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 "multipath.h"
44 #include "ofp-print.h"
45 #include "ofproto-dpif-sflow.h"
46 #include "poll-loop.h"
48 #include "unaligned.h"
50 #include "vlan-bitmap.h"
53 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
55 COVERAGE_DEFINE(ofproto_dpif_ctlr_action);
56 COVERAGE_DEFINE(ofproto_dpif_expired);
57 COVERAGE_DEFINE(ofproto_dpif_no_packet_in);
58 COVERAGE_DEFINE(ofproto_dpif_xlate);
59 COVERAGE_DEFINE(facet_changed_rule);
60 COVERAGE_DEFINE(facet_invalidated);
61 COVERAGE_DEFINE(facet_revalidate);
62 COVERAGE_DEFINE(facet_unexpected);
64 /* Maximum depth of flow table recursion (due to resubmit actions) in a
65 * flow translation. */
66 #define MAX_RESUBMIT_RECURSION 32
68 /* Number of implemented OpenFlow tables. */
69 enum { N_TABLES = 255 };
70 BUILD_ASSERT_DECL(N_TABLES >= 1 && N_TABLES <= 255);
78 long long int used; /* Time last used; time created if not used. */
82 * - Do include packets and bytes from facets that have been deleted or
83 * whose own statistics have been folded into the rule.
85 * - Do include packets and bytes sent "by hand" that were accounted to
86 * the rule without any facet being involved (this is a rare corner
87 * case in rule_execute()).
89 * - Do not include packet or bytes that can be obtained from any facet's
90 * packet_count or byte_count member or that can be obtained from the
91 * datapath by, e.g., dpif_flow_get() for any subfacet.
93 uint64_t packet_count; /* Number of packets received. */
94 uint64_t byte_count; /* Number of bytes received. */
96 tag_type tag; /* Caches rule_calculate_tag() result. */
98 struct list facets; /* List of "struct facet"s. */
101 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
103 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
106 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
107 const struct flow *, uint8_t table);
109 static void flow_push_stats(const struct rule_dpif *, const struct flow *,
110 uint64_t packets, uint64_t bytes,
113 static uint32_t rule_calculate_tag(const struct flow *,
114 const struct flow_wildcards *,
116 static void rule_invalidate(const struct rule_dpif *);
118 #define MAX_MIRRORS 32
119 typedef uint32_t mirror_mask_t;
120 #define MIRROR_MASK_C(X) UINT32_C(X)
121 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
123 struct ofproto_dpif *ofproto; /* Owning ofproto. */
124 size_t idx; /* In ofproto's "mirrors" array. */
125 void *aux; /* Key supplied by ofproto's client. */
126 char *name; /* Identifier for log messages. */
128 /* Selection criteria. */
129 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
130 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
131 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
133 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
134 struct ofbundle *out; /* Output port or NULL. */
135 int out_vlan; /* Output VLAN or -1. */
136 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
139 int64_t packet_count; /* Number of packets sent. */
140 int64_t byte_count; /* Number of bytes sent. */
143 static void mirror_destroy(struct ofmirror *);
144 static void update_mirror_stats(struct ofproto_dpif *ofproto,
145 mirror_mask_t mirrors,
146 uint64_t packets, uint64_t bytes);
149 struct ofproto_dpif *ofproto; /* Owning ofproto. */
150 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
151 void *aux; /* Key supplied by ofproto's client. */
152 char *name; /* Identifier for log messages. */
155 struct list ports; /* Contains "struct ofport"s. */
156 enum port_vlan_mode vlan_mode; /* VLAN mode */
157 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
158 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
159 * NULL if all VLANs are trunked. */
160 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
161 struct bond *bond; /* Nonnull iff more than one port. */
162 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
165 bool floodable; /* True if no port has OFPPC_NO_FLOOD set. */
167 /* Port mirroring info. */
168 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
169 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
170 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
173 static void bundle_remove(struct ofport *);
174 static void bundle_update(struct ofbundle *);
175 static void bundle_destroy(struct ofbundle *);
176 static void bundle_del_port(struct ofport_dpif *);
177 static void bundle_run(struct ofbundle *);
178 static void bundle_wait(struct ofbundle *);
179 static struct ofbundle *lookup_input_bundle(struct ofproto_dpif *,
180 uint16_t in_port, bool warn);
182 /* A controller may use OFPP_NONE as the ingress port to indicate that
183 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
184 * when an input bundle is needed for validation (e.g., mirroring or
185 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
186 * any 'port' structs, so care must be taken when dealing with it. */
187 static struct ofbundle ofpp_none_bundle = {
189 .vlan_mode = PORT_VLAN_TRUNK
192 static void stp_run(struct ofproto_dpif *ofproto);
193 static void stp_wait(struct ofproto_dpif *ofproto);
195 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
197 struct action_xlate_ctx {
198 /* action_xlate_ctx_init() initializes these members. */
201 struct ofproto_dpif *ofproto;
203 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
204 * this flow when actions change header fields. */
207 /* The packet corresponding to 'flow', or a null pointer if we are
208 * revalidating without a packet to refer to. */
209 const struct ofpbuf *packet;
211 /* Should OFPP_NORMAL MAC learning and NXAST_LEARN actions execute? We
212 * want to execute them if we are actually processing a packet, or if we
213 * are accounting for packets that the datapath has processed, but not if
214 * we are just revalidating. */
217 /* Cookie of the currently matching rule, or 0. */
220 /* If nonnull, called just before executing a resubmit action.
222 * This is normally null so the client has to set it manually after
223 * calling action_xlate_ctx_init(). */
224 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *);
226 /* xlate_actions() initializes and uses these members. The client might want
227 * to look at them after it returns. */
229 struct ofpbuf *odp_actions; /* Datapath actions. */
230 tag_type tags; /* Tags associated with actions. */
231 bool may_set_up_flow; /* True ordinarily; false if the actions must
232 * be reassessed for every packet. */
233 bool has_learn; /* Actions include NXAST_LEARN? */
234 bool has_normal; /* Actions output to OFPP_NORMAL? */
235 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
236 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
238 /* xlate_actions() initializes and uses these members, but the client has no
239 * reason to look at them. */
241 int recurse; /* Recursion level, via xlate_table_action. */
242 struct flow base_flow; /* Flow at the last commit. */
243 uint32_t orig_skb_priority; /* Priority when packet arrived. */
244 uint8_t table_id; /* OpenFlow table ID where flow was found. */
245 uint32_t sflow_n_outputs; /* Number of output ports. */
246 uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
247 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
248 bool exit; /* No further actions should be processed. */
251 static void action_xlate_ctx_init(struct action_xlate_ctx *,
252 struct ofproto_dpif *, const struct flow *,
253 ovs_be16 initial_tci, ovs_be64 cookie,
254 const struct ofpbuf *);
255 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
256 const union ofp_action *in, size_t n_in);
258 /* An exact-match instantiation of an OpenFlow flow.
260 * A facet associates a "struct flow", which represents the Open vSwitch
261 * userspace idea of an exact-match flow, with one or more subfacets. Each
262 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
263 * the facet. When the kernel module (or other dpif implementation) and Open
264 * vSwitch userspace agree on the definition of a flow key, there is exactly
265 * one subfacet per facet. If the dpif implementation supports more-specific
266 * flow matching than userspace, however, a facet can have more than one
267 * subfacet, each of which corresponds to some distinction in flow that
268 * userspace simply doesn't understand.
270 * Flow expiration works in terms of subfacets, so a facet must have at least
271 * one subfacet or it will never expire, leaking memory. */
274 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
275 struct list list_node; /* In owning rule's 'facets' list. */
276 struct rule_dpif *rule; /* Owning rule. */
279 struct list subfacets;
280 long long int used; /* Time last used; time created if not used. */
287 * - Do include packets and bytes sent "by hand", e.g. with
290 * - Do include packets and bytes that were obtained from the datapath
291 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
292 * DPIF_FP_ZERO_STATS).
294 * - Do not include packets or bytes that can be obtained from the
295 * datapath for any existing subfacet.
297 uint64_t packet_count; /* Number of packets received. */
298 uint64_t byte_count; /* Number of bytes received. */
300 /* Resubmit statistics. */
301 uint64_t prev_packet_count; /* Number of packets from last stats push. */
302 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
303 long long int prev_used; /* Used time from last stats push. */
306 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
307 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
309 /* Properties of datapath actions.
311 * Every subfacet has its own actions because actions can differ slightly
312 * between splintered and non-splintered subfacets due to the VLAN tag
313 * being initially different (present vs. absent). All of them have these
314 * properties in common so we just store one copy of them here. */
315 bool may_install; /* Reassess actions for every packet? */
316 bool has_learn; /* Actions include NXAST_LEARN? */
317 bool has_normal; /* Actions output to OFPP_NORMAL? */
318 tag_type tags; /* Tags that would require revalidation. */
319 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
322 static struct facet *facet_create(struct rule_dpif *, const struct flow *);
323 static void facet_remove(struct ofproto_dpif *, struct facet *);
324 static void facet_free(struct facet *);
326 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
327 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
328 const struct flow *);
329 static bool facet_revalidate(struct ofproto_dpif *, struct facet *);
331 static void facet_flush_stats(struct ofproto_dpif *, struct facet *);
333 static void facet_update_time(struct ofproto_dpif *, struct facet *,
335 static void facet_reset_counters(struct facet *);
336 static void facet_push_stats(struct facet *);
337 static void facet_account(struct ofproto_dpif *, struct facet *);
339 static bool facet_is_controller_flow(struct facet *);
341 /* A dpif flow and actions associated with a facet.
343 * See also the large comment on struct facet. */
346 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
347 struct list list_node; /* In struct facet's 'facets' list. */
348 struct facet *facet; /* Owning facet. */
352 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
353 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
354 * regenerate the ODP flow key from ->facet->flow. */
355 enum odp_key_fitness key_fitness;
359 long long int used; /* Time last used; time created if not used. */
361 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
362 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
366 * These should be essentially identical for every subfacet in a facet, but
367 * may differ in trivial ways due to VLAN splinters. */
368 size_t actions_len; /* Number of bytes in actions[]. */
369 struct nlattr *actions; /* Datapath actions. */
371 bool installed; /* Installed in datapath? */
373 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
374 * splinters can cause it to differ. This value should be removed when
375 * the VLAN splinters feature is no longer needed. */
376 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
379 static struct subfacet *subfacet_create(struct ofproto_dpif *, struct facet *,
380 enum odp_key_fitness,
381 const struct nlattr *key,
382 size_t key_len, ovs_be16 initial_tci);
383 static struct subfacet *subfacet_find(struct ofproto_dpif *,
384 const struct nlattr *key, size_t key_len);
385 static void subfacet_destroy(struct ofproto_dpif *, struct subfacet *);
386 static void subfacet_destroy__(struct ofproto_dpif *, struct subfacet *);
387 static void subfacet_reset_dp_stats(struct subfacet *,
388 struct dpif_flow_stats *);
389 static void subfacet_update_time(struct ofproto_dpif *, struct subfacet *,
391 static void subfacet_update_stats(struct ofproto_dpif *, struct subfacet *,
392 const struct dpif_flow_stats *);
393 static void subfacet_make_actions(struct ofproto_dpif *, struct subfacet *,
394 const struct ofpbuf *packet);
395 static int subfacet_install(struct ofproto_dpif *, struct subfacet *,
396 const struct nlattr *actions, size_t actions_len,
397 struct dpif_flow_stats *);
398 static void subfacet_uninstall(struct ofproto_dpif *, struct subfacet *);
404 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
405 struct list bundle_node; /* In struct ofbundle's "ports" list. */
406 struct cfm *cfm; /* Connectivity Fault Management, if any. */
407 tag_type tag; /* Tag associated with this port. */
408 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
409 bool may_enable; /* May be enabled in bonds. */
412 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
413 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
414 long long int stp_state_entered;
416 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
418 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
420 * This is deprecated. It is only for compatibility with broken device
421 * drivers in old versions of Linux that do not properly support VLANs when
422 * VLAN devices are not used. When broken device drivers are no longer in
423 * widespread use, we will delete these interfaces. */
424 uint16_t realdev_ofp_port;
428 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
429 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
430 * traffic egressing the 'ofport' with that priority should be marked with. */
431 struct priority_to_dscp {
432 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
433 uint32_t priority; /* Priority of this queue (see struct flow). */
435 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
438 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
440 * This is deprecated. It is only for compatibility with broken device drivers
441 * in old versions of Linux that do not properly support VLANs when VLAN
442 * devices are not used. When broken device drivers are no longer in
443 * widespread use, we will delete these interfaces. */
444 struct vlan_splinter {
445 struct hmap_node realdev_vid_node;
446 struct hmap_node vlandev_node;
447 uint16_t realdev_ofp_port;
448 uint16_t vlandev_ofp_port;
452 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
453 uint32_t realdev, ovs_be16 vlan_tci);
454 static uint16_t vsp_vlandev_to_realdev(const struct ofproto_dpif *,
455 uint16_t vlandev, int *vid);
456 static void vsp_remove(struct ofport_dpif *);
457 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
459 static struct ofport_dpif *
460 ofport_dpif_cast(const struct ofport *ofport)
462 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
463 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
466 static void port_run(struct ofport_dpif *);
467 static void port_wait(struct ofport_dpif *);
468 static int set_cfm(struct ofport *, const struct cfm_settings *);
469 static void ofport_clear_priorities(struct ofport_dpif *);
471 struct dpif_completion {
472 struct list list_node;
473 struct ofoperation *op;
476 /* Extra information about a classifier table.
477 * Currently used just for optimized flow revalidation. */
479 /* If either of these is nonnull, then this table has a form that allows
480 * flows to be tagged to avoid revalidating most flows for the most common
481 * kinds of flow table changes. */
482 struct cls_table *catchall_table; /* Table that wildcards all fields. */
483 struct cls_table *other_table; /* Table with any other wildcard set. */
484 uint32_t basis; /* Keeps each table's tags separate. */
487 struct ofproto_dpif {
488 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
497 struct netflow *netflow;
498 struct dpif_sflow *sflow;
499 struct hmap bundles; /* Contains "struct ofbundle"s. */
500 struct mac_learning *ml;
501 struct ofmirror *mirrors[MAX_MIRRORS];
502 bool has_bonded_bundles;
505 struct timer next_expiration;
509 struct hmap subfacets;
512 struct table_dpif tables[N_TABLES];
513 bool need_revalidate;
514 struct tag_set revalidate_set;
516 /* Support for debugging async flow mods. */
517 struct list completions;
519 bool has_bundle_action; /* True when the first bundle action appears. */
520 struct netdev_stats stats; /* To account packets generated and consumed in
525 long long int stp_last_tick;
527 /* VLAN splinters. */
528 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
529 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
532 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
533 * for debugging the asynchronous flow_mod implementation.) */
536 /* All existing ofproto_dpif instances, indexed by ->up.name. */
537 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
539 static void ofproto_dpif_unixctl_init(void);
541 static struct ofproto_dpif *
542 ofproto_dpif_cast(const struct ofproto *ofproto)
544 assert(ofproto->ofproto_class == &ofproto_dpif_class);
545 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
548 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
550 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
553 /* Packet processing. */
554 static void update_learning_table(struct ofproto_dpif *,
555 const struct flow *, int vlan,
558 #define FLOW_MISS_MAX_BATCH 50
559 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
561 /* Flow expiration. */
562 static int expire(struct ofproto_dpif *);
565 static void send_netflow_active_timeouts(struct ofproto_dpif *);
568 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
570 compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions,
571 const struct flow *, uint32_t odp_port);
572 static void add_mirror_actions(struct action_xlate_ctx *ctx,
573 const struct flow *flow);
574 /* Global variables. */
575 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
577 /* Factory functions. */
580 enumerate_types(struct sset *types)
582 dp_enumerate_types(types);
586 enumerate_names(const char *type, struct sset *names)
588 return dp_enumerate_names(type, names);
592 del(const char *type, const char *name)
597 error = dpif_open(name, type, &dpif);
599 error = dpif_delete(dpif);
605 /* Basic life-cycle. */
607 static struct ofproto *
610 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
615 dealloc(struct ofproto *ofproto_)
617 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
622 construct(struct ofproto *ofproto_, int *n_tablesp)
624 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
625 const char *name = ofproto->up.name;
629 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
631 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
635 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
636 ofproto->n_matches = 0;
638 dpif_flow_flush(ofproto->dpif);
639 dpif_recv_purge(ofproto->dpif);
641 error = dpif_recv_set(ofproto->dpif, true);
643 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
644 dpif_close(ofproto->dpif);
648 ofproto->netflow = NULL;
649 ofproto->sflow = NULL;
651 hmap_init(&ofproto->bundles);
652 ofproto->ml = mac_learning_create();
653 for (i = 0; i < MAX_MIRRORS; i++) {
654 ofproto->mirrors[i] = NULL;
656 ofproto->has_bonded_bundles = false;
658 timer_set_duration(&ofproto->next_expiration, 1000);
660 hmap_init(&ofproto->facets);
661 hmap_init(&ofproto->subfacets);
663 for (i = 0; i < N_TABLES; i++) {
664 struct table_dpif *table = &ofproto->tables[i];
666 table->catchall_table = NULL;
667 table->other_table = NULL;
668 table->basis = random_uint32();
670 ofproto->need_revalidate = false;
671 tag_set_init(&ofproto->revalidate_set);
673 list_init(&ofproto->completions);
675 ofproto_dpif_unixctl_init();
677 ofproto->has_bundle_action = false;
679 hmap_init(&ofproto->vlandev_map);
680 hmap_init(&ofproto->realdev_vid_map);
682 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
683 hash_string(ofproto->up.name, 0));
685 *n_tablesp = N_TABLES;
686 memset(&ofproto->stats, 0, sizeof ofproto->stats);
691 complete_operations(struct ofproto_dpif *ofproto)
693 struct dpif_completion *c, *next;
695 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
696 ofoperation_complete(c->op, 0);
697 list_remove(&c->list_node);
703 destruct(struct ofproto *ofproto_)
705 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
706 struct rule_dpif *rule, *next_rule;
707 struct classifier *table;
710 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
711 complete_operations(ofproto);
713 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
714 struct cls_cursor cursor;
716 cls_cursor_init(&cursor, table, NULL);
717 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
718 ofproto_rule_destroy(&rule->up);
722 for (i = 0; i < MAX_MIRRORS; i++) {
723 mirror_destroy(ofproto->mirrors[i]);
726 netflow_destroy(ofproto->netflow);
727 dpif_sflow_destroy(ofproto->sflow);
728 hmap_destroy(&ofproto->bundles);
729 mac_learning_destroy(ofproto->ml);
731 hmap_destroy(&ofproto->facets);
732 hmap_destroy(&ofproto->subfacets);
734 hmap_destroy(&ofproto->vlandev_map);
735 hmap_destroy(&ofproto->realdev_vid_map);
737 dpif_close(ofproto->dpif);
741 run_fast(struct ofproto *ofproto_)
743 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
746 /* Handle one or more batches of upcalls, until there's nothing left to do
747 * or until we do a fixed total amount of work.
