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 tag_type 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);
194 static int set_stp_port(struct ofport *,
195 const struct ofproto_port_stp_settings *);
197 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
199 struct action_xlate_ctx {
200 /* action_xlate_ctx_init() initializes these members. */
203 struct ofproto_dpif *ofproto;
205 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
206 * this flow when actions change header fields. */
209 /* The packet corresponding to 'flow', or a null pointer if we are
210 * revalidating without a packet to refer to. */
211 const struct ofpbuf *packet;
213 /* Should OFPP_NORMAL update the MAC learning table? We want to update it
214 * if we are actually processing a packet, or if we are accounting for
215 * packets that the datapath has processed, but not if we are just
219 /* Should "learn" actions update the flow table? We want to update it if
220 * we are actually processing a packet, or in most cases if we are
221 * accounting for packets that the datapath has processed, but not if we
222 * are just revalidating. */
225 /* Cookie of the currently matching rule, or 0. */
228 /* If nonnull, called just before executing a resubmit action.
230 * This is normally null so the client has to set it manually after
231 * calling action_xlate_ctx_init(). */
232 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *);
234 /* xlate_actions() initializes and uses these members. The client might want
235 * to look at them after it returns. */
237 struct ofpbuf *odp_actions; /* Datapath actions. */
238 tag_type tags; /* Tags associated with actions. */
239 bool may_set_up_flow; /* True ordinarily; false if the actions must
240 * be reassessed for every packet. */
241 bool has_learn; /* Actions include NXAST_LEARN? */
242 bool has_normal; /* Actions output to OFPP_NORMAL? */
243 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
244 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
246 /* xlate_actions() initializes and uses these members, but the client has no
247 * reason to look at them. */
249 int recurse; /* Recursion level, via xlate_table_action. */
250 struct flow base_flow; /* Flow at the last commit. */
251 uint32_t orig_skb_priority; /* Priority when packet arrived. */
252 uint8_t table_id; /* OpenFlow table ID where flow was found. */
253 uint32_t sflow_n_outputs; /* Number of output ports. */
254 uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
255 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
256 bool exit; /* No further actions should be processed. */
259 static void action_xlate_ctx_init(struct action_xlate_ctx *,
260 struct ofproto_dpif *, const struct flow *,
261 ovs_be16 initial_tci, ovs_be64 cookie,
262 const struct ofpbuf *);
263 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
264 const union ofp_action *in, size_t n_in);
266 /* An exact-match instantiation of an OpenFlow flow.
268 * A facet associates a "struct flow", which represents the Open vSwitch
269 * userspace idea of an exact-match flow, with one or more subfacets. Each
270 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
271 * the facet. When the kernel module (or other dpif implementation) and Open
272 * vSwitch userspace agree on the definition of a flow key, there is exactly
273 * one subfacet per facet. If the dpif implementation supports more-specific
274 * flow matching than userspace, however, a facet can have more than one
275 * subfacet, each of which corresponds to some distinction in flow that
276 * userspace simply doesn't understand.
278 * Flow expiration works in terms of subfacets, so a facet must have at least
279 * one subfacet or it will never expire, leaking memory. */
282 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
283 struct list list_node; /* In owning rule's 'facets' list. */
284 struct rule_dpif *rule; /* Owning rule. */
287 struct list subfacets;
288 long long int used; /* Time last used; time created if not used. */
295 * - Do include packets and bytes sent "by hand", e.g. with
298 * - Do include packets and bytes that were obtained from the datapath
299 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
300 * DPIF_FP_ZERO_STATS).
302 * - Do not include packets or bytes that can be obtained from the
303 * datapath for any existing subfacet.
305 uint64_t packet_count; /* Number of packets received. */
306 uint64_t byte_count; /* Number of bytes received. */
308 /* Resubmit statistics. */
309 uint64_t prev_packet_count; /* Number of packets from last stats push. */
310 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
311 long long int prev_used; /* Used time from last stats push. */
314 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
315 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
317 /* Properties of datapath actions.
319 * Every subfacet has its own actions because actions can differ slightly
320 * between splintered and non-splintered subfacets due to the VLAN tag
321 * being initially different (present vs. absent). All of them have these
322 * properties in common so we just store one copy of them here. */
323 bool may_install; /* Reassess actions for every packet? */
324 bool has_learn; /* Actions include NXAST_LEARN? */
325 bool has_normal; /* Actions output to OFPP_NORMAL? */
326 tag_type tags; /* Tags that would require revalidation. */
327 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
330 static struct facet *facet_create(struct rule_dpif *, const struct flow *);
331 static void facet_remove(struct ofproto_dpif *, struct facet *);
332 static void facet_free(struct facet *);
334 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
335 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
336 const struct flow *);
337 static bool facet_revalidate(struct ofproto_dpif *, struct facet *);
339 static void facet_flush_stats(struct ofproto_dpif *, struct facet *);
341 static void facet_update_time(struct ofproto_dpif *, struct facet *,
343 static void facet_reset_counters(struct facet *);
344 static void facet_push_stats(struct facet *);
345 static void facet_account(struct ofproto_dpif *, struct facet *,
348 static bool facet_is_controller_flow(struct facet *);
350 /* A dpif flow and actions associated with a facet.
352 * See also the large comment on struct facet. */
355 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
356 struct list list_node; /* In struct facet's 'facets' list. */
357 struct facet *facet; /* Owning facet. */
361 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
362 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
363 * regenerate the ODP flow key from ->facet->flow. */
364 enum odp_key_fitness key_fitness;
368 long long int used; /* Time last used; time created if not used. */
370 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
371 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
375 * These should be essentially identical for every subfacet in a facet, but
376 * may differ in trivial ways due to VLAN splinters. */
377 size_t actions_len; /* Number of bytes in actions[]. */
378 struct nlattr *actions; /* Datapath actions. */
380 bool installed; /* Installed in datapath? */
382 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
383 * splinters can cause it to differ. This value should be removed when
384 * the VLAN splinters feature is no longer needed. */
385 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
388 static struct subfacet *subfacet_create(struct ofproto_dpif *, struct facet *,
389 enum odp_key_fitness,
390 const struct nlattr *key,
391 size_t key_len, ovs_be16 initial_tci);
392 static struct subfacet *subfacet_find(struct ofproto_dpif *,
393 const struct nlattr *key, size_t key_len);
394 static void subfacet_destroy(struct ofproto_dpif *, struct subfacet *);
395 static void subfacet_destroy__(struct ofproto_dpif *, struct subfacet *);
396 static void subfacet_reset_dp_stats(struct subfacet *,
397 struct dpif_flow_stats *);
398 static void subfacet_update_time(struct ofproto_dpif *, struct subfacet *,
400 static void subfacet_update_stats(struct ofproto_dpif *, struct subfacet *,
401 const struct dpif_flow_stats *);
402 static void subfacet_make_actions(struct ofproto_dpif *, struct subfacet *,
403 const struct ofpbuf *packet);
404 static int subfacet_install(struct ofproto_dpif *, struct subfacet *,
405 const struct nlattr *actions, size_t actions_len,
406 struct dpif_flow_stats *);
407 static void subfacet_uninstall(struct ofproto_dpif *, struct subfacet *);
413 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
414 struct list bundle_node; /* In struct ofbundle's "ports" list. */
415 struct cfm *cfm; /* Connectivity Fault Management, if any. */
416 tag_type tag; /* Tag associated with this port. */
417 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
418 bool may_enable; /* May be enabled in bonds. */
421 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
422 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
423 long long int stp_state_entered;
425 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
427 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
429 * This is deprecated. It is only for compatibility with broken device
430 * drivers in old versions of Linux that do not properly support VLANs when
431 * VLAN devices are not used. When broken device drivers are no longer in
432 * widespread use, we will delete these interfaces. */
433 uint16_t realdev_ofp_port;
437 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
438 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
439 * traffic egressing the 'ofport' with that priority should be marked with. */
440 struct priority_to_dscp {
441 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
442 uint32_t priority; /* Priority of this queue (see struct flow). */
444 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
447 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
449 * This is deprecated. It is only for compatibility with broken device drivers
450 * in old versions of Linux that do not properly support VLANs when VLAN
451 * devices are not used. When broken device drivers are no longer in
452 * widespread use, we will delete these interfaces. */
453 struct vlan_splinter {
454 struct hmap_node realdev_vid_node;
455 struct hmap_node vlandev_node;
456 uint16_t realdev_ofp_port;
457 uint16_t vlandev_ofp_port;
461 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
462 uint32_t realdev, ovs_be16 vlan_tci);
463 static uint16_t vsp_vlandev_to_realdev(const struct ofproto_dpif *,
464 uint16_t vlandev, int *vid);
465 static void vsp_remove(struct ofport_dpif *);
466 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
468 static struct ofport_dpif *
469 ofport_dpif_cast(const struct ofport *ofport)
471 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
472 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
475 static void port_run(struct ofport_dpif *);
476 static void port_wait(struct ofport_dpif *);
477 static int set_cfm(struct ofport *, const struct cfm_settings *);
478 static void ofport_clear_priorities(struct ofport_dpif *);
480 struct dpif_completion {
481 struct list list_node;
482 struct ofoperation *op;
485 /* Extra information about a classifier table.
486 * Currently used just for optimized flow revalidation. */
488 /* If either of these is nonnull, then this table has a form that allows
489 * flows to be tagged to avoid revalidating most flows for the most common
490 * kinds of flow table changes. */
491 struct cls_table *catchall_table; /* Table that wildcards all fields. */
492 struct cls_table *other_table; /* Table with any other wildcard set. */
493 uint32_t basis; /* Keeps each table's tags separate. */
496 struct ofproto_dpif {
497 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
506 struct netflow *netflow;
507 struct dpif_sflow *sflow;
508 struct hmap bundles; /* Contains "struct ofbundle"s. */
509 struct mac_learning *ml;
510 struct ofmirror *mirrors[MAX_MIRRORS];
511 bool has_bonded_bundles;
514 struct timer next_expiration;
518 struct hmap subfacets;
521 struct table_dpif tables[N_TABLES];
522 bool need_revalidate;
523 struct tag_set revalidate_set;
525 /* Support for debugging async flow mods. */
526 struct list completions;
528 bool has_bundle_action; /* True when the first bundle action appears. */
529 struct netdev_stats stats; /* To account packets generated and consumed in
534 long long int stp_last_tick;
536 /* VLAN splinters. */
537 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
538 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
541 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
542 * for debugging the asynchronous flow_mod implementation.) */
545 /* All existing ofproto_dpif instances, indexed by ->up.name. */
546 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
548 static void ofproto_dpif_unixctl_init(void);
550 static struct ofproto_dpif *
551 ofproto_dpif_cast(const struct ofproto *ofproto)
553 assert(ofproto->ofproto_class == &ofproto_dpif_class);
554 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
557 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
559 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
562 /* Packet processing. */
563 static void update_learning_table(struct ofproto_dpif *,
564 const struct flow *, int vlan,
567 #define FLOW_MISS_MAX_BATCH 50
568 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
570 /* Flow expiration. */
571 static int expire(struct ofproto_dpif *);
574 static void send_netflow_active_timeouts(struct ofproto_dpif *);
577 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
579 compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions,
580 const struct flow *, uint32_t odp_port);
581 static void add_mirror_actions(struct action_xlate_ctx *ctx,
582 const struct flow *flow);
583 /* Global variables. */
584 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
586 /* Factory functions. */
589 enumerate_types(struct sset *types)
591 dp_enumerate_types(types);
595 enumerate_names(const char *type, struct sset *names)
597 return dp_enumerate_names(type, names);
601 del(const char *type, const char *name)
606 error = dpif_open(name, type, &dpif);
608 error = dpif_delete(dpif);
614 /* Basic life-cycle. */
616 static struct ofproto *
619 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
624 dealloc(struct ofproto *ofproto_)
626 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
631 construct(struct ofproto *ofproto_, int *n_tablesp)
633 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
634 const char *name = ofproto->up.name;
638 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
640 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
644 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
645 ofproto->n_matches = 0;
647 dpif_flow_flush(ofproto->dpif);
648 dpif_recv_purge(ofproto->dpif);
650 error = dpif_recv_set_mask(ofproto->dpif,
651 ((1u << DPIF_UC_MISS) |
652 (1u << DPIF_UC_ACTION)));
654 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
655 dpif_close(ofproto->dpif);
659 ofproto->netflow = NULL;
660 ofproto->sflow = NULL;
662 hmap_init(&ofproto->bundles);
663 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
664 for (i = 0; i < MAX_MIRRORS; i++) {
665 ofproto->mirrors[i] = NULL;
667 ofproto->has_bonded_bundles = false;
669 timer_set_duration(&ofproto->next_expiration, 1000);
671 hmap_init(&ofproto->facets);
672 hmap_init(&ofproto->subfacets);
674 for (i = 0; i < N_TABLES; i++) {
675 struct table_dpif *table = &ofproto->tables[i];
677 table->catchall_table = NULL;
678 table->other_table = NULL;
679 table->basis = random_uint32();
681 ofproto->need_revalidate = false;
682 tag_set_init(&ofproto->revalidate_set);
684 list_init(&ofproto->completions);
686 ofproto_dpif_unixctl_init();
688 ofproto->has_bundle_action = false;
690 hmap_init(&ofproto->vlandev_map);
691 hmap_init(&ofproto->realdev_vid_map);
693 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
694 hash_string(ofproto->up.name, 0));
696 *n_tablesp = N_TABLES;
697 memset(&ofproto->stats, 0, sizeof ofproto->stats);
702 complete_operations(struct ofproto_dpif *ofproto)
704 struct dpif_completion *c, *next;
706 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
707 ofoperation_complete(c->op, 0);
708 list_remove(&c->list_node);
714 destruct(struct ofproto *ofproto_)
716 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
717 struct rule_dpif *rule, *next_rule;
718 struct classifier *table;
721 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
722 complete_operations(ofproto);
724 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
725 struct cls_cursor cursor;
727 cls_cursor_init(&cursor, table, NULL);
728 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
729 ofproto_rule_destroy(&rule->up);
733 for (i = 0; i < MAX_MIRRORS; i++) {
734 mirror_destroy(ofproto->mirrors[i]);
737 netflow_destroy(ofproto->netflow);
738 dpif_sflow_destroy(ofproto->sflow);
739 hmap_destroy(&ofproto->bundles);
740 mac_learning_destroy(ofproto->ml);
742 hmap_destroy(&ofproto->facets);
743 hmap_destroy(&ofproto->subfacets);
745 hmap_destroy(&ofproto->vlandev_map);
746 hmap_destroy(&ofproto->realdev_vid_map);
748 dpif_close(ofproto->dpif);
752 run_fast(struct ofproto *ofproto_)
754 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
757 /* Handle one or more batches of upcalls, until there's nothing left to do
758 * or until we do a fixed total amount of work.
760 * We do work in batches because it can be much cheaper to set up a number
761 * of flows and fire off their patches all at once. We do multiple batches
762 * because in some cases handling a packet can cause another packet to be
763 * queued almost immediately as part of the return flow. Both
764 * optimizations can make major improvements on some benchmarks and
765 * presumably for real traffic as well. */
767 while (work < FLOW_MISS_MAX_BATCH) {
768 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
778 run(struct ofproto *ofproto_)
780 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
781 struct ofport_dpif *ofport;
782 struct ofbundle *bundle;
786 complete_operations(ofproto);
788 dpif_run(ofproto->dpif);
790 error = run_fast(ofproto_);
795 if (timer_expired(&ofproto->next_expiration)) {
796 int delay = expire(ofproto);
797 timer_set_duration(&ofproto->next_expiration, delay);
800 if (ofproto->netflow) {
801 if (netflow_run(ofproto->netflow)) {
802 send_netflow_active_timeouts(ofproto);
805 if (ofproto->sflow) {
806 dpif_sflow_run(ofproto->sflow);
809 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
812 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
817 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
819 /* Now revalidate if there's anything to do. */
820 if (ofproto->need_revalidate
821 || !tag_set_is_empty(&ofproto->revalidate_set)) {
822 struct tag_set revalidate_set = ofproto->revalidate_set;
823 bool revalidate_all = ofproto->need_revalidate;
824 struct facet *facet, *next;
826 /* Clear the revalidation flags. */
827 tag_set_init(&ofproto->revalidate_set);
828 ofproto->need_revalidate = false;
830 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
832 || tag_set_intersects(&revalidate_set, facet->tags)) {
833 facet_revalidate(ofproto, facet);
842 wait(struct ofproto *ofproto_)
844 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
845 struct ofport_dpif *ofport;
846 struct ofbundle *bundle;
848 if (!clogged && !list_is_empty(&ofproto->completions)) {
849 poll_immediate_wake();
852 dpif_wait(ofproto->dpif);
853 dpif_recv_wait(ofproto->dpif);
854 if (ofproto->sflow) {
855 dpif_sflow_wait(ofproto->sflow);
857 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
858 poll_immediate_wake();
860 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
863 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
866 if (ofproto->netflow) {
867 netflow_wait(ofproto->netflow);
869 mac_learning_wait(ofproto->ml);
871 if (ofproto->need_revalidate) {
872 /* Shouldn't happen, but if it does just go around again. */
873 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
874 poll_immediate_wake();
876 timer_wait(&ofproto->next_expiration);
881 flush(struct ofproto *ofproto_)
883 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
884 struct facet *facet, *next_facet;
886 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
887 /* Mark the facet as not installed so that facet_remove() doesn't
888 * bother trying to uninstall it. There is no point in uninstalling it
889 * individually since we are about to blow away all the facets with
890 * dpif_flow_flush(). */
891 struct subfacet *subfacet;
893 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
894 subfacet->installed = false;
895 subfacet->dp_packet_count = 0;
896 subfacet->dp_byte_count = 0;
898 facet_remove(ofproto, facet);
900 dpif_flow_flush(ofproto->dpif);
904 get_features(struct ofproto *ofproto_ OVS_UNUSED,
905 bool *arp_match_ip, uint32_t *actions)
907 *arp_match_ip = true;
908 *actions = ((1u << OFPAT_OUTPUT) |
909 (1u << OFPAT_SET_VLAN_VID) |
910 (1u << OFPAT_SET_VLAN_PCP) |
911 (1u << OFPAT_STRIP_VLAN) |
912 (1u << OFPAT_SET_DL_SRC) |
913 (1u << OFPAT_SET_DL_DST) |
914 (1u << OFPAT_SET_NW_SRC) |
915 (1u << OFPAT_SET_NW_DST) |
916 (1u << OFPAT_SET_NW_TOS) |
917 (1u << OFPAT_SET_TP_SRC) |
918 (1u << OFPAT_SET_TP_DST) |
919 (1u << OFPAT_ENQUEUE));
923 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
925 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
926 struct dpif_dp_stats s;
928 strcpy(ots->name, "classifier");
930 dpif_get_dp_stats(ofproto->dpif, &s);
931 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
932 put_32aligned_be64(&ots->matched_count,
933 htonll(s.n_hit + ofproto->n_matches));
936 static struct ofport *
939 struct ofport_dpif *port = xmalloc(sizeof *port);
944 port_dealloc(struct ofport *port_)
946 struct ofport_dpif *port = ofport_dpif_cast(port_);
951 port_construct(struct ofport *port_)
953 struct ofport_dpif *port = ofport_dpif_cast(port_);
954 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
956 ofproto->need_revalidate = true;
957 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
960 port->tag = tag_create_random();
961 port->may_enable = true;
962 port->stp_port = NULL;
963 port->stp_state = STP_DISABLED;
964 hmap_init(&port->priorities);
965 port->realdev_ofp_port = 0;
966 port->vlandev_vid = 0;
968 if (ofproto->sflow) {
969 dpif_sflow_add_port(ofproto->sflow, port_);
976 port_destruct(struct ofport *port_)
978 struct ofport_dpif *port = ofport_dpif_cast(port_);
979 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
981 ofproto->need_revalidate = true;
982 bundle_remove(port_);
983 set_cfm(port_, NULL);
984 if (ofproto->sflow) {
985 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
988 ofport_clear_priorities(port);
989 hmap_destroy(&port->priorities);
993 port_modified(struct ofport *port_)
995 struct ofport_dpif *port = ofport_dpif_cast(port_);
997 if (port->bundle && port->bundle->bond) {
998 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1003 port_reconfigured(struct ofport *port_, ovs_be32 old_config)
1005 struct ofport_dpif *port = ofport_dpif_cast(port_);
1006 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1007 ovs_be32 changed = old_config ^ port->up.opp.config;
1009 if (changed & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
1010 OFPPC_NO_FWD | OFPPC_NO_FLOOD)) {
1011 ofproto->need_revalidate = true;
1013 if (changed & htonl(OFPPC_NO_FLOOD) && port->bundle) {
1014 bundle_update(port->bundle);
1020 set_sflow(struct ofproto *ofproto_,
1021 const struct ofproto_sflow_options *sflow_options)
1023 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1024 struct dpif_sflow *ds = ofproto->sflow;
1026 if (sflow_options) {
1028 struct ofport_dpif *ofport;
1030 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1031 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1032 dpif_sflow_add_port(ds, &ofport->up);
1034 ofproto->need_revalidate = true;
1036 dpif_sflow_set_options(ds, sflow_options);
1039 dpif_sflow_destroy(ds);
1040 ofproto->need_revalidate = true;
1041 ofproto->sflow = NULL;
1048 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1050 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1057 struct ofproto_dpif *ofproto;
1059 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1060 ofproto->need_revalidate = true;
1061 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1064 if (cfm_configure(ofport->cfm, s)) {
1070 cfm_destroy(ofport->cfm);
1076 get_cfm_fault(const struct ofport *ofport_)
1078 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1080 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1084 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1087 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1090 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1097 /* Spanning Tree. */
1100 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1102 struct ofproto_dpif *ofproto = ofproto_;
1103 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1104 struct ofport_dpif *ofport;
1106 ofport = stp_port_get_aux(sp);
1108 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1109 ofproto->up.name, port_num);
1111 struct eth_header *eth = pkt->l2;
1113 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1114 if (eth_addr_is_zero(eth->eth_src)) {
1115 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1116 "with unknown MAC", ofproto->up.name, port_num);
1118 send_packet(ofport, pkt);
1124 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1126 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1128 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1130 /* Only revalidate flows if the configuration changed. */
1131 if (!s != !ofproto->stp) {
1132 ofproto->need_revalidate = true;
1136 if (!ofproto->stp) {
1137 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1138 send_bpdu_cb, ofproto);
1139 ofproto->stp_last_tick = time_msec();
1142 stp_set_bridge_id(ofproto->stp, s->system_id);
1143 stp_set_bridge_priority(ofproto->stp, s->priority);
1144 stp_set_hello_time(ofproto->stp, s->hello_time);
1145 stp_set_max_age(ofproto->stp, s->max_age);
1146 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1148 struct ofport *ofport;
1150 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1151 set_stp_port(ofport, NULL);
1154 stp_destroy(ofproto->stp);
1155 ofproto->stp = NULL;
1162 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1164 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1168 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1169 s->designated_root = stp_get_designated_root(ofproto->stp);
1170 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1179 update_stp_port_state(struct ofport_dpif *ofport)
1181 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1182 enum stp_state state;
1184 /* Figure out new state. */
1185 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1189 if (ofport->stp_state != state) {
1193 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1194 netdev_get_name(ofport->up.netdev),
1195 stp_state_name(ofport->stp_state),
1196 stp_state_name(state));
1197 if (stp_learn_in_state(ofport->stp_state)
1198 != stp_learn_in_state(state)) {
1199 /* xxx Learning action flows should also be flushed. */
1200 mac_learning_flush(ofproto->ml);
1202 fwd_change = stp_forward_in_state(ofport->stp_state)
1203 != stp_forward_in_state(state);
1205 ofproto->need_revalidate = true;
1206 ofport->stp_state = state;
1207 ofport->stp_state_entered = time_msec();
1209 if (fwd_change && ofport->bundle) {
1210 bundle_update(ofport->bundle);
1213 /* Update the STP state bits in the OpenFlow port description. */
1214 of_state = (ofport->up.opp.state & htonl(~OFPPS_STP_MASK))
1215 | htonl(state == STP_LISTENING ? OFPPS_STP_LISTEN
1216 : state == STP_LEARNING ? OFPPS_STP_LEARN
1217 : state == STP_FORWARDING ? OFPPS_STP_FORWARD
1218 : state == STP_BLOCKING ? OFPPS_STP_BLOCK
1220 ofproto_port_set_state(&ofport->up, of_state);
1224 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1225 * caller is responsible for assigning STP port numbers and ensuring
1226 * there are no duplicates. */
1228 set_stp_port(struct ofport *ofport_,
1229 const struct ofproto_port_stp_settings *s)
1231 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1232 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1233 struct stp_port *sp = ofport->stp_port;
1235 if (!s || !s->enable) {
1237 ofport->stp_port = NULL;
1238 stp_port_disable(sp);
1239 update_stp_port_state(ofport);
1242 } else if (sp && stp_port_no(sp) != s->port_num
1243 && ofport == stp_port_get_aux(sp)) {
1244 /* The port-id changed, so disable the old one if it's not
1245 * already in use by another port. */
1246 stp_port_disable(sp);
1249 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1250 stp_port_enable(sp);
1252 stp_port_set_aux(sp, ofport);
1253 stp_port_set_priority(sp, s->priority);
1254 stp_port_set_path_cost(sp, s->path_cost);
1256 update_stp_port_state(ofport);
1262 get_stp_port_status(struct ofport *ofport_,
1263 struct ofproto_port_stp_status *s)
1265 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1266 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1267 struct stp_port *sp = ofport->stp_port;
1269 if (!ofproto->stp || !sp) {
1275 s->port_id = stp_port_get_id(sp);
1276 s->state = stp_port_get_state(sp);
1277 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1278 s->role = stp_port_get_role(sp);
1279 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1285 stp_run(struct ofproto_dpif *ofproto)
1288 long long int now = time_msec();
1289 long long int elapsed = now - ofproto->stp_last_tick;
1290 struct stp_port *sp;
1293 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1294 ofproto->stp_last_tick = now;
1296 while (stp_get_changed_port(ofproto->stp, &sp)) {
1297 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1300 update_stp_port_state(ofport);
1307 stp_wait(struct ofproto_dpif *ofproto)
1310 poll_timer_wait(1000);
1314 /* Returns true if STP should process 'flow'. */
1316 stp_should_process_flow(const struct flow *flow)
1318 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1322 stp_process_packet(const struct ofport_dpif *ofport,
1323 const struct ofpbuf *packet)
1325 struct ofpbuf payload = *packet;
1326 struct eth_header *eth = payload.data;
1327 struct stp_port *sp = ofport->stp_port;
1329 /* Sink packets on ports that have STP disabled when the bridge has
1331 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1335 /* Trim off padding on payload. */
1336 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1337 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1340 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1341 stp_received_bpdu(sp, payload.data, payload.size);
1345 static struct priority_to_dscp *
1346 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1348 struct priority_to_dscp *pdscp;
1351 hash = hash_int(priority, 0);
1352 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1353 if (pdscp->priority == priority) {
1361 ofport_clear_priorities(struct ofport_dpif *ofport)
1363 struct priority_to_dscp *pdscp, *next;
1365 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1366 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1372 set_queues(struct ofport *ofport_,
1373 const struct ofproto_port_queue *qdscp_list,
1376 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1377 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1378 struct hmap new = HMAP_INITIALIZER(&new);
1381 for (i = 0; i < n_qdscp; i++) {
1382 struct priority_to_dscp *pdscp;
1386 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1387 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1392 pdscp = get_priority(ofport, priority);
1394 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1396 pdscp = xmalloc(sizeof *pdscp);
1397 pdscp->priority = priority;
1399 ofproto->need_revalidate = true;
1402 if (pdscp->dscp != dscp) {
1404 ofproto->need_revalidate = true;
1407 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1410 if (!hmap_is_empty(&ofport->priorities)) {
1411 ofport_clear_priorities(ofport);
1412 ofproto->need_revalidate = true;
1415 hmap_swap(&new, &ofport->priorities);
1423 /* Expires all MAC learning entries associated with 'bundle' and forces its
1424 * ofproto to revalidate every flow.
1426 * Normally MAC learning entries are removed only from the ofproto associated
1427 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1428 * are removed from every ofproto. When patch ports and SLB bonds are in use
1429 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1430 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1431 * with the host from which it migrated. */
1433 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1435 struct ofproto_dpif *ofproto = bundle->ofproto;
1436 struct mac_learning *ml = ofproto->ml;
1437 struct mac_entry *mac, *next_mac;
1439 ofproto->need_revalidate = true;
1440 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1441 if (mac->port.p == bundle) {
1443 struct ofproto_dpif *o;
1445 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1447 struct mac_entry *e;
1449 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1452 tag_set_add(&o->revalidate_set, e->tag);
1453 mac_learning_expire(o->ml, e);
1459 mac_learning_expire(ml, mac);
1464 static struct ofbundle *
1465 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1467 struct ofbundle *bundle;
1469 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1470 &ofproto->bundles) {
1471 if (bundle->aux == aux) {
1478 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1479 * ones that are found to 'bundles'. */
1481 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1482 void **auxes, size_t n_auxes,
1483 struct hmapx *bundles)
1487 hmapx_init(bundles);
1488 for (i = 0; i < n_auxes; i++) {
1489 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1491 hmapx_add(bundles, bundle);
1497 bundle_update(struct ofbundle *bundle)
1499 struct ofport_dpif *port;
1501 bundle->floodable = true;
1502 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1503 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
1504 bundle->floodable = false;
1511 bundle_del_port(struct ofport_dpif *port)
1513 struct ofbundle *bundle = port->bundle;
1515 bundle->ofproto->need_revalidate = true;
1517 list_remove(&port->bundle_node);
1518 port->bundle = NULL;
1521 lacp_slave_unregister(bundle->lacp, port);
1524 bond_slave_unregister(bundle->bond, port);
1527 bundle_update(bundle);
1531 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1532 struct lacp_slave_settings *lacp,
1533 uint32_t bond_stable_id)
1535 struct ofport_dpif *port;
1537 port = get_ofp_port(bundle->ofproto, ofp_port);
1542 if (port->bundle != bundle) {
1543 bundle->ofproto->need_revalidate = true;
1545 bundle_del_port(port);
1548 port->bundle = bundle;
1549 list_push_back(&bundle->ports, &port->bundle_node);
1550 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
1551 bundle->floodable = false;
1555 port->bundle->ofproto->need_revalidate = true;
1556 lacp_slave_register(bundle->lacp, port, lacp);
1559 port->bond_stable_id = bond_stable_id;
1565 bundle_destroy(struct ofbundle *bundle)
1567 struct ofproto_dpif *ofproto;
1568 struct ofport_dpif *port, *next_port;
1575 ofproto = bundle->ofproto;
1576 for (i = 0; i < MAX_MIRRORS; i++) {
1577 struct ofmirror *m = ofproto->mirrors[i];
1579 if (m->out == bundle) {
1581 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1582 || hmapx_find_and_delete(&m->dsts, bundle)) {
1583 ofproto->need_revalidate = true;
1588 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1589 bundle_del_port(port);
1592 bundle_flush_macs(bundle, true);
1593 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1595 free(bundle->trunks);
1596 lacp_destroy(bundle->lacp);
1597 bond_destroy(bundle->bond);
1602 bundle_set(struct ofproto *ofproto_, void *aux,
1603 const struct ofproto_bundle_settings *s)
1605 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1606 bool need_flush = false;
1607 struct ofport_dpif *port;
1608 struct ofbundle *bundle;
1609 unsigned long *trunks;
1615 bundle_destroy(bundle_lookup(ofproto, aux));
1619 assert(s->n_slaves == 1 || s->bond != NULL);
1620 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1622 bundle = bundle_lookup(ofproto, aux);
1624 bundle = xmalloc(sizeof *bundle);
1626 bundle->ofproto = ofproto;
1627 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1628 hash_pointer(aux, 0));
1630 bundle->name = NULL;
1632 list_init(&bundle->ports);
1633 bundle->vlan_mode = PORT_VLAN_TRUNK;
1635 bundle->trunks = NULL;
1636 bundle->use_priority_tags = s->use_priority_tags;
1637 bundle->lacp = NULL;
1638 bundle->bond = NULL;
1640 bundle->floodable = true;
1642 bundle->src_mirrors = 0;
1643 bundle->dst_mirrors = 0;
1644 bundle->mirror_out = 0;
1647 if (!bundle->name || strcmp(s->name, bundle->name)) {
1649 bundle->name = xstrdup(s->name);
1654 if (!bundle->lacp) {
1655 ofproto->need_revalidate = true;
1656 bundle->lacp = lacp_create();
1658 lacp_configure(bundle->lacp, s->lacp);
1660 lacp_destroy(bundle->lacp);
1661 bundle->lacp = NULL;
1664 /* Update set of ports. */
1666 for (i = 0; i < s->n_slaves; i++) {
1667 if (!bundle_add_port(bundle, s->slaves[i],
1668 s->lacp ? &s->lacp_slaves[i] : NULL,
1669 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1673 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1674 struct ofport_dpif *next_port;
1676 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1677 for (i = 0; i < s->n_slaves; i++) {
1678 if (s->slaves[i] == port->up.ofp_port) {
1683 bundle_del_port(port);
1687 assert(list_size(&bundle->ports) <= s->n_slaves);
1689 if (list_is_empty(&bundle->ports)) {
1690 bundle_destroy(bundle);
1694 /* Set VLAN tagging mode */
1695 if (s->vlan_mode != bundle->vlan_mode
1696 || s->use_priority_tags != bundle->use_priority_tags) {
1697 bundle->vlan_mode = s->vlan_mode;
1698 bundle->use_priority_tags = s->use_priority_tags;
1703 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1704 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1706 if (vlan != bundle->vlan) {
1707 bundle->vlan = vlan;
1711 /* Get trunked VLANs. */
1712 switch (s->vlan_mode) {
1713 case PORT_VLAN_ACCESS:
1717 case PORT_VLAN_TRUNK:
1718 trunks = (unsigned long *) s->trunks;
1721 case PORT_VLAN_NATIVE_UNTAGGED:
1722 case PORT_VLAN_NATIVE_TAGGED:
1723 if (vlan != 0 && (!s->trunks
1724 || !bitmap_is_set(s->trunks, vlan)
1725 || bitmap_is_set(s->trunks, 0))) {
1726 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1728 trunks = bitmap_clone(s->trunks, 4096);
1730 trunks = bitmap_allocate1(4096);
1732 bitmap_set1(trunks, vlan);
1733 bitmap_set0(trunks, 0);
1735 trunks = (unsigned long *) s->trunks;
1742 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1743 free(bundle->trunks);
1744 if (trunks == s->trunks) {
1745 bundle->trunks = vlan_bitmap_clone(trunks);
1747 bundle->trunks = trunks;
1752 if (trunks != s->trunks) {
1757 if (!list_is_short(&bundle->ports)) {
1758 bundle->ofproto->has_bonded_bundles = true;
1760 if (bond_reconfigure(bundle->bond, s->bond)) {
1761 ofproto->need_revalidate = true;
1764 bundle->bond = bond_create(s->bond);
1765 ofproto->need_revalidate = true;
1768 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1769 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1773 bond_destroy(bundle->bond);
1774 bundle->bond = NULL;
1777 /* If we changed something that would affect MAC learning, un-learn
1778 * everything on this port and force flow revalidation. */
1780 bundle_flush_macs(bundle, false);
1787 bundle_remove(struct ofport *port_)
1789 struct ofport_dpif *port = ofport_dpif_cast(port_);
1790 struct ofbundle *bundle = port->bundle;
1793 bundle_del_port(port);
1794 if (list_is_empty(&bundle->ports)) {
1795 bundle_destroy(bundle);
1796 } else if (list_is_short(&bundle->ports)) {
1797 bond_destroy(bundle->bond);
1798 bundle->bond = NULL;
1804 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
1806 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1807 struct ofport_dpif *port = port_;
1808 uint8_t ea[ETH_ADDR_LEN];
1811 error = netdev_get_etheraddr(port->up.netdev, ea);
1813 struct ofpbuf packet;
1816 ofpbuf_init(&packet, 0);
1817 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1819 memcpy(packet_pdu, pdu, pdu_size);
1821 send_packet(port, &packet);
1822 ofpbuf_uninit(&packet);
1824 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1825 "%s (%s)", port->bundle->name,
1826 netdev_get_name(port->up.netdev), strerror(error));
1831 bundle_send_learning_packets(struct ofbundle *bundle)
1833 struct ofproto_dpif *ofproto = bundle->ofproto;
1834 int error, n_packets, n_errors;
1835 struct mac_entry *e;
1837 error = n_packets = n_errors = 0;
1838 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1839 if (e->port.p != bundle) {
1840 struct ofpbuf *learning_packet;
1841 struct ofport_dpif *port;
1845 /* The assignment to "port" is unnecessary but makes "grep"ing for
1846 * struct ofport_dpif more effective. */
1847 learning_packet = bond_compose_learning_packet(bundle->bond,
1851 ret = send_packet(port, learning_packet);
1852 ofpbuf_delete(learning_packet);
1862 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1863 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1864 "packets, last error was: %s",
1865 bundle->name, n_errors, n_packets, strerror(error));
1867 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1868 bundle->name, n_packets);
1873 bundle_run(struct ofbundle *bundle)
1876 lacp_run(bundle->lacp, send_pdu_cb);
1879 struct ofport_dpif *port;
1881 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1882 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1885 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1886 lacp_negotiated(bundle->lacp));
1887 if (bond_should_send_learning_packets(bundle->bond)) {
1888 bundle_send_learning_packets(bundle);
1894 bundle_wait(struct ofbundle *bundle)
1897 lacp_wait(bundle->lacp);
1900 bond_wait(bundle->bond);
1907 mirror_scan(struct ofproto_dpif *ofproto)
1911 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1912 if (!ofproto->mirrors[idx]) {
1919 static struct ofmirror *
1920 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1924 for (i = 0; i < MAX_MIRRORS; i++) {
1925 struct ofmirror *mirror = ofproto->mirrors[i];
1926 if (mirror && mirror->aux == aux) {
1934 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
1936 mirror_update_dups(struct ofproto_dpif *ofproto)
1940 for (i = 0; i < MAX_MIRRORS; i++) {
1941 struct ofmirror *m = ofproto->mirrors[i];
1944 m->dup_mirrors = MIRROR_MASK_C(1) << i;
1948 for (i = 0; i < MAX_MIRRORS; i++) {
1949 struct ofmirror *m1 = ofproto->mirrors[i];
1956 for (j = i + 1; j < MAX_MIRRORS; j++) {
1957 struct ofmirror *m2 = ofproto->mirrors[j];
1959 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
1960 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
1961 m2->dup_mirrors |= m1->dup_mirrors;
1968 mirror_set(struct ofproto *ofproto_, void *aux,
1969 const struct ofproto_mirror_settings *s)
1971 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1972 mirror_mask_t mirror_bit;
1973 struct ofbundle *bundle;
1974 struct ofmirror *mirror;
1975 struct ofbundle *out;
1976 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
1977 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
1980 mirror = mirror_lookup(ofproto, aux);
1982 mirror_destroy(mirror);
1988 idx = mirror_scan(ofproto);
1990 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
1992 ofproto->up.name, MAX_MIRRORS, s->name);
1996 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
1997 mirror->ofproto = ofproto;
2000 mirror->out_vlan = -1;
2001 mirror->name = NULL;
2004 if (!mirror->name || strcmp(s->name, mirror->name)) {
2006 mirror->name = xstrdup(s->name);
2009 /* Get the new configuration. */
2010 if (s->out_bundle) {
2011 out = bundle_lookup(ofproto, s->out_bundle);
2013 mirror_destroy(mirror);
2019 out_vlan = s->out_vlan;
2021 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2022 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2024 /* If the configuration has not changed, do nothing. */
2025 if (hmapx_equals(&srcs, &mirror->srcs)
2026 && hmapx_equals(&dsts, &mirror->dsts)
2027 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2028 && mirror->out == out
2029 && mirror->out_vlan == out_vlan)
2031 hmapx_destroy(&srcs);
2032 hmapx_destroy(&dsts);
2036 hmapx_swap(&srcs, &mirror->srcs);
2037 hmapx_destroy(&srcs);
2039 hmapx_swap(&dsts, &mirror->dsts);
2040 hmapx_destroy(&dsts);
2042 free(mirror->vlans);
2043 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2046 mirror->out_vlan = out_vlan;
2048 /* Update bundles. */
2049 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2050 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2051 if (hmapx_contains(&mirror->srcs, bundle)) {
2052 bundle->src_mirrors |= mirror_bit;
2054 bundle->src_mirrors &= ~mirror_bit;
2057 if (hmapx_contains(&mirror->dsts, bundle)) {
2058 bundle->dst_mirrors |= mirror_bit;
2060 bundle->dst_mirrors &= ~mirror_bit;
2063 if (mirror->out == bundle) {
2064 bundle->mirror_out |= mirror_bit;
2066 bundle->mirror_out &= ~mirror_bit;
2070 ofproto->need_revalidate = true;
2071 mac_learning_flush(ofproto->ml);
2072 mirror_update_dups(ofproto);
2078 mirror_destroy(struct ofmirror *mirror)
2080 struct ofproto_dpif *ofproto;
2081 mirror_mask_t mirror_bit;
2082 struct ofbundle *bundle;
2088 ofproto = mirror->ofproto;
2089 ofproto->need_revalidate = true;
2090 mac_learning_flush(ofproto->ml);
2092 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2093 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2094 bundle->src_mirrors &= ~mirror_bit;
2095 bundle->dst_mirrors &= ~mirror_bit;
2096 bundle->mirror_out &= ~mirror_bit;
2099 hmapx_destroy(&mirror->srcs);
2100 hmapx_destroy(&mirror->dsts);
2101 free(mirror->vlans);
2103 ofproto->mirrors[mirror->idx] = NULL;
2107 mirror_update_dups(ofproto);
2111 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2112 uint64_t *packets, uint64_t *bytes)
2114 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2115 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2118 *packets = *bytes = UINT64_MAX;
2122 *packets = mirror->packet_count;
2123 *bytes = mirror->byte_count;
2129 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2131 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2132 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2133 ofproto->need_revalidate = true;
2134 mac_learning_flush(ofproto->ml);
2140 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2142 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2143 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2144 return bundle && bundle->mirror_out != 0;
2148 forward_bpdu_changed(struct ofproto *ofproto_)
2150 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2151 /* Revalidate cached flows whenever forward_bpdu option changes. */
2152 ofproto->need_revalidate = true;
2156 set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time)
2158 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2159 mac_learning_set_idle_time(ofproto->ml, idle_time);
2164 static struct ofport_dpif *
2165 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
2167 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2168 return ofport ? ofport_dpif_cast(ofport) : NULL;
2171 static struct ofport_dpif *
2172 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
2174 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2178 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2179 struct dpif_port *dpif_port)
2181 ofproto_port->name = dpif_port->name;
2182 ofproto_port->type = dpif_port->type;
2183 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2187 port_run(struct ofport_dpif *ofport)
2189 bool enable = netdev_get_carrier(ofport->up.netdev);
2192 cfm_run(ofport->cfm);
2194 if (cfm_should_send_ccm(ofport->cfm)) {
2195 struct ofpbuf packet;
2197 ofpbuf_init(&packet, 0);
2198 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr);
2199 send_packet(ofport, &packet);
2200 ofpbuf_uninit(&packet);
2203 enable = enable && !cfm_get_fault(ofport->cfm)
2204 && cfm_get_opup(ofport->cfm);
2207 if (ofport->bundle) {
2208 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2211 if (ofport->may_enable != enable) {
2212 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2214 if (ofproto->has_bundle_action) {
2215 ofproto->need_revalidate = true;
2219 ofport->may_enable = enable;
2223 port_wait(struct ofport_dpif *ofport)
2226 cfm_wait(ofport->cfm);
2231 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2232 struct ofproto_port *ofproto_port)
2234 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2235 struct dpif_port dpif_port;
2238 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2240 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2246 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2248 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2252 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2254 *ofp_portp = odp_port_to_ofp_port(odp_port);
2260 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2262 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2265 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2267 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2269 /* The caller is going to close ofport->up.netdev. If this is a
2270 * bonded port, then the bond is using that netdev, so remove it
2271 * from the bond. The client will need to reconfigure everything
2272 * after deleting ports, so then the slave will get re-added. */
2273 bundle_remove(&ofport->up);
2280 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2282 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2285 error = netdev_get_stats(ofport->up.netdev, stats);
2287 if (!error && ofport->odp_port == OVSP_LOCAL) {
2288 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2290 /* ofproto->stats.tx_packets represents packets that we created
2291 * internally and sent to some port (e.g. packets sent with
2292 * send_packet()). Account for them as if they had come from
2293 * OFPP_LOCAL and got forwarded. */
2295 if (stats->rx_packets != UINT64_MAX) {
2296 stats->rx_packets += ofproto->stats.tx_packets;
2299 if (stats->rx_bytes != UINT64_MAX) {
2300 stats->rx_bytes += ofproto->stats.tx_bytes;
2303 /* ofproto->stats.rx_packets represents packets that were received on
2304 * some port and we processed internally and dropped (e.g. STP).
2305 * Account fro them as if they had been forwarded to OFPP_LOCAL. */
2307 if (stats->tx_packets != UINT64_MAX) {
2308 stats->tx_packets += ofproto->stats.rx_packets;
2311 if (stats->tx_bytes != UINT64_MAX) {
2312 stats->tx_bytes += ofproto->stats.rx_bytes;
2319 /* Account packets for LOCAL port. */
2321 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2322 size_t tx_size, size_t rx_size)
2324 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2327 ofproto->stats.rx_packets++;
2328 ofproto->stats.rx_bytes += rx_size;
2331 ofproto->stats.tx_packets++;
2332 ofproto->stats.tx_bytes += tx_size;
2336 struct port_dump_state {
2337 struct dpif_port_dump dump;
2342 port_dump_start(const struct ofproto *ofproto_, void **statep)
2344 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2345 struct port_dump_state *state;
2347 *statep = state = xmalloc(sizeof *state);
2348 dpif_port_dump_start(&state->dump, ofproto->dpif);
2349 state->done = false;
2354 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2355 struct ofproto_port *port)
2357 struct port_dump_state *state = state_;
2358 struct dpif_port dpif_port;
2360 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2361 ofproto_port_from_dpif_port(port, &dpif_port);
2364 int error = dpif_port_dump_done(&state->dump);
2366 return error ? error : EOF;
2371 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2373 struct port_dump_state *state = state_;
2376 dpif_port_dump_done(&state->dump);
2383 port_poll(const struct ofproto *ofproto_, char **devnamep)
2385 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2386 return dpif_port_poll(ofproto->dpif, devnamep);
2390 port_poll_wait(const struct ofproto *ofproto_)
2392 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2393 dpif_port_poll_wait(ofproto->dpif);
2397 port_is_lacp_current(const struct ofport *ofport_