complete_operations(ofproto);
737 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
738 struct cls_cursor cursor;
740 cls_cursor_init(&cursor, &table->cls, NULL);
741 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
742 ofproto_rule_destroy(&rule->up);
746 for (i = 0; i < MAX_MIRRORS; i++) {
747 mirror_destroy(ofproto->mirrors[i]);
750 netflow_destroy(ofproto->netflow);
751 dpif_sflow_destroy(ofproto->sflow);
752 hmap_destroy(&ofproto->bundles);
753 mac_learning_destroy(ofproto->ml);
755 hmap_destroy(&ofproto->facets);
756 hmap_destroy(&ofproto->subfacets);
758 hmap_destroy(&ofproto->vlandev_map);
759 hmap_destroy(&ofproto->realdev_vid_map);
761 dpif_close(ofproto->dpif);
765 run_fast(struct ofproto *ofproto_)
767 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
770 /* Handle one or more batches of upcalls, until there's nothing left to do
771 * or until we do a fixed total amount of work.
773 * We do work in batches because it can be much cheaper to set up a number
774 * of flows and fire off their patches all at once. We do multiple batches
775 * because in some cases handling a packet can cause another packet to be
776 * queued almost immediately as part of the return flow. Both
777 * optimizations can make major improvements on some benchmarks and
778 * presumably for real traffic as well. */
780 while (work < FLOW_MISS_MAX_BATCH) {
781 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
791 run(struct ofproto *ofproto_)
793 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
794 struct ofport_dpif *ofport;
795 struct ofbundle *bundle;
799 complete_operations(ofproto);
801 dpif_run(ofproto->dpif);
803 error = run_fast(ofproto_);
808 if (timer_expired(&ofproto->next_expiration)) {
809 int delay = expire(ofproto);
810 timer_set_duration(&ofproto->next_expiration, delay);
813 if (ofproto->netflow) {
814 if (netflow_run(ofproto->netflow)) {
815 send_netflow_active_timeouts(ofproto);
818 if (ofproto->sflow) {
819 dpif_sflow_run(ofproto->sflow);
822 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
825 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
830 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
832 /* Now revalidate if there's anything to do. */
833 if (ofproto->need_revalidate
834 || !tag_set_is_empty(&ofproto->revalidate_set)) {
835 struct tag_set revalidate_set = ofproto->revalidate_set;
836 bool revalidate_all = ofproto->need_revalidate;
837 struct facet *facet, *next;
839 /* Clear the revalidation flags. */
840 tag_set_init(&ofproto->revalidate_set);
841 ofproto->need_revalidate = false;
843 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
845 || tag_set_intersects(&revalidate_set, facet->tags)) {
846 facet_revalidate(facet);
851 /* Check the consistency of a random facet, to aid debugging. */
852 if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) {
855 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
856 struct facet, hmap_node);
857 if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) {
858 if (!facet_check_consistency(facet)) {
859 ofproto->need_revalidate = true;
868 wait(struct ofproto *ofproto_)
870 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
871 struct ofport_dpif *ofport;
872 struct ofbundle *bundle;
874 if (!clogged && !list_is_empty(&ofproto->completions)) {
875 poll_immediate_wake();
878 dpif_wait(ofproto->dpif);
879 dpif_recv_wait(ofproto->dpif);
880 if (ofproto->sflow) {
881 dpif_sflow_wait(ofproto->sflow);
883 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
884 poll_immediate_wake();
886 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
889 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
892 if (ofproto->netflow) {
893 netflow_wait(ofproto->netflow);
895 mac_learning_wait(ofproto->ml);
897 if (ofproto->need_revalidate) {
898 /* Shouldn't happen, but if it does just go around again. */
899 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
900 poll_immediate_wake();
902 timer_wait(&ofproto->next_expiration);
907 flush(struct ofproto *ofproto_)
909 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
910 struct facet *facet, *next_facet;
912 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
913 /* Mark the facet as not installed so that facet_remove() doesn't
914 * bother trying to uninstall it. There is no point in uninstalling it
915 * individually since we are about to blow away all the facets with
916 * dpif_flow_flush(). */
917 struct subfacet *subfacet;
919 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
920 subfacet->installed = false;
921 subfacet->dp_packet_count = 0;
922 subfacet->dp_byte_count = 0;
926 dpif_flow_flush(ofproto->dpif);
930 get_features(struct ofproto *ofproto_ OVS_UNUSED,
931 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
933 *arp_match_ip = true;
934 *actions = (OFPUTIL_A_OUTPUT |
935 OFPUTIL_A_SET_VLAN_VID |
936 OFPUTIL_A_SET_VLAN_PCP |
937 OFPUTIL_A_STRIP_VLAN |
938 OFPUTIL_A_SET_DL_SRC |
939 OFPUTIL_A_SET_DL_DST |
940 OFPUTIL_A_SET_NW_SRC |
941 OFPUTIL_A_SET_NW_DST |
942 OFPUTIL_A_SET_NW_TOS |
943 OFPUTIL_A_SET_TP_SRC |
944 OFPUTIL_A_SET_TP_DST |
949 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
951 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
952 struct dpif_dp_stats s;
954 strcpy(ots->name, "classifier");
956 dpif_get_dp_stats(ofproto->dpif, &s);
957 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
958 put_32aligned_be64(&ots->matched_count,
959 htonll(s.n_hit + ofproto->n_matches));
962 static struct ofport *
965 struct ofport_dpif *port = xmalloc(sizeof *port);
970 port_dealloc(struct ofport *port_)
972 struct ofport_dpif *port = ofport_dpif_cast(port_);
977 port_construct(struct ofport *port_)
979 struct ofport_dpif *port = ofport_dpif_cast(port_);
980 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
982 ofproto->need_revalidate = true;
983 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
986 port->tag = tag_create_random();
987 port->may_enable = true;
988 port->stp_port = NULL;
989 port->stp_state = STP_DISABLED;
990 hmap_init(&port->priorities);
991 port->realdev_ofp_port = 0;
992 port->vlandev_vid = 0;
993 port->carrier_seq = netdev_get_carrier_resets(port->up.netdev);
995 if (ofproto->sflow) {
996 dpif_sflow_add_port(ofproto->sflow, port_);
1003 port_destruct(struct ofport *port_)
1005 struct ofport_dpif *port = ofport_dpif_cast(port_);
1006 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1008 ofproto->need_revalidate = true;
1009 bundle_remove(port_);
1010 set_cfm(port_, NULL);
1011 if (ofproto->sflow) {
1012 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1015 ofport_clear_priorities(port);
1016 hmap_destroy(&port->priorities);
1020 port_modified(struct ofport *port_)
1022 struct ofport_dpif *port = ofport_dpif_cast(port_);
1024 if (port->bundle && port->bundle->bond) {
1025 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1030 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1032 struct ofport_dpif *port = ofport_dpif_cast(port_);
1033 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1034 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1036 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1037 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD)) {
1038 ofproto->need_revalidate = true;
1040 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1041 bundle_update(port->bundle);
1047 set_sflow(struct ofproto *ofproto_,
1048 const struct ofproto_sflow_options *sflow_options)
1050 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1051 struct dpif_sflow *ds = ofproto->sflow;
1053 if (sflow_options) {
1055 struct ofport_dpif *ofport;
1057 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1058 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1059 dpif_sflow_add_port(ds, &ofport->up);
1061 ofproto->need_revalidate = true;
1063 dpif_sflow_set_options(ds, sflow_options);
1066 dpif_sflow_destroy(ds);
1067 ofproto->need_revalidate = true;
1068 ofproto->sflow = NULL;
1075 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1077 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1084 struct ofproto_dpif *ofproto;
1086 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1087 ofproto->need_revalidate = true;
1088 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1091 if (cfm_configure(ofport->cfm, s)) {
1097 cfm_destroy(ofport->cfm);
1103 get_cfm_fault(const struct ofport *ofport_)
1105 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1107 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1111 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1114 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1117 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1125 get_cfm_health(const struct ofport *ofport_)
1127 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1129 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1132 /* Spanning Tree. */
1135 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1137 struct ofproto_dpif *ofproto = ofproto_;
1138 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1139 struct ofport_dpif *ofport;
1141 ofport = stp_port_get_aux(sp);
1143 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1144 ofproto->up.name, port_num);
1146 struct eth_header *eth = pkt->l2;
1148 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1149 if (eth_addr_is_zero(eth->eth_src)) {
1150 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1151 "with unknown MAC", ofproto->up.name, port_num);
1153 send_packet(ofport, pkt);
1159 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1161 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1163 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1165 /* Only revalidate flows if the configuration changed. */
1166 if (!s != !ofproto->stp) {
1167 ofproto->need_revalidate = true;
1171 if (!ofproto->stp) {
1172 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1173 send_bpdu_cb, ofproto);
1174 ofproto->stp_last_tick = time_msec();
1177 stp_set_bridge_id(ofproto->stp, s->system_id);
1178 stp_set_bridge_priority(ofproto->stp, s->priority);
1179 stp_set_hello_time(ofproto->stp, s->hello_time);
1180 stp_set_max_age(ofproto->stp, s->max_age);
1181 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1183 struct ofport *ofport;
1185 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1186 set_stp_port(ofport, NULL);
1189 stp_destroy(ofproto->stp);
1190 ofproto->stp = NULL;
1197 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1199 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1203 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1204 s->designated_root = stp_get_designated_root(ofproto->stp);
1205 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1214 update_stp_port_state(struct ofport_dpif *ofport)
1216 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1217 enum stp_state state;
1219 /* Figure out new state. */
1220 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1224 if (ofport->stp_state != state) {
1225 enum ofputil_port_state of_state;
1228 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1229 netdev_get_name(ofport->up.netdev),
1230 stp_state_name(ofport->stp_state),
1231 stp_state_name(state));
1232 if (stp_learn_in_state(ofport->stp_state)
1233 != stp_learn_in_state(state)) {
1234 /* xxx Learning action flows should also be flushed. */
1235 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1237 fwd_change = stp_forward_in_state(ofport->stp_state)
1238 != stp_forward_in_state(state);
1240 ofproto->need_revalidate = true;
1241 ofport->stp_state = state;
1242 ofport->stp_state_entered = time_msec();
1244 if (fwd_change && ofport->bundle) {
1245 bundle_update(ofport->bundle);
1248 /* Update the STP state bits in the OpenFlow port description. */
1249 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1250 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1251 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1252 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1253 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1255 ofproto_port_set_state(&ofport->up, of_state);
1259 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1260 * caller is responsible for assigning STP port numbers and ensuring
1261 * there are no duplicates. */
1263 set_stp_port(struct ofport *ofport_,
1264 const struct ofproto_port_stp_settings *s)
1266 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1267 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1268 struct stp_port *sp = ofport->stp_port;
1270 if (!s || !s->enable) {
1272 ofport->stp_port = NULL;
1273 stp_port_disable(sp);
1274 update_stp_port_state(ofport);
1277 } else if (sp && stp_port_no(sp) != s->port_num
1278 && ofport == stp_port_get_aux(sp)) {
1279 /* The port-id changed, so disable the old one if it's not
1280 * already in use by another port. */
1281 stp_port_disable(sp);
1284 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1285 stp_port_enable(sp);
1287 stp_port_set_aux(sp, ofport);
1288 stp_port_set_priority(sp, s->priority);
1289 stp_port_set_path_cost(sp, s->path_cost);
1291 update_stp_port_state(ofport);
1297 get_stp_port_status(struct ofport *ofport_,
1298 struct ofproto_port_stp_status *s)
1300 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1301 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1302 struct stp_port *sp = ofport->stp_port;
1304 if (!ofproto->stp || !sp) {
1310 s->port_id = stp_port_get_id(sp);
1311 s->state = stp_port_get_state(sp);
1312 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1313 s->role = stp_port_get_role(sp);
1314 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1320 stp_run(struct ofproto_dpif *ofproto)
1323 long long int now = time_msec();
1324 long long int elapsed = now - ofproto->stp_last_tick;
1325 struct stp_port *sp;
1328 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1329 ofproto->stp_last_tick = now;
1331 while (stp_get_changed_port(ofproto->stp, &sp)) {
1332 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1335 update_stp_port_state(ofport);
1339 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1340 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1346 stp_wait(struct ofproto_dpif *ofproto)
1349 poll_timer_wait(1000);
1353 /* Returns true if STP should process 'flow'. */
1355 stp_should_process_flow(const struct flow *flow)
1357 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1361 stp_process_packet(const struct ofport_dpif *ofport,
1362 const struct ofpbuf *packet)
1364 struct ofpbuf payload = *packet;
1365 struct eth_header *eth = payload.data;
1366 struct stp_port *sp = ofport->stp_port;
1368 /* Sink packets on ports that have STP disabled when the bridge has
1370 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1374 /* Trim off padding on payload. */
1375 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1376 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1379 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1380 stp_received_bpdu(sp, payload.data, payload.size);
1384 static struct priority_to_dscp *
1385 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1387 struct priority_to_dscp *pdscp;
1390 hash = hash_int(priority, 0);
1391 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1392 if (pdscp->priority == priority) {
1400 ofport_clear_priorities(struct ofport_dpif *ofport)
1402 struct priority_to_dscp *pdscp, *next;
1404 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1405 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1411 set_queues(struct ofport *ofport_,
1412 const struct ofproto_port_queue *qdscp_list,
1415 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1416 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1417 struct hmap new = HMAP_INITIALIZER(&new);
1420 for (i = 0; i < n_qdscp; i++) {
1421 struct priority_to_dscp *pdscp;
1425 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1426 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1431 pdscp = get_priority(ofport, priority);
1433 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1435 pdscp = xmalloc(sizeof *pdscp);
1436 pdscp->priority = priority;
1438 ofproto->need_revalidate = true;
1441 if (pdscp->dscp != dscp) {
1443 ofproto->need_revalidate = true;
1446 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1449 if (!hmap_is_empty(&ofport->priorities)) {
1450 ofport_clear_priorities(ofport);
1451 ofproto->need_revalidate = true;
1454 hmap_swap(&new, &ofport->priorities);
1462 /* Expires all MAC learning entries associated with 'bundle' and forces its
1463 * ofproto to revalidate every flow.
1465 * Normally MAC learning entries are removed only from the ofproto associated
1466 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1467 * are removed from every ofproto. When patch ports and SLB bonds are in use
1468 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1469 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1470 * with the host from which it migrated. */
1472 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1474 struct ofproto_dpif *ofproto = bundle->ofproto;
1475 struct mac_learning *ml = ofproto->ml;
1476 struct mac_entry *mac, *next_mac;
1478 ofproto->need_revalidate = true;
1479 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1480 if (mac->port.p == bundle) {
1482 struct ofproto_dpif *o;
1484 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1486 struct mac_entry *e;
1488 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1491 tag_set_add(&o->revalidate_set, e->tag);
1492 mac_learning_expire(o->ml, e);
1498 mac_learning_expire(ml, mac);
1503 static struct ofbundle *
1504 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1506 struct ofbundle *bundle;
1508 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1509 &ofproto->bundles) {
1510 if (bundle->aux == aux) {
1517 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1518 * ones that are found to 'bundles'. */
1520 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1521 void **auxes, size_t n_auxes,
1522 struct hmapx *bundles)
1526 hmapx_init(bundles);
1527 for (i = 0; i < n_auxes; i++) {
1528 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1530 hmapx_add(bundles, bundle);
1536 bundle_update(struct ofbundle *bundle)
1538 struct ofport_dpif *port;
1540 bundle->floodable = true;
1541 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1542 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1543 || !stp_forward_in_state(port->stp_state)) {
1544 bundle->floodable = false;
1551 bundle_del_port(struct ofport_dpif *port)
1553 struct ofbundle *bundle = port->bundle;
1555 bundle->ofproto->need_revalidate = true;
1557 list_remove(&port->bundle_node);
1558 port->bundle = NULL;
1561 lacp_slave_unregister(bundle->lacp, port);
1564 bond_slave_unregister(bundle->bond, port);
1567 bundle_update(bundle);
1571 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1572 struct lacp_slave_settings *lacp,
1573 uint32_t bond_stable_id)
1575 struct ofport_dpif *port;
1577 port = get_ofp_port(bundle->ofproto, ofp_port);
1582 if (port->bundle != bundle) {
1583 bundle->ofproto->need_revalidate = true;
1585 bundle_del_port(port);
1588 port->bundle = bundle;
1589 list_push_back(&bundle->ports, &port->bundle_node);
1590 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1591 || !stp_forward_in_state(port->stp_state)) {
1592 bundle->floodable = false;
1596 port->bundle->ofproto->need_revalidate = true;
1597 lacp_slave_register(bundle->lacp, port, lacp);
1600 port->bond_stable_id = bond_stable_id;
1606 bundle_destroy(struct ofbundle *bundle)
1608 struct ofproto_dpif *ofproto;
1609 struct ofport_dpif *port, *next_port;
1616 ofproto = bundle->ofproto;
1617 for (i = 0; i < MAX_MIRRORS; i++) {
1618 struct ofmirror *m = ofproto->mirrors[i];
1620 if (m->out == bundle) {
1622 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1623 || hmapx_find_and_delete(&m->dsts, bundle)) {
1624 ofproto->need_revalidate = true;
1629 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1630 bundle_del_port(port);
1633 bundle_flush_macs(bundle, true);
1634 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1636 free(bundle->trunks);
1637 lacp_destroy(bundle->lacp);
1638 bond_destroy(bundle->bond);
1643 bundle_set(struct ofproto *ofproto_, void *aux,
1644 const struct ofproto_bundle_settings *s)
1646 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1647 bool need_flush = false;
1648 struct ofport_dpif *port;
1649 struct ofbundle *bundle;
1650 unsigned long *trunks;
1656 bundle_destroy(bundle_lookup(ofproto, aux));
1660 assert(s->n_slaves == 1 || s->bond != NULL);
1661 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1663 bundle = bundle_lookup(ofproto, aux);
1665 bundle = xmalloc(sizeof *bundle);
1667 bundle->ofproto = ofproto;
1668 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1669 hash_pointer(aux, 0));
1671 bundle->name = NULL;
1673 list_init(&bundle->ports);
1674 bundle->vlan_mode = PORT_VLAN_TRUNK;
1676 bundle->trunks = NULL;
1677 bundle->use_priority_tags = s->use_priority_tags;
1678 bundle->lacp = NULL;
1679 bundle->bond = NULL;
1681 bundle->floodable = true;
1683 bundle->src_mirrors = 0;
1684 bundle->dst_mirrors = 0;
1685 bundle->mirror_out = 0;
1688 if (!bundle->name || strcmp(s->name, bundle->name)) {
1690 bundle->name = xstrdup(s->name);
1695 if (!bundle->lacp) {
1696 ofproto->need_revalidate = true;
1697 bundle->lacp = lacp_create();
1699 lacp_configure(bundle->lacp, s->lacp);
1701 lacp_destroy(bundle->lacp);
1702 bundle->lacp = NULL;
1705 /* Update set of ports. */
1707 for (i = 0; i < s->n_slaves; i++) {
1708 if (!bundle_add_port(bundle, s->slaves[i],
1709 s->lacp ? &s->lacp_slaves[i] : NULL,
1710 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1714 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1715 struct ofport_dpif *next_port;
1717 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1718 for (i = 0; i < s->n_slaves; i++) {
1719 if (s->slaves[i] == port->up.ofp_port) {
1724 bundle_del_port(port);
1728 assert(list_size(&bundle->ports) <= s->n_slaves);
1730 if (list_is_empty(&bundle->ports)) {
1731 bundle_destroy(bundle);
1735 /* Set VLAN tagging mode */
1736 if (s->vlan_mode != bundle->vlan_mode
1737 || s->use_priority_tags != bundle->use_priority_tags) {
1738 bundle->vlan_mode = s->vlan_mode;
1739 bundle->use_priority_tags = s->use_priority_tags;
1744 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1745 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1747 if (vlan != bundle->vlan) {
1748 bundle->vlan = vlan;
1752 /* Get trunked VLANs. */
1753 switch (s->vlan_mode) {
1754 case PORT_VLAN_ACCESS:
1758 case PORT_VLAN_TRUNK:
1759 trunks = (unsigned long *) s->trunks;
1762 case PORT_VLAN_NATIVE_UNTAGGED:
1763 case PORT_VLAN_NATIVE_TAGGED:
1764 if (vlan != 0 && (!s->trunks
1765 || !bitmap_is_set(s->trunks, vlan)
1766 || bitmap_is_set(s->trunks, 0))) {
1767 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1769 trunks = bitmap_clone(s->trunks, 4096);
1771 trunks = bitmap_allocate1(4096);
1773 bitmap_set1(trunks, vlan);
1774 bitmap_set0(trunks, 0);
1776 trunks = (unsigned long *) s->trunks;
1783 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1784 free(bundle->trunks);
1785 if (trunks == s->trunks) {
1786 bundle->trunks = vlan_bitmap_clone(trunks);
1788 bundle->trunks = trunks;
1793 if (trunks != s->trunks) {
1798 if (!list_is_short(&bundle->ports)) {
1799 bundle->ofproto->has_bonded_bundles = true;
1801 if (bond_reconfigure(bundle->bond, s->bond)) {
1802 ofproto->need_revalidate = true;
1805 bundle->bond = bond_create(s->bond);
1806 ofproto->need_revalidate = true;
1809 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1810 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1814 bond_destroy(bundle->bond);
1815 bundle->bond = NULL;
1818 /* If we changed something that would affect MAC learning, un-learn
1819 * everything on this port and force flow revalidation. */
1821 bundle_flush_macs(bundle, false);
1828 bundle_remove(struct ofport *port_)
1830 struct ofport_dpif *port = ofport_dpif_cast(port_);
1831 struct ofbundle *bundle = port->bundle;
1834 bundle_del_port(port);
1835 if (list_is_empty(&bundle->ports)) {
1836 bundle_destroy(bundle);
1837 } else if (list_is_short(&bundle->ports)) {
1838 bond_destroy(bundle->bond);
1839 bundle->bond = NULL;
1845 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
1847 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1848 struct ofport_dpif *port = port_;
1849 uint8_t ea[ETH_ADDR_LEN];
1852 error = netdev_get_etheraddr(port->up.netdev, ea);
1854 struct ofpbuf packet;
1857 ofpbuf_init(&packet, 0);
1858 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1860 memcpy(packet_pdu, pdu, pdu_size);
1862 send_packet(port, &packet);
1863 ofpbuf_uninit(&packet);
1865 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1866 "%s (%s)", port->bundle->name,
1867 netdev_get_name(port->up.netdev), strerror(error));
1872 bundle_send_learning_packets(struct ofbundle *bundle)
1874 struct ofproto_dpif *ofproto = bundle->ofproto;
1875 int error, n_packets, n_errors;
1876 struct mac_entry *e;
1878 error = n_packets = n_errors = 0;
1879 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1880 if (e->port.p != bundle) {
1881 struct ofpbuf *learning_packet;
1882 struct ofport_dpif *port;
1886 /* The assignment to "port" is unnecessary but makes "grep"ing for
1887 * struct ofport_dpif more effective. */
1888 learning_packet = bond_compose_learning_packet(bundle->bond,
1892 ret = send_packet(port, learning_packet);
1893 ofpbuf_delete(learning_packet);
1903 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1904 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1905 "packets, last error was: %s",
1906 bundle->name, n_errors, n_packets, strerror(error));
1908 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1909 bundle->name, n_packets);
1914 bundle_run(struct ofbundle *bundle)
1917 lacp_run(bundle->lacp, send_pdu_cb);
1920 struct ofport_dpif *port;
1922 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1923 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1926 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1927 lacp_status(bundle->lacp));
1928 if (bond_should_send_learning_packets(bundle->bond)) {
1929 bundle_send_learning_packets(bundle);
1935 bundle_wait(struct ofbundle *bundle)
1938 lacp_wait(bundle->lacp);
1941 bond_wait(bundle->bond);
1948 mirror_scan(struct ofproto_dpif *ofproto)
1952 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1953 if (!ofproto->mirrors[idx]) {
1960 static struct ofmirror *
1961 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1965 for (i = 0; i < MAX_MIRRORS; i++) {
1966 struct ofmirror *mirror = ofproto->mirrors[i];
1967 if (mirror && mirror->aux == aux) {
1975 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
1977 mirror_update_dups(struct ofproto_dpif *ofproto)
1981 for (i = 0; i < MAX_MIRRORS; i++) {
1982 struct ofmirror *m = ofproto->mirrors[i];
1985 m->dup_mirrors = MIRROR_MASK_C(1) << i;
1989 for (i = 0; i < MAX_MIRRORS; i++) {
1990 struct ofmirror *m1 = ofproto->mirrors[i];
1997 for (j = i + 1; j < MAX_MIRRORS; j++) {
1998 struct ofmirror *m2 = ofproto->mirrors[j];
2000 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2001 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2002 m2->dup_mirrors |= m1->dup_mirrors;
2009 mirror_set(struct ofproto *ofproto_, void *aux,
2010 const struct ofproto_mirror_settings *s)
2012 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2013 mirror_mask_t mirror_bit;
2014 struct ofbundle *bundle;
2015 struct ofmirror *mirror;
2016 struct ofbundle *out;
2017 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2018 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2021 mirror = mirror_lookup(ofproto, aux);
2023 mirror_destroy(mirror);
2029 idx = mirror_scan(ofproto);
2031 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2033 ofproto->up.name, MAX_MIRRORS, s->name);
2037 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2038 mirror->ofproto = ofproto;
2041 mirror->out_vlan = -1;
2042 mirror->name = NULL;
2045 if (!mirror->name || strcmp(s->name, mirror->name)) {
2047 mirror->name = xstrdup(s->name);
2050 /* Get the new configuration. */
2051 if (s->out_bundle) {
2052 out = bundle_lookup(ofproto, s->out_bundle);
2054 mirror_destroy(mirror);
2060 out_vlan = s->out_vlan;
2062 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2063 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2065 /* If the configuration has not changed, do nothing. */
2066 if (hmapx_equals(&srcs, &mirror->srcs)
2067 && hmapx_equals(&dsts, &mirror->dsts)
2068 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2069 && mirror->out == out
2070 && mirror->out_vlan == out_vlan)
2072 hmapx_destroy(&srcs);
2073 hmapx_destroy(&dsts);
2077 hmapx_swap(&srcs, &mirror->srcs);
2078 hmapx_destroy(&srcs);
2080 hmapx_swap(&dsts, &mirror->dsts);
2081 hmapx_destroy(&dsts);
2083 free(mirror->vlans);
2084 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2087 mirror->out_vlan = out_vlan;
2089 /* Update bundles. */
2090 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2091 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2092 if (hmapx_contains(&mirror->srcs, bundle)) {
2093 bundle->src_mirrors |= mirror_bit;
2095 bundle->src_mirrors &= ~mirror_bit;
2098 if (hmapx_contains(&mirror->dsts, bundle)) {
2099 bundle->dst_mirrors |= mirror_bit;
2101 bundle->dst_mirrors &= ~mirror_bit;
2104 if (mirror->out == bundle) {
2105 bundle->mirror_out |= mirror_bit;
2107 bundle->mirror_out &= ~mirror_bit;
2111 ofproto->need_revalidate = true;
2112 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2113 mirror_update_dups(ofproto);
2119 mirror_destroy(struct ofmirror *mirror)
2121 struct ofproto_dpif *ofproto;
2122 mirror_mask_t mirror_bit;
2123 struct ofbundle *bundle;
2129 ofproto = mirror->ofproto;
2130 ofproto->need_revalidate = true;
2131 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2133 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2134 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2135 bundle->src_mirrors &= ~mirror_bit;
2136 bundle->dst_mirrors &= ~mirror_bit;
2137 bundle->mirror_out &= ~mirror_bit;
2140 hmapx_destroy(&mirror->srcs);
2141 hmapx_destroy(&mirror->dsts);
2142 free(mirror->vlans);
2144 ofproto->mirrors[mirror->idx] = NULL;
2148 mirror_update_dups(ofproto);
2152 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2153 uint64_t *packets, uint64_t *bytes)
2155 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2156 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2159 *packets = *bytes = UINT64_MAX;
2163 *packets = mirror->packet_count;
2164 *bytes = mirror->byte_count;
2170 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2172 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2173 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2174 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2180 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2182 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2183 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2184 return bundle && bundle->mirror_out != 0;
2188 forward_bpdu_changed(struct ofproto *ofproto_)
2190 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2191 /* Revalidate cached flows whenever forward_bpdu option changes. */
2192 ofproto->need_revalidate = true;
2196 set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time)
2198 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2199 mac_learning_set_idle_time(ofproto->ml, idle_time);
2204 static struct ofport_dpif *
2205 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
2207 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2208 return ofport ? ofport_dpif_cast(ofport) : NULL;
2211 static struct ofport_dpif *
2212 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
2214 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2218 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2219 struct dpif_port *dpif_port)
2221 ofproto_port->name = dpif_port->name;
2222 ofproto_port->type = dpif_port->type;
2223 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2227 port_run(struct ofport_dpif *ofport)
2229 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2230 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2231 bool enable = netdev_get_carrier(ofport->up.netdev);
2233 ofport->carrier_seq = carrier_seq;
2236 cfm_run(ofport->cfm);
2238 if (cfm_should_send_ccm(ofport->cfm)) {
2239 struct ofpbuf packet;
2241 ofpbuf_init(&packet, 0);
2242 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2243 send_packet(ofport, &packet);
2244 ofpbuf_uninit(&packet);
2247 enable = enable && !cfm_get_fault(ofport->cfm)
2248 && cfm_get_opup(ofport->cfm);
2251 if (ofport->bundle) {
2252 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2253 if (carrier_changed) {
2254 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2258 if (ofport->may_enable != enable) {
2259 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2261 if (ofproto->has_bundle_action) {
2262 ofproto->need_revalidate = true;
2266 ofport->may_enable = enable;
2270 port_wait(struct ofport_dpif *ofport)
2273 cfm_wait(ofport->cfm);
2278 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2279 struct ofproto_port *ofproto_port)
2281 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2282 struct dpif_port dpif_port;
2285 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2287 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2293 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2295 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2299 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2301 *ofp_portp = odp_port_to_ofp_port(odp_port);
2307 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2309 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2312 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2314 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2316 /* The caller is going to close ofport->up.netdev. If this is a
2317 * bonded port, then the bond is using that netdev, so remove it
2318 * from the bond. The client will need to reconfigure everything
2319 * after deleting ports, so then the slave will get re-added. */
2320 bundle_remove(&ofport->up);
2327 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2329 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2332 error = netdev_get_stats(ofport->up.netdev, stats);
2334 if (!error && ofport->odp_port == OVSP_LOCAL) {
2335 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2337 /* ofproto->stats.tx_packets represents packets that we created
2338 * internally and sent to some port (e.g. packets sent with
2339 * send_packet()). Account for them as if they had come from
2340 * OFPP_LOCAL and got forwarded. */
2342 if (stats->rx_packets != UINT64_MAX) {
2343 stats->rx_packets += ofproto->stats.tx_packets;
2346 if (stats->rx_bytes != UINT64_MAX) {
2347 stats->rx_bytes += ofproto->stats.tx_bytes;
2350 /* ofproto->stats.rx_packets represents packets that were received on
2351 * some port and we processed internally and dropped (e.g. STP).
2352 * Account fro them as if they had been forwarded to OFPP_LOCAL. */
2354 if (stats->tx_packets != UINT64_MAX) {
2355 stats->tx_packets += ofproto->stats.rx_packets;
2358 if (stats->tx_bytes != UINT64_MAX) {
2359 stats->tx_bytes += ofproto->stats.rx_bytes;
2366 /* Account packets for LOCAL port. */
2368 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2369 size_t tx_size, size_t rx_size)
2371 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2374 ofproto->stats.rx_packets++;
2375 ofproto->stats.rx_bytes += rx_size;
2378 ofproto->stats.tx_packets++;
2379 ofproto->stats.tx_bytes += tx_size;
2383 struct port_dump_state {
2384 struct dpif_port_dump dump;
2389 port_dump_start(const struct ofproto *ofproto_, void **statep)
2391 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2392 struct port_dump_state *state;
2394 *statep = state = xmalloc(sizeof *state);
2395 dpif_port_dump_start(&state->dump, ofproto->dpif);
2396 state->done = false;
2401 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2402 struct ofproto_port *port)
2404 struct port_dump_state *state = state_;
2405 struct dpif_port dpif_port;
2407 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2408 ofproto_port_from_dpif_port(port, &dpif_port);
2411 int error = dpif_port_dump_done(&state->dump);
2413 return error ? error : EOF;
2418 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2420 struct port_dump_state *state = state_;
2423 dpif_port_dump_done(&state->dump);
2430 port_poll(const struct ofproto *ofproto_, char **devnamep)
2432 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2433 return dpif_port_poll(ofproto->dpif, devnamep);
2437 port_poll_wait(const struct ofproto *ofproto_)
2439 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2440 dpif_port_poll_wait(ofproto->dpif);
2444 port_is_lacp_current(const struct ofport *ofport_)
2446 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2447 return (ofport->bundle && ofport->bundle->lacp
2448 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2452 /* Upcall handling. */
2454 /* Flow miss batching.
2456 * Some dpifs implement operations faster when you hand them off in a batch.
2457 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2458 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2459 * more packets, plus possibly installing the flow in the dpif.
2461 * So far we only batch the operations that affect flow setup time the most.
2462 * It's possible to batch more than that, but the benefit might be minimal. */
2464 struct hmap_node hmap_node;
2466 enum odp_key_fitness key_fitness;
2467 const struct nlattr *key;
2469 ovs_be16 initial_tci;
2470 struct list packets;
2473 struct flow_miss_op {
2474 struct dpif_op dpif_op;
2475 struct subfacet *subfacet;
2478 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2479 * OpenFlow controller as necessary according to their individual
2480 * configurations. */
2482 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
2483 const struct flow *flow)
2485 struct ofputil_packet_in pin;
2487 pin.packet = packet->data;
2488 pin.packet_len = packet->size;
2489 pin.reason = OFPR_NO_MATCH;
2490 pin.controller_id = 0;
2495 pin.send_len = 0; /* not used for flow table misses */
2497 flow_get_metadata(flow, &pin.fmd);
2499 /* Registers aren't meaningful on a miss. */
2500 memset(pin.fmd.reg_masks, 0, sizeof pin.fmd.reg_masks);
2502 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
2506 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2507 const struct ofpbuf *packet)
2509 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2515 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2517 cfm_process_heartbeat(ofport->cfm, packet);
2520 } else if (ofport->bundle && ofport->bundle->lacp
2521 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2523 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2526 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2528 stp_process_packet(ofport, packet);
2535 static struct flow_miss *
2536 flow_miss_create(struct hmap *todo, const struct flow *flow,
2537 enum odp_key_fitness key_fitness,
2538 const struct nlattr *key, size_t key_len,
2539 ovs_be16 initial_tci)
2541 uint32_t hash = flow_hash(flow, 0);
2542 struct flow_miss *miss;
2544 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2545 if (flow_equal(&miss->flow, flow)) {
2550 miss = xmalloc(sizeof *miss);
2551 hmap_insert(todo, &miss->hmap_node, hash);
2553 miss->key_fitness = key_fitness;
2555 miss->key_len = key_len;
2556 miss->initial_tci = initial_tci;
2557 list_init(&miss->packets);
2562 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2563 struct flow_miss_op *ops, size_t *n_ops)
2565 const struct flow *flow = &miss->flow;
2566 struct subfacet *subfacet;
2567 struct ofpbuf *packet;
2568 struct facet *facet;
2570 facet = facet_lookup_valid(ofproto, flow);
2572 struct rule_dpif *rule;
2574 rule = rule_dpif_lookup(ofproto, flow, 0);
2576 /* Don't send a packet-in if OFPUTIL_PC_NO_PACKET_IN asserted. */
2577 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
2579 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
2580 COVERAGE_INC(ofproto_dpif_no_packet_in);
2581 /* XXX install 'drop' flow entry */
2585 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
2589 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2590 send_packet_in_miss(ofproto, packet, flow);
2596 facet = facet_create(rule, flow);
2599 subfacet = subfacet_create(facet,
2600 miss->key_fitness, miss->key, miss->key_len,
2603 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2604 struct dpif_flow_stats stats;
2605 struct flow_miss_op *op;
2606 struct dpif_execute *execute;
2608 ofproto->n_matches++;
2610 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2612 * Extra-special case for fail-open mode.
2614 * We are in fail-open mode and the packet matched the fail-open
2615 * rule, but we are connected to a controller too. We should send
2616 * the packet up to the controller in the hope that it will try to
2617 * set up a flow and thereby allow us to exit fail-open.
2619 * See the top-level comment in fail-open.c for more information.
2621 send_packet_in_miss(ofproto, packet, flow);
2624 if (!facet->may_install || !subfacet->actions) {
2625 subfacet_make_actions(subfacet, packet);
2628 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2629 subfacet_update_stats(subfacet, &stats);
2631 if (!subfacet->actions_len) {
2632 /* No actions to execute, so skip talking to the dpif. */
2636 if (flow->vlan_tci != subfacet->initial_tci) {
2637 /* This packet was received on a VLAN splinter port. We added
2638 * a VLAN to the packet to make the packet resemble the flow,
2639 * but the actions were composed assuming that the packet
2640 * contained no VLAN. So, we must remove the VLAN header from
2641 * the packet before trying to execute the actions. */
2642 eth_pop_vlan(packet);
2645 op = &ops[(*n_ops)++];
2646 execute = &op->dpif_op.u.execute;
2647 op->subfacet = subfacet;
2648 op->dpif_op.type = DPIF_OP_EXECUTE;
2649 execute->key = miss->key;
2650 execute->key_len = miss->key_len;
2651 execute->actions = (facet->may_install
2653 : xmemdup(subfacet->actions,
2654 subfacet->actions_len));
2655 execute->actions_len = subfacet->actions_len;
2656 execute->packet = packet;
2659 if (facet->may_install && subfacet->key_fitness != ODP_FIT_TOO_LITTLE) {
2660 struct flow_miss_op *op = &ops[(*n_ops)++];
2661 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
2663 op->subfacet = subfacet;
2664 op->dpif_op.type = DPIF_OP_FLOW_PUT;
2665 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2666 put->key = miss->key;
2667 put->key_len = miss->key_len;
2668 put->actions = subfacet->actions;
2669 put->actions_len = subfacet->actions_len;
2674 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
2675 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
2676 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
2677 * what a flow key should contain.
2679 * This function also includes some logic to help make VLAN splinters
2680 * transparent to the rest of the upcall processing logic. In particular, if
2681 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
2682 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
2683 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
2685 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
2686 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
2687 * (This differs from the value returned in flow->vlan_tci only for packets
2688 * received on VLAN splinters.)
2690 static enum odp_key_fitness
2691 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
2692 const struct nlattr *key, size_t key_len,
2693 struct flow *flow, ovs_be16 *initial_tci,
2694 struct ofpbuf *packet)
2696 enum odp_key_fitness fitness;
2700 fitness = odp_flow_key_to_flow(key, key_len, flow);
2701 if (fitness == ODP_FIT_ERROR) {
2704 *initial_tci = flow->vlan_tci;
2706 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
2708 /* Cause the flow to be processed as if it came in on the real device
2709 * with the VLAN device's VLAN ID. */
2710 flow->in_port = realdev;
2711 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
2713 /* Make the packet resemble the flow, so that it gets sent to an
2714 * OpenFlow controller properly, so that it looks correct for
2715 * sFlow, and so that flow_extract() will get the correct vlan_tci
2716 * if it is called on 'packet'.
2718 * The allocated space inside 'packet' probably also contains
2719 * 'key', that is, both 'packet' and 'key' are probably part of a
2720 * struct dpif_upcall (see the large comment on that structure
2721 * definition), so pushing data on 'packet' is in general not a
2722 * good idea since it could overwrite 'key' or free it as a side
2723 * effect. However, it's OK in this special case because we know
2724 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
2725 * will just overwrite the 4-byte "struct nlattr", which is fine
2726 * since we don't need that header anymore. */
2727 eth_push_vlan(packet, flow->vlan_tci);
2730 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
2731 if (fitness == ODP_FIT_PERFECT) {
2732 fitness = ODP_FIT_TOO_MUCH;
2740 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
2743 struct dpif_upcall *upcall;
2744 struct flow_miss *miss, *next_miss;
2745 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
2746 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
2755 /* Construct the to-do list.
2757 * This just amounts to extracting the flow from each packet and sticking
2758 * the packets that have the same flow in the same "flow_miss" structure so
2759 * that we can process them together. */
2761 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
2762 enum odp_key_fitness fitness;
2763 struct flow_miss *miss;
2764 ovs_be16 initial_tci;
2767 /* Obtain metadata and check userspace/kernel agreement on flow match,
2768 * then set 'flow''s header pointers. */
2769 fitness = ofproto_dpif_extract_flow_key(ofproto,
2770 upcall->key, upcall->key_len,
2771 &flow, &initial_tci,
2773 if (fitness == ODP_FIT_ERROR) {
2774 ofpbuf_delete(upcall->packet);
2777 flow_extract(upcall->packet, flow.skb_priority, flow.tun_id,
2778 flow.in_port, &flow);
2780 /* Handle 802.1ag, LACP, and STP specially. */
2781 if (process_special(ofproto, &flow, upcall->packet)) {
2782 ofproto_update_local_port_stats(&ofproto->up,
2783 0, upcall->packet->size);
2784 ofpbuf_delete(upcall->packet);
2785 ofproto->n_matches++;
2789 /* Add other packets to a to-do list. */
2790 miss = flow_miss_create(&todo, &flow, fitness,
2791 upcall->key, upcall->key_len, initial_tci);
2792 list_push_back(&miss->packets, &upcall->packet->list_node);
2795 /* Process each element in the to-do list, constructing the set of
2796 * operations to batch. */
2798 HMAP_FOR_EACH (miss, hmap_node, &todo) {
2799 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
2801 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
2803 /* Execute batch. */
2804 for (i = 0; i < n_ops; i++) {
2805 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
2807 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
2809 /* Free memory and update facets. */
2810 for (i = 0; i < n_ops; i++) {
2811 struct flow_miss_op *op = &flow_miss_ops[i];
2812 struct dpif_execute *execute;
2814 switch (op->dpif_op.type) {
2815 case DPIF_OP_EXECUTE:
2816 execute = &op->dpif_op.u.execute;
2817 if (op->subfacet->actions != execute->actions) {
2818 free((struct nlattr *) execute->actions);
2822 case DPIF_OP_FLOW_PUT:
2823 if (!op->dpif_op.error) {
2824 op->subfacet->installed = true;
2828 case DPIF_OP_FLOW_DEL:
2832 HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) {
2833 ofpbuf_list_delete(&miss->packets);
2834 hmap_remove(&todo, &miss->hmap_node);
2837 hmap_destroy(&todo);
2841 handle_userspace_upcall(struct ofproto_dpif *ofproto,
2842 struct dpif_upcall *upcall)
2844 struct user_action_cookie cookie;
2845 enum odp_key_fitness fitness;
2846 ovs_be16 initial_tci;
2849 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
2851 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
2852 upcall->key_len, &flow,
2853 &initial_tci, upcall->packet);
2854 if (fitness == ODP_FIT_ERROR) {
2855 ofpbuf_delete(upcall->packet);
2859 if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
2860 if (ofproto->sflow) {
2861 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
2865 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
2867 ofpbuf_delete(upcall->packet);
2871 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
2873 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
2877 assert (max_batch <= FLOW_MISS_MAX_BATCH);
2880 for (i = 0; i < max_batch; i++) {
2881 struct dpif_upcall *upcall = &misses[n_misses];
2884 error = dpif_recv(ofproto->dpif, upcall);
2889 switch (upcall->type) {
2890 case DPIF_UC_ACTION:
2891 handle_userspace_upcall(ofproto, upcall);
2895 /* Handle it later. */
2899 case DPIF_N_UC_TYPES:
2901 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32,
2907 handle_miss_upcalls(ofproto, misses, n_misses);
2912 /* Flow expiration. */
2914 static int subfacet_max_idle(const struct ofproto_dpif *);
2915 static void update_stats(struct ofproto_dpif *);
2916 static void rule_expire(struct rule_dpif *);
2917 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
2919 /* This function is called periodically by run(). Its job is to collect
2920 * updates for the flows that have been installed into the datapath, most
2921 * importantly when they last were used, and then use that information to
2922 * expire flows that have not been used recently.
2924 * Returns the number of milliseconds after which it should be called again. */
2926 expire(struct ofproto_dpif *ofproto)
2928 struct rule_dpif *rule, *next_rule;
2929 struct oftable *table;
2932 /* Update stats for each flow in the datapath. */
2933 update_stats(ofproto);
2935 /* Expire subfacets that have been idle too long. */
2936 dp_max_idle = subfacet_max_idle(ofproto);
2937 expire_subfacets(ofproto, dp_max_idle);
2939 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
2940 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
2941 struct cls_cursor cursor;
2943 cls_cursor_init(&cursor, &table->cls, NULL);
2944 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
2949 /* All outstanding data in existing flows has been accounted, so it's a
2950 * good time to do bond rebalancing. */
2951 if (ofproto->has_bonded_bundles) {
2952 struct ofbundle *bundle;
2954 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2956 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
2961 return MIN(dp_max_idle, 1000);
2964 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
2966 * This function also pushes statistics updates to rules which each facet
2967 * resubmits into. Generally these statistics will be accurate. However, if a
2968 * facet changes the rule it resubmits into at some time in between
2969 * update_stats() runs, it is possible that statistics accrued to the
2970 * old rule will be incorrectly attributed to the new rule. This could be
2971 * avoided by calling update_stats() whenever rules are created or
2972 * deleted. However, the performance impact of making so many calls to the
2973 * datapath do not justify the benefit of having perfectly accurate statistics.
2976 update_stats(struct ofproto_dpif *p)
2978 const struct dpif_flow_stats *stats;
2979 struct dpif_flow_dump dump;
2980 const struct nlattr *key;
2983 dpif_flow_dump_start(&dump, p->dpif);
2984 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
2985 struct subfacet *subfacet;
2987 subfacet = subfacet_find(p, key, key_len);
2988 if (subfacet && subfacet->installed) {
2989 struct facet *facet = subfacet->facet;
2991 if (stats->n_packets >= subfacet->dp_packet_count) {
2992 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
2993 facet->packet_count += extra;
2995 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
2998 if (stats->n_bytes >= subfacet->dp_byte_count) {
2999 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
3001 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3004 subfacet->dp_packet_count = stats->n_packets;
3005 subfacet->dp_byte_count = stats->n_bytes;
3007 facet->tcp_flags |= stats->tcp_flags;
3009 subfacet_update_time(subfacet, stats->used);
3010 if (facet->accounted_bytes < facet->byte_count) {
3012 facet_account(facet);
3013 facet->accounted_bytes = facet->byte_count;
3015 facet_push_stats(facet);
3017 if (!VLOG_DROP_WARN(&rl)) {
3021 odp_flow_key_format(key, key_len, &s);
3022 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
3026 COVERAGE_INC(facet_unexpected);
3027 /* There's a flow in the datapath that we know nothing about, or a
3028 * flow that shouldn't be installed but was anyway. Delete it. */
3029 dpif_flow_del(p->dpif, key, key_len, NULL);
3032 dpif_flow_dump_done(&dump);
3035 /* Calculates and returns the number of milliseconds of idle time after which
3036 * subfacets should expire from the datapath. When a subfacet expires, we fold
3037 * its statistics into its facet, and when a facet's last subfacet expires, we
3038 * fold its statistic into its rule. */
3040 subfacet_max_idle(const struct ofproto_dpif *ofproto)
3043 * Idle time histogram.
3045 * Most of the time a switch has a relatively small number of subfacets.
3046 * When this is the case we might as well keep statistics for all of them
3047 * in userspace and to cache them in the kernel datapath for performance as
3050 * As the number of subfacets increases, the memory required to maintain
3051 * statistics about them in userspace and in the kernel becomes
3052 * significant. However, with a large number of subfacets it is likely
3053 * that only a few of them are "heavy hitters" that consume a large amount
3054 * of bandwidth. At this point, only heavy hitters are worth caching in
3055 * the kernel and maintaining in userspaces; other subfacets we can
3058 * The technique used to compute the idle time is to build a histogram with
3059 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3060 * that is installed in the kernel gets dropped in the appropriate bucket.
3061 * After the histogram has been built, we compute the cutoff so that only
3062 * the most-recently-used 1% of subfacets (but at least
3063 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3064 * the most-recently-used bucket of subfacets is kept, so actually an
3065 * arbitrary number of subfacets can be kept in any given expiration run
3066 * (though the next run will delete most of those unless they receive
3069 * This requires a second pass through the subfacets, in addition to the
3070 * pass made by update_stats(), because the former function never looks at
3071 * uninstallable subfacets.
3073 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3074 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3075 int buckets[N_BUCKETS] = { 0 };
3076 int total, subtotal, bucket;
3077 struct subfacet *subfacet;
3081 total = hmap_count(&ofproto->subfacets);
3082 if (total <= ofproto->up.flow_eviction_threshold) {
3083 return N_BUCKETS * BUCKET_WIDTH;
3086 /* Build histogram. */
3088 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3089 long long int idle = now - subfacet->used;
3090 int bucket = (idle <= 0 ? 0
3091 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3092 : (unsigned int) idle / BUCKET_WIDTH);
3096 /* Find the first bucket whose flows should be expired. */
3097 subtotal = bucket = 0;
3099 subtotal += buckets[bucket++];
3100 } while (bucket < N_BUCKETS &&
3101 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3103 if (VLOG_IS_DBG_ENABLED()) {
3107 ds_put_cstr(&s, "keep");
3108 for (i = 0; i < N_BUCKETS; i++) {
3110 ds_put_cstr(&s, ", drop");
3113 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3116 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3120 return bucket * BUCKET_WIDTH;
3123 enum { EXPIRE_MAX_BATCH = 50 };
3126 expire_batch(struct ofproto_dpif *ofproto, struct subfacet **subfacets, int n)
3128 struct odputil_keybuf keybufs[EXPIRE_MAX_BATCH];
3129 struct dpif_op ops[EXPIRE_MAX_BATCH];
3130 struct dpif_op *opsp[EXPIRE_MAX_BATCH];
3131 struct ofpbuf keys[EXPIRE_MAX_BATCH];
3132 struct dpif_flow_stats stats[EXPIRE_MAX_BATCH];
3135 for (i = 0; i < n; i++) {
3136 ops[i].type = DPIF_OP_FLOW_DEL;
3137 subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]);
3138 ops[i].u.flow_del.key = keys[i].data;
3139 ops[i].u.flow_del.key_len = keys[i].size;
3140 ops[i].u.flow_del.stats = &stats[i];
3144 dpif_operate(ofproto->dpif, opsp, n);
3145 for (i = 0; i < n; i++) {
3146 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
3147 subfacets[i]->installed = false;
3148 subfacet_destroy(subfacets[i]);
3153 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3155 long long int cutoff = time_msec() - dp_max_idle;
3157 struct subfacet *subfacet, *next_subfacet;
3158 struct subfacet *batch[EXPIRE_MAX_BATCH];
3162 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3163 &ofproto->subfacets) {
3164 if (subfacet->used < cutoff) {
3165 if (subfacet->installed) {
3166 batch[n_batch++] = subfacet;
3167 if (n_batch >= EXPIRE_MAX_BATCH) {
3168 expire_batch(ofproto, batch, n_batch);
3172 subfacet_destroy(subfacet);
3178 expire_batch(ofproto, batch, n_batch);
3182 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3183 * then delete it entirely. */
3185 rule_expire(struct rule_dpif *rule)
3187 struct facet *facet, *next_facet;
3191 /* Has 'rule' expired? */
3193 if (rule->up.hard_timeout
3194 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3195 reason = OFPRR_HARD_TIMEOUT;
3196 } else if (rule->up.idle_timeout
3197 && now > rule->up.used + rule->up.idle_timeout * 1000) {
3198 reason = OFPRR_IDLE_TIMEOUT;
3203 COVERAGE_INC(ofproto_dpif_expired);
3205 /* Update stats. (This is a no-op if the rule expired due to an idle
3206 * timeout, because that only happens when the rule has no facets left.) */
3207 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3208 facet_remove(facet);
3211 /* Get rid of the rule. */
3212 ofproto_rule_expire(&rule->up, reason);
3217 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3219 * The caller must already have determined that no facet with an identical
3220 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3221 * the ofproto's classifier table.
3223 * The facet will initially have no subfacets. The caller should create (at
3224 * least) one subfacet with subfacet_create(). */
3225 static struct facet *
3226 facet_create(struct rule_dpif *rule, const struct flow *flow)
3228 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3229 struct facet *facet;
3231 facet = xzalloc(sizeof *facet);
3232 facet->used = time_msec();
3233 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
3234 list_push_back(&rule->facets, &facet->list_node);
3236 facet->flow = *flow;
3237 list_init(&facet->subfacets);
3238 netflow_flow_init(&facet->nf_flow);
3239 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3245 facet_free(struct facet *facet)
3250 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3251 * 'packet', which arrived on 'in_port'.
3253 * Takes ownership of 'packet'. */
3255 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3256 const struct nlattr *odp_actions, size_t actions_len,
3257 struct ofpbuf *packet)
3259 struct odputil_keybuf keybuf;
3263 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3264 odp_flow_key_from_flow(&key, flow);
3266 error = dpif_execute(ofproto->dpif, key.data, key.size,
3267 odp_actions, actions_len, packet);
3269 ofpbuf_delete(packet);
3273 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3275 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3276 * rule's statistics, via subfacet_uninstall().
3278 * - Removes 'facet' from its rule and from ofproto->facets.
3281 facet_remove(struct facet *facet)
3283 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3284 struct subfacet *subfacet, *next_subfacet;
3286 assert(!list_is_empty(&facet->subfacets));
3288 /* First uninstall all of the subfacets to get final statistics. */
3289 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3290 subfacet_uninstall(subfacet);
3293 /* Flush the final stats to the rule.
3295 * This might require us to have at least one subfacet around so that we
3296 * can use its actions for accounting in facet_account(), which is why we
3297 * have uninstalled but not yet destroyed the subfacets. */
3298 facet_flush_stats(facet);
3300 /* Now we're really all done so destroy everything. */
3301 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3302 &facet->subfacets) {
3303 subfacet_destroy__(subfacet);
3305 hmap_remove(&ofproto->facets, &facet->hmap_node);
3306 list_remove(&facet->list_node);
3310 /* Feed information from 'facet' back into the learning table to keep it in
3311 * sync with what is actually flowing through the datapath. */
3313 facet_learn(struct facet *facet)
3315 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3316 struct action_xlate_ctx ctx;
3318 if (!facet->has_learn
3319 && !facet->has_normal
3320 && (!facet->has_fin_timeout
3321 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
3325 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3326 facet->flow.vlan_tci,
3327 facet->rule, facet->tcp_flags, NULL);
3328 ctx.may_learn = true;
3329 xlate_actions_for_side_effects(&ctx, facet->rule->up.actions,
3330 facet->rule->up.n_actions);
3334 facet_account(struct facet *facet)
3336 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3337 struct subfacet *subfacet;
3338 const struct nlattr *a;
3343 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3346 n_bytes = facet->byte_count - facet->accounted_bytes;
3348 /* This loop feeds byte counters to bond_account() for rebalancing to use
3349 * as a basis. We also need to track the actual VLAN on which the packet
3350 * is going to be sent to ensure that it matches the one passed to
3351 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3354 * We use the actions from an arbitrary subfacet because they should all
3355 * be equally valid for our purpose. */
3356 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3357 struct subfacet, list_node);
3358 vlan_tci = facet->flow.vlan_tci;
3359 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3360 subfacet->actions, subfacet->actions_len) {
3361 const struct ovs_action_push_vlan *vlan;
3362 struct ofport_dpif *port;
3364 switch (nl_attr_type(a)) {
3365 case OVS_ACTION_ATTR_OUTPUT:
3366 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3367 if (port && port->bundle && port->bundle->bond) {
3368 bond_account(port->bundle->bond, &facet->flow,
3369 vlan_tci_to_vid(vlan_tci), n_bytes);
3373 case OVS_ACTION_ATTR_POP_VLAN:
3374 vlan_tci = htons(0);
3377 case OVS_ACTION_ATTR_PUSH_VLAN:
3378 vlan = nl_attr_get(a);
3379 vlan_tci = vlan->vlan_tci;
3385 /* Returns true if the only action for 'facet' is to send to the controller.
3386 * (We don't report NetFlow expiration messages for such facets because they
3387 * are just part of the control logic for the network, not real traffic). */
3389 facet_is_controller_flow(struct facet *facet)
3392 && facet->rule->up.n_actions == 1
3393 && action_outputs_to_port(&facet->rule->up.actions[0],
3394 htons(OFPP_CONTROLLER)));
3397 /* Folds all of 'facet''s statistics into its rule. Also updates the
3398 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3399 * 'facet''s statistics in the datapath should have been zeroed and folded into
3400 * its packet and byte counts before this function is called. */
3402 facet_flush_stats(struct facet *facet)
3404 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3405 struct subfacet *subfacet;
3407 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3408 assert(!subfacet->dp_byte_count);
3409 assert(!subfacet->dp_packet_count);
3412 facet_push_stats(facet);
3413 if (facet->accounted_bytes < facet->byte_count) {
3414 facet_account(facet);
3415 facet->accounted_bytes = facet->byte_count;
3418 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3419 struct ofexpired expired;
3420 expired.flow = facet->flow;
3421 expired.packet_count = facet->packet_count;
3422 expired.byte_count = facet->byte_count;
3423 expired.used = facet->used;
3424 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3427 facet->rule->packet_count += facet->packet_count;
3428 facet->rule->byte_count += facet->byte_count;
3430 /* Reset counters to prevent double counting if 'facet' ever gets
3432 facet_reset_counters(facet);
3434 netflow_flow_clear(&facet->nf_flow);
3435 facet->tcp_flags = 0;
3438 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3439 * Returns it if found, otherwise a null pointer.
3441 * The returned facet might need revalidation; use facet_lookup_valid()
3442 * instead if that is important. */
3443 static struct facet *
3444 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
3446 struct facet *facet;
3448 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
3450 if (flow_equal(flow, &facet->flow)) {
3458 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3459 * Returns it if found, otherwise a null pointer.
3461 * The returned facet is guaranteed to be valid. */
3462 static struct facet *
3463 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
3465 struct facet *facet = facet_find(ofproto, flow);
3467 /* The facet we found might not be valid, since we could be in need of
3468 * revalidation. If it is not valid, don't return it. */
3470 && (ofproto->need_revalidate
3471 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))
3472 && !facet_revalidate(facet)) {
3473 COVERAGE_INC(facet_invalidated);
3481 facet_check_consistency(struct facet *facet)
3483 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3485 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3487 uint64_t odp_actions_stub[1024 / 8];
3488 struct ofpbuf odp_actions;
3490 struct rule_dpif *rule;
3491 struct subfacet *subfacet;
3492 bool may_log = false;
3495 /* Check the rule for consistency. */
3496 rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3498 if (!VLOG_DROP_WARN(&rl)) {
3499 char *s = flow_to_string(&facet->flow);
3500 VLOG_WARN("%s: facet should not exist", s);
3504 } else if (rule != facet->rule) {
3505 may_log = !VLOG_DROP_WARN(&rl);
3511 flow_format(&s, &facet->flow);
3512 ds_put_format(&s, ": facet associated with wrong rule (was "
3513 "table=%"PRIu8",", facet->rule->up.table_id);
3514 cls_rule_format(&facet->rule->up.cr, &s);
3515 ds_put_format(&s, ") (should have been table=%"PRIu8",",
3517 cls_rule_format(&rule->up.cr, &s);
3518 ds_put_char(&s, ')');
3520 VLOG_WARN("%s", ds_cstr(&s));
3527 /* Check the datapath actions for consistency. */
3528 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
3529 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3530 struct action_xlate_ctx ctx;
3531 bool actions_changed;
3532 bool should_install;
3534 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3535 subfacet->initial_tci, rule, 0, NULL);
3536 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions,
3539 should_install = (ctx.may_set_up_flow
3540 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3541 if (!should_install && !subfacet->installed) {
3542 /* The actions for uninstallable flows may vary from one packet to
3543 * the next, so don't compare the actions. */
3547 actions_changed = (subfacet->actions_len != odp_actions.size
3548 || memcmp(subfacet->actions, odp_actions.data,
3549 subfacet->actions_len));
3550 if (should_install != subfacet->installed || actions_changed) {
3552 may_log = !VLOG_DROP_WARN(&rl);
3557 struct odputil_keybuf keybuf;
3562 subfacet_get_key(subfacet, &keybuf, &key);
3563 odp_flow_key_format(key.data, key.size, &s);
3565 ds_put_cstr(&s, ": inconsistency in subfacet");
3566 if (should_install != subfacet->installed) {
3567 enum odp_key_fitness fitness = subfacet->key_fitness;
3569 ds_put_format(&s, " (should%s have been installed)",
3570 should_install ? "" : " not");
3571 ds_put_format(&s, " (may_set_up_flow=%s, fitness=%s)",
3572 ctx.may_set_up_flow ? "true" : "false",
3573 odp_key_fitness_to_string(fitness));
3575 if (actions_changed) {
3576 ds_put_cstr(&s, " (actions were: ");
3577 format_odp_actions(&s, subfacet->actions,
3578 subfacet->actions_len);
3579 ds_put_cstr(&s, ") (correct actions: ");
3580 format_odp_actions(&s, odp_actions.data, odp_actions.size);
3581 ds_put_char(&s, ')');
3583 ds_put_cstr(&s, " (actions: ");
3584 format_odp_actions(&s, subfacet->actions,
3585 subfacet->actions_len);
3586 ds_put_char(&s, ')');
3588 VLOG_WARN("%s", ds_cstr(&s));
3593 ofpbuf_uninit(&odp_actions);
3598 /* Re-searches the classifier for 'facet':
3600 * - If the rule found is different from 'facet''s current rule, moves
3601 * 'facet' to the new rule and recompiles its actions.
3603 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3604 * where it is and recompiles its actions anyway.
3606 * - If there is none, destroys 'facet'.
3608 * Returns true if 'facet' still exists, false if it has been destroyed. */
3610 facet_revalidate(struct facet *facet)
3612 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3614 struct nlattr *odp_actions;
3617 struct actions *new_actions;
3619 struct action_xlate_ctx ctx;
3620 uint64_t odp_actions_stub[1024 / 8];
3621 struct ofpbuf odp_actions;
3623 struct rule_dpif *new_rule;
3624 struct subfacet *subfacet;
3625 bool actions_changed;
3628 COVERAGE_INC(facet_revalidate);
3630 /* Determine the new rule. */
3631 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3633 /* No new rule, so delete the facet. */
3634 facet_remove(facet);
3638 /* Calculate new datapath actions.
3640 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3641 * emit a NetFlow expiration and, if so, we need to have the old state
3642 * around to properly compose it. */
3644 /* If the datapath actions changed or the installability changed,
3645 * then we need to talk to the datapath. */
3648 memset(&ctx, 0, sizeof ctx);
3649 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
3650 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3651 bool should_install;
3653 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3654 subfacet->initial_tci, new_rule, 0, NULL);
3655 xlate_actions(&ctx, new_rule->up.actions, new_rule->up.n_actions,
3657 actions_changed = (subfacet->actions_len != odp_actions.size
3658 || memcmp(subfacet->actions, odp_actions.data,
3659 subfacet->actions_len));
3661 should_install = (ctx.may_set_up_flow
3662 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3663 if (actions_changed || should_install != subfacet->installed) {
3664 if (should_install) {
3665 struct dpif_flow_stats stats;
3667 subfacet_install(subfacet,
3668 odp_actions.data, odp_actions.size, &stats);
3669 subfacet_update_stats(subfacet, &stats);
3671 subfacet_uninstall(subfacet);
3675 new_actions = xcalloc(list_size(&facet->subfacets),
3676 sizeof *new_actions);
3678 new_actions[i].odp_actions = xmemdup(odp_actions.data,
3680 new_actions[i].actions_len = odp_actions.size;
3685 ofpbuf_uninit(&odp_actions);
3688 facet_flush_stats(facet);
3691 /* Update 'facet' now that we've taken care of all the old state. */
3692 facet->tags = ctx.tags;
3693 facet->nf_flow.output_iface = ctx.nf_output_iface;
3694 facet->may_install = ctx.may_set_up_flow;
3695 facet->has_learn = ctx.has_learn;
3696 facet->has_normal = ctx.has_normal;
3697 facet->has_fin_timeout = ctx.has_fin_timeout;
3698 facet->mirrors = ctx.mirrors;
3701 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3702 if (new_actions[i].odp_actions) {
3703 free(subfacet->actions);
3704 subfacet->actions = new_actions[i].odp_actions;
3705 subfacet->actions_len = new_actions[i].actions_len;
3711 if (facet->rule != new_rule) {
3712 COVERAGE_INC(facet_changed_rule);
3713 list_remove(&facet->list_node);
3714 list_push_back(&new_rule->facets, &facet->list_node);
3715 facet->rule = new_rule;
3716 facet->used = new_rule->up.created;
3717 facet->prev_used = facet->used;
3723 /* Updates 'facet''s used time. Caller is responsible for calling
3724 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3726 facet_update_time(struct facet *facet, long long int used)
3728 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3729 if (used > facet->used) {
3731 ofproto_rule_update_used(&facet->rule->up, used);
3732 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3737 facet_reset_counters(struct facet *facet)
3739 facet->packet_count = 0;
3740 facet->byte_count = 0;
3741 facet->prev_packet_count = 0;
3742 facet->prev_byte_count = 0;
3743 facet->accounted_bytes = 0;
3747 facet_push_stats(struct facet *facet)
3749 uint64_t new_packets, new_bytes;
3751 assert(facet->packet_count >= facet->prev_packet_count);
3752 assert(facet->byte_count >= facet->prev_byte_count);
3753 assert(facet->used >= facet->prev_used);
3755 new_packets = facet->packet_count - facet->prev_packet_count;
3756 new_bytes = facet->byte_count - facet->prev_byte_count;
3758 if (new_packets || new_bytes || facet->used > facet->prev_used) {
3759 facet->prev_packet_count = facet->packet_count;
3760 facet->prev_byte_count = facet->byte_count;
3761 facet->prev_used = facet->used;
3763 flow_push_stats(facet->rule, &facet->flow,
3764 new_packets, new_bytes, facet->used);
3766 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
3767 facet->mirrors, new_packets, new_bytes);
3771 struct ofproto_push {
3772 struct action_xlate_ctx ctx;
3779 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
3781 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3784 rule->packet_count += push->packets;
3785 rule->byte_count += push->bytes;
3786 ofproto_rule_update_used(&rule->up, push->used);
3790 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3791 * 'rule''s actions and mirrors. */
3793 flow_push_stats(struct rule_dpif *rule,
3794 const struct flow *flow, uint64_t packets, uint64_t bytes,
3797 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3798 struct ofproto_push push;
3800 push.packets = packets;
3804 ofproto_rule_update_used(&rule->up, used);
3806 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, rule,
3808 push.ctx.resubmit_hook = push_resubmit;
3809 xlate_actions_for_side_effects(&push.ctx,
3810 rule->up.actions, rule->up.n_actions);
3815 static struct subfacet *
3816 subfacet_find__(struct ofproto_dpif *ofproto,
3817 const struct nlattr *key, size_t key_len, uint32_t key_hash,
3818 const struct flow *flow)
3820 struct subfacet *subfacet;
3822 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
3823 &ofproto->subfacets) {
3825 ? (subfacet->key_len == key_len
3826 && !memcmp(key, subfacet->key, key_len))
3827 : flow_equal(flow, &subfacet->facet->flow)) {
3835 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
3836 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
3837 * there is one, otherwise creates and returns a new subfacet.
3839 * If the returned subfacet is new, then subfacet->actions will be NULL, in
3840 * which case the caller must populate the actions with
3841 * subfacet_make_actions(). */
3842 static struct subfacet *
3843 subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness,
3844 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
3846 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3847 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3848 struct subfacet *subfacet;
3850 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
3852 if (subfacet->facet == facet) {
3856 /* This shouldn't happen. */
3857 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
3858 subfacet_destroy(subfacet);
3861 subfacet = xzalloc(sizeof *subfacet);
3862 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
3863 list_push_back(&facet->subfacets, &subfacet->list_node);
3864 subfacet->facet = facet;
3865 subfacet->used = time_msec();
3866 subfacet->key_fitness = key_fitness;
3867 if (key_fitness != ODP_FIT_PERFECT) {
3868 subfacet->key = xmemdup(key, key_len);
3869 subfacet->key_len = key_len;
3871 subfacet->installed = false;
3872 subfacet->initial_tci = initial_tci;
3877 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
3878 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
3879 static struct subfacet *
3880 subfacet_find(struct ofproto_dpif *ofproto,
3881 const struct nlattr *key, size_t key_len)
3883 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3884 enum odp_key_fitness fitness;
3887 fitness = odp_flow_key_to_flow(key, key_len, &flow);
3888 if (fitness == ODP_FIT_ERROR) {
3892 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
3895 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
3896 * its facet within 'ofproto', and frees it. */
3898 subfacet_destroy__(struct subfacet *subfacet)
3900 struct facet *facet = subfacet->facet;
3901 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3903 subfacet_uninstall(subfacet);
3904 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
3905 list_remove(&subfacet->list_node);
3906 free(subfacet->key);
3907 free(subfacet->actions);
3911 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
3912 * last remaining subfacet in its facet destroys the facet too. */
3914 subfacet_destroy(struct subfacet *subfacet)
3916 struct facet *facet = subfacet->facet;
3918 if (list_is_singleton(&facet->subfacets)) {
3919 /* facet_remove() needs at least one subfacet (it will remove it). */
3920 facet_remove(facet);
3922 subfacet_destroy__(subfacet);
3926 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
3927 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
3928 * for use as temporary storage. */
3930 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
3933 if (!subfacet->key) {
3934 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
3935 odp_flow_key_from_flow(key, &subfacet->facet->flow);
3937 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
3941 /* Composes the datapath actions for 'subfacet' based on its rule's actions. */
3943 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet)
3945 struct facet *facet = subfacet->facet;
3946 struct rule_dpif *rule = facet->rule;
3947 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3949 struct action_xlate_ctx ctx;
3950 uint64_t odp_actions_stub[1024 / 8];
3951 struct ofpbuf odp_actions;
3953 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
3954 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
3956 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions, &odp_actions);
3957 facet->tags = ctx.tags;
3958 facet->may_install = ctx.may_set_up_flow;
3959 facet->has_learn = ctx.has_learn;
3960 facet->has_normal = ctx.has_normal;
3961 facet->has_fin_timeout = ctx.has_fin_timeout;
3962 facet->nf_flow.output_iface = ctx.nf_output_iface;
3963 facet->mirrors = ctx.mirrors;
3965 if (subfacet->actions_len != odp_actions.size
3966 || memcmp(subfacet->actions, odp_actions.data, odp_actions.size)) {
3967 free(subfacet->actions);
3968 subfacet->actions_len = odp_actions.size;
3969 subfacet->actions = xmemdup(odp_actions.data, odp_actions.size);
3972 ofpbuf_uninit(&odp_actions);
3975 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
3976 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
3977 * in the datapath will be zeroed and 'stats' will be updated with traffic new
3978 * since 'subfacet' was last updated.
3980 * Returns 0 if successful, otherwise a positive errno value. */
3982 subfacet_install(struct subfacet *subfacet,
3983 const struct nlattr *actions, size_t actions_len,
3984 struct dpif_flow_stats *stats)
3986 struct facet *facet = subfacet->facet;
3987 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3988 struct odputil_keybuf keybuf;
3989 enum dpif_flow_put_flags flags;
3993 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3995 flags |= DPIF_FP_ZERO_STATS;
3998 subfacet_get_key(subfacet, &keybuf, &key);
3999 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
4000 actions, actions_len, stats);
4003 subfacet_reset_dp_stats(subfacet, stats);
4009 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4011 subfacet_uninstall(struct subfacet *subfacet)
4013 if (subfacet->installed) {
4014 struct rule_dpif *rule = subfacet->facet->rule;
4015 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4016 struct odputil_keybuf keybuf;
4017 struct dpif_flow_stats stats;
4021 subfacet_get_key(subfacet, &keybuf, &key);
4022 error = dpif_flow_del(ofproto->dpif, key.data, key.size, &stats);
4023 subfacet_reset_dp_stats(subfacet, &stats);
4025 subfacet_update_stats(subfacet, &stats);
4027 subfacet->installed = false;
4029 assert(subfacet->dp_packet_count == 0);
4030 assert(subfacet->dp_byte_count == 0);
4034 /* Resets 'subfacet''s datapath statistics counters. This should be called
4035 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4036 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4037 * was reset in the datapath. 'stats' will be modified to include only
4038 * statistics new since 'subfacet' was last updated. */
4040 subfacet_reset_dp_stats(struct subfacet *subfacet,
4041 struct dpif_flow_stats *stats)
4044 && subfacet->dp_packet_count <= stats->n_packets
4045 && subfacet->dp_byte_count <= stats->n_bytes) {
4046 stats->n_packets -= subfacet->dp_packet_count;
4047 stats->n_bytes -= subfacet->dp_byte_count;
4050 subfacet->dp_packet_count = 0;
4051 subfacet->dp_byte_count = 0;
4054 /* Updates 'subfacet''s used time. The caller is responsible for calling
4055 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
4057 subfacet_update_time(struct subfacet *subfacet, long long int used)
4059 if (used > subfacet->used) {
4060 subfacet->used = used;
4061 facet_update_time(subfacet->facet, used);
4065 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4067 * Because of the meaning of a subfacet's counters, it only makes sense to do
4068 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4069 * represents a packet that was sent by hand or if it represents statistics
4070 * that have been cleared out of the datapath. */
4072 subfacet_update_stats(struct subfacet *subfacet,
4073 const struct dpif_flow_stats *stats)
4075 if (stats->n_packets || stats->used > subfacet->used) {
4076 struct facet *facet = subfacet->facet;
4078 subfacet_update_time(subfacet, stats->used);
4079 facet->packet_count += stats->n_packets;
4080 facet->byte_count += stats->n_bytes;
4081 facet->tcp_flags |= stats->tcp_flags;
4082 facet_push_stats(facet);
4083 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
4089 static struct rule_dpif *
4090 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4093 struct cls_rule *cls_rule;
4094 struct classifier *cls;
4096 if (table_id >= N_TABLES) {
4100 cls = &ofproto->up.tables[table_id].cls;
4101 if (flow->nw_frag & FLOW_NW_FRAG_ANY
4102 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4103 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
4104 * are unavailable. */
4105 struct flow ofpc_normal_flow = *flow;
4106 ofpc_normal_flow.tp_src = htons(0);
4107 ofpc_normal_flow.tp_dst = htons(0);
4108 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
4110 cls_rule = classifier_lookup(cls, flow);
4112 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
4116 complete_operation(struct rule_dpif *rule)
4118 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4120 rule_invalidate(rule);
4122 struct dpif_completion *c = xmalloc(sizeof *c);
4123 c->op = rule->up.pending;
4124 list_push_back(&ofproto->completions, &c->list_node);
4126 ofoperation_complete(rule->up.pending, 0);
4130 static struct rule *
4133 struct rule_dpif *rule = xmalloc(sizeof *rule);
4138 rule_dealloc(struct rule *rule_)
4140 struct rule_dpif *rule = rule_dpif_cast(rule_);
4145 rule_construct(struct rule *rule_)
4147 struct rule_dpif *rule = rule_dpif_cast(rule_);
4148 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4149 struct rule_dpif *victim;
4153 error = validate_actions(rule->up.actions, rule->up.n_actions,
4154 &rule->up.cr.flow, ofproto->max_ports);
4159 rule->packet_count = 0;
4160 rule->byte_count = 0;
4162 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
4163 if (victim && !list_is_empty(&victim->facets)) {
4164 struct facet *facet;
4166 rule->facets = victim->facets;
4167 list_moved(&rule->facets);
4168 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4169 /* XXX: We're only clearing our local counters here. It's possible
4170 * that quite a few packets are unaccounted for in the datapath
4171 * statistics. These will be accounted to the new rule instead of
4172 * cleared as required. This could be fixed by clearing out the
4173 * datapath statistics for this facet, but currently it doesn't
4175 facet_reset_counters(facet);
4179 /* Must avoid list_moved() in this case. */
4180 list_init(&rule->facets);
4183 table_id = rule->up.table_id;
4184 rule->tag = (victim ? victim->tag
4186 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
4187 ofproto->tables[table_id].basis));
4189 complete_operation(rule);
4194 rule_destruct(struct rule *rule_)
4196 struct rule_dpif *rule = rule_dpif_cast(rule_);
4197 struct facet *facet, *next_facet;
4199 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4200 facet_revalidate(facet);
4203 complete_operation(rule);
4207 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4209 struct rule_dpif *rule = rule_dpif_cast(rule_);
4210 struct facet *facet;
4212 /* Start from historical data for 'rule' itself that are no longer tracked
4213 * in facets. This counts, for example, facets that have expired. */
4214 *packets = rule->packet_count;
4215 *bytes = rule->byte_count;
4217 /* Add any statistics that are tracked by facets. This includes
4218 * statistical data recently updated by ofproto_update_stats() as well as
4219 * stats for packets that were executed "by hand" via dpif_execute(). */
4220 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4221 *packets += facet->packet_count;
4222 *bytes += facet->byte_count;
4227 rule_execute(struct rule *rule_, const struct flow *flow,
4228 struct ofpbuf *packet)
4230 struct rule_dpif *rule = rule_dpif_cast(rule_);
4231 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4233 size_t size = packet->size;
4235 struct action_xlate_ctx ctx;
4236 uint64_t odp_actions_stub[1024 / 8];
4237 struct ofpbuf odp_actions;
4239 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4240 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
4241 rule, packet_get_tcp_flags(packet, flow), packet);
4242 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions, &odp_actions);
4243 if (execute_odp_actions(ofproto, flow, odp_actions.data,
4244 odp_actions.size, packet)) {
4245 rule->packet_count++;
4246 rule->byte_count += size;
4247 flow_push_stats(rule, flow, 1, size, time_msec());
4249 ofpbuf_uninit(&odp_actions);
4255 rule_modify_actions(struct rule *rule_)
4257 struct rule_dpif *rule = rule_dpif_cast(rule_);
4258 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4261 error = validate_actions(rule->up.actions, rule->up.n_actions,
4262 &rule->up.cr.flow, ofproto->max_ports);
4264 ofoperation_complete(rule->up.pending, error);
4268 complete_operation(rule);
4271 /* Sends 'packet' out 'ofport'.
4272 * May modify 'packet'.
4273 * Returns 0 if successful, otherwise a positive errno value. */
4275 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4277 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4278 struct ofpbuf key, odp_actions;
4279 struct odputil_keybuf keybuf;
4284 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
4285 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4287 if (odp_port != ofport->odp_port) {
4288 eth_pop_vlan(packet);
4289 flow.vlan_tci = htons(0);
4292 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4293 odp_flow_key_from_flow(&key, &flow);
4295 ofpbuf_init(&odp_actions, 32);
4296 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4298 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4299 error = dpif_execute(ofproto->dpif,
4301 odp_actions.data, odp_actions.size,
4303 ofpbuf_uninit(&odp_actions);
4306 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4307 ofproto->up.name, odp_port, strerror(error));
4309 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
4313 /* OpenFlow to datapath action translation. */
4315 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
4316 struct action_xlate_ctx *ctx);
4317 static void xlate_normal(struct action_xlate_ctx *);
4320 put_userspace_action(const struct ofproto_dpif *ofproto,
4321 struct ofpbuf *odp_actions,
4322 const struct flow *flow,
4323 const struct user_action_cookie *cookie)
4327 pid = dpif_port_get_pid(ofproto->dpif,
4328 ofp_port_to_odp_port(flow->in_port));
4330 return odp_put_userspace_action(pid, cookie, odp_actions);
4333 /* Compose SAMPLE action for sFlow. */
4335 compose_sflow_action(const struct ofproto_dpif *ofproto,
4336 struct ofpbuf *odp_actions,
4337 const struct flow *flow,
4340 uint32_t port_ifindex;
4341 uint32_t probability;
4342 struct user_action_cookie cookie;
4343 size_t sample_offset, actions_offset;
4344 int cookie_offset, n_output;
4346 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4350 if (odp_port == OVSP_NONE) {
4354 port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4358 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4360 /* Number of packets out of UINT_MAX to sample. */
4361 probability = dpif_sflow_get_probability(ofproto->sflow);
4362 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4364 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4366 cookie.type = USER_ACTION_COOKIE_SFLOW;
4367 cookie.data = port_ifindex;
4368 cookie.n_output = n_output;
4369 cookie.vlan_tci = 0;
4370 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4372 nl_msg_end_nested(odp_actions, actions_offset);
4373 nl_msg_end_nested(odp_actions, sample_offset);
4374 return cookie_offset;
4377 /* SAMPLE action must be first action in any given list of actions.
4378 * At this point we do not have all information required to build it. So try to
4379 * build sample action as complete as possible. */
4381 add_sflow_action(struct action_xlate_ctx *ctx)
4383 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4385 &ctx->flow, OVSP_NONE);
4386 ctx->sflow_odp_port = 0;
4387 ctx->sflow_n_outputs = 0;
4390 /* Fix SAMPLE action according to data collected while composing ODP actions.
4391 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4392 * USERSPACE action's user-cookie which is required for sflow. */
4394 fix_sflow_action(struct action_xlate_ctx *ctx)
4396 const struct flow *base = &ctx->base_flow;
4397 struct user_action_cookie *cookie;
4399 if (!ctx->user_cookie_offset) {
4403 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4405 assert(cookie != NULL);
4406 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4408 if (ctx->sflow_n_outputs) {
4409 cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow,
4410 ctx->sflow_odp_port);
4412 if (ctx->sflow_n_outputs >= 255) {
4413 cookie->n_output = 255;
4415 cookie->n_output = ctx->sflow_n_outputs;
4417 cookie->vlan_tci = base->vlan_tci;
4421 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4424 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4425 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4426 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4427 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4431 struct priority_to_dscp *pdscp;
4433 if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD
4434 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4438 pdscp = get_priority(ofport, ctx->flow.skb_priority);
4440 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4441 ctx->flow.nw_tos |= pdscp->dscp;
4444 /* We may not have an ofport record for this port, but it doesn't hurt
4445 * to allow forwarding to it anyhow. Maybe such a port will appear
4446 * later and we're pre-populating the flow table. */
4449 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4450 ctx->flow.vlan_tci);
4451 if (out_port != odp_port) {
4452 ctx->flow.vlan_tci = htons(0);
4454 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4455 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4457 ctx->sflow_odp_port = odp_port;
4458 ctx->sflow_n_outputs++;
4459 ctx->nf_output_iface = ofp_port;
4460 ctx->flow.vlan_tci = flow_vlan_tci;
4461 ctx->flow.nw_tos = flow_nw_tos;
4465 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4467 compose_output_action__(ctx, ofp_port, true);
4471 xlate_table_action(struct action_xlate_ctx *ctx,
4472 uint16_t in_port, uint8_t table_id)
4474 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4475 struct ofproto_dpif *ofproto = ctx->ofproto;
4476 struct rule_dpif *rule;
4477 uint16_t old_in_port;
4478 uint8_t old_table_id;
4480 old_table_id = ctx->table_id;
4481 ctx->table_id = table_id;
4483 /* Look up a flow with 'in_port' as the input port. */
4484 old_in_port = ctx->flow.in_port;
4485 ctx->flow.in_port = in_port;
4486 rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
4489 if (table_id > 0 && table_id < N_TABLES) {
4490 struct table_dpif *table = &ofproto->tables[table_id];
4491 if (table->other_table) {
4492 ctx->tags |= (rule && rule->tag
4494 : rule_calculate_tag(&ctx->flow,
4495 &table->other_table->wc,
4500 /* Restore the original input port. Otherwise OFPP_NORMAL and
4501 * OFPP_IN_PORT will have surprising behavior. */
4502 ctx->flow.in_port = old_in_port;
4504 if (ctx->resubmit_hook) {
4505 ctx->resubmit_hook(ctx, rule);
4509 struct rule_dpif *old_rule = ctx->rule;
4513 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
4514 ctx->rule = old_rule;
4518 ctx->table_id = old_table_id;
4520 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4522 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
4523 MAX_RESUBMIT_RECURSION);
4524 ctx->max_resubmit_trigger = true;
4529 xlate_resubmit_table(struct action_xlate_ctx *ctx,
4530 const struct nx_action_resubmit *nar)
4535 in_port = (nar->in_port == htons(OFPP_IN_PORT)
4537 : ntohs(nar->in_port));
4538 table_id = nar->table == 255 ? ctx->table_id : nar->table;
4540 xlate_table_action(ctx, in_port, table_id);
4544 flood_packets(struct action_xlate_ctx *ctx, bool all)
4546 struct ofport_dpif *ofport;
4548 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
4549 uint16_t ofp_port = ofport->up.ofp_port;
4551 if (ofp_port == ctx->flow.in_port) {
4556 compose_output_action__(ctx, ofp_port, false);
4557 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
4558 compose_output_action(ctx, ofp_port);
4562 ctx->nf_output_iface = NF_OUT_FLOOD;
4566 execute_controller_action(struct action_xlate_ctx *ctx, int len,
4567 enum ofp_packet_in_reason reason,
4568 uint16_t controller_id)
4570 struct ofputil_packet_in pin;
4571 struct ofpbuf *packet;
4573 ctx->may_set_up_flow = false;
4578 packet = ofpbuf_clone(ctx->packet);
4580 if (packet->l2 && packet->l3) {
4581 struct eth_header *eh;
4583 eth_pop_vlan(packet);
4586 /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
4587 * LLC frame. Calculating the Ethernet type of these frames is more
4588 * trouble than seems appropriate for a simple assertion. */
4589 assert(ntohs(eh->eth_type) < ETH_TYPE_MIN
4590 || eh->eth_type == ctx->flow.dl_type);
4592 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
4593 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
4595 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
4596 eth_push_vlan(packet, ctx->flow.vlan_tci);
4600 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4601 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
4602 ctx->flow.nw_tos, ctx->flow.nw_ttl);
4606 if (ctx->flow.nw_proto == IPPROTO_TCP) {
4607 packet_set_tcp_port(packet, ctx->flow.tp_src,
4609 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
4610 packet_set_udp_port(packet, ctx->flow.tp_src,
4617 pin.packet = packet->data;
4618 pin.packet_len = packet->size;
4619 pin.reason = reason;
4620 pin.controller_id = controller_id;
4621 pin.table_id = ctx->table_id;
4622 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
4625 flow_get_metadata(&ctx->flow, &pin.fmd);
4627 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
4628 ofpbuf_delete(packet);
4632 compose_dec_ttl(struct action_xlate_ctx *ctx)
4634 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
4635 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
4639 if (ctx->flow.nw_ttl > 1) {
4643 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
4645 /* Stop processing for current table. */
4651 xlate_output_action__(struct action_xlate_ctx *ctx,
4652 uint16_t port, uint16_t max_len)
4654 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
4656 ctx->nf_output_iface = NF_OUT_DROP;
4660 compose_output_action(ctx, ctx->flow.in_port);
4663 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
4669 flood_packets(ctx, false);
4672 flood_packets(ctx, true);
4674 case OFPP_CONTROLLER:
4675 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
4681 if (port != ctx->flow.in_port) {
4682 compose_output_action(ctx, port);
4687 if (prev_nf_output_iface == NF_OUT_FLOOD) {
4688 ctx->nf_output_iface = NF_OUT_FLOOD;
4689 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
4690 ctx->nf_output_iface = prev_nf_output_iface;
4691 } else if (prev_nf_output_iface != NF_OUT_DROP &&
4692 ctx->nf_output_iface != NF_OUT_FLOOD) {
4693 ctx->nf_output_iface = NF_OUT_MULTI;
4698 xlate_output_reg_action(struct action_xlate_ctx *ctx,
4699 const struct nx_action_output_reg *naor)
4701 struct mf_subfield src;
4704 nxm_decode(&src, naor->src, naor->ofs_nbits);
4705 ofp_port = mf_get_subfield(&src, &ctx->flow);
4707 if (ofp_port <= UINT16_MAX) {
4708 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
4713 xlate_output_action(struct action_xlate_ctx *ctx,
4714 const struct ofp_action_output *oao)
4716 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
4720 xlate_enqueue_action(struct action_xlate_ctx *ctx,
4721 const struct ofp_action_enqueue *oae)
4724 uint32_t flow_priority, priority;
4727 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
4730 /* Fall back to ordinary output action. */
4731 xlate_output_action__(ctx, ntohs(oae->port), 0);
4735 /* Figure out datapath output port. */
4736 ofp_port = ntohs(oae->port);
4737 if (ofp_port == OFPP_IN_PORT) {
4738 ofp_port = ctx->flow.in_port;
4739 } else if (ofp_port == ctx->flow.in_port) {
4743 /* Add datapath actions. */
4744 flow_priority = ctx->flow.skb_priority;
4745 ctx->flow.skb_priority = priority;
4746 compose_output_action(ctx, ofp_port);
4747 ctx->flow.skb_priority = flow_priority;
4749 /* Update NetFlow output port. */
4750 if (ctx->nf_output_iface == NF_OUT_DROP) {
4751 ctx->nf_output_iface = ofp_port;
4752 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
4753 ctx->nf_output_iface = NF_OUT_MULTI;
4758 xlate_set_queue_action(struct action_xlate_ctx *ctx,
4759 const struct nx_action_set_queue *nasq)
4764 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
4767 /* Couldn't translate queue to a priority, so ignore. A warning
4768 * has already been logged. */
4772 ctx->flow.skb_priority = priority;
4775 struct xlate_reg_state {
4781 xlate_autopath(struct action_xlate_ctx *ctx,
4782 const struct nx_action_autopath *naa)
4784 uint16_t ofp_port = ntohl(naa->id);
4785 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
4787 if (!port || !port->bundle) {
4788 ofp_port = OFPP_NONE;
4789 } else if (port->bundle->bond) {
4790 /* Autopath does not support VLAN hashing. */
4791 struct ofport_dpif *slave = bond_choose_output_slave(
4792 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
4794 ofp_port = slave->up.ofp_port;
4797 autopath_execute(naa, &ctx->flow, ofp_port);
4801 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
4803 struct ofproto_dpif *ofproto = ofproto_;
4804 struct ofport_dpif *port;
4814 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
4817 port = get_ofp_port(ofproto, ofp_port);
4818 return port ? port->may_enable : false;
4823 xlate_learn_action(struct action_xlate_ctx *ctx,
4824 const struct nx_action_learn *learn)
4826 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
4827 struct ofputil_flow_mod fm;
4830 learn_execute(learn, &ctx->flow, &fm);
4832 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
4833 if (error && !VLOG_DROP_WARN(&rl)) {
4834 VLOG_WARN("learning action failed to modify flow table (%s)",
4835 ofperr_get_name(error));
4841 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
4842 * means "infinite". */
4844 reduce_timeout(uint16_t max, uint16_t *timeout)
4846 if (max && (!*timeout || *timeout > max)) {
4852 xlate_fin_timeout(struct action_xlate_ctx *ctx,
4853 const struct nx_action_fin_timeout *naft)
4855 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
4856 struct rule_dpif *rule = ctx->rule;
4858 reduce_timeout(ntohs(naft->fin_idle_timeout), &rule->up.idle_timeout);
4859 reduce_timeout(ntohs(naft->fin_hard_timeout), &rule->up.hard_timeout);
4864 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
4866 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
4867 ? OFPUTIL_PC_NO_RECV_STP
4868 : OFPUTIL_PC_NO_RECV)) {
4872 /* Only drop packets here if both forwarding and learning are
4873 * disabled. If just learning is enabled, we need to have
4874 * OFPP_NORMAL and the learning action have a look at the packet
4875 * before we can drop it. */
4876 if (!stp_forward_in_state(port->stp_state)
4877 && !stp_learn_in_state(port->stp_state)) {
4885 do_xlate_actions(const union ofp_action *in, size_t n_in,
4886 struct action_xlate_ctx *ctx)
4888 const struct ofport_dpif *port;
4889 const union ofp_action *ia;
4890 bool was_evictable = true;
4893 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
4894 if (port && !may_receive(port, ctx)) {
4895 /* Drop this flow. */
4900 /* Don't let the rule we're working on get evicted underneath us. */
4901 was_evictable = ctx->rule->up.evictable;
4902 ctx->rule->up.evictable = false;
4904 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
4905 const struct ofp_action_dl_addr *oada;
4906 const struct nx_action_resubmit *nar;
4907 const struct nx_action_set_tunnel *nast;
4908 const struct nx_action_set_queue *nasq;
4909 const struct nx_action_multipath *nam;
4910 const struct nx_action_autopath *naa;
4911 const struct nx_action_bundle *nab;
4912 const struct nx_action_output_reg *naor;
4913 const struct nx_action_controller *nac;
4914 enum ofputil_action_code code;
4921 code = ofputil_decode_action_unsafe(ia);
4923 case OFPUTIL_OFPAT10_OUTPUT:
4924 xlate_output_action(ctx, &ia->output);
4927 case OFPUTIL_OFPAT10_SET_VLAN_VID:
4928 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
4929 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
4932 case OFPUTIL_OFPAT10_SET_VLAN_PCP:
4933 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
4934 ctx->flow.vlan_tci |= htons(
4935 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
4938 case OFPUTIL_OFPAT10_STRIP_VLAN:
4939 ctx->flow.vlan_tci = htons(0);
4942 case OFPUTIL_OFPAT10_SET_DL_SRC:
4943 oada = ((struct ofp_action_dl_addr *) ia);
4944 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
4947 case OFPUTIL_OFPAT10_SET_DL_DST:
4948 oada = ((struct ofp_action_dl_addr *) ia);
4949 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
4952 case OFPUTIL_OFPAT10_SET_NW_SRC:
4953 ctx->flow.nw_src = ia->nw_addr.nw_addr;
4956 case OFPUTIL_OFPAT10_SET_NW_DST:
4957 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
4960 case OFPUTIL_OFPAT10_SET_NW_TOS:
4961 /* OpenFlow 1.0 only supports IPv4. */
4962 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4963 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4964 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
4968 case OFPUTIL_OFPAT10_SET_TP_SRC:
4969 ctx->flow.tp_src = ia->tp_port.tp_port;
4972 case OFPUTIL_OFPAT10_SET_TP_DST:
4973 ctx->flow.tp_dst = ia->tp_port.tp_port;
4976 case OFPUTIL_OFPAT10_ENQUEUE:
4977 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
4980 case OFPUTIL_NXAST_RESUBMIT:
4981 nar = (const struct nx_action_resubmit *) ia;
4982 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
4985 case OFPUTIL_NXAST_RESUBMIT_TABLE:
4986 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
4989 case OFPUTIL_NXAST_SET_TUNNEL:
4990 nast = (const struct nx_action_set_tunnel *) ia;
4991 tun_id = htonll(ntohl(nast->tun_id));
4992 ctx->flow.tun_id = tun_id;
4995 case OFPUTIL_NXAST_SET_QUEUE:
4996 nasq = (const struct nx_action_set_queue *) ia;
4997 xlate_set_queue_action(ctx, nasq);
5000 case OFPUTIL_NXAST_POP_QUEUE:
5001 ctx->flow.skb_priority = ctx->orig_skb_priority;
5004 case OFPUTIL_NXAST_REG_MOVE:
5005 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
5009 case OFPUTIL_NXAST_REG_LOAD:
5010 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
5014 case OFPUTIL_NXAST_NOTE:
5015 /* Nothing to do. */
5018 case OFPUTIL_NXAST_SET_TUNNEL64:
5019 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
5020 ctx->flow.tun_id = tun_id;
5023 case OFPUTIL_NXAST_MULTIPATH:
5024 nam = (const struct nx_action_multipath *) ia;
5025 multipath_execute(nam, &ctx->flow);
5028 case OFPUTIL_NXAST_AUTOPATH:
5029 naa = (const struct nx_action_autopath *) ia;
5030 xlate_autopath(ctx, naa);
5033 case OFPUTIL_NXAST_BUNDLE:
5034 ctx->ofproto->has_bundle_action = true;
5035 nab = (const struct nx_action_bundle *) ia;
5036 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
5041 case OFPUTIL_NXAST_BUNDLE_LOAD:
5042 ctx->ofproto->has_bundle_action = true;
5043 nab = (const struct nx_action_bundle *) ia;
5044 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
5048 case OFPUTIL_NXAST_OUTPUT_REG:
5049 naor = (const struct nx_action_output_reg *) ia;
5050 xlate_output_reg_action(ctx, naor);
5053 case OFPUTIL_NXAST_LEARN:
5054 ctx->has_learn = true;
5055 if (ctx->may_learn) {
5056 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
5060 case OFPUTIL_NXAST_DEC_TTL:
5061 if (compose_dec_ttl(ctx)) {
5066 case OFPUTIL_NXAST_EXIT:
5070 case OFPUTIL_NXAST_FIN_TIMEOUT:
5071 ctx->has_fin_timeout = true;
5072 xlate_fin_timeout(ctx, (const struct nx_action_fin_timeout *) ia);
5075 case OFPUTIL_NXAST_CONTROLLER:
5076 nac = (const struct nx_action_controller *) ia;
5077 execute_controller_action(ctx, ntohs(nac->max_len), nac->reason,
5078 ntohs(nac->controller_id));
5084 /* We've let OFPP_NORMAL and the learning action look at the packet,
5085 * so drop it now if forwarding is disabled. */
5086 if (port && !stp_forward_in_state(port->stp_state)) {
5087 ofpbuf_clear(ctx->odp_actions);
5088 add_sflow_action(ctx);
5091 ctx->rule->up.evictable = was_evictable;
5096 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
5097 struct ofproto_dpif *ofproto, const struct flow *flow,
5098 ovs_be16 initial_tci, struct rule_dpif *rule,
5099 uint8_t tcp_flags, const struct ofpbuf *packet)
5101 ctx->ofproto = ofproto;
5103 ctx->base_flow = ctx->flow;
5104 ctx->base_flow.tun_id = 0;
5105 ctx->base_flow.vlan_tci = initial_tci;
5107 ctx->packet = packet;
5108 ctx->may_learn = packet != NULL;
5109 ctx->tcp_flags = tcp_flags;
5110 ctx->resubmit_hook = NULL;
5113 /* Translates the 'n_in' "union ofp_action"s in 'in' into datapath actions in
5114 * 'odp_actions', using 'ctx'. */
5116 xlate_actions(struct action_xlate_ctx *ctx,
5117 const union ofp_action *in, size_t n_in,
5118 struct ofpbuf *odp_actions)
5120 struct flow orig_flow = ctx->flow;
5122 COVERAGE_INC(ofproto_dpif_xlate);
5124 ofpbuf_clear(odp_actions);
5125 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
5127 ctx->odp_actions = odp_actions;
5129 ctx->may_set_up_flow = true;
5130 ctx->has_learn = false;
5131 ctx->has_normal = false;
5132 ctx->has_fin_timeout = false;
5133 ctx->nf_output_iface = NF_OUT_DROP;
5136 ctx->max_resubmit_trigger = false;
5137 ctx->orig_skb_priority = ctx->flow.skb_priority;
5141 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
5142 switch (ctx->ofproto->up.frag_handling) {
5143 case OFPC_FRAG_NORMAL:
5144 /* We must pretend that transport ports are unavailable. */
5145 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
5146 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
5149 case OFPC_FRAG_DROP:
5152 case OFPC_FRAG_REASM:
5155 case OFPC_FRAG_NX_MATCH:
5156 /* Nothing to do. */
5159 case OFPC_INVALID_TTL_TO_CONTROLLER:
5164 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
5165 ctx->may_set_up_flow = false;
5167 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5168 struct flow original_flow = ctx->flow;
5169 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
5171 add_sflow_action(ctx);
5172 do_xlate_actions(in, n_in, ctx);
5174 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook
5175 && !VLOG_DROP_ERR(&trace_rl)) {
5176 struct ds ds = DS_EMPTY_INITIALIZER;
5178 ofproto_trace(ctx->ofproto, &original_flow, ctx->packet,
5180 VLOG_ERR("Trace triggered by excessive resubmit recursion:\n%s",
5185 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
5186 ctx->odp_actions->data,
5187 ctx->odp_actions->size)) {
5188 ctx->may_set_up_flow = false;
5190 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
5192 compose_output_action(ctx, OFPP_LOCAL);
5195 add_mirror_actions(ctx, &orig_flow);
5196 fix_sflow_action(ctx);
5200 /* Translates the 'n_in' "union ofp_action"s in 'in' into datapath actions,
5201 * using 'ctx', and discards the datapath actions. */
5203 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
5204 const union ofp_action *in, size_t n_in)
5206 uint64_t odp_actions_stub[1024 / 8];
5207 struct ofpbuf odp_actions;
5209 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5210 xlate_actions(ctx, in, n_in, &odp_actions);
5211 ofpbuf_uninit(&odp_actions);
5214 /* OFPP_NORMAL implementation. */
5216 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
5218 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
5219 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
5220 * the bundle on which the packet was received, returns the VLAN to which the
5223 * Both 'vid' and the return value are in the range 0...4095. */
5225 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
5227 switch (in_bundle->vlan_mode) {
5228 case PORT_VLAN_ACCESS:
5229 return in_bundle->vlan;
5232 case PORT_VLAN_TRUNK:
5235 case PORT_VLAN_NATIVE_UNTAGGED:
5236 case PORT_VLAN_NATIVE_TAGGED:
5237 return vid ? vid : in_bundle->vlan;
5244 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
5245 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
5248 * 'vid' should be 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), in the range
5252 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
5254 /* Allow any VID on the OFPP_NONE port. */
5255 if (in_bundle == &ofpp_none_bundle) {
5259 switch (in_bundle->vlan_mode) {
5260 case PORT_VLAN_ACCESS:
5263 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5264 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
5265 "packet received on port %s configured as VLAN "
5266 "%"PRIu16" access port",
5267 in_bundle->ofproto->up.name, vid,
5268 in_bundle->name, in_bundle->vlan);
5274 case PORT_VLAN_NATIVE_UNTAGGED:
5275 case PORT_VLAN_NATIVE_TAGGED:
5277 /* Port must always carry its native VLAN. */
5281 case PORT_VLAN_TRUNK:
5282 if (!ofbundle_includes_vlan(in_bundle, vid)) {
5284 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5285 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
5286 "received on port %s not configured for trunking "
5288 in_bundle->ofproto->up.name, vid,
5289 in_bundle->name, vid);
5301 /* Given 'vlan', the VLAN that a packet belongs to, and
5302 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
5303 * that should be included in the 802.1Q header. (If the return value is 0,
5304 * then the 802.1Q header should only be included in the packet if there is a
5307 * Both 'vlan' and the return value are in the range 0...4095. */
5309 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
5311 switch (out_bundle->vlan_mode) {
5312 case PORT_VLAN_ACCESS:
5315 case PORT_VLAN_TRUNK:
5316 case PORT_VLAN_NATIVE_TAGGED:
5319 case PORT_VLAN_NATIVE_UNTAGGED:
5320 return vlan == out_bundle->vlan ? 0 : vlan;
5328 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
5331 struct ofport_dpif *port;
5333 ovs_be16 tci, old_tci;
5335 vid = output_vlan_to_vid(out_bundle, vlan);
5336 if (!out_bundle->bond) {
5337 port = ofbundle_get_a_port(out_bundle);
5339 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
5342 /* No slaves enabled, so drop packet. */
5347 old_tci = ctx->flow.vlan_tci;
5349 if (tci || out_bundle->use_priority_tags) {
5350 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
5352 tci |= htons(VLAN_CFI);
5355 ctx->flow.vlan_tci = tci;
5357 compose_output_action(ctx, port->up.ofp_port);
5358 ctx->flow.vlan_tci = old_tci;
5362 mirror_mask_ffs(mirror_mask_t mask)
5364 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
5369 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
5371 return (bundle->vlan_mode != PORT_VLAN_ACCESS
5372 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
5376 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
5378 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
5381 /* Returns an arbitrary interface within 'bundle'. */
5382 static struct ofport_dpif *
5383 ofbundle_get_a_port(const struct ofbundle *bundle)
5385 return CONTAINER_OF(list_front(&bundle->ports),
5386 struct ofport_dpif, bundle_node);
5390 vlan_is_mirrored(const struct ofmirror *m, int vlan)
5392 return !m->vlans || bitmap_is_set(m->vlans, vlan);
5395 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
5396 * to a VLAN. In general most packets may be mirrored but we want to drop
5397 * protocols that may confuse switches. */
5399 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
5401 /* If you change this function's behavior, please update corresponding
5402 * documentation in vswitch.xml at the same time. */
5403 if (dst[0] != 0x01) {
5404 /* All the currently banned MACs happen to start with 01 currently, so
5405 * this is a quick way to eliminate most of the good ones. */
5407 if (eth_addr_is_reserved(dst)) {
5408 /* Drop STP, IEEE pause frames, and other reserved protocols
5409 * (01-80-c2-00-00-0x). */
5413 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
5415 if ((dst[3] & 0xfe) == 0xcc &&
5416 (dst[4] & 0xfe) == 0xcc &&
5417 (dst[5] & 0xfe) == 0xcc) {
5418 /* Drop the following protocols plus others following the same
5421 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
5422 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
5423 STP Uplink Fast (01-00-0c-cd-cd-cd) */
5427 if (!(dst[3] | dst[4] | dst[5])) {
5428 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
5437 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5439 struct ofproto_dpif *ofproto = ctx->ofproto;
5440 mirror_mask_t mirrors;
5441 struct ofbundle *in_bundle;
5444 const struct nlattr *a;
5447 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5448 ctx->packet != NULL);
5452 mirrors = in_bundle->src_mirrors;
5454 /* Drop frames on bundles reserved for mirroring. */
5455 if (in_bundle->mirror_out) {
5456 if (ctx->packet != NULL) {
5457 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5458 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5459 "%s, which is reserved exclusively for mirroring",
5460 ctx->ofproto->up.name, in_bundle->name);
5466 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5467 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5470 vlan = input_vid_to_vlan(in_bundle, vid);
5472 /* Look at the output ports to check for destination selections. */
5474 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5475 ctx->odp_actions->size) {
5476 enum ovs_action_attr type = nl_attr_type(a);
5477 struct ofport_dpif *ofport;
5479 if (type != OVS_ACTION_ATTR_OUTPUT) {
5483 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5484 if (ofport && ofport->bundle) {
5485 mirrors |= ofport->bundle->dst_mirrors;
5493 /* Restore the original packet before adding the mirror actions. */
5494 ctx->flow = *orig_flow;
5499 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5501 if (!vlan_is_mirrored(m, vlan)) {
5502 mirrors &= mirrors - 1;
5506 mirrors &= ~m->dup_mirrors;
5507 ctx->mirrors |= m->dup_mirrors;
5509 output_normal(ctx, m->out, vlan);
5510 } else if (eth_dst_may_rspan(orig_flow->dl_dst)
5511 && vlan != m->out_vlan) {
5512 struct ofbundle *bundle;
5514 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5515 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5516 && !bundle->mirror_out) {
5517 output_normal(ctx, bundle, m->out_vlan);
5525 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5526 uint64_t packets, uint64_t bytes)
5532 for (; mirrors; mirrors &= mirrors - 1) {
5535 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5538 /* In normal circumstances 'm' will not be NULL. However,
5539 * if mirrors are reconfigured, we can temporarily get out
5540 * of sync in facet_revalidate(). We could "correct" the
5541 * mirror list before reaching here, but doing that would
5542 * not properly account the traffic stats we've currently
5543 * accumulated for previous mirror configuration. */
5547 m->packet_count += packets;
5548 m->byte_count += bytes;
5552 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
5553 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
5554 * indicate this; newer upstream kernels use gratuitous ARP requests. */
5556 is_gratuitous_arp(const struct flow *flow)
5558 return (flow->dl_type == htons(ETH_TYPE_ARP)
5559 && eth_addr_is_broadcast(flow->dl_dst)
5560 && (flow->nw_proto == ARP_OP_REPLY
5561 || (flow->nw_proto == ARP_OP_REQUEST
5562 && flow->nw_src == flow->nw_dst)));
5566 update_learning_table(struct ofproto_dpif *ofproto,
5567 const struct flow *flow, int vlan,
5568 struct ofbundle *in_bundle)
5570 struct mac_entry *mac;
5572 /* Don't learn the OFPP_NONE port. */
5573 if (in_bundle == &ofpp_none_bundle) {
5577 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
5581 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
5582 if (is_gratuitous_arp(flow)) {
5583 /* We don't want to learn from gratuitous ARP packets that are
5584 * reflected back over bond slaves so we lock the learning table. */
5585 if (!in_bundle->bond) {
5586 mac_entry_set_grat_arp_lock(mac);
5587 } else if (mac_entry_is_grat_arp_locked(mac)) {
5592 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
5593 /* The log messages here could actually be useful in debugging,
5594 * so keep the rate limit relatively high. */
5595 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5596 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
5597 "on port %s in VLAN %d",
5598 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
5599 in_bundle->name, vlan);
5601 mac->port.p = in_bundle;
5602 tag_set_add(&ofproto->revalidate_set,
5603 mac_learning_changed(ofproto->ml, mac));
5607 static struct ofbundle *
5608 lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn)
5610 struct ofport_dpif *ofport;
5612 /* Special-case OFPP_NONE, which a controller may use as the ingress
5613 * port for traffic that it is sourcing. */
5614 if (in_port == OFPP_NONE) {
5615 return &ofpp_none_bundle;
5618 /* Find the port and bundle for the received packet. */
5619 ofport = get_ofp_port(ofproto, in_port);
5620 if (ofport && ofport->bundle) {
5621 return ofport->bundle;
5624 /* Odd. A few possible reasons here:
5626 * - We deleted a port but there are still a few packets queued up
5629 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
5630 * we don't know about.
5632 * - The ofproto client didn't configure the port as part of a bundle.
5635 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5637 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
5638 "port %"PRIu16, ofproto->up.name, in_port);
5643 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
5644 * dropped. Returns true if they may be forwarded, false if they should be
5647 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
5648 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
5650 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
5651 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
5652 * checked by input_vid_is_valid().
5654 * May also add tags to '*tags', although the current implementation only does
5655 * so in one special case.
5658 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
5659 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
5661 struct ofbundle *in_bundle = in_port->bundle;
5663 /* Drop frames for reserved multicast addresses
5664 * only if forward_bpdu option is absent. */
5665 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
5669 if (in_bundle->bond) {
5670 struct mac_entry *mac;
5672 switch (bond_check_admissibility(in_bundle->bond, in_port,
5673 flow->dl_dst, tags)) {
5680 case BV_DROP_IF_MOVED:
5681 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
5682 if (mac && mac->port.p != in_bundle &&
5683 (!is_gratuitous_arp(flow)
5684 || mac_entry_is_grat_arp_locked(mac))) {
5695 xlate_normal(struct action_xlate_ctx *ctx)
5697 struct ofport_dpif *in_port;
5698 struct ofbundle *in_bundle;
5699 struct mac_entry *mac;
5703 ctx->has_normal = true;
5705 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
5706 ctx->packet != NULL);
5711 /* We know 'in_port' exists unless it is "ofpp_none_bundle",
5712 * since lookup_input_bundle() succeeded. */
5713 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5715 /* Drop malformed frames. */
5716 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
5717 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
5718 if (ctx->packet != NULL) {
5719 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5720 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
5721 "VLAN tag received on port %s",
5722 ctx->ofproto->up.name, in_bundle->name);
5727 /* Drop frames on bundles reserved for mirroring. */
5728 if (in_bundle->mirror_out) {
5729 if (ctx->packet != NULL) {
5730 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5731 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5732 "%s, which is reserved exclusively for mirroring",
5733 ctx->ofproto->up.name, in_bundle->name);
5739 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
5740 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5743 vlan = input_vid_to_vlan(in_bundle, vid);
5745 /* Check other admissibility requirements. */
5747 !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
5751 /* Learn source MAC. */
5752 if (ctx->may_learn) {
5753 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
5756 /* Determine output bundle. */
5757 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
5760 if (mac->port.p != in_bundle) {
5761 output_normal(ctx, mac->port.p, vlan);
5764 struct ofbundle *bundle;
5766 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
5767 if (bundle != in_bundle
5768 && ofbundle_includes_vlan(bundle, vlan)
5769 && bundle->floodable
5770 && !bundle->mirror_out) {
5771 output_normal(ctx, bundle, vlan);
5774 ctx->nf_output_iface = NF_OUT_FLOOD;
5778 /* Optimized flow revalidation.
5780 * It's a difficult problem, in general, to tell which facets need to have
5781 * their actions recalculated whenever the OpenFlow flow table changes. We
5782 * don't try to solve that general problem: for most kinds of OpenFlow flow
5783 * table changes, we recalculate the actions for every facet. This is
5784 * relatively expensive, but it's good enough if the OpenFlow flow table
5785 * doesn't change very often.
5787 * However, we can expect one particular kind of OpenFlow flow table change to
5788 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5789 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5790 * table, we add a special case that applies to flow tables in which every rule
5791 * has the same form (that is, the same wildcards), except that the table is
5792 * also allowed to have a single "catch-all" flow that matches all packets. We
5793 * optimize this case by tagging all of the facets that resubmit into the table
5794 * and invalidating the same tag whenever a flow changes in that table. The
5795 * end result is that we revalidate just the facets that need it (and sometimes
5796 * a few more, but not all of the facets or even all of the facets that
5797 * resubmit to the table modified by MAC learning). */
5799 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
5800 * into an OpenFlow table with the given 'basis'. */
5802 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
5805 if (flow_wildcards_is_catchall(wc)) {
5808 struct flow tag_flow = *flow;
5809 flow_zero_wildcards(&tag_flow, wc);
5810 return tag_create_deterministic(flow_hash(&tag_flow, secret));
5814 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5815 * taggability of that table.
5817 * This function must be called after *each* change to a flow table. If you
5818 * skip calling it on some changes then the pointer comparisons at the end can
5819 * be invalid if you get unlucky. For example, if a flow removal causes a
5820 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5821 * different wildcards to be created with the same address, then this function
5822 * will incorrectly skip revalidation. */
5824 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5826 struct table_dpif *table = &ofproto->tables[table_id];
5827 const struct oftable *oftable = &ofproto->up.tables[table_id];
5828 struct cls_table *catchall, *other;
5829 struct cls_table *t;
5831 catchall = other = NULL;
5833 switch (hmap_count(&oftable->cls.tables)) {
5835 /* We could tag this OpenFlow table but it would make the logic a
5836 * little harder and it's a corner case that doesn't seem worth it
5842 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
5843 if (cls_table_is_catchall(t)) {
5845 } else if (!other) {
5848 /* Indicate that we can't tag this by setting both tables to
5849 * NULL. (We know that 'catchall' is already NULL.) */
5856 /* Can't tag this table. */
5860 if (table->catchall_table != catchall || table->other_table != other) {
5861 table->catchall_table = catchall;
5862 table->other_table = other;
5863 ofproto->need_revalidate = true;
5867 /* Given 'rule' that has changed in some way (either it is a rule being
5868 * inserted, a rule being deleted, or a rule whose actions are being
5869 * modified), marks facets for revalidation to ensure that packets will be
5870 * forwarded correctly according to the new state of the flow table.
5872 * This function must be called after *each* change to a flow table. See
5873 * the comment on table_update_taggable() for more information. */
5875 rule_invalidate(const struct rule_dpif *rule)
5877 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5879 table_update_taggable(ofproto, rule->up.table_id);
5881 if (!ofproto->need_revalidate) {
5882 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5884 if (table->other_table && rule->tag) {
5885 tag_set_add(&ofproto->revalidate_set, rule->tag);
5887 ofproto->need_revalidate = true;
5893 set_frag_handling(struct ofproto *ofproto_,
5894 enum ofp_config_flags frag_handling)
5896 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5898 if (frag_handling != OFPC_FRAG_REASM) {
5899 ofproto->need_revalidate = true;
5907 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5908 const struct flow *flow,
5909 const union ofp_action *ofp_actions, size_t n_ofp_actions)
5911 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5914 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
5915 return OFPERR_NXBRC_BAD_IN_PORT;
5918 error = validate_actions(ofp_actions, n_ofp_actions, flow,
5919 ofproto->max_ports);
5921 struct odputil_keybuf keybuf;
5924 uint64_t odp_actions_stub[1024 / 8];
5925 struct ofpbuf odp_actions;
5926 struct ofproto_push push;
5928 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5929 odp_flow_key_from_flow(&key, flow);
5931 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, NULL,
5932 packet_get_tcp_flags(packet, flow), packet);
5934 /* Ensure that resubmits in 'ofp_actions' get accounted to their
5935 * matching rules. */
5937 push.bytes = packet->size;
5938 push.used = time_msec();
5939 push.ctx.resubmit_hook = push_resubmit;
5941 ofpbuf_use_stub(&odp_actions,
5942 odp_actions_stub, sizeof odp_actions_stub);
5943 xlate_actions(&push.ctx, ofp_actions, n_ofp_actions, &odp_actions);
5944 dpif_execute(ofproto->dpif, key.data, key.size,
5945 odp_actions.data, odp_actions.size, packet);
5946 ofpbuf_uninit(&odp_actions);
5954 set_netflow(struct ofproto *ofproto_,
5955 const struct netflow_options *netflow_options)
5957 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5959 if (netflow_options) {
5960 if (!ofproto->netflow) {
5961 ofproto->netflow = netflow_create();
5963 return netflow_set_options(ofproto->netflow, netflow_options);
5965 netflow_destroy(ofproto->netflow);
5966 ofproto->netflow = NULL;
5972 get_netflow_ids(const struct ofproto *ofproto_,
5973 uint8_t *engine_type, uint8_t *engine_id)
5975 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5977 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
5981 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5983 if (!facet_is_controller_flow(facet) &&
5984 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5985 struct subfacet *subfacet;
5986 struct ofexpired expired;
5988 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5989 if (subfacet->installed) {
5990 struct dpif_flow_stats stats;
5992 subfacet_install(subfacet, subfacet->actions,
5993 subfacet->actions_len, &stats);
5994 subfacet_update_stats(subfacet, &stats);
5998 expired.flow = facet->flow;
5999 expired.packet_count = facet->packet_count;
6000 expired.byte_count = facet->byte_count;
6001 expired.used = facet->used;
6002 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
6007 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
6009 struct facet *facet;
6011 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6012 send_active_timeout(ofproto, facet);
6016 static struct ofproto_dpif *
6017 ofproto_dpif_lookup(const char *name)
6019 struct ofproto_dpif *ofproto;
6021 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
6022 hash_string(name, 0), &all_ofproto_dpifs) {
6023 if (!strcmp(ofproto->up.name, name)) {
6031 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
6032 const char *argv[], void *aux OVS_UNUSED)
6034 struct ofproto_dpif *ofproto;
6037 ofproto = ofproto_dpif_lookup(argv[1]);
6039 unixctl_command_reply_error(conn, "no such bridge");
6042 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6044 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6045 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6049 unixctl_command_reply(conn, "table successfully flushed");
6053 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6054 const char *argv[], void *aux OVS_UNUSED)
6056 struct ds ds = DS_EMPTY_INITIALIZER;
6057 const struct ofproto_dpif *ofproto;
6058 const struct mac_entry *e;
6060 ofproto = ofproto_dpif_lookup(argv[1]);
6062 unixctl_command_reply_error(conn, "no such bridge");
6066 ds_put_cstr(&ds, " port VLAN MAC Age\n");
6067 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
6068 struct ofbundle *bundle = e->port.p;
6069 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
6070 ofbundle_get_a_port(bundle)->odp_port,
6071 e->vlan, ETH_ADDR_ARGS(e->mac),
6072 mac_entry_age(ofproto->ml, e));
6074 unixctl_command_reply(conn, ds_cstr(&ds));
6079 struct action_xlate_ctx ctx;
6085 trace_format_rule(struct ds *result, uint8_t table_id, int level,
6086 const struct rule_dpif *rule)
6088 ds_put_char_multiple(result, '\t', level);
6090 ds_put_cstr(result, "No match\n");
6094 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
6095 table_id, ntohll(rule->up.flow_cookie));
6096 cls_rule_format(&rule->up.cr, result);
6097 ds_put_char(result, '\n');
6099 ds_put_char_multiple(result, '\t', level);
6100 ds_put_cstr(result, "OpenFlow ");
6101 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
6102 ds_put_char(result, '\n');
6106 trace_format_flow(struct ds *result, int level, const char *title,
6107 struct trace_ctx *trace)
6109 ds_put_char_multiple(result, '\t', level);
6110 ds_put_format(result, "%s: ", title);
6111 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
6112 ds_put_cstr(result, "unchanged");
6114 flow_format(result, &trace->ctx.flow);
6115 trace->flow = trace->ctx.flow;
6117 ds_put_char(result, '\n');
6121 trace_format_regs(struct ds *result, int level, const char *title,
6122 struct trace_ctx *trace)
6126 ds_put_char_multiple(result, '\t', level);
6127 ds_put_format(result, "%s:", title);
6128 for (i = 0; i < FLOW_N_REGS; i++) {
6129 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
6131 ds_put_char(result, '\n');
6135 trace_format_odp(struct ds *result, int level, const char *title,
6136 struct trace_ctx *trace)
6138 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
6140 ds_put_char_multiple(result, '\t', level);
6141 ds_put_format(result, "%s: ", title);
6142 format_odp_actions(result, odp_actions->data, odp_actions->size);
6143 ds_put_char(result, '\n');
6147 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
6149 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
6150 struct ds *result = trace->result;
6152 ds_put_char(result, '\n');
6153 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
6154 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
6155 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
6156 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
6160 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
6161 void *aux OVS_UNUSED)
6163 const char *dpname = argv[1];
6164 struct ofproto_dpif *ofproto;
6165 struct ofpbuf odp_key;
6166 struct ofpbuf *packet;
6167 ovs_be16 initial_tci;
6173 ofpbuf_init(&odp_key, 0);
6176 ofproto = ofproto_dpif_lookup(dpname);
6178 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
6182 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
6183 /* ofproto/trace dpname flow [-generate] */
6184 const char *flow_s = argv[2];
6185 const char *generate_s = argv[3];
6188 /* Convert string to datapath key. */
6189 ofpbuf_init(&odp_key, 0);
6190 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
6192 unixctl_command_reply_error(conn, "Bad flow syntax");
6196 /* Convert odp_key to flow. */
6197 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
6198 odp_key.size, &flow,
6199 &initial_tci, NULL);
6200 if (error == ODP_FIT_ERROR) {
6201 unixctl_command_reply_error(conn, "Invalid flow");
6205 /* Generate a packet, if requested. */
6207 packet = ofpbuf_new(0);
6208 flow_compose(packet, &flow);
6210 } else if (argc == 6) {
6211 /* ofproto/trace dpname priority tun_id in_port packet */
6212 const char *priority_s = argv[2];
6213 const char *tun_id_s = argv[3];
6214 const char *in_port_s = argv[4];
6215 const char *packet_s = argv[5];
6216 uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s));
6217 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
6218 uint32_t priority = atoi(priority_s);
6221 msg = eth_from_hex(packet_s, &packet);
6223 unixctl_command_reply_error(conn, msg);
6227 ds_put_cstr(&result, "Packet: ");
6228 s = ofp_packet_to_string(packet->data, packet->size);
6229 ds_put_cstr(&result, s);
6232 flow_extract(packet, priority, tun_id, in_port, &flow);
6233 initial_tci = flow.vlan_tci;
6235 unixctl_command_reply_error(conn, "Bad command syntax");
6239 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
6240 unixctl_command_reply(conn, ds_cstr(&result));
6243 ds_destroy(&result);
6244 ofpbuf_delete(packet);
6245 ofpbuf_uninit(&odp_key);
6249 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
6250 const struct ofpbuf *packet, ovs_be16 initial_tci,
6253 struct rule_dpif *rule;
6255 ds_put_cstr(ds, "Flow: ");
6256 flow_format(ds, flow);
6257 ds_put_char(ds, '\n');
6259 rule = rule_dpif_lookup(ofproto, flow, 0);
6260 trace_format_rule(ds, 0, 0, rule);
6262 uint64_t odp_actions_stub[1024 / 8];
6263 struct ofpbuf odp_actions;
6265 struct trace_ctx trace;
6268 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
6271 ofpbuf_use_stub(&odp_actions,
6272 odp_actions_stub, sizeof odp_actions_stub);
6273 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
6274 rule, tcp_flags, packet);
6275 trace.ctx.resubmit_hook = trace_resubmit;
6276 xlate_actions(&trace.ctx, rule->up.actions, rule->up.n_actions,
6279 ds_put_char(ds, '\n');
6280 trace_format_flow(ds, 0, "Final flow", &trace);
6281 ds_put_cstr(ds, "Datapath actions: ");
6282 format_odp_actions(ds, odp_actions.data, odp_actions.size);
6283 ofpbuf_uninit(&odp_actions);
6285 if (!trace.ctx.may_set_up_flow) {
6287 ds_put_cstr(ds, "\nThis flow is not cachable.");
6289 ds_put_cstr(ds, "\nThe datapath actions are incomplete--"
6290 "for complete actions, please supply a packet.");
6297 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6298 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6301 unixctl_command_reply(conn, NULL);
6305 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6306 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6309 unixctl_command_reply(conn, NULL);
6312 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
6313 * 'reply' describing the results. */
6315 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
6317 struct facet *facet;
6321 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6322 if (!facet_check_consistency(facet)) {
6327 ofproto->need_revalidate = true;
6331 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
6332 ofproto->up.name, errors);
6334 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
6339 ofproto_dpif_self_check(struct unixctl_conn *conn,
6340 int argc, const char *argv[], void *aux OVS_UNUSED)
6342 struct ds reply = DS_EMPTY_INITIALIZER;
6343 struct ofproto_dpif *ofproto;
6346 ofproto = ofproto_dpif_lookup(argv[1]);
6348 unixctl_command_reply_error(conn, "Unknown ofproto (use "
6349 "ofproto/list for help)");
6352 ofproto_dpif_self_check__(ofproto, &reply);
6354 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6355 ofproto_dpif_self_check__(ofproto, &reply);
6359 unixctl_command_reply(conn, ds_cstr(&reply));
6364 ofproto_dpif_unixctl_init(void)
6366 static bool registered;
6372 unixctl_command_register(
6374 "bridge {tun_id in_port packet | odp_flow [-generate]}",
6375 2, 5, ofproto_unixctl_trace, NULL);
6376 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6377 ofproto_unixctl_fdb_flush, NULL);
6378 unixctl_command_register("fdb/show", "bridge", 1, 1,
6379 ofproto_unixctl_fdb_show, NULL);
6380 unixctl_command_register("ofproto/clog", "", 0, 0,
6381 ofproto_dpif_clog, NULL);
6382 unixctl_command_register("ofproto/unclog", "", 0, 0,
6383 ofproto_dpif_unclog, NULL);
6384 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6385 ofproto_dpif_self_check, NULL);
6388 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6390 * This is deprecated. It is only for compatibility with broken device drivers
6391 * in old versions of Linux that do not properly support VLANs when VLAN
6392 * devices are not used. When broken device drivers are no longer in
6393 * widespread use, we will delete these interfaces. */
6396 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
6398 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6399 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6401 if (realdev_ofp_port == ofport->realdev_ofp_port
6402 && vid == ofport->vlandev_vid) {
6406 ofproto->need_revalidate = true;
6408 if (ofport->realdev_ofp_port) {
6411 if (realdev_ofp_port && ofport->bundle) {
6412 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6413 * themselves be part of a bundle. */
6414 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6417 ofport->realdev_ofp_port = realdev_ofp_port;
6418 ofport->vlandev_vid = vid;
6420 if (realdev_ofp_port) {
6421 vsp_add(ofport, realdev_ofp_port, vid);
6428 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
6430 return hash_2words(realdev_ofp_port, vid);
6433 /* Returns the ODP port number of the Linux VLAN device that corresponds to
6434 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
6435 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
6436 * it would return the port number of eth0.9.
6438 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
6439 * function just returns its 'realdev_odp_port' argument. */
6441 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6442 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
6444 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6445 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
6446 int vid = vlan_tci_to_vid(vlan_tci);
6447 const struct vlan_splinter *vsp;
6449 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6450 hash_realdev_vid(realdev_ofp_port, vid),
6451 &ofproto->realdev_vid_map) {
6452 if (vsp->realdev_ofp_port == realdev_ofp_port
6453 && vsp->vid == vid) {
6454 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
6458 return realdev_odp_port;
6461 static struct vlan_splinter *
6462 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
6464 struct vlan_splinter *vsp;
6466 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
6467 &ofproto->vlandev_map) {
6468 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6476 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6477 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6478 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6479 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6480 * eth0 and store 9 in '*vid'.
6482 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6483 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6486 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6487 uint16_t vlandev_ofp_port, int *vid)
6489 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6490 const struct vlan_splinter *vsp;
6492 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6497 return vsp->realdev_ofp_port;
6504 vsp_remove(struct ofport_dpif *port)
6506 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6507 struct vlan_splinter *vsp;
6509 vsp = vlandev_find(ofproto, port->up.ofp_port);
6511 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6512 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6515 port->realdev_ofp_port = 0;
6517 VLOG_ERR("missing vlan device record");
6522 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
6524 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6526 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6527 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6528 == realdev_ofp_port)) {
6529 struct vlan_splinter *vsp;
6531 vsp = xmalloc(sizeof *vsp);
6532 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6533 hash_int(port->up.ofp_port, 0));
6534 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6535 hash_realdev_vid(realdev_ofp_port, vid));
6536 vsp->realdev_ofp_port = realdev_ofp_port;
6537 vsp->vlandev_ofp_port = port->up.ofp_port;
6540 port->realdev_ofp_port = realdev_ofp_port;
6542 VLOG_ERR("duplicate vlan device record");
6546 const struct ofproto_class ofproto_dpif_class = {
6575 port_is_lacp_current,
6576 NULL, /* rule_choose_table */
6583 rule_modify_actions,
6591 get_cfm_remote_mpids,
6596 get_stp_port_status,
6603 is_mirror_output_bundle,
6604 forward_bpdu_changed,