749 * We do work in batches because it can be much cheaper to set up a number
750 * of flows and fire off their patches all at once. We do multiple batches
751 * because in some cases handling a packet can cause another packet to be
752 * queued almost immediately as part of the return flow. Both
753 * optimizations can make major improvements on some benchmarks and
754 * presumably for real traffic as well. */
756 while (work < FLOW_MISS_MAX_BATCH) {
757 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
767 run(struct ofproto *ofproto_)
769 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
770 struct ofport_dpif *ofport;
771 struct ofbundle *bundle;
775 complete_operations(ofproto);
777 dpif_run(ofproto->dpif);
779 error = run_fast(ofproto_);
784 if (timer_expired(&ofproto->next_expiration)) {
785 int delay = expire(ofproto);
786 timer_set_duration(&ofproto->next_expiration, delay);
789 if (ofproto->netflow) {
790 if (netflow_run(ofproto->netflow)) {
791 send_netflow_active_timeouts(ofproto);
794 if (ofproto->sflow) {
795 dpif_sflow_run(ofproto->sflow);
798 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
801 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
806 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
808 /* Now revalidate if there's anything to do. */
809 if (ofproto->need_revalidate
810 || !tag_set_is_empty(&ofproto->revalidate_set)) {
811 struct tag_set revalidate_set = ofproto->revalidate_set;
812 bool revalidate_all = ofproto->need_revalidate;
813 struct facet *facet, *next;
815 /* Clear the revalidation flags. */
816 tag_set_init(&ofproto->revalidate_set);
817 ofproto->need_revalidate = false;
819 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
821 || tag_set_intersects(&revalidate_set, facet->tags)) {
822 facet_revalidate(ofproto, facet);
831 wait(struct ofproto *ofproto_)
833 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
834 struct ofport_dpif *ofport;
835 struct ofbundle *bundle;
837 if (!clogged && !list_is_empty(&ofproto->completions)) {
838 poll_immediate_wake();
841 dpif_wait(ofproto->dpif);
842 dpif_recv_wait(ofproto->dpif);
843 if (ofproto->sflow) {
844 dpif_sflow_wait(ofproto->sflow);
846 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
847 poll_immediate_wake();
849 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
852 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
855 if (ofproto->netflow) {
856 netflow_wait(ofproto->netflow);
858 mac_learning_wait(ofproto->ml);
860 if (ofproto->need_revalidate) {
861 /* Shouldn't happen, but if it does just go around again. */
862 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
863 poll_immediate_wake();
865 timer_wait(&ofproto->next_expiration);
870 flush(struct ofproto *ofproto_)
872 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
873 struct facet *facet, *next_facet;
875 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
876 /* Mark the facet as not installed so that facet_remove() doesn't
877 * bother trying to uninstall it. There is no point in uninstalling it
878 * individually since we are about to blow away all the facets with
879 * dpif_flow_flush(). */
880 struct subfacet *subfacet;
882 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
883 subfacet->installed = false;
884 subfacet->dp_packet_count = 0;
885 subfacet->dp_byte_count = 0;
887 facet_remove(ofproto, facet);
889 dpif_flow_flush(ofproto->dpif);
893 get_features(struct ofproto *ofproto_ OVS_UNUSED,
894 bool *arp_match_ip, uint32_t *actions)
896 *arp_match_ip = true;
897 *actions = ((1u << OFPAT_OUTPUT) |
898 (1u << OFPAT_SET_VLAN_VID) |
899 (1u << OFPAT_SET_VLAN_PCP) |
900 (1u << OFPAT_STRIP_VLAN) |
901 (1u << OFPAT_SET_DL_SRC) |
902 (1u << OFPAT_SET_DL_DST) |
903 (1u << OFPAT_SET_NW_SRC) |
904 (1u << OFPAT_SET_NW_DST) |
905 (1u << OFPAT_SET_NW_TOS) |
906 (1u << OFPAT_SET_TP_SRC) |
907 (1u << OFPAT_SET_TP_DST) |
908 (1u << OFPAT_ENQUEUE));
912 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
914 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
915 struct dpif_dp_stats s;
917 strcpy(ots->name, "classifier");
919 dpif_get_dp_stats(ofproto->dpif, &s);
920 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
921 put_32aligned_be64(&ots->matched_count,
922 htonll(s.n_hit + ofproto->n_matches));
925 static struct ofport *
928 struct ofport_dpif *port = xmalloc(sizeof *port);
933 port_dealloc(struct ofport *port_)
935 struct ofport_dpif *port = ofport_dpif_cast(port_);
940 port_construct(struct ofport *port_)
942 struct ofport_dpif *port = ofport_dpif_cast(port_);
943 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
945 ofproto->need_revalidate = true;
946 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
949 port->tag = tag_create_random();
950 port->may_enable = true;
951 port->stp_port = NULL;
952 port->stp_state = STP_DISABLED;
953 hmap_init(&port->priorities);
954 port->realdev_ofp_port = 0;
955 port->vlandev_vid = 0;
957 if (ofproto->sflow) {
958 dpif_sflow_add_port(ofproto->sflow, port_);
965 port_destruct(struct ofport *port_)
967 struct ofport_dpif *port = ofport_dpif_cast(port_);
968 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
970 ofproto->need_revalidate = true;
971 bundle_remove(port_);
972 set_cfm(port_, NULL);
973 if (ofproto->sflow) {
974 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
977 ofport_clear_priorities(port);
978 hmap_destroy(&port->priorities);
982 port_modified(struct ofport *port_)
984 struct ofport_dpif *port = ofport_dpif_cast(port_);
986 if (port->bundle && port->bundle->bond) {
987 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
992 port_reconfigured(struct ofport *port_, ovs_be32 old_config)
994 struct ofport_dpif *port = ofport_dpif_cast(port_);
995 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
996 ovs_be32 changed = old_config ^ port->up.opp.config;
998 if (changed & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
999 OFPPC_NO_FWD | OFPPC_NO_FLOOD)) {
1000 ofproto->need_revalidate = true;
1002 if (changed & htonl(OFPPC_NO_FLOOD) && port->bundle) {
1003 bundle_update(port->bundle);
1009 set_sflow(struct ofproto *ofproto_,
1010 const struct ofproto_sflow_options *sflow_options)
1012 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1013 struct dpif_sflow *ds = ofproto->sflow;
1015 if (sflow_options) {
1017 struct ofport_dpif *ofport;
1019 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1020 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1021 dpif_sflow_add_port(ds, &ofport->up);
1023 ofproto->need_revalidate = true;
1025 dpif_sflow_set_options(ds, sflow_options);
1028 dpif_sflow_destroy(ds);
1029 ofproto->need_revalidate = true;
1030 ofproto->sflow = NULL;
1037 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1039 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1046 struct ofproto_dpif *ofproto;
1048 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1049 ofproto->need_revalidate = true;
1050 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1053 if (cfm_configure(ofport->cfm, s)) {
1059 cfm_destroy(ofport->cfm);
1065 get_cfm_fault(const struct ofport *ofport_)
1067 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1069 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1073 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1076 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1079 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1086 /* Spanning Tree. */
1089 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1091 struct ofproto_dpif *ofproto = ofproto_;
1092 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1093 struct ofport_dpif *ofport;
1095 ofport = stp_port_get_aux(sp);
1097 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1098 ofproto->up.name, port_num);
1100 struct eth_header *eth = pkt->l2;
1102 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1103 if (eth_addr_is_zero(eth->eth_src)) {
1104 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1105 "with unknown MAC", ofproto->up.name, port_num);
1107 send_packet(ofport, pkt);
1113 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1115 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1117 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1119 /* Only revalidate flows if the configuration changed. */
1120 if (!s != !ofproto->stp) {
1121 ofproto->need_revalidate = true;
1125 if (!ofproto->stp) {
1126 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1127 send_bpdu_cb, ofproto);
1128 ofproto->stp_last_tick = time_msec();
1131 stp_set_bridge_id(ofproto->stp, s->system_id);
1132 stp_set_bridge_priority(ofproto->stp, s->priority);
1133 stp_set_hello_time(ofproto->stp, s->hello_time);
1134 stp_set_max_age(ofproto->stp, s->max_age);
1135 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1137 stp_destroy(ofproto->stp);
1138 ofproto->stp = NULL;
1145 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1147 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1151 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1152 s->designated_root = stp_get_designated_root(ofproto->stp);
1153 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1162 update_stp_port_state(struct ofport_dpif *ofport)
1164 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1165 enum stp_state state;
1167 /* Figure out new state. */
1168 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1172 if (ofport->stp_state != state) {
1176 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1177 netdev_get_name(ofport->up.netdev),
1178 stp_state_name(ofport->stp_state),
1179 stp_state_name(state));
1180 if (stp_learn_in_state(ofport->stp_state)
1181 != stp_learn_in_state(state)) {
1182 /* xxx Learning action flows should also be flushed. */
1183 mac_learning_flush(ofproto->ml);
1185 fwd_change = stp_forward_in_state(ofport->stp_state)
1186 != stp_forward_in_state(state);
1188 ofproto->need_revalidate = true;
1189 ofport->stp_state = state;
1190 ofport->stp_state_entered = time_msec();
1192 if (fwd_change && ofport->bundle) {
1193 bundle_update(ofport->bundle);
1196 /* Update the STP state bits in the OpenFlow port description. */
1197 of_state = (ofport->up.opp.state & htonl(~OFPPS_STP_MASK))
1198 | htonl(state == STP_LISTENING ? OFPPS_STP_LISTEN
1199 : state == STP_LEARNING ? OFPPS_STP_LEARN
1200 : state == STP_FORWARDING ? OFPPS_STP_FORWARD
1201 : state == STP_BLOCKING ? OFPPS_STP_BLOCK
1203 ofproto_port_set_state(&ofport->up, of_state);
1207 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1208 * caller is responsible for assigning STP port numbers and ensuring
1209 * there are no duplicates. */
1211 set_stp_port(struct ofport *ofport_,
1212 const struct ofproto_port_stp_settings *s)
1214 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1215 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1216 struct stp_port *sp = ofport->stp_port;
1218 if (!s || !s->enable) {
1220 ofport->stp_port = NULL;
1221 stp_port_disable(sp);
1222 update_stp_port_state(ofport);
1225 } else if (sp && stp_port_no(sp) != s->port_num
1226 && ofport == stp_port_get_aux(sp)) {
1227 /* The port-id changed, so disable the old one if it's not
1228 * already in use by another port. */
1229 stp_port_disable(sp);
1232 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1233 stp_port_enable(sp);
1235 stp_port_set_aux(sp, ofport);
1236 stp_port_set_priority(sp, s->priority);
1237 stp_port_set_path_cost(sp, s->path_cost);
1239 update_stp_port_state(ofport);
1245 get_stp_port_status(struct ofport *ofport_,
1246 struct ofproto_port_stp_status *s)
1248 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1249 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1250 struct stp_port *sp = ofport->stp_port;
1252 if (!ofproto->stp || !sp) {
1258 s->port_id = stp_port_get_id(sp);
1259 s->state = stp_port_get_state(sp);
1260 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1261 s->role = stp_port_get_role(sp);
1262 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1268 stp_run(struct ofproto_dpif *ofproto)
1271 long long int now = time_msec();
1272 long long int elapsed = now - ofproto->stp_last_tick;
1273 struct stp_port *sp;
1276 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1277 ofproto->stp_last_tick = now;
1279 while (stp_get_changed_port(ofproto->stp, &sp)) {
1280 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1283 update_stp_port_state(ofport);
1290 stp_wait(struct ofproto_dpif *ofproto)
1293 poll_timer_wait(1000);
1297 /* Returns true if STP should process 'flow'. */
1299 stp_should_process_flow(const struct flow *flow)
1301 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1305 stp_process_packet(const struct ofport_dpif *ofport,
1306 const struct ofpbuf *packet)
1308 struct ofpbuf payload = *packet;
1309 struct eth_header *eth = payload.data;
1310 struct stp_port *sp = ofport->stp_port;
1312 /* Sink packets on ports that have STP disabled when the bridge has
1314 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1318 /* Trim off padding on payload. */
1319 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1320 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1323 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1324 stp_received_bpdu(sp, payload.data, payload.size);
1328 static struct priority_to_dscp *
1329 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1331 struct priority_to_dscp *pdscp;
1334 hash = hash_int(priority, 0);
1335 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1336 if (pdscp->priority == priority) {
1344 ofport_clear_priorities(struct ofport_dpif *ofport)
1346 struct priority_to_dscp *pdscp, *next;
1348 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1349 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1355 set_queues(struct ofport *ofport_,
1356 const struct ofproto_port_queue *qdscp_list,
1359 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1360 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1361 struct hmap new = HMAP_INITIALIZER(&new);
1364 for (i = 0; i < n_qdscp; i++) {
1365 struct priority_to_dscp *pdscp;
1369 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1370 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1375 pdscp = get_priority(ofport, priority);
1377 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1379 pdscp = xmalloc(sizeof *pdscp);
1380 pdscp->priority = priority;
1382 ofproto->need_revalidate = true;
1385 if (pdscp->dscp != dscp) {
1387 ofproto->need_revalidate = true;
1390 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1393 if (!hmap_is_empty(&ofport->priorities)) {
1394 ofport_clear_priorities(ofport);
1395 ofproto->need_revalidate = true;
1398 hmap_swap(&new, &ofport->priorities);
1406 /* Expires all MAC learning entries associated with 'bundle' and forces its
1407 * ofproto to revalidate every flow.
1409 * Normally MAC learning entries are removed only from the ofproto associated
1410 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1411 * are removed from every ofproto. When patch ports and SLB bonds are in use
1412 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1413 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1414 * with the host from which it migrated. */
1416 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1418 struct ofproto_dpif *ofproto = bundle->ofproto;
1419 struct mac_learning *ml = ofproto->ml;
1420 struct mac_entry *mac, *next_mac;
1422 ofproto->need_revalidate = true;
1423 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1424 if (mac->port.p == bundle) {
1426 struct ofproto_dpif *o;
1428 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1430 struct mac_entry *e;
1432 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1435 tag_set_add(&o->revalidate_set, e->tag);
1436 mac_learning_expire(o->ml, e);
1442 mac_learning_expire(ml, mac);
1447 static struct ofbundle *
1448 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1450 struct ofbundle *bundle;
1452 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1453 &ofproto->bundles) {
1454 if (bundle->aux == aux) {
1461 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1462 * ones that are found to 'bundles'. */
1464 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1465 void **auxes, size_t n_auxes,
1466 struct hmapx *bundles)
1470 hmapx_init(bundles);
1471 for (i = 0; i < n_auxes; i++) {
1472 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1474 hmapx_add(bundles, bundle);
1480 bundle_update(struct ofbundle *bundle)
1482 struct ofport_dpif *port;
1484 bundle->floodable = true;
1485 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1486 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
1487 bundle->floodable = false;
1494 bundle_del_port(struct ofport_dpif *port)
1496 struct ofbundle *bundle = port->bundle;
1498 bundle->ofproto->need_revalidate = true;
1500 list_remove(&port->bundle_node);
1501 port->bundle = NULL;
1504 lacp_slave_unregister(bundle->lacp, port);
1507 bond_slave_unregister(bundle->bond, port);
1510 bundle_update(bundle);
1514 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1515 struct lacp_slave_settings *lacp,
1516 uint32_t bond_stable_id)
1518 struct ofport_dpif *port;
1520 port = get_ofp_port(bundle->ofproto, ofp_port);
1525 if (port->bundle != bundle) {
1526 bundle->ofproto->need_revalidate = true;
1528 bundle_del_port(port);
1531 port->bundle = bundle;
1532 list_push_back(&bundle->ports, &port->bundle_node);
1533 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
1534 bundle->floodable = false;
1538 port->bundle->ofproto->need_revalidate = true;
1539 lacp_slave_register(bundle->lacp, port, lacp);
1542 port->bond_stable_id = bond_stable_id;
1548 bundle_destroy(struct ofbundle *bundle)
1550 struct ofproto_dpif *ofproto;
1551 struct ofport_dpif *port, *next_port;
1558 ofproto = bundle->ofproto;
1559 for (i = 0; i < MAX_MIRRORS; i++) {
1560 struct ofmirror *m = ofproto->mirrors[i];
1562 if (m->out == bundle) {
1564 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1565 || hmapx_find_and_delete(&m->dsts, bundle)) {
1566 ofproto->need_revalidate = true;
1571 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1572 bundle_del_port(port);
1575 bundle_flush_macs(bundle, true);
1576 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1578 free(bundle->trunks);
1579 lacp_destroy(bundle->lacp);
1580 bond_destroy(bundle->bond);
1585 bundle_set(struct ofproto *ofproto_, void *aux,
1586 const struct ofproto_bundle_settings *s)
1588 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1589 bool need_flush = false;
1590 struct ofport_dpif *port;
1591 struct ofbundle *bundle;
1592 unsigned long *trunks;
1598 bundle_destroy(bundle_lookup(ofproto, aux));
1602 assert(s->n_slaves == 1 || s->bond != NULL);
1603 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1605 bundle = bundle_lookup(ofproto, aux);
1607 bundle = xmalloc(sizeof *bundle);
1609 bundle->ofproto = ofproto;
1610 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1611 hash_pointer(aux, 0));
1613 bundle->name = NULL;
1615 list_init(&bundle->ports);
1616 bundle->vlan_mode = PORT_VLAN_TRUNK;
1618 bundle->trunks = NULL;
1619 bundle->use_priority_tags = s->use_priority_tags;
1620 bundle->lacp = NULL;
1621 bundle->bond = NULL;
1623 bundle->floodable = true;
1625 bundle->src_mirrors = 0;
1626 bundle->dst_mirrors = 0;
1627 bundle->mirror_out = 0;
1630 if (!bundle->name || strcmp(s->name, bundle->name)) {
1632 bundle->name = xstrdup(s->name);
1637 if (!bundle->lacp) {
1638 ofproto->need_revalidate = true;
1639 bundle->lacp = lacp_create();
1641 lacp_configure(bundle->lacp, s->lacp);
1643 lacp_destroy(bundle->lacp);
1644 bundle->lacp = NULL;
1647 /* Update set of ports. */
1649 for (i = 0; i < s->n_slaves; i++) {
1650 if (!bundle_add_port(bundle, s->slaves[i],
1651 s->lacp ? &s->lacp_slaves[i] : NULL,
1652 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1656 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1657 struct ofport_dpif *next_port;
1659 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1660 for (i = 0; i < s->n_slaves; i++) {
1661 if (s->slaves[i] == port->up.ofp_port) {
1666 bundle_del_port(port);
1670 assert(list_size(&bundle->ports) <= s->n_slaves);
1672 if (list_is_empty(&bundle->ports)) {
1673 bundle_destroy(bundle);
1677 /* Set VLAN tagging mode */
1678 if (s->vlan_mode != bundle->vlan_mode
1679 || s->use_priority_tags != bundle->use_priority_tags) {
1680 bundle->vlan_mode = s->vlan_mode;
1681 bundle->use_priority_tags = s->use_priority_tags;
1686 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1687 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1689 if (vlan != bundle->vlan) {
1690 bundle->vlan = vlan;
1694 /* Get trunked VLANs. */
1695 switch (s->vlan_mode) {
1696 case PORT_VLAN_ACCESS:
1700 case PORT_VLAN_TRUNK:
1701 trunks = (unsigned long *) s->trunks;
1704 case PORT_VLAN_NATIVE_UNTAGGED:
1705 case PORT_VLAN_NATIVE_TAGGED:
1706 if (vlan != 0 && (!s->trunks
1707 || !bitmap_is_set(s->trunks, vlan)
1708 || bitmap_is_set(s->trunks, 0))) {
1709 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1711 trunks = bitmap_clone(s->trunks, 4096);
1713 trunks = bitmap_allocate1(4096);
1715 bitmap_set1(trunks, vlan);
1716 bitmap_set0(trunks, 0);
1718 trunks = (unsigned long *) s->trunks;
1725 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1726 free(bundle->trunks);
1727 if (trunks == s->trunks) {
1728 bundle->trunks = vlan_bitmap_clone(trunks);
1730 bundle->trunks = trunks;
1735 if (trunks != s->trunks) {
1740 if (!list_is_short(&bundle->ports)) {
1741 bundle->ofproto->has_bonded_bundles = true;
1743 if (bond_reconfigure(bundle->bond, s->bond)) {
1744 ofproto->need_revalidate = true;
1747 bundle->bond = bond_create(s->bond);
1748 ofproto->need_revalidate = true;
1751 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1752 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1756 bond_destroy(bundle->bond);
1757 bundle->bond = NULL;
1760 /* If we changed something that would affect MAC learning, un-learn
1761 * everything on this port and force flow revalidation. */
1763 bundle_flush_macs(bundle, false);
1770 bundle_remove(struct ofport *port_)
1772 struct ofport_dpif *port = ofport_dpif_cast(port_);
1773 struct ofbundle *bundle = port->bundle;
1776 bundle_del_port(port);
1777 if (list_is_empty(&bundle->ports)) {
1778 bundle_destroy(bundle);
1779 } else if (list_is_short(&bundle->ports)) {
1780 bond_destroy(bundle->bond);
1781 bundle->bond = NULL;
1787 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
1789 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1790 struct ofport_dpif *port = port_;
1791 uint8_t ea[ETH_ADDR_LEN];
1794 error = netdev_get_etheraddr(port->up.netdev, ea);
1796 struct ofpbuf packet;
1799 ofpbuf_init(&packet, 0);
1800 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1802 memcpy(packet_pdu, pdu, pdu_size);
1804 send_packet(port, &packet);
1805 ofpbuf_uninit(&packet);
1807 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1808 "%s (%s)", port->bundle->name,
1809 netdev_get_name(port->up.netdev), strerror(error));
1814 bundle_send_learning_packets(struct ofbundle *bundle)
1816 struct ofproto_dpif *ofproto = bundle->ofproto;
1817 int error, n_packets, n_errors;
1818 struct mac_entry *e;
1820 error = n_packets = n_errors = 0;
1821 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1822 if (e->port.p != bundle) {
1823 struct ofpbuf *learning_packet;
1824 struct ofport_dpif *port;
1828 /* The assignment to "port" is unnecessary but makes "grep"ing for
1829 * struct ofport_dpif more effective. */
1830 learning_packet = bond_compose_learning_packet(bundle->bond,
1834 ret = send_packet(port, learning_packet);
1835 ofpbuf_delete(learning_packet);
1845 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1846 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1847 "packets, last error was: %s",
1848 bundle->name, n_errors, n_packets, strerror(error));
1850 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1851 bundle->name, n_packets);
1856 bundle_run(struct ofbundle *bundle)
1859 lacp_run(bundle->lacp, send_pdu_cb);
1862 struct ofport_dpif *port;
1864 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1865 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1868 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1869 lacp_negotiated(bundle->lacp));
1870 if (bond_should_send_learning_packets(bundle->bond)) {
1871 bundle_send_learning_packets(bundle);
1877 bundle_wait(struct ofbundle *bundle)
1880 lacp_wait(bundle->lacp);
1883 bond_wait(bundle->bond);
1890 mirror_scan(struct ofproto_dpif *ofproto)
1894 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1895 if (!ofproto->mirrors[idx]) {
1902 static struct ofmirror *
1903 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1907 for (i = 0; i < MAX_MIRRORS; i++) {
1908 struct ofmirror *mirror = ofproto->mirrors[i];
1909 if (mirror && mirror->aux == aux) {
1917 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
1919 mirror_update_dups(struct ofproto_dpif *ofproto)
1923 for (i = 0; i < MAX_MIRRORS; i++) {
1924 struct ofmirror *m = ofproto->mirrors[i];
1927 m->dup_mirrors = MIRROR_MASK_C(1) << i;
1931 for (i = 0; i < MAX_MIRRORS; i++) {
1932 struct ofmirror *m1 = ofproto->mirrors[i];
1939 for (j = i + 1; j < MAX_MIRRORS; j++) {
1940 struct ofmirror *m2 = ofproto->mirrors[j];
1942 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
1943 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
1944 m2->dup_mirrors |= m1->dup_mirrors;
1951 mirror_set(struct ofproto *ofproto_, void *aux,
1952 const struct ofproto_mirror_settings *s)
1954 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1955 mirror_mask_t mirror_bit;
1956 struct ofbundle *bundle;
1957 struct ofmirror *mirror;
1958 struct ofbundle *out;
1959 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
1960 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
1963 mirror = mirror_lookup(ofproto, aux);
1965 mirror_destroy(mirror);
1971 idx = mirror_scan(ofproto);
1973 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
1975 ofproto->up.name, MAX_MIRRORS, s->name);
1979 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
1980 mirror->ofproto = ofproto;
1983 mirror->out_vlan = -1;
1984 mirror->name = NULL;
1987 if (!mirror->name || strcmp(s->name, mirror->name)) {
1989 mirror->name = xstrdup(s->name);
1992 /* Get the new configuration. */
1993 if (s->out_bundle) {
1994 out = bundle_lookup(ofproto, s->out_bundle);
1996 mirror_destroy(mirror);
2002 out_vlan = s->out_vlan;
2004 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2005 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2007 /* If the configuration has not changed, do nothing. */
2008 if (hmapx_equals(&srcs, &mirror->srcs)
2009 && hmapx_equals(&dsts, &mirror->dsts)
2010 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2011 && mirror->out == out
2012 && mirror->out_vlan == out_vlan)
2014 hmapx_destroy(&srcs);
2015 hmapx_destroy(&dsts);
2019 hmapx_swap(&srcs, &mirror->srcs);
2020 hmapx_destroy(&srcs);
2022 hmapx_swap(&dsts, &mirror->dsts);
2023 hmapx_destroy(&dsts);
2025 free(mirror->vlans);
2026 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2029 mirror->out_vlan = out_vlan;
2031 /* Update bundles. */
2032 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2033 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2034 if (hmapx_contains(&mirror->srcs, bundle)) {
2035 bundle->src_mirrors |= mirror_bit;
2037 bundle->src_mirrors &= ~mirror_bit;
2040 if (hmapx_contains(&mirror->dsts, bundle)) {
2041 bundle->dst_mirrors |= mirror_bit;
2043 bundle->dst_mirrors &= ~mirror_bit;
2046 if (mirror->out == bundle) {
2047 bundle->mirror_out |= mirror_bit;
2049 bundle->mirror_out &= ~mirror_bit;
2053 ofproto->need_revalidate = true;
2054 mac_learning_flush(ofproto->ml);
2055 mirror_update_dups(ofproto);
2061 mirror_destroy(struct ofmirror *mirror)
2063 struct ofproto_dpif *ofproto;
2064 mirror_mask_t mirror_bit;
2065 struct ofbundle *bundle;
2071 ofproto = mirror->ofproto;
2072 ofproto->need_revalidate = true;
2073 mac_learning_flush(ofproto->ml);
2075 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2076 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2077 bundle->src_mirrors &= ~mirror_bit;
2078 bundle->dst_mirrors &= ~mirror_bit;
2079 bundle->mirror_out &= ~mirror_bit;
2082 hmapx_destroy(&mirror->srcs);
2083 hmapx_destroy(&mirror->dsts);
2084 free(mirror->vlans);
2086 ofproto->mirrors[mirror->idx] = NULL;
2090 mirror_update_dups(ofproto);
2094 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2095 uint64_t *packets, uint64_t *bytes)
2097 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2098 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2101 *packets = *bytes = UINT64_MAX;
2105 *packets = mirror->packet_count;
2106 *bytes = mirror->byte_count;
2112 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2114 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2115 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2116 ofproto->need_revalidate = true;
2117 mac_learning_flush(ofproto->ml);
2123 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2125 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2126 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2127 return bundle && bundle->mirror_out != 0;
2131 forward_bpdu_changed(struct ofproto *ofproto_)
2133 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2134 /* Revalidate cached flows whenever forward_bpdu option changes. */
2135 ofproto->need_revalidate = true;
2140 static struct ofport_dpif *
2141 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
2143 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2144 return ofport ? ofport_dpif_cast(ofport) : NULL;
2147 static struct ofport_dpif *
2148 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
2150 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2154 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2155 struct dpif_port *dpif_port)
2157 ofproto_port->name = dpif_port->name;
2158 ofproto_port->type = dpif_port->type;
2159 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2163 port_run(struct ofport_dpif *ofport)
2165 bool enable = netdev_get_carrier(ofport->up.netdev);
2168 cfm_run(ofport->cfm);
2170 if (cfm_should_send_ccm(ofport->cfm)) {
2171 struct ofpbuf packet;
2173 ofpbuf_init(&packet, 0);
2174 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr);
2175 send_packet(ofport, &packet);
2176 ofpbuf_uninit(&packet);
2179 enable = enable && !cfm_get_fault(ofport->cfm)
2180 && cfm_get_opup(ofport->cfm);
2183 if (ofport->bundle) {
2184 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2187 if (ofport->may_enable != enable) {
2188 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2190 if (ofproto->has_bundle_action) {
2191 ofproto->need_revalidate = true;
2195 ofport->may_enable = enable;
2199 port_wait(struct ofport_dpif *ofport)
2202 cfm_wait(ofport->cfm);
2207 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2208 struct ofproto_port *ofproto_port)
2210 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2211 struct dpif_port dpif_port;
2214 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2216 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2222 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2224 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2228 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2230 *ofp_portp = odp_port_to_ofp_port(odp_port);
2236 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2238 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2241 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2243 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2245 /* The caller is going to close ofport->up.netdev. If this is a
2246 * bonded port, then the bond is using that netdev, so remove it
2247 * from the bond. The client will need to reconfigure everything
2248 * after deleting ports, so then the slave will get re-added. */
2249 bundle_remove(&ofport->up);
2256 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2258 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2261 error = netdev_get_stats(ofport->up.netdev, stats);
2263 if (!error && ofport->odp_port == OVSP_LOCAL) {
2264 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2266 /* ofproto->stats.tx_packets represents packets that we created
2267 * internally and sent to some port (e.g. packets sent with
2268 * send_packet()). Account for them as if they had come from
2269 * OFPP_LOCAL and got forwarded. */
2271 if (stats->rx_packets != UINT64_MAX) {
2272 stats->rx_packets += ofproto->stats.tx_packets;
2275 if (stats->rx_bytes != UINT64_MAX) {
2276 stats->rx_bytes += ofproto->stats.tx_bytes;
2279 /* ofproto->stats.rx_packets represents packets that were received on
2280 * some port and we processed internally and dropped (e.g. STP).
2281 * Account fro them as if they had been forwarded to OFPP_LOCAL. */
2283 if (stats->tx_packets != UINT64_MAX) {
2284 stats->tx_packets += ofproto->stats.rx_packets;
2287 if (stats->tx_bytes != UINT64_MAX) {
2288 stats->tx_bytes += ofproto->stats.rx_bytes;
2295 /* Account packets for LOCAL port. */
2297 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2298 size_t tx_size, size_t rx_size)
2300 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2303 ofproto->stats.rx_packets++;
2304 ofproto->stats.rx_bytes += rx_size;
2307 ofproto->stats.tx_packets++;
2308 ofproto->stats.tx_bytes += tx_size;
2312 struct port_dump_state {
2313 struct dpif_port_dump dump;
2318 port_dump_start(const struct ofproto *ofproto_, void **statep)
2320 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2321 struct port_dump_state *state;
2323 *statep = state = xmalloc(sizeof *state);
2324 dpif_port_dump_start(&state->dump, ofproto->dpif);
2325 state->done = false;
2330 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2331 struct ofproto_port *port)
2333 struct port_dump_state *state = state_;
2334 struct dpif_port dpif_port;
2336 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2337 ofproto_port_from_dpif_port(port, &dpif_port);
2340 int error = dpif_port_dump_done(&state->dump);
2342 return error ? error : EOF;
2347 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2349 struct port_dump_state *state = state_;
2352 dpif_port_dump_done(&state->dump);
2359 port_poll(const struct ofproto *ofproto_, char **devnamep)
2361 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2362 return dpif_port_poll(ofproto->dpif, devnamep);
2366 port_poll_wait(const struct ofproto *ofproto_)
2368 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2369 dpif_port_poll_wait(ofproto->dpif);
2373 port_is_lacp_current(const struct ofport *ofport_)
2375 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2376 return (ofport->bundle && ofport->bundle->lacp
2377 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2381 /* Upcall handling. */
2383 /* Flow miss batching.
2385 * Some dpifs implement operations faster when you hand them off in a batch.
2386 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2387 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2388 * more packets, plus possibly installing the flow in the dpif.
2390 * So far we only batch the operations that affect flow setup time the most.
2391 * It's possible to batch more than that, but the benefit might be minimal. */
2393 struct hmap_node hmap_node;
2395 enum odp_key_fitness key_fitness;
2396 const struct nlattr *key;
2398 ovs_be16 initial_tci;
2399 struct list packets;
2402 struct flow_miss_op {
2403 union dpif_op dpif_op;
2404 struct subfacet *subfacet;
2407 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2408 * OpenFlow controller as necessary according to their individual
2409 * configurations. */
2411 send_packet_in_miss(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
2412 const struct flow *flow)
2414 struct ofputil_packet_in pin;
2416 pin.packet = packet->data;
2417 pin.packet_len = packet->size;
2418 pin.total_len = packet->size;
2419 pin.reason = OFPR_NO_MATCH;
2424 pin.buffer_id = 0; /* not yet known */
2425 pin.send_len = 0; /* not used for flow table misses */
2427 flow_get_metadata(flow, &pin.fmd);
2429 /* Registers aren't meaningful on a miss. */
2430 memset(pin.fmd.reg_masks, 0, sizeof pin.fmd.reg_masks);
2432 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow);
2436 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2437 const struct ofpbuf *packet)
2439 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2445 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2447 cfm_process_heartbeat(ofport->cfm, packet);
2450 } else if (ofport->bundle && ofport->bundle->lacp
2451 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2453 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2456 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2458 stp_process_packet(ofport, packet);
2465 static struct flow_miss *
2466 flow_miss_create(struct hmap *todo, const struct flow *flow,
2467 enum odp_key_fitness key_fitness,
2468 const struct nlattr *key, size_t key_len,
2469 ovs_be16 initial_tci)
2471 uint32_t hash = flow_hash(flow, 0);
2472 struct flow_miss *miss;
2474 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2475 if (flow_equal(&miss->flow, flow)) {
2480 miss = xmalloc(sizeof *miss);
2481 hmap_insert(todo, &miss->hmap_node, hash);
2483 miss->key_fitness = key_fitness;
2485 miss->key_len = key_len;
2486 miss->initial_tci = initial_tci;
2487 list_init(&miss->packets);
2492 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2493 struct flow_miss_op *ops, size_t *n_ops)
2495 const struct flow *flow = &miss->flow;
2496 struct ofpbuf *packet, *next_packet;
2497 struct subfacet *subfacet;
2498 struct facet *facet;
2500 facet = facet_lookup_valid(ofproto, flow);
2502 struct rule_dpif *rule;
2504 rule = rule_dpif_lookup(ofproto, flow, 0);
2506 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
2507 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
2509 if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) {
2510 COVERAGE_INC(ofproto_dpif_no_packet_in);
2511 /* XXX install 'drop' flow entry */
2515 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
2519 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2520 send_packet_in_miss(ofproto, packet, flow);
2526 facet = facet_create(rule, flow);
2529 subfacet = subfacet_create(ofproto, facet,
2530 miss->key_fitness, miss->key, miss->key_len,
2533 LIST_FOR_EACH_SAFE (packet, next_packet, list_node, &miss->packets) {
2534 struct dpif_flow_stats stats;
2535 struct flow_miss_op *op;
2536 struct dpif_execute *execute;
2538 list_remove(&packet->list_node);
2539 ofproto->n_matches++;
2541 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2543 * Extra-special case for fail-open mode.
2545 * We are in fail-open mode and the packet matched the fail-open
2546 * rule, but we are connected to a controller too. We should send
2547 * the packet up to the controller in the hope that it will try to
2548 * set up a flow and thereby allow us to exit fail-open.
2550 * See the top-level comment in fail-open.c for more information.
2552 send_packet_in_miss(ofproto, packet, flow);
2555 if (!facet->may_install || !subfacet->actions) {
2556 subfacet_make_actions(ofproto, subfacet, packet);
2559 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2560 subfacet_update_stats(ofproto, subfacet, &stats);
2562 if (flow->vlan_tci != subfacet->initial_tci) {
2563 /* This packet was received on a VLAN splinter port. We added
2564 * a VLAN to the packet to make the packet resemble the flow,
2565 * but the actions were composed assuming that the packet
2566 * contained no VLAN. So, we must remove the VLAN header from
2567 * the packet before trying to execute the actions. */
2568 eth_pop_vlan(packet);
2571 op = &ops[(*n_ops)++];
2572 execute = &op->dpif_op.execute;
2573 op->subfacet = subfacet;
2574 execute->type = DPIF_OP_EXECUTE;
2575 execute->key = miss->key;
2576 execute->key_len = miss->key_len;
2577 execute->actions = (facet->may_install
2579 : xmemdup(subfacet->actions,
2580 subfacet->actions_len));
2581 execute->actions_len = subfacet->actions_len;
2582 execute->packet = packet;
2585 if (facet->may_install && subfacet->key_fitness != ODP_FIT_TOO_LITTLE) {
2586 struct flow_miss_op *op = &ops[(*n_ops)++];
2587 struct dpif_flow_put *put = &op->dpif_op.flow_put;
2589 op->subfacet = subfacet;
2590 put->type = DPIF_OP_FLOW_PUT;
2591 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2592 put->key = miss->key;
2593 put->key_len = miss->key_len;
2594 put->actions = subfacet->actions;
2595 put->actions_len = subfacet->actions_len;
2600 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
2601 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
2602 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
2603 * what a flow key should contain.
2605 * This function also includes some logic to help make VLAN splinters
2606 * transparent to the rest of the upcall processing logic. In particular, if
2607 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
2608 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
2609 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
2611 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
2612 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
2613 * (This differs from the value returned in flow->vlan_tci only for packets
2614 * received on VLAN splinters.)
2616 static enum odp_key_fitness
2617 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
2618 const struct nlattr *key, size_t key_len,
2619 struct flow *flow, ovs_be16 *initial_tci,
2620 struct ofpbuf *packet)
2622 enum odp_key_fitness fitness;
2626 fitness = odp_flow_key_to_flow(key, key_len, flow);
2627 if (fitness == ODP_FIT_ERROR) {
2630 *initial_tci = flow->vlan_tci;
2632 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
2634 /* Cause the flow to be processed as if it came in on the real device
2635 * with the VLAN device's VLAN ID. */
2636 flow->in_port = realdev;
2637 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
2639 /* Make the packet resemble the flow, so that it gets sent to an
2640 * OpenFlow controller properly, so that it looks correct for
2641 * sFlow, and so that flow_extract() will get the correct vlan_tci
2642 * if it is called on 'packet'.
2644 * The allocated space inside 'packet' probably also contains
2645 * 'key', that is, both 'packet' and 'key' are probably part of a
2646 * struct dpif_upcall (see the large comment on that structure
2647 * definition), so pushing data on 'packet' is in general not a
2648 * good idea since it could overwrite 'key' or free it as a side
2649 * effect. However, it's OK in this special case because we know
2650 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
2651 * will just overwrite the 4-byte "struct nlattr", which is fine
2652 * since we don't need that header anymore. */
2653 eth_push_vlan(packet, flow->vlan_tci);
2656 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
2657 if (fitness == ODP_FIT_PERFECT) {
2658 fitness = ODP_FIT_TOO_MUCH;
2666 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
2669 struct dpif_upcall *upcall;
2670 struct flow_miss *miss, *next_miss;
2671 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
2672 union dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
2681 /* Construct the to-do list.
2683 * This just amounts to extracting the flow from each packet and sticking
2684 * the packets that have the same flow in the same "flow_miss" structure so
2685 * that we can process them together. */
2687 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
2688 enum odp_key_fitness fitness;
2689 struct flow_miss *miss;
2690 ovs_be16 initial_tci;
2693 /* Obtain metadata and check userspace/kernel agreement on flow match,
2694 * then set 'flow''s header pointers. */
2695 fitness = ofproto_dpif_extract_flow_key(ofproto,
2696 upcall->key, upcall->key_len,
2697 &flow, &initial_tci,
2699 if (fitness == ODP_FIT_ERROR) {
2700 ofpbuf_delete(upcall->packet);
2703 flow_extract(upcall->packet, flow.skb_priority, flow.tun_id,
2704 flow.in_port, &flow);
2706 /* Handle 802.1ag, LACP, and STP specially. */
2707 if (process_special(ofproto, &flow, upcall->packet)) {
2708 ofproto_update_local_port_stats(&ofproto->up,
2709 0, upcall->packet->size);
2710 ofpbuf_delete(upcall->packet);
2711 ofproto->n_matches++;
2715 /* Add other packets to a to-do list. */
2716 miss = flow_miss_create(&todo, &flow, fitness,
2717 upcall->key, upcall->key_len, initial_tci);
2718 list_push_back(&miss->packets, &upcall->packet->list_node);
2721 /* Process each element in the to-do list, constructing the set of
2722 * operations to batch. */
2724 HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) {
2725 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
2726 ofpbuf_list_delete(&miss->packets);
2727 hmap_remove(&todo, &miss->hmap_node);
2730 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
2731 hmap_destroy(&todo);
2733 /* Execute batch. */
2734 for (i = 0; i < n_ops; i++) {
2735 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
2737 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
2739 /* Free memory and update facets. */
2740 for (i = 0; i < n_ops; i++) {
2741 struct flow_miss_op *op = &flow_miss_ops[i];
2742 struct dpif_execute *execute;
2743 struct dpif_flow_put *put;
2745 switch (op->dpif_op.type) {
2746 case DPIF_OP_EXECUTE:
2747 execute = &op->dpif_op.execute;
2748 if (op->subfacet->actions != execute->actions) {
2749 free((struct nlattr *) execute->actions);
2751 ofpbuf_delete((struct ofpbuf *) execute->packet);
2754 case DPIF_OP_FLOW_PUT:
2755 put = &op->dpif_op.flow_put;
2757 op->subfacet->installed = true;
2765 handle_userspace_upcall(struct ofproto_dpif *ofproto,
2766 struct dpif_upcall *upcall)
2768 struct user_action_cookie cookie;
2769 enum odp_key_fitness fitness;
2770 ovs_be16 initial_tci;
2773 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
2775 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
2776 upcall->key_len, &flow,
2777 &initial_tci, upcall->packet);
2778 if (fitness == ODP_FIT_ERROR) {
2779 ofpbuf_delete(upcall->packet);
2783 if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
2784 if (ofproto->sflow) {
2785 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
2789 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
2791 ofpbuf_delete(upcall->packet);
2795 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
2797 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
2801 assert (max_batch <= FLOW_MISS_MAX_BATCH);
2804 for (i = 0; i < max_batch; i++) {
2805 struct dpif_upcall *upcall = &misses[n_misses];
2808 error = dpif_recv(ofproto->dpif, upcall);
2813 switch (upcall->type) {
2814 case DPIF_UC_ACTION:
2815 handle_userspace_upcall(ofproto, upcall);
2819 /* Handle it later. */
2823 case DPIF_N_UC_TYPES:
2825 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32,
2831 handle_miss_upcalls(ofproto, misses, n_misses);
2836 /* Flow expiration. */
2838 static int subfacet_max_idle(const struct ofproto_dpif *);
2839 static void update_stats(struct ofproto_dpif *);
2840 static void rule_expire(struct rule_dpif *);
2841 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
2843 /* This function is called periodically by run(). Its job is to collect
2844 * updates for the flows that have been installed into the datapath, most
2845 * importantly when they last were used, and then use that information to
2846 * expire flows that have not been used recently.
2848 * Returns the number of milliseconds after which it should be called again. */
2850 expire(struct ofproto_dpif *ofproto)
2852 struct rule_dpif *rule, *next_rule;
2853 struct classifier *table;
2856 /* Update stats for each flow in the datapath. */
2857 update_stats(ofproto);
2859 /* Expire subfacets that have been idle too long. */
2860 dp_max_idle = subfacet_max_idle(ofproto);
2861 expire_subfacets(ofproto, dp_max_idle);
2863 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
2864 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
2865 struct cls_cursor cursor;
2867 cls_cursor_init(&cursor, table, NULL);
2868 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
2873 /* All outstanding data in existing flows has been accounted, so it's a
2874 * good time to do bond rebalancing. */
2875 if (ofproto->has_bonded_bundles) {
2876 struct ofbundle *bundle;
2878 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2880 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
2885 return MIN(dp_max_idle, 1000);
2888 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
2890 * This function also pushes statistics updates to rules which each facet
2891 * resubmits into. Generally these statistics will be accurate. However, if a
2892 * facet changes the rule it resubmits into at some time in between
2893 * update_stats() runs, it is possible that statistics accrued to the
2894 * old rule will be incorrectly attributed to the new rule. This could be
2895 * avoided by calling update_stats() whenever rules are created or
2896 * deleted. However, the performance impact of making so many calls to the
2897 * datapath do not justify the benefit of having perfectly accurate statistics.
2900 update_stats(struct ofproto_dpif *p)
2902 const struct dpif_flow_stats *stats;
2903 struct dpif_flow_dump dump;
2904 const struct nlattr *key;
2907 dpif_flow_dump_start(&dump, p->dpif);
2908 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
2909 struct subfacet *subfacet;
2911 subfacet = subfacet_find(p, key, key_len);
2912 if (subfacet && subfacet->installed) {
2913 struct facet *facet = subfacet->facet;
2915 if (stats->n_packets >= subfacet->dp_packet_count) {
2916 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
2917 facet->packet_count += extra;
2919 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
2922 if (stats->n_bytes >= subfacet->dp_byte_count) {
2923 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
2925 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
2928 subfacet->dp_packet_count = stats->n_packets;
2929 subfacet->dp_byte_count = stats->n_bytes;
2931 subfacet_update_time(p, subfacet, stats->used);
2932 facet_account(p, facet);
2933 facet_push_stats(facet);
2935 if (!VLOG_DROP_WARN(&rl)) {
2939 odp_flow_key_format(key, key_len, &s);
2940 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
2944 COVERAGE_INC(facet_unexpected);
2945 /* There's a flow in the datapath that we know nothing about, or a
2946 * flow that shouldn't be installed but was anyway. Delete it. */
2947 dpif_flow_del(p->dpif, key, key_len, NULL);
2950 dpif_flow_dump_done(&dump);
2953 /* Calculates and returns the number of milliseconds of idle time after which
2954 * subfacets should expire from the datapath. When a subfacet expires, we fold
2955 * its statistics into its facet, and when a facet's last subfacet expires, we
2956 * fold its statistic into its rule. */
2958 subfacet_max_idle(const struct ofproto_dpif *ofproto)
2961 * Idle time histogram.
2963 * Most of the time a switch has a relatively small number of subfacets.
2964 * When this is the case we might as well keep statistics for all of them
2965 * in userspace and to cache them in the kernel datapath for performance as
2968 * As the number of subfacets increases, the memory required to maintain
2969 * statistics about them in userspace and in the kernel becomes
2970 * significant. However, with a large number of subfacets it is likely
2971 * that only a few of them are "heavy hitters" that consume a large amount
2972 * of bandwidth. At this point, only heavy hitters are worth caching in
2973 * the kernel and maintaining in userspaces; other subfacets we can
2976 * The technique used to compute the idle time is to build a histogram with
2977 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
2978 * that is installed in the kernel gets dropped in the appropriate bucket.
2979 * After the histogram has been built, we compute the cutoff so that only
2980 * the most-recently-used 1% of subfacets (but at least
2981 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
2982 * the most-recently-used bucket of subfacets is kept, so actually an
2983 * arbitrary number of subfacets can be kept in any given expiration run
2984 * (though the next run will delete most of those unless they receive
2987 * This requires a second pass through the subfacets, in addition to the
2988 * pass made by update_stats(), because the former function never looks at
2989 * uninstallable subfacets.
2991 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
2992 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
2993 int buckets[N_BUCKETS] = { 0 };
2994 int total, subtotal, bucket;
2995 struct subfacet *subfacet;
2999 total = hmap_count(&ofproto->subfacets);
3000 if (total <= ofproto->up.flow_eviction_threshold) {
3001 return N_BUCKETS * BUCKET_WIDTH;
3004 /* Build histogram. */
3006 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3007 long long int idle = now - subfacet->used;
3008 int bucket = (idle <= 0 ? 0
3009 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3010 : (unsigned int) idle / BUCKET_WIDTH);
3014 /* Find the first bucket whose flows should be expired. */
3015 subtotal = bucket = 0;
3017 subtotal += buckets[bucket++];
3018 } while (bucket < N_BUCKETS &&
3019 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3021 if (VLOG_IS_DBG_ENABLED()) {
3025 ds_put_cstr(&s, "keep");
3026 for (i = 0; i < N_BUCKETS; i++) {
3028 ds_put_cstr(&s, ", drop");
3031 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3034 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3038 return bucket * BUCKET_WIDTH;
3042 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3044 long long int cutoff = time_msec() - dp_max_idle;
3045 struct subfacet *subfacet, *next_subfacet;
3047 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3048 &ofproto->subfacets) {
3049 if (subfacet->used < cutoff) {
3050 subfacet_destroy(ofproto, subfacet);
3055 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3056 * then delete it entirely. */
3058 rule_expire(struct rule_dpif *rule)
3060 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3061 struct facet *facet, *next_facet;
3065 /* Has 'rule' expired? */
3067 if (rule->up.hard_timeout
3068 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3069 reason = OFPRR_HARD_TIMEOUT;
3070 } else if (rule->up.idle_timeout && list_is_empty(&rule->facets)
3071 && now > rule->used + rule->up.idle_timeout * 1000) {
3072 reason = OFPRR_IDLE_TIMEOUT;
3077 COVERAGE_INC(ofproto_dpif_expired);
3079 /* Update stats. (This is a no-op if the rule expired due to an idle
3080 * timeout, because that only happens when the rule has no facets left.) */
3081 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3082 facet_remove(ofproto, facet);
3085 /* Get rid of the rule. */
3086 ofproto_rule_expire(&rule->up, reason);
3091 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3093 * The caller must already have determined that no facet with an identical
3094 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3095 * the ofproto's classifier table.
3097 * The facet will initially have no subfacets. The caller should create (at
3098 * least) one subfacet with subfacet_create(). */
3099 static struct facet *
3100 facet_create(struct rule_dpif *rule, const struct flow *flow)
3102 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3103 struct facet *facet;
3105 facet = xzalloc(sizeof *facet);
3106 facet->used = time_msec();
3107 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
3108 list_push_back(&rule->facets, &facet->list_node);
3110 facet->flow = *flow;
3111 list_init(&facet->subfacets);
3112 netflow_flow_init(&facet->nf_flow);
3113 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3119 facet_free(struct facet *facet)
3124 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3125 * 'packet', which arrived on 'in_port'.
3127 * Takes ownership of 'packet'. */
3129 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3130 const struct nlattr *odp_actions, size_t actions_len,
3131 struct ofpbuf *packet)
3133 struct odputil_keybuf keybuf;
3137 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3138 odp_flow_key_from_flow(&key, flow);
3140 error = dpif_execute(ofproto->dpif, key.data, key.size,
3141 odp_actions, actions_len, packet);
3143 ofpbuf_delete(packet);
3147 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3149 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3150 * rule's statistics, via subfacet_uninstall().
3152 * - Removes 'facet' from its rule and from ofproto->facets.
3155 facet_remove(struct ofproto_dpif *ofproto, struct facet *facet)
3157 struct subfacet *subfacet, *next_subfacet;
3159 assert(!list_is_empty(&facet->subfacets));
3161 /* First uninstall all of the subfacets to get final statistics. */
3162 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3163 subfacet_uninstall(ofproto, subfacet);
3166 /* Flush the final stats to the rule.
3168 * This might require us to have at least one subfacet around so that we
3169 * can use its actions for accounting in facet_account(), which is why we
3170 * have uninstalled but not yet destroyed the subfacets. */
3171 facet_flush_stats(ofproto, facet);
3173 /* Now we're really all done so destroy everything. */
3174 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3175 &facet->subfacets) {
3176 subfacet_destroy__(ofproto, subfacet);
3178 hmap_remove(&ofproto->facets, &facet->hmap_node);
3179 list_remove(&facet->list_node);
3184 facet_account(struct ofproto_dpif *ofproto, struct facet *facet)
3187 struct subfacet *subfacet;
3188 const struct nlattr *a;
3192 if (facet->byte_count <= facet->accounted_bytes) {
3195 n_bytes = facet->byte_count - facet->accounted_bytes;
3196 facet->accounted_bytes = facet->byte_count;
3198 /* Feed information from the active flows back into the learning table to
3199 * ensure that table is always in sync with what is actually flowing
3200 * through the datapath. */
3201 if (facet->has_learn || facet->has_normal) {
3202 struct action_xlate_ctx ctx;
3204 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3205 facet->flow.vlan_tci,
3206 facet->rule->up.flow_cookie, NULL);
3207 ctx.may_learn = true;
3208 ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions,
3209 facet->rule->up.n_actions));
3212 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3216 /* This loop feeds byte counters to bond_account() for rebalancing to use
3217 * as a basis. We also need to track the actual VLAN on which the packet
3218 * is going to be sent to ensure that it matches the one passed to
3219 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3222 * We use the actions from an arbitrary subfacet because they should all
3223 * be equally valid for our purpose. */
3224 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3225 struct subfacet, list_node);
3226 vlan_tci = facet->flow.vlan_tci;
3227 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3228 subfacet->actions, subfacet->actions_len) {
3229 const struct ovs_action_push_vlan *vlan;
3230 struct ofport_dpif *port;
3232 switch (nl_attr_type(a)) {
3233 case OVS_ACTION_ATTR_OUTPUT:
3234 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3235 if (port && port->bundle && port->bundle->bond) {
3236 bond_account(port->bundle->bond, &facet->flow,
3237 vlan_tci_to_vid(vlan_tci), n_bytes);
3241 case OVS_ACTION_ATTR_POP_VLAN:
3242 vlan_tci = htons(0);
3245 case OVS_ACTION_ATTR_PUSH_VLAN:
3246 vlan = nl_attr_get(a);
3247 vlan_tci = vlan->vlan_tci;
3253 /* Returns true if the only action for 'facet' is to send to the controller.
3254 * (We don't report NetFlow expiration messages for such facets because they
3255 * are just part of the control logic for the network, not real traffic). */
3257 facet_is_controller_flow(struct facet *facet)
3260 && facet->rule->up.n_actions == 1
3261 && action_outputs_to_port(&facet->rule->up.actions[0],
3262 htons(OFPP_CONTROLLER)));
3265 /* Folds all of 'facet''s statistics into its rule. Also updates the
3266 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3267 * 'facet''s statistics in the datapath should have been zeroed and folded into
3268 * its packet and byte counts before this function is called. */
3270 facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet)
3272 struct subfacet *subfacet;
3274 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3275 assert(!subfacet->dp_byte_count);
3276 assert(!subfacet->dp_packet_count);
3279 facet_push_stats(facet);
3280 facet_account(ofproto, facet);
3282 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3283 struct ofexpired expired;
3284 expired.flow = facet->flow;
3285 expired.packet_count = facet->packet_count;
3286 expired.byte_count = facet->byte_count;
3287 expired.used = facet->used;
3288 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3291 facet->rule->packet_count += facet->packet_count;
3292 facet->rule->byte_count += facet->byte_count;
3294 /* Reset counters to prevent double counting if 'facet' ever gets
3296 facet_reset_counters(facet);
3298 netflow_flow_clear(&facet->nf_flow);
3301 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3302 * Returns it if found, otherwise a null pointer.
3304 * The returned facet might need revalidation; use facet_lookup_valid()
3305 * instead if that is important. */
3306 static struct facet *
3307 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
3309 struct facet *facet;
3311 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
3313 if (flow_equal(flow, &facet->flow)) {
3321 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3322 * Returns it if found, otherwise a null pointer.
3324 * The returned facet is guaranteed to be valid. */
3325 static struct facet *
3326 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
3328 struct facet *facet = facet_find(ofproto, flow);
3330 /* The facet we found might not be valid, since we could be in need of
3331 * revalidation. If it is not valid, don't return it. */
3333 && (ofproto->need_revalidate
3334 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))
3335 && !facet_revalidate(ofproto, facet)) {
3336 COVERAGE_INC(facet_invalidated);
3343 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
3345 * - If the rule found is different from 'facet''s current rule, moves
3346 * 'facet' to the new rule and recompiles its actions.
3348 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3349 * where it is and recompiles its actions anyway.
3351 * - If there is none, destroys 'facet'.
3353 * Returns true if 'facet' still exists, false if it has been destroyed. */
3355 facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet)
3358 struct nlattr *odp_actions;
3361 struct actions *new_actions;
3363 struct action_xlate_ctx ctx;
3364 struct rule_dpif *new_rule;
3365 struct subfacet *subfacet;
3366 bool actions_changed;
3369 COVERAGE_INC(facet_revalidate);
3371 /* Determine the new rule. */
3372 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3374 /* No new rule, so delete the facet. */
3375 facet_remove(ofproto, facet);
3379 /* Calculate new datapath actions.
3381 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3382 * emit a NetFlow expiration and, if so, we need to have the old state
3383 * around to properly compose it. */
3385 /* If the datapath actions changed or the installability changed,
3386 * then we need to talk to the datapath. */
3389 memset(&ctx, 0, sizeof ctx);
3390 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3391 struct ofpbuf *odp_actions;
3392 bool should_install;
3394 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3395 subfacet->initial_tci, new_rule->up.flow_cookie,
3397 odp_actions = xlate_actions(&ctx, new_rule->up.actions,
3398 new_rule->up.n_actions);
3399 actions_changed = (subfacet->actions_len != odp_actions->size
3400 || memcmp(subfacet->actions, odp_actions->data,
3401 subfacet->actions_len));
3403 should_install = (ctx.may_set_up_flow
3404 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3405 if (actions_changed || should_install != subfacet->installed) {
3406 if (should_install) {
3407 struct dpif_flow_stats stats;
3409 subfacet_install(ofproto, subfacet,
3410 odp_actions->data, odp_actions->size, &stats);
3411 subfacet_update_stats(ofproto, subfacet, &stats);
3413 subfacet_uninstall(ofproto, subfacet);
3417 new_actions = xcalloc(list_size(&facet->subfacets),
3418 sizeof *new_actions);
3420 new_actions[i].odp_actions = xmemdup(odp_actions->data,
3422 new_actions[i].actions_len = odp_actions->size;
3425 ofpbuf_delete(odp_actions);
3429 facet_flush_stats(ofproto, facet);
3432 /* Update 'facet' now that we've taken care of all the old state. */
3433 facet->tags = ctx.tags;
3434 facet->nf_flow.output_iface = ctx.nf_output_iface;
3435 facet->may_install = ctx.may_set_up_flow;
3436 facet->has_learn = ctx.has_learn;
3437 facet->has_normal = ctx.has_normal;
3438 facet->mirrors = ctx.mirrors;
3441 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3442 if (new_actions[i].odp_actions) {
3443 free(subfacet->actions);
3444 subfacet->actions = new_actions[i].odp_actions;
3445 subfacet->actions_len = new_actions[i].actions_len;
3451 if (facet->rule != new_rule) {
3452 COVERAGE_INC(facet_changed_rule);
3453 list_remove(&facet->list_node);
3454 list_push_back(&new_rule->facets, &facet->list_node);
3455 facet->rule = new_rule;
3456 facet->used = new_rule->up.created;
3457 facet->prev_used = facet->used;
3463 /* Updates 'facet''s used time. Caller is responsible for calling
3464 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3466 facet_update_time(struct ofproto_dpif *ofproto, struct facet *facet,
3469 if (used > facet->used) {
3471 if (used > facet->rule->used) {
3472 facet->rule->used = used;
3474 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3479 facet_reset_counters(struct facet *facet)
3481 facet->packet_count = 0;
3482 facet->byte_count = 0;
3483 facet->prev_packet_count = 0;
3484 facet->prev_byte_count = 0;
3485 facet->accounted_bytes = 0;
3489 facet_push_stats(struct facet *facet)
3491 uint64_t new_packets, new_bytes;
3493 assert(facet->packet_count >= facet->prev_packet_count);
3494 assert(facet->byte_count >= facet->prev_byte_count);
3495 assert(facet->used >= facet->prev_used);
3497 new_packets = facet->packet_count - facet->prev_packet_count;
3498 new_bytes = facet->byte_count - facet->prev_byte_count;
3500 if (new_packets || new_bytes || facet->used > facet->prev_used) {
3501 facet->prev_packet_count = facet->packet_count;
3502 facet->prev_byte_count = facet->byte_count;
3503 facet->prev_used = facet->used;
3505 flow_push_stats(facet->rule, &facet->flow,
3506 new_packets, new_bytes, facet->used);
3508 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
3509 facet->mirrors, new_packets, new_bytes);
3513 struct ofproto_push {
3514 struct action_xlate_ctx ctx;
3521 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
3523 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3526 rule->packet_count += push->packets;
3527 rule->byte_count += push->bytes;
3528 rule->used = MAX(push->used, rule->used);
3532 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3533 * 'rule''s actions and mirrors. */
3535 flow_push_stats(const struct rule_dpif *rule,
3536 const struct flow *flow, uint64_t packets, uint64_t bytes,
3539 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3540 struct ofproto_push push;
3542 push.packets = packets;
3546 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci,
3547 rule->up.flow_cookie, NULL);
3548 push.ctx.resubmit_hook = push_resubmit;
3549 ofpbuf_delete(xlate_actions(&push.ctx,
3550 rule->up.actions, rule->up.n_actions));
3555 static struct subfacet *
3556 subfacet_find__(struct ofproto_dpif *ofproto,
3557 const struct nlattr *key, size_t key_len, uint32_t key_hash,
3558 const struct flow *flow)
3560 struct subfacet *subfacet;
3562 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
3563 &ofproto->subfacets) {
3565 ? (subfacet->key_len == key_len
3566 && !memcmp(key, subfacet->key, key_len))
3567 : flow_equal(flow, &subfacet->facet->flow)) {
3575 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
3576 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
3577 * there is one, otherwise creates and returns a new subfacet.
3579 * If the returned subfacet is new, then subfacet->actions will be NULL, in
3580 * which case the caller must populate the actions with
3581 * subfacet_make_actions(). */
3582 static struct subfacet *
3583 subfacet_create(struct ofproto_dpif *ofproto, struct facet *facet,
3584 enum odp_key_fitness key_fitness,
3585 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
3587 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3588 struct subfacet *subfacet;
3590 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
3592 if (subfacet->facet == facet) {
3596 /* This shouldn't happen. */
3597 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
3598 subfacet_destroy(ofproto, subfacet);
3601 subfacet = xzalloc(sizeof *subfacet);
3602 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
3603 list_push_back(&facet->subfacets, &subfacet->list_node);
3604 subfacet->facet = facet;
3605 subfacet->used = time_msec();
3606 subfacet->key_fitness = key_fitness;
3607 if (key_fitness != ODP_FIT_PERFECT) {
3608 subfacet->key = xmemdup(key, key_len);
3609 subfacet->key_len = key_len;
3611 subfacet->installed = false;
3612 subfacet->initial_tci = initial_tci;
3617 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
3618 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
3619 static struct subfacet *
3620 subfacet_find(struct ofproto_dpif *ofproto,
3621 const struct nlattr *key, size_t key_len)
3623 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3624 enum odp_key_fitness fitness;
3627 fitness = odp_flow_key_to_flow(key, key_len, &flow);
3628 if (fitness == ODP_FIT_ERROR) {
3632 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
3635 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
3636 * its facet within 'ofproto', and frees it. */
3638 subfacet_destroy__(struct ofproto_dpif *ofproto, struct subfacet *subfacet)
3640 subfacet_uninstall(ofproto, subfacet);
3641 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
3642 list_remove(&subfacet->list_node);
3643 free(subfacet->key);
3644 free(subfacet->actions);
3648 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
3649 * last remaining subfacet in its facet destroys the facet too. */
3651 subfacet_destroy(struct ofproto_dpif *ofproto, struct subfacet *subfacet)
3653 struct facet *facet = subfacet->facet;
3655 if (list_is_singleton(&facet->subfacets)) {
3656 /* facet_remove() needs at least one subfacet (it will remove it). */
3657 facet_remove(ofproto, facet);
3659 subfacet_destroy__(ofproto, subfacet);
3663 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
3664 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
3665 * for use as temporary storage. */
3667 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
3670 if (!subfacet->key) {
3671 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
3672 odp_flow_key_from_flow(key, &subfacet->facet->flow);
3674 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
3678 /* Composes the datapath actions for 'subfacet' based on its rule's actions. */
3680 subfacet_make_actions(struct ofproto_dpif *p, struct subfacet *subfacet,
3681 const struct ofpbuf *packet)
3683 struct facet *facet = subfacet->facet;
3684 const struct rule_dpif *rule = facet->rule;
3685 struct ofpbuf *odp_actions;
3686 struct action_xlate_ctx ctx;
3688 action_xlate_ctx_init(&ctx, p, &facet->flow, subfacet->initial_tci,
3689 rule->up.flow_cookie, packet);
3690 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3691 facet->tags = ctx.tags;
3692 facet->may_install = ctx.may_set_up_flow;
3693 facet->has_learn = ctx.has_learn;
3694 facet->has_normal = ctx.has_normal;
3695 facet->nf_flow.output_iface = ctx.nf_output_iface;
3696 facet->mirrors = ctx.mirrors;
3698 if (subfacet->actions_len != odp_actions->size
3699 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
3700 free(subfacet->actions);
3701 subfacet->actions_len = odp_actions->size;
3702 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
3705 ofpbuf_delete(odp_actions);
3708 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
3709 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
3710 * in the datapath will be zeroed and 'stats' will be updated with traffic new
3711 * since 'subfacet' was last updated.
3713 * Returns 0 if successful, otherwise a positive errno value. */
3715 subfacet_install(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3716 const struct nlattr *actions, size_t actions_len,
3717 struct dpif_flow_stats *stats)
3719 struct odputil_keybuf keybuf;
3720 enum dpif_flow_put_flags flags;
3724 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3726 flags |= DPIF_FP_ZERO_STATS;
3729 subfacet_get_key(subfacet, &keybuf, &key);
3730 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
3731 actions, actions_len, stats);
3734 subfacet_reset_dp_stats(subfacet, stats);
3740 /* If 'subfacet' is installed in the datapath, uninstalls it. */
3742 subfacet_uninstall(struct ofproto_dpif *p, struct subfacet *subfacet)
3744 if (subfacet->installed) {
3745 struct odputil_keybuf keybuf;
3746 struct dpif_flow_stats stats;
3750 subfacet_get_key(subfacet, &keybuf, &key);
3751 error = dpif_flow_del(p->dpif, key.data, key.size, &stats);
3752 subfacet_reset_dp_stats(subfacet, &stats);
3754 subfacet_update_stats(p, subfacet, &stats);
3756 subfacet->installed = false;
3758 assert(subfacet->dp_packet_count == 0);
3759 assert(subfacet->dp_byte_count == 0);
3763 /* Resets 'subfacet''s datapath statistics counters. This should be called
3764 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
3765 * non-null, it should contain the statistics returned by dpif when 'subfacet'
3766 * was reset in the datapath. 'stats' will be modified to include only
3767 * statistics new since 'subfacet' was last updated. */
3769 subfacet_reset_dp_stats(struct subfacet *subfacet,
3770 struct dpif_flow_stats *stats)
3773 && subfacet->dp_packet_count <= stats->n_packets
3774 && subfacet->dp_byte_count <= stats->n_bytes) {
3775 stats->n_packets -= subfacet->dp_packet_count;
3776 stats->n_bytes -= subfacet->dp_byte_count;
3779 subfacet->dp_packet_count = 0;
3780 subfacet->dp_byte_count = 0;
3783 /* Updates 'subfacet''s used time. The caller is responsible for calling
3784 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
3786 subfacet_update_time(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3789 if (used > subfacet->used) {
3790 subfacet->used = used;
3791 facet_update_time(ofproto, subfacet->facet, used);
3795 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
3797 * Because of the meaning of a subfacet's counters, it only makes sense to do
3798 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
3799 * represents a packet that was sent by hand or if it represents statistics
3800 * that have been cleared out of the datapath. */
3802 subfacet_update_stats(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3803 const struct dpif_flow_stats *stats)
3805 if (stats->n_packets || stats->used > subfacet->used) {
3806 struct facet *facet = subfacet->facet;
3808 subfacet_update_time(ofproto, subfacet, stats->used);
3809 facet->packet_count += stats->n_packets;
3810 facet->byte_count += stats->n_bytes;
3811 facet_push_stats(facet);
3812 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3818 static struct rule_dpif *
3819 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
3822 struct cls_rule *cls_rule;
3823 struct classifier *cls;
3825 if (table_id >= N_TABLES) {
3829 cls = &ofproto->up.tables[table_id];
3830 if (flow->nw_frag & FLOW_NW_FRAG_ANY
3831 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3832 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
3833 * are unavailable. */
3834 struct flow ofpc_normal_flow = *flow;
3835 ofpc_normal_flow.tp_src = htons(0);
3836 ofpc_normal_flow.tp_dst = htons(0);
3837 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
3839 cls_rule = classifier_lookup(cls, flow);
3841 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
3845 complete_operation(struct rule_dpif *rule)
3847 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3849 rule_invalidate(rule);
3851 struct dpif_completion *c = xmalloc(sizeof *c);
3852 c->op = rule->up.pending;
3853 list_push_back(&ofproto->completions, &c->list_node);
3855 ofoperation_complete(rule->up.pending, 0);
3859 static struct rule *
3862 struct rule_dpif *rule = xmalloc(sizeof *rule);
3867 rule_dealloc(struct rule *rule_)
3869 struct rule_dpif *rule = rule_dpif_cast(rule_);
3874 rule_construct(struct rule *rule_)
3876 struct rule_dpif *rule = rule_dpif_cast(rule_);
3877 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3878 struct rule_dpif *victim;
3882 error = validate_actions(rule->up.actions, rule->up.n_actions,
3883 &rule->up.cr.flow, ofproto->max_ports);
3888 rule->used = rule->up.created;
3889 rule->packet_count = 0;
3890 rule->byte_count = 0;
3892 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
3893 if (victim && !list_is_empty(&victim->facets)) {
3894 struct facet *facet;
3896 rule->facets = victim->facets;
3897 list_moved(&rule->facets);
3898 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3899 /* XXX: We're only clearing our local counters here. It's possible
3900 * that quite a few packets are unaccounted for in the datapath
3901 * statistics. These will be accounted to the new rule instead of
3902 * cleared as required. This could be fixed by clearing out the
3903 * datapath statistics for this facet, but currently it doesn't
3905 facet_reset_counters(facet);
3909 /* Must avoid list_moved() in this case. */
3910 list_init(&rule->facets);
3913 table_id = rule->up.table_id;
3914 rule->tag = (victim ? victim->tag
3916 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
3917 ofproto->tables[table_id].basis));
3919 complete_operation(rule);
3924 rule_destruct(struct rule *rule_)
3926 struct rule_dpif *rule = rule_dpif_cast(rule_);
3927 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3928 struct facet *facet, *next_facet;
3930 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3931 facet_revalidate(ofproto, facet);
3934 complete_operation(rule);
3938 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
3940 struct rule_dpif *rule = rule_dpif_cast(rule_);
3941 struct facet *facet;
3943 /* Start from historical data for 'rule' itself that are no longer tracked
3944 * in facets. This counts, for example, facets that have expired. */
3945 *packets = rule->packet_count;
3946 *bytes = rule->byte_count;
3948 /* Add any statistics that are tracked by facets. This includes
3949 * statistical data recently updated by ofproto_update_stats() as well as
3950 * stats for packets that were executed "by hand" via dpif_execute(). */
3951 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3952 *packets += facet->packet_count;
3953 *bytes += facet->byte_count;
3958 rule_execute(struct rule *rule_, const struct flow *flow,
3959 struct ofpbuf *packet)
3961 struct rule_dpif *rule = rule_dpif_cast(rule_);
3962 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3963 struct action_xlate_ctx ctx;
3964 struct ofpbuf *odp_actions;
3967 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
3968 rule->up.flow_cookie, packet);
3969 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3970 size = packet->size;
3971 if (execute_odp_actions(ofproto, flow, odp_actions->data,
3972 odp_actions->size, packet)) {
3973 rule->used = time_msec();
3974 rule->packet_count++;
3975 rule->byte_count += size;
3976 flow_push_stats(rule, flow, 1, size, rule->used);
3978 ofpbuf_delete(odp_actions);
3984 rule_modify_actions(struct rule *rule_)
3986 struct rule_dpif *rule = rule_dpif_cast(rule_);
3987 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3990 error = validate_actions(rule->up.actions, rule->up.n_actions,
3991 &rule->up.cr.flow, ofproto->max_ports);
3993 ofoperation_complete(rule->up.pending, error);
3997 complete_operation(rule);
4000 /* Sends 'packet' out 'ofport'.
4001 * May modify 'packet'.
4002 * Returns 0 if successful, otherwise a positive errno value. */
4004 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4006 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4007 struct ofpbuf key, odp_actions;
4008 struct odputil_keybuf keybuf;
4013 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
4014 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4016 if (odp_port != ofport->odp_port) {
4017 eth_pop_vlan(packet);
4018 flow.vlan_tci = htons(0);
4021 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4022 odp_flow_key_from_flow(&key, &flow);
4024 ofpbuf_init(&odp_actions, 32);
4025 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4027 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4028 error = dpif_execute(ofproto->dpif,
4030 odp_actions.data, odp_actions.size,
4032 ofpbuf_uninit(&odp_actions);
4035 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4036 ofproto->up.name, odp_port, strerror(error));
4038 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
4042 /* OpenFlow to datapath action translation. */
4044 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
4045 struct action_xlate_ctx *ctx);
4046 static void xlate_normal(struct action_xlate_ctx *);
4049 put_userspace_action(const struct ofproto_dpif *ofproto,
4050 struct ofpbuf *odp_actions,
4051 const struct flow *flow,
4052 const struct user_action_cookie *cookie)
4056 pid = dpif_port_get_pid(ofproto->dpif,
4057 ofp_port_to_odp_port(flow->in_port));
4059 return odp_put_userspace_action(pid, cookie, odp_actions);
4062 /* Compose SAMPLE action for sFlow. */
4064 compose_sflow_action(const struct ofproto_dpif *ofproto,
4065 struct ofpbuf *odp_actions,
4066 const struct flow *flow,
4069 uint32_t port_ifindex;
4070 uint32_t probability;
4071 struct user_action_cookie cookie;
4072 size_t sample_offset, actions_offset;
4073 int cookie_offset, n_output;
4075 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4079 if (odp_port == OVSP_NONE) {
4083 port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4087 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4089 /* Number of packets out of UINT_MAX to sample. */
4090 probability = dpif_sflow_get_probability(ofproto->sflow);
4091 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4093 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4095 cookie.type = USER_ACTION_COOKIE_SFLOW;
4096 cookie.data = port_ifindex;
4097 cookie.n_output = n_output;
4098 cookie.vlan_tci = 0;
4099 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4101 nl_msg_end_nested(odp_actions, actions_offset);
4102 nl_msg_end_nested(odp_actions, sample_offset);
4103 return cookie_offset;
4106 /* SAMPLE action must be first action in any given list of actions.
4107 * At this point we do not have all information required to build it. So try to
4108 * build sample action as complete as possible. */
4110 add_sflow_action(struct action_xlate_ctx *ctx)
4112 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4114 &ctx->flow, OVSP_NONE);
4115 ctx->sflow_odp_port = 0;
4116 ctx->sflow_n_outputs = 0;
4119 /* Fix SAMPLE action according to data collected while composing ODP actions.
4120 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4121 * USERSPACE action's user-cookie which is required for sflow. */
4123 fix_sflow_action(struct action_xlate_ctx *ctx)
4125 const struct flow *base = &ctx->base_flow;
4126 struct user_action_cookie *cookie;
4128 if (!ctx->user_cookie_offset) {
4132 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4134 assert(cookie != NULL);
4135 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4137 if (ctx->sflow_n_outputs) {
4138 cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow,
4139 ctx->sflow_odp_port);
4141 if (ctx->sflow_n_outputs >= 255) {
4142 cookie->n_output = 255;
4144 cookie->n_output = ctx->sflow_n_outputs;
4146 cookie->vlan_tci = base->vlan_tci;
4150 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4153 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4154 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4155 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4156 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4160 struct priority_to_dscp *pdscp;
4162 if (ofport->up.opp.config & htonl(OFPPC_NO_FWD)
4163 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4167 pdscp = get_priority(ofport, ctx->flow.skb_priority);
4169 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4170 ctx->flow.nw_tos |= pdscp->dscp;
4173 /* We may not have an ofport record for this port, but it doesn't hurt
4174 * to allow forwarding to it anyhow. Maybe such a port will appear
4175 * later and we're pre-populating the flow table. */
4178 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4179 ctx->flow.vlan_tci);
4180 if (out_port != odp_port) {
4181 ctx->flow.vlan_tci = htons(0);
4183 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4184 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4186 ctx->sflow_odp_port = odp_port;
4187 ctx->sflow_n_outputs++;
4188 ctx->nf_output_iface = ofp_port;
4189 ctx->flow.vlan_tci = flow_vlan_tci;
4190 ctx->flow.nw_tos = flow_nw_tos;
4194 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4196 compose_output_action__(ctx, ofp_port, true);
4200 xlate_table_action(struct action_xlate_ctx *ctx,
4201 uint16_t in_port, uint8_t table_id)
4203 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4204 struct ofproto_dpif *ofproto = ctx->ofproto;
4205 struct rule_dpif *rule;
4206 uint16_t old_in_port;
4207 uint8_t old_table_id;
4209 old_table_id = ctx->table_id;
4210 ctx->table_id = table_id;
4212 /* Look up a flow with 'in_port' as the input port. */
4213 old_in_port = ctx->flow.in_port;
4214 ctx->flow.in_port = in_port;
4215 rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
4218 if (table_id > 0 && table_id < N_TABLES) {
4219 struct table_dpif *table = &ofproto->tables[table_id];
4220 if (table->other_table) {
4223 : rule_calculate_tag(&ctx->flow,
4224 &table->other_table->wc,
4229 /* Restore the original input port. Otherwise OFPP_NORMAL and
4230 * OFPP_IN_PORT will have surprising behavior. */
4231 ctx->flow.in_port = old_in_port;
4233 if (ctx->resubmit_hook) {
4234 ctx->resubmit_hook(ctx, rule);
4238 ovs_be64 old_cookie = ctx->cookie;
4241 ctx->cookie = rule->up.flow_cookie;
4242 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
4243 ctx->cookie = old_cookie;
4247 ctx->table_id = old_table_id;
4249 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4251 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
4252 MAX_RESUBMIT_RECURSION);
4257 xlate_resubmit_table(struct action_xlate_ctx *ctx,
4258 const struct nx_action_resubmit *nar)
4263 in_port = (nar->in_port == htons(OFPP_IN_PORT)
4265 : ntohs(nar->in_port));
4266 table_id = nar->table == 255 ? ctx->table_id : nar->table;
4268 xlate_table_action(ctx, in_port, table_id);
4272 flood_packets(struct action_xlate_ctx *ctx, bool all)
4274 struct ofport_dpif *ofport;
4276 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
4277 uint16_t ofp_port = ofport->up.ofp_port;
4279 if (ofp_port == ctx->flow.in_port) {
4284 compose_output_action__(ctx, ofp_port, false);
4285 } else if (!(ofport->up.opp.config & htonl(OFPPC_NO_FLOOD))) {
4286 compose_output_action(ctx, ofp_port);
4290 ctx->nf_output_iface = NF_OUT_FLOOD;
4294 execute_controller_action(struct action_xlate_ctx *ctx, int len,
4295 enum ofp_packet_in_reason reason)
4297 struct ofputil_packet_in pin;
4298 struct ofpbuf *packet;
4300 ctx->may_set_up_flow = false;
4305 packet = ofpbuf_clone(ctx->packet);
4307 if (packet->l2 && packet->l3) {
4308 struct eth_header *eh;
4310 eth_pop_vlan(packet);
4312 assert(eh->eth_type == ctx->flow.dl_type);
4313 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
4314 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
4316 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
4317 eth_push_vlan(packet, ctx->flow.vlan_tci);
4321 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4322 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
4323 ctx->flow.nw_tos, ctx->flow.nw_ttl);
4327 if (ctx->flow.nw_proto == IPPROTO_TCP) {
4328 packet_set_tcp_port(packet, ctx->flow.tp_src,
4330 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
4331 packet_set_udp_port(packet, ctx->flow.tp_src,
4338 pin.packet = packet->data;
4339 pin.packet_len = packet->size;
4340 pin.reason = reason;
4341 pin.table_id = ctx->table_id;
4342 pin.cookie = ctx->cookie;
4346 pin.total_len = packet->size;
4347 flow_get_metadata(&ctx->flow, &pin.fmd);
4349 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin, &ctx->flow);
4350 ofpbuf_delete(packet);
4354 compose_dec_ttl(struct action_xlate_ctx *ctx)
4356 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
4357 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
4361 if (ctx->flow.nw_ttl > 1) {
4365 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL);
4367 /* Stop processing for current table. */
4373 xlate_output_action__(struct action_xlate_ctx *ctx,
4374 uint16_t port, uint16_t max_len)
4376 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
4378 ctx->nf_output_iface = NF_OUT_DROP;
4382 compose_output_action(ctx, ctx->flow.in_port);
4385 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
4391 flood_packets(ctx, false);
4394 flood_packets(ctx, true);
4396 case OFPP_CONTROLLER:
4397 execute_controller_action(ctx, max_len, OFPR_ACTION);
4400 compose_output_action(ctx, OFPP_LOCAL);
4405 if (port != ctx->flow.in_port) {
4406 compose_output_action(ctx, port);
4411 if (prev_nf_output_iface == NF_OUT_FLOOD) {
4412 ctx->nf_output_iface = NF_OUT_FLOOD;
4413 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
4414 ctx->nf_output_iface = prev_nf_output_iface;
4415 } else if (prev_nf_output_iface != NF_OUT_DROP &&
4416 ctx->nf_output_iface != NF_OUT_FLOOD) {
4417 ctx->nf_output_iface = NF_OUT_MULTI;
4422 xlate_output_reg_action(struct action_xlate_ctx *ctx,
4423 const struct nx_action_output_reg *naor)
4427 ofp_port = nxm_read_field_bits(naor->src, naor->ofs_nbits, &ctx->flow);
4429 if (ofp_port <= UINT16_MAX) {
4430 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
4435 xlate_output_action(struct action_xlate_ctx *ctx,
4436 const struct ofp_action_output *oao)
4438 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
4442 xlate_enqueue_action(struct action_xlate_ctx *ctx,
4443 const struct ofp_action_enqueue *oae)
4446 uint32_t flow_priority, priority;
4449 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
4452 /* Fall back to ordinary output action. */
4453 xlate_output_action__(ctx, ntohs(oae->port), 0);
4457 /* Figure out datapath output port. */
4458 ofp_port = ntohs(oae->port);
4459 if (ofp_port == OFPP_IN_PORT) {
4460 ofp_port = ctx->flow.in_port;
4461 } else if (ofp_port == ctx->flow.in_port) {
4465 /* Add datapath actions. */
4466 flow_priority = ctx->flow.skb_priority;
4467 ctx->flow.skb_priority = priority;
4468 compose_output_action(ctx, ofp_port);
4469 ctx->flow.skb_priority = flow_priority;
4471 /* Update NetFlow output port. */
4472 if (ctx->nf_output_iface == NF_OUT_DROP) {
4473 ctx->nf_output_iface = ofp_port;
4474 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
4475 ctx->nf_output_iface = NF_OUT_MULTI;
4480 xlate_set_queue_action(struct action_xlate_ctx *ctx,
4481 const struct nx_action_set_queue *nasq)
4486 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
4489 /* Couldn't translate queue to a priority, so ignore. A warning
4490 * has already been logged. */
4494 ctx->flow.skb_priority = priority;
4497 struct xlate_reg_state {
4503 xlate_autopath(struct action_xlate_ctx *ctx,
4504 const struct nx_action_autopath *naa)
4506 uint16_t ofp_port = ntohl(naa->id);
4507 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
4509 if (!port || !port->bundle) {
4510 ofp_port = OFPP_NONE;
4511 } else if (port->bundle->bond) {
4512 /* Autopath does not support VLAN hashing. */
4513 struct ofport_dpif *slave = bond_choose_output_slave(
4514 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
4516 ofp_port = slave->up.ofp_port;
4519 autopath_execute(naa, &ctx->flow, ofp_port);
4523 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
4525 struct ofproto_dpif *ofproto = ofproto_;
4526 struct ofport_dpif *port;
4536 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
4539 port = get_ofp_port(ofproto, ofp_port);
4540 return port ? port->may_enable : false;
4545 xlate_learn_action(struct action_xlate_ctx *ctx,
4546 const struct nx_action_learn *learn)
4548 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
4549 struct ofputil_flow_mod fm;
4552 learn_execute(learn, &ctx->flow, &fm);
4554 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
4555 if (error && !VLOG_DROP_WARN(&rl)) {
4556 VLOG_WARN("learning action failed to modify flow table (%s)",
4557 ofperr_get_name(error));
4564 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
4566 if (port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
4567 ? htonl(OFPPC_NO_RECV_STP)
4568 : htonl(OFPPC_NO_RECV))) {
4572 /* Only drop packets here if both forwarding and learning are
4573 * disabled. If just learning is enabled, we need to have
4574 * OFPP_NORMAL and the learning action have a look at the packet
4575 * before we can drop it. */
4576 if (!stp_forward_in_state(port->stp_state)
4577 && !stp_learn_in_state(port->stp_state)) {
4585 do_xlate_actions(const union ofp_action *in, size_t n_in,
4586 struct action_xlate_ctx *ctx)
4588 const struct ofport_dpif *port;
4589 const union ofp_action *ia;
4592 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
4593 if (port && !may_receive(port, ctx)) {
4594 /* Drop this flow. */
4598 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
4599 const struct ofp_action_dl_addr *oada;
4600 const struct nx_action_resubmit *nar;
4601 const struct nx_action_set_tunnel *nast;
4602 const struct nx_action_set_queue *nasq;
4603 const struct nx_action_multipath *nam;
4604 const struct nx_action_autopath *naa;
4605 const struct nx_action_bundle *nab;
4606 const struct nx_action_output_reg *naor;
4607 enum ofputil_action_code code;
4614 code = ofputil_decode_action_unsafe(ia);
4616 case OFPUTIL_OFPAT_OUTPUT:
4617 xlate_output_action(ctx, &ia->output);
4620 case OFPUTIL_OFPAT_SET_VLAN_VID:
4621 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
4622 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
4625 case OFPUTIL_OFPAT_SET_VLAN_PCP:
4626 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
4627 ctx->flow.vlan_tci |= htons(
4628 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
4631 case OFPUTIL_OFPAT_STRIP_VLAN:
4632 ctx->flow.vlan_tci = htons(0);
4635 case OFPUTIL_OFPAT_SET_DL_SRC:
4636 oada = ((struct ofp_action_dl_addr *) ia);
4637 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
4640 case OFPUTIL_OFPAT_SET_DL_DST:
4641 oada = ((struct ofp_action_dl_addr *) ia);
4642 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
4645 case OFPUTIL_OFPAT_SET_NW_SRC:
4646 ctx->flow.nw_src = ia->nw_addr.nw_addr;
4649 case OFPUTIL_OFPAT_SET_NW_DST:
4650 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
4653 case OFPUTIL_OFPAT_SET_NW_TOS:
4654 /* OpenFlow 1.0 only supports IPv4. */
4655 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4656 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4657 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
4661 case OFPUTIL_OFPAT_SET_TP_SRC:
4662 ctx->flow.tp_src = ia->tp_port.tp_port;
4665 case OFPUTIL_OFPAT_SET_TP_DST:
4666 ctx->flow.tp_dst = ia->tp_port.tp_port;
4669 case OFPUTIL_OFPAT_ENQUEUE:
4670 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
4673 case OFPUTIL_NXAST_RESUBMIT:
4674 nar = (const struct nx_action_resubmit *) ia;
4675 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
4678 case OFPUTIL_NXAST_RESUBMIT_TABLE:
4679 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
4682 case OFPUTIL_NXAST_SET_TUNNEL:
4683 nast = (const struct nx_action_set_tunnel *) ia;
4684 tun_id = htonll(ntohl(nast->tun_id));
4685 ctx->flow.tun_id = tun_id;
4688 case OFPUTIL_NXAST_SET_QUEUE:
4689 nasq = (const struct nx_action_set_queue *) ia;
4690 xlate_set_queue_action(ctx, nasq);
4693 case OFPUTIL_NXAST_POP_QUEUE:
4694 ctx->flow.skb_priority = ctx->orig_skb_priority;
4697 case OFPUTIL_NXAST_REG_MOVE:
4698 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
4702 case OFPUTIL_NXAST_REG_LOAD:
4703 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
4707 case OFPUTIL_NXAST_NOTE:
4708 /* Nothing to do. */
4711 case OFPUTIL_NXAST_SET_TUNNEL64:
4712 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
4713 ctx->flow.tun_id = tun_id;
4716 case OFPUTIL_NXAST_MULTIPATH:
4717 nam = (const struct nx_action_multipath *) ia;
4718 multipath_execute(nam, &ctx->flow);
4721 case OFPUTIL_NXAST_AUTOPATH:
4722 naa = (const struct nx_action_autopath *) ia;
4723 xlate_autopath(ctx, naa);
4726 case OFPUTIL_NXAST_BUNDLE:
4727 ctx->ofproto->has_bundle_action = true;
4728 nab = (const struct nx_action_bundle *) ia;
4729 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
4734 case OFPUTIL_NXAST_BUNDLE_LOAD:
4735 ctx->ofproto->has_bundle_action = true;
4736 nab = (const struct nx_action_bundle *) ia;
4737 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
4741 case OFPUTIL_NXAST_OUTPUT_REG:
4742 naor = (const struct nx_action_output_reg *) ia;
4743 xlate_output_reg_action(ctx, naor);
4746 case OFPUTIL_NXAST_LEARN:
4747 ctx->has_learn = true;
4748 if (ctx->may_learn) {
4749 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
4753 case OFPUTIL_NXAST_DEC_TTL:
4754 if (compose_dec_ttl(ctx)) {
4759 case OFPUTIL_NXAST_EXIT:
4766 /* We've let OFPP_NORMAL and the learning action look at the packet,
4767 * so drop it now if forwarding is disabled. */
4768 if (port && !stp_forward_in_state(port->stp_state)) {
4769 ofpbuf_clear(ctx->odp_actions);
4770 add_sflow_action(ctx);
4775 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
4776 struct ofproto_dpif *ofproto, const struct flow *flow,
4777 ovs_be16 initial_tci, ovs_be64 cookie,
4778 const struct ofpbuf *packet)
4780 ctx->ofproto = ofproto;
4782 ctx->base_flow = ctx->flow;
4783 ctx->base_flow.tun_id = 0;
4784 ctx->base_flow.vlan_tci = initial_tci;
4785 ctx->cookie = cookie;
4786 ctx->packet = packet;
4787 ctx->may_learn = packet != NULL;
4788 ctx->resubmit_hook = NULL;
4791 static struct ofpbuf *
4792 xlate_actions(struct action_xlate_ctx *ctx,
4793 const union ofp_action *in, size_t n_in)
4795 struct flow orig_flow = ctx->flow;
4797 COVERAGE_INC(ofproto_dpif_xlate);
4799 ctx->odp_actions = ofpbuf_new(512);
4800 ofpbuf_reserve(ctx->odp_actions, NL_A_U32_SIZE);
4802 ctx->may_set_up_flow = true;
4803 ctx->has_learn = false;
4804 ctx->has_normal = false;
4805 ctx->nf_output_iface = NF_OUT_DROP;
4808 ctx->orig_skb_priority = ctx->flow.skb_priority;
4812 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
4813 switch (ctx->ofproto->up.frag_handling) {
4814 case OFPC_FRAG_NORMAL:
4815 /* We must pretend that transport ports are unavailable. */
4816 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
4817 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
4820 case OFPC_FRAG_DROP:
4821 return ctx->odp_actions;
4823 case OFPC_FRAG_REASM:
4826 case OFPC_FRAG_NX_MATCH:
4827 /* Nothing to do. */
4830 case OFPC_INVALID_TTL_TO_CONTROLLER:
4835 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
4836 ctx->may_set_up_flow = false;
4837 return ctx->odp_actions;
4839 add_sflow_action(ctx);
4840 do_xlate_actions(in, n_in, ctx);
4842 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
4843 ctx->odp_actions->data,
4844 ctx->odp_actions->size)) {
4845 ctx->may_set_up_flow = false;
4847 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
4849 compose_output_action(ctx, OFPP_LOCAL);
4852 add_mirror_actions(ctx, &orig_flow);
4853 fix_sflow_action(ctx);
4856 return ctx->odp_actions;
4859 /* OFPP_NORMAL implementation. */
4861 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
4863 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
4864 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
4865 * the bundle on which the packet was received, returns the VLAN to which the
4868 * Both 'vid' and the return value are in the range 0...4095. */
4870 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
4872 switch (in_bundle->vlan_mode) {
4873 case PORT_VLAN_ACCESS:
4874 return in_bundle->vlan;
4877 case PORT_VLAN_TRUNK:
4880 case PORT_VLAN_NATIVE_UNTAGGED:
4881 case PORT_VLAN_NATIVE_TAGGED:
4882 return vid ? vid : in_bundle->vlan;
4889 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
4890 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
4893 * 'vid' should be the VID obtained from the 802.1Q header that was received as
4894 * part of a packet (specify 0 if there was no 802.1Q header), in the range
4897 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
4899 /* Allow any VID on the OFPP_NONE port. */
4900 if (in_bundle == &ofpp_none_bundle) {
4904 switch (in_bundle->vlan_mode) {
4905 case PORT_VLAN_ACCESS:
4908 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4909 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
4910 "packet received on port %s configured as VLAN "
4911 "%"PRIu16" access port",
4912 in_bundle->ofproto->up.name, vid,
4913 in_bundle->name, in_bundle->vlan);
4919 case PORT_VLAN_NATIVE_UNTAGGED:
4920 case PORT_VLAN_NATIVE_TAGGED:
4922 /* Port must always carry its native VLAN. */
4926 case PORT_VLAN_TRUNK:
4927 if (!ofbundle_includes_vlan(in_bundle, vid)) {
4929 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4930 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
4931 "received on port %s not configured for trunking "
4933 in_bundle->ofproto->up.name, vid,
4934 in_bundle->name, vid);
4946 /* Given 'vlan', the VLAN that a packet belongs to, and
4947 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
4948 * that should be included in the 802.1Q header. (If the return value is 0,
4949 * then the 802.1Q header should only be included in the packet if there is a
4952 * Both 'vlan' and the return value are in the range 0...4095. */
4954 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
4956 switch (out_bundle->vlan_mode) {
4957 case PORT_VLAN_ACCESS:
4960 case PORT_VLAN_TRUNK:
4961 case PORT_VLAN_NATIVE_TAGGED:
4964 case PORT_VLAN_NATIVE_UNTAGGED:
4965 return vlan == out_bundle->vlan ? 0 : vlan;
4973 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
4976 struct ofport_dpif *port;
4978 ovs_be16 tci, old_tci;
4980 vid = output_vlan_to_vid(out_bundle, vlan);
4981 if (!out_bundle->bond) {
4982 port = ofbundle_get_a_port(out_bundle);
4984 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
4987 /* No slaves enabled, so drop packet. */
4992 old_tci = ctx->flow.vlan_tci;
4994 if (tci || out_bundle->use_priority_tags) {
4995 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
4997 tci |= htons(VLAN_CFI);
5000 ctx->flow.vlan_tci = tci;
5002 compose_output_action(ctx, port->up.ofp_port);
5003 ctx->flow.vlan_tci = old_tci;
5007 mirror_mask_ffs(mirror_mask_t mask)
5009 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
5014 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
5016 return (bundle->vlan_mode != PORT_VLAN_ACCESS
5017 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
5021 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
5023 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
5026 /* Returns an arbitrary interface within 'bundle'. */
5027 static struct ofport_dpif *
5028 ofbundle_get_a_port(const struct ofbundle *bundle)
5030 return CONTAINER_OF(list_front(&bundle->ports),
5031 struct ofport_dpif, bundle_node);
5035 vlan_is_mirrored(const struct ofmirror *m, int vlan)
5037 return !m->vlans || bitmap_is_set(m->vlans, vlan);
5040 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
5041 * to a VLAN. In general most packets may be mirrored but we want to drop
5042 * protocols that may confuse switches. */
5044 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
5046 /* If you change this function's behavior, please update corresponding
5047 * documentation in vswitch.xml at the same time. */
5048 if (dst[0] != 0x01) {
5049 /* All the currently banned MACs happen to start with 01 currently, so
5050 * this is a quick way to eliminate most of the good ones. */
5052 if (eth_addr_is_reserved(dst)) {
5053 /* Drop STP, IEEE pause frames, and other reserved protocols
5054 * (01-80-c2-00-00-0x). */
5058 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
5060 if ((dst[3] & 0xfe) == 0xcc &&
5061 (dst[4] & 0xfe) == 0xcc &&
5062 (dst[5] & 0xfe) == 0xcc) {
5063 /* Drop the following protocols plus others following the same
5066 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
5067 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
5068 STP Uplink Fast (01-00-0c-cd-cd-cd) */
5072 if (!(dst[3] | dst[4] | dst[5])) {
5073 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
5082 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5084 struct ofproto_dpif *ofproto = ctx->ofproto;
5085 mirror_mask_t mirrors;
5086 struct ofbundle *in_bundle;
5089 const struct nlattr *a;
5092 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5093 ctx->packet != NULL);
5097 mirrors = in_bundle->src_mirrors;
5099 /* Drop frames on bundles reserved for mirroring. */
5100 if (in_bundle->mirror_out) {
5101 if (ctx->packet != NULL) {
5102 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5103 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5104 "%s, which is reserved exclusively for mirroring",
5105 ctx->ofproto->up.name, in_bundle->name);
5111 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5112 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5115 vlan = input_vid_to_vlan(in_bundle, vid);
5117 /* Look at the output ports to check for destination selections. */
5119 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5120 ctx->odp_actions->size) {
5121 enum ovs_action_attr type = nl_attr_type(a);
5122 struct ofport_dpif *ofport;
5124 if (type != OVS_ACTION_ATTR_OUTPUT) {
5128 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5129 if (ofport && ofport->bundle) {
5130 mirrors |= ofport->bundle->dst_mirrors;
5138 /* Restore the original packet before adding the mirror actions. */
5139 ctx->flow = *orig_flow;
5144 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5146 if (!vlan_is_mirrored(m, vlan)) {
5147 mirrors &= mirrors - 1;
5151 mirrors &= ~m->dup_mirrors;
5152 ctx->mirrors |= m->dup_mirrors;
5154 output_normal(ctx, m->out, vlan);
5155 } else if (eth_dst_may_rspan(orig_flow->dl_dst)
5156 && vlan != m->out_vlan) {
5157 struct ofbundle *bundle;
5159 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5160 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5161 && !bundle->mirror_out) {
5162 output_normal(ctx, bundle, m->out_vlan);
5170 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5171 uint64_t packets, uint64_t bytes)
5177 for (; mirrors; mirrors &= mirrors - 1) {
5180 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5183 /* In normal circumstances 'm' will not be NULL. However,
5184 * if mirrors are reconfigured, we can temporarily get out
5185 * of sync in facet_revalidate(). We could "correct" the
5186 * mirror list before reaching here, but doing that would
5187 * not properly account the traffic stats we've currently
5188 * accumulated for previous mirror configuration. */
5192 m->packet_count += packets;
5193 m->byte_count += bytes;
5197 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
5198 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
5199 * indicate this; newer upstream kernels use gratuitous ARP requests. */
5201 is_gratuitous_arp(const struct flow *flow)
5203 return (flow->dl_type == htons(ETH_TYPE_ARP)
5204 && eth_addr_is_broadcast(flow->dl_dst)
5205 && (flow->nw_proto == ARP_OP_REPLY
5206 || (flow->nw_proto == ARP_OP_REQUEST
5207 && flow->nw_src == flow->nw_dst)));
5211 update_learning_table(struct ofproto_dpif *ofproto,
5212 const struct flow *flow, int vlan,
5213 struct ofbundle *in_bundle)
5215 struct mac_entry *mac;
5217 /* Don't learn the OFPP_NONE port. */
5218 if (in_bundle == &ofpp_none_bundle) {
5222 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
5226 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
5227 if (is_gratuitous_arp(flow)) {
5228 /* We don't want to learn from gratuitous ARP packets that are
5229 * reflected back over bond slaves so we lock the learning table. */
5230 if (!in_bundle->bond) {
5231 mac_entry_set_grat_arp_lock(mac);
5232 } else if (mac_entry_is_grat_arp_locked(mac)) {
5237 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
5238 /* The log messages here could actually be useful in debugging,
5239 * so keep the rate limit relatively high. */
5240 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5241 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
5242 "on port %s in VLAN %d",
5243 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
5244 in_bundle->name, vlan);
5246 mac->port.p = in_bundle;
5247 tag_set_add(&ofproto->revalidate_set,
5248 mac_learning_changed(ofproto->ml, mac));
5252 static struct ofbundle *
5253 lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn)
5255 struct ofport_dpif *ofport;
5257 /* Special-case OFPP_NONE, which a controller may use as the ingress
5258 * port for traffic that it is sourcing. */
5259 if (in_port == OFPP_NONE) {
5260 return &ofpp_none_bundle;
5263 /* Find the port and bundle for the received packet. */
5264 ofport = get_ofp_port(ofproto, in_port);
5265 if (ofport && ofport->bundle) {
5266 return ofport->bundle;
5269 /* Odd. A few possible reasons here:
5271 * - We deleted a port but there are still a few packets queued up
5274 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
5275 * we don't know about.
5277 * - The ofproto client didn't configure the port as part of a bundle.
5280 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5282 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
5283 "port %"PRIu16, ofproto->up.name, in_port);
5288 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
5289 * dropped. Returns true if they may be forwarded, false if they should be
5292 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
5293 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
5295 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
5296 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
5297 * checked by input_vid_is_valid().
5299 * May also add tags to '*tags', although the current implementation only does
5300 * so in one special case.
5303 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
5304 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
5306 struct ofbundle *in_bundle = in_port->bundle;
5308 /* Drop frames for reserved multicast addresses
5309 * only if forward_bpdu option is absent. */
5310 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
5314 if (in_bundle->bond) {
5315 struct mac_entry *mac;
5317 switch (bond_check_admissibility(in_bundle->bond, in_port,
5318 flow->dl_dst, tags)) {
5325 case BV_DROP_IF_MOVED:
5326 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
5327 if (mac && mac->port.p != in_bundle &&
5328 (!is_gratuitous_arp(flow)
5329 || mac_entry_is_grat_arp_locked(mac))) {
5340 xlate_normal(struct action_xlate_ctx *ctx)
5342 struct ofport_dpif *in_port;
5343 struct ofbundle *in_bundle;
5344 struct mac_entry *mac;
5348 ctx->has_normal = true;
5350 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
5351 ctx->packet != NULL);
5356 /* We know 'in_port' exists unless it is "ofpp_none_bundle",
5357 * since lookup_input_bundle() succeeded. */
5358 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5360 /* Drop malformed frames. */
5361 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
5362 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
5363 if (ctx->packet != NULL) {
5364 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5365 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
5366 "VLAN tag received on port %s",
5367 ctx->ofproto->up.name, in_bundle->name);
5372 /* Drop frames on bundles reserved for mirroring. */
5373 if (in_bundle->mirror_out) {
5374 if (ctx->packet != NULL) {
5375 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5376 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5377 "%s, which is reserved exclusively for mirroring",
5378 ctx->ofproto->up.name, in_bundle->name);
5384 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
5385 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5388 vlan = input_vid_to_vlan(in_bundle, vid);
5390 /* Check other admissibility requirements. */
5392 !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
5396 /* Learn source MAC. */
5397 if (ctx->may_learn) {
5398 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
5401 /* Determine output bundle. */
5402 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
5405 if (mac->port.p != in_bundle) {
5406 output_normal(ctx, mac->port.p, vlan);
5409 struct ofbundle *bundle;
5411 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
5412 if (bundle != in_bundle
5413 && ofbundle_includes_vlan(bundle, vlan)
5414 && bundle->floodable
5415 && !bundle->mirror_out) {
5416 output_normal(ctx, bundle, vlan);
5419 ctx->nf_output_iface = NF_OUT_FLOOD;
5423 /* Optimized flow revalidation.
5425 * It's a difficult problem, in general, to tell which facets need to have
5426 * their actions recalculated whenever the OpenFlow flow table changes. We
5427 * don't try to solve that general problem: for most kinds of OpenFlow flow
5428 * table changes, we recalculate the actions for every facet. This is
5429 * relatively expensive, but it's good enough if the OpenFlow flow table
5430 * doesn't change very often.
5432 * However, we can expect one particular kind of OpenFlow flow table change to
5433 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5434 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5435 * table, we add a special case that applies to flow tables in which every rule
5436 * has the same form (that is, the same wildcards), except that the table is
5437 * also allowed to have a single "catch-all" flow that matches all packets. We
5438 * optimize this case by tagging all of the facets that resubmit into the table
5439 * and invalidating the same tag whenever a flow changes in that table. The
5440 * end result is that we revalidate just the facets that need it (and sometimes
5441 * a few more, but not all of the facets or even all of the facets that
5442 * resubmit to the table modified by MAC learning). */
5444 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
5445 * into an OpenFlow table with the given 'basis'. */
5447 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
5450 if (flow_wildcards_is_catchall(wc)) {
5453 struct flow tag_flow = *flow;
5454 flow_zero_wildcards(&tag_flow, wc);
5455 return tag_create_deterministic(flow_hash(&tag_flow, secret));
5459 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5460 * taggability of that table.
5462 * This function must be called after *each* change to a flow table. If you
5463 * skip calling it on some changes then the pointer comparisons at the end can
5464 * be invalid if you get unlucky. For example, if a flow removal causes a
5465 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5466 * different wildcards to be created with the same address, then this function
5467 * will incorrectly skip revalidation. */
5469 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5471 struct table_dpif *table = &ofproto->tables[table_id];
5472 const struct classifier *cls = &ofproto->up.tables[table_id];
5473 struct cls_table *catchall, *other;
5474 struct cls_table *t;
5476 catchall = other = NULL;
5478 switch (hmap_count(&cls->tables)) {
5480 /* We could tag this OpenFlow table but it would make the logic a
5481 * little harder and it's a corner case that doesn't seem worth it
5487 HMAP_FOR_EACH (t, hmap_node, &cls->tables) {
5488 if (cls_table_is_catchall(t)) {
5490 } else if (!other) {
5493 /* Indicate that we can't tag this by setting both tables to
5494 * NULL. (We know that 'catchall' is already NULL.) */
5501 /* Can't tag this table. */
5505 if (table->catchall_table != catchall || table->other_table != other) {
5506 table->catchall_table = catchall;
5507 table->other_table = other;
5508 ofproto->need_revalidate = true;
5512 /* Given 'rule' that has changed in some way (either it is a rule being
5513 * inserted, a rule being deleted, or a rule whose actions are being
5514 * modified), marks facets for revalidation to ensure that packets will be
5515 * forwarded correctly according to the new state of the flow table.
5517 * This function must be called after *each* change to a flow table. See
5518 * the comment on table_update_taggable() for more information. */
5520 rule_invalidate(const struct rule_dpif *rule)
5522 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5524 table_update_taggable(ofproto, rule->up.table_id);
5526 if (!ofproto->need_revalidate) {
5527 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5529 if (table->other_table && rule->tag) {
5530 tag_set_add(&ofproto->revalidate_set, rule->tag);
5532 ofproto->need_revalidate = true;
5538 set_frag_handling(struct ofproto *ofproto_,
5539 enum ofp_config_flags frag_handling)
5541 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5543 if (frag_handling != OFPC_FRAG_REASM) {
5544 ofproto->need_revalidate = true;
5552 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5553 const struct flow *flow,
5554 const union ofp_action *ofp_actions, size_t n_ofp_actions)
5556 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5559 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
5560 return OFPERR_NXBRC_BAD_IN_PORT;
5563 error = validate_actions(ofp_actions, n_ofp_actions, flow,
5564 ofproto->max_ports);
5566 struct odputil_keybuf keybuf;
5567 struct ofpbuf *odp_actions;
5568 struct ofproto_push push;
5571 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5572 odp_flow_key_from_flow(&key, flow);
5574 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, 0,
5577 /* Ensure that resubmits in 'ofp_actions' get accounted to their
5578 * matching rules. */
5580 push.bytes = packet->size;
5581 push.used = time_msec();
5582 push.ctx.resubmit_hook = push_resubmit;
5584 odp_actions = xlate_actions(&push.ctx, ofp_actions, n_ofp_actions);
5585 dpif_execute(ofproto->dpif, key.data, key.size,
5586 odp_actions->data, odp_actions->size, packet);
5587 ofpbuf_delete(odp_actions);
5595 set_netflow(struct ofproto *ofproto_,
5596 const struct netflow_options *netflow_options)
5598 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5600 if (netflow_options) {
5601 if (!ofproto->netflow) {
5602 ofproto->netflow = netflow_create();
5604 return netflow_set_options(ofproto->netflow, netflow_options);
5606 netflow_destroy(ofproto->netflow);
5607 ofproto->netflow = NULL;
5613 get_netflow_ids(const struct ofproto *ofproto_,
5614 uint8_t *engine_type, uint8_t *engine_id)
5616 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5618 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
5622 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5624 if (!facet_is_controller_flow(facet) &&
5625 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5626 struct subfacet *subfacet;
5627 struct ofexpired expired;
5629 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5630 if (subfacet->installed) {
5631 struct dpif_flow_stats stats;
5633 subfacet_install(ofproto, subfacet, subfacet->actions,
5634 subfacet->actions_len, &stats);
5635 subfacet_update_stats(ofproto, subfacet, &stats);
5639 expired.flow = facet->flow;
5640 expired.packet_count = facet->packet_count;
5641 expired.byte_count = facet->byte_count;
5642 expired.used = facet->used;
5643 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5648 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5650 struct facet *facet;
5652 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5653 send_active_timeout(ofproto, facet);
5657 static struct ofproto_dpif *
5658 ofproto_dpif_lookup(const char *name)
5660 struct ofproto_dpif *ofproto;
5662 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5663 hash_string(name, 0), &all_ofproto_dpifs) {
5664 if (!strcmp(ofproto->up.name, name)) {
5672 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc OVS_UNUSED,
5673 const char *argv[], void *aux OVS_UNUSED)
5675 const struct ofproto_dpif *ofproto;
5677 ofproto = ofproto_dpif_lookup(argv[1]);
5679 unixctl_command_reply(conn, 501, "no such bridge");
5682 mac_learning_flush(ofproto->ml);
5684 unixctl_command_reply(conn, 200, "table successfully flushed");
5688 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5689 const char *argv[], void *aux OVS_UNUSED)
5691 struct ds ds = DS_EMPTY_INITIALIZER;
5692 const struct ofproto_dpif *ofproto;
5693 const struct mac_entry *e;
5695 ofproto = ofproto_dpif_lookup(argv[1]);
5697 unixctl_command_reply(conn, 501, "no such bridge");
5701 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5702 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5703 struct ofbundle *bundle = e->port.p;
5704 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
5705 ofbundle_get_a_port(bundle)->odp_port,
5706 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
5708 unixctl_command_reply(conn, 200, ds_cstr(&ds));
5712 struct ofproto_trace {
5713 struct action_xlate_ctx ctx;
5719 trace_format_rule(struct ds *result, uint8_t table_id, int level,
5720 const struct rule_dpif *rule)
5722 ds_put_char_multiple(result, '\t', level);
5724 ds_put_cstr(result, "No match\n");
5728 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5729 table_id, ntohll(rule->up.flow_cookie));
5730 cls_rule_format(&rule->up.cr, result);
5731 ds_put_char(result, '\n');
5733 ds_put_char_multiple(result, '\t', level);
5734 ds_put_cstr(result, "OpenFlow ");
5735 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
5736 ds_put_char(result, '\n');
5740 trace_format_flow(struct ds *result, int level, const char *title,
5741 struct ofproto_trace *trace)
5743 ds_put_char_multiple(result, '\t', level);
5744 ds_put_format(result, "%s: ", title);
5745 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
5746 ds_put_cstr(result, "unchanged");
5748 flow_format(result, &trace->ctx.flow);
5749 trace->flow = trace->ctx.flow;
5751 ds_put_char(result, '\n');
5755 trace_format_regs(struct ds *result, int level, const char *title,
5756 struct ofproto_trace *trace)
5760 ds_put_char_multiple(result, '\t', level);
5761 ds_put_format(result, "%s:", title);
5762 for (i = 0; i < FLOW_N_REGS; i++) {
5763 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5765 ds_put_char(result, '\n');
5769 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
5771 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
5772 struct ds *result = trace->result;
5774 ds_put_char(result, '\n');
5775 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
5776 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
5777 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
5781 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5782 void *aux OVS_UNUSED)
5784 const char *dpname = argv[1];
5785 struct ofproto_dpif *ofproto;
5786 struct ofpbuf odp_key;
5787 struct ofpbuf *packet;
5788 struct rule_dpif *rule;
5789 ovs_be16 initial_tci;
5795 ofpbuf_init(&odp_key, 0);
5798 ofproto = ofproto_dpif_lookup(dpname);
5800 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
5804 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
5805 /* ofproto/trace dpname flow [-generate] */
5806 const char *flow_s = argv[2];
5807 const char *generate_s = argv[3];
5810 /* Convert string to datapath key. */
5811 ofpbuf_init(&odp_key, 0);
5812 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
5814 unixctl_command_reply(conn, 501, "Bad flow syntax");
5818 /* Convert odp_key to flow. */
5819 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
5820 odp_key.size, &flow,
5821 &initial_tci, NULL);
5822 if (error == ODP_FIT_ERROR) {
5823 unixctl_command_reply(conn, 501, "Invalid flow");
5827 /* Generate a packet, if requested. */
5829 packet = ofpbuf_new(0);
5830 flow_compose(packet, &flow);
5832 } else if (argc == 6) {
5833 /* ofproto/trace dpname priority tun_id in_port packet */
5834 const char *priority_s = argv[2];
5835 const char *tun_id_s = argv[3];
5836 const char *in_port_s = argv[4];
5837 const char *packet_s = argv[5];
5838 uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s));
5839 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
5840 uint32_t priority = atoi(priority_s);
5843 msg = eth_from_hex(packet_s, &packet);
5845 unixctl_command_reply(conn, 501, msg);
5849 ds_put_cstr(&result, "Packet: ");
5850 s = ofp_packet_to_string(packet->data, packet->size);
5851 ds_put_cstr(&result, s);
5854 flow_extract(packet, priority, tun_id, in_port, &flow);
5855 initial_tci = flow.vlan_tci;
5857 unixctl_command_reply(conn, 501, "Bad command syntax");
5861 ds_put_cstr(&result, "Flow: ");
5862 flow_format(&result, &flow);
5863 ds_put_char(&result, '\n');
5865 rule = rule_dpif_lookup(ofproto, &flow, 0);
5866 trace_format_rule(&result, 0, 0, rule);
5868 struct ofproto_trace trace;
5869 struct ofpbuf *odp_actions;
5871 trace.result = &result;
5873 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, initial_tci,
5874 rule->up.flow_cookie, packet);
5875 trace.ctx.resubmit_hook = trace_resubmit;
5876 odp_actions = xlate_actions(&trace.ctx,
5877 rule->up.actions, rule->up.n_actions);
5879 ds_put_char(&result, '\n');
5880 trace_format_flow(&result, 0, "Final flow", &trace);
5881 ds_put_cstr(&result, "Datapath actions: ");
5882 format_odp_actions(&result, odp_actions->data, odp_actions->size);
5883 ofpbuf_delete(odp_actions);
5885 if (!trace.ctx.may_set_up_flow) {
5887 ds_put_cstr(&result, "\nThis flow is not cachable.");
5889 ds_put_cstr(&result, "\nThe datapath actions are incomplete--"
5890 "for complete actions, please supply a packet.");
5895 unixctl_command_reply(conn, 200, ds_cstr(&result));
5898 ds_destroy(&result);
5899 ofpbuf_delete(packet);
5900 ofpbuf_uninit(&odp_key);
5904 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5905 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5908 unixctl_command_reply(conn, 200, NULL);
5912 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5913 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5916 unixctl_command_reply(conn, 200, NULL);
5920 ofproto_dpif_unixctl_init(void)
5922 static bool registered;
5928 unixctl_command_register(
5930 "bridge {tun_id in_port packet | odp_flow [-generate]}",
5931 2, 4, ofproto_unixctl_trace, NULL);
5932 unixctl_command_register("fdb/flush", "bridge", 1, 1,
5933 ofproto_unixctl_fdb_flush, NULL);
5934 unixctl_command_register("fdb/show", "bridge", 1, 1,
5935 ofproto_unixctl_fdb_show, NULL);
5936 unixctl_command_register("ofproto/clog", "", 0, 0,
5937 ofproto_dpif_clog, NULL);
5938 unixctl_command_register("ofproto/unclog", "", 0, 0,
5939 ofproto_dpif_unclog, NULL);
5942 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5944 * This is deprecated. It is only for compatibility with broken device drivers
5945 * in old versions of Linux that do not properly support VLANs when VLAN
5946 * devices are not used. When broken device drivers are no longer in
5947 * widespread use, we will delete these interfaces. */
5950 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
5952 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
5953 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
5955 if (realdev_ofp_port == ofport->realdev_ofp_port
5956 && vid == ofport->vlandev_vid) {
5960 ofproto->need_revalidate = true;
5962 if (ofport->realdev_ofp_port) {
5965 if (realdev_ofp_port && ofport->bundle) {
5966 /* vlandevs are enslaved to their realdevs, so they are not allowed to
5967 * themselves be part of a bundle. */
5968 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
5971 ofport->realdev_ofp_port = realdev_ofp_port;
5972 ofport->vlandev_vid = vid;
5974 if (realdev_ofp_port) {
5975 vsp_add(ofport, realdev_ofp_port, vid);
5982 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
5984 return hash_2words(realdev_ofp_port, vid);
5988 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
5989 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
5991 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
5992 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
5993 int vid = vlan_tci_to_vid(vlan_tci);
5994 const struct vlan_splinter *vsp;
5996 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
5997 hash_realdev_vid(realdev_ofp_port, vid),
5998 &ofproto->realdev_vid_map) {
5999 if (vsp->realdev_ofp_port == realdev_ofp_port
6000 && vsp->vid == vid) {
6001 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
6005 return realdev_odp_port;
6008 static struct vlan_splinter *
6009 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
6011 struct vlan_splinter *vsp;
6013 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
6014 &ofproto->vlandev_map) {
6015 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6024 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6025 uint16_t vlandev_ofp_port, int *vid)
6027 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6028 const struct vlan_splinter *vsp;
6030 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6035 return vsp->realdev_ofp_port;
6042 vsp_remove(struct ofport_dpif *port)
6044 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6045 struct vlan_splinter *vsp;
6047 vsp = vlandev_find(ofproto, port->up.ofp_port);
6049 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6050 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6053 port->realdev_ofp_port = 0;
6055 VLOG_ERR("missing vlan device record");
6060 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
6062 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6064 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6065 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6066 == realdev_ofp_port)) {
6067 struct vlan_splinter *vsp;
6069 vsp = xmalloc(sizeof *vsp);
6070 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6071 hash_int(port->up.ofp_port, 0));
6072 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6073 hash_realdev_vid(realdev_ofp_port, vid));
6074 vsp->realdev_ofp_port = realdev_ofp_port;
6075 vsp->vlandev_ofp_port = port->up.ofp_port;
6078 port->realdev_ofp_port = realdev_ofp_port;
6080 VLOG_ERR("duplicate vlan device record");
6084 const struct ofproto_class ofproto_dpif_class = {
6113 port_is_lacp_current,
6114 NULL, /* rule_choose_table */
6121 rule_modify_actions,
6129 get_cfm_remote_mpids,
6133 get_stp_port_status,
6140 is_mirror_output_bundle,
6141 forward_bpdu_changed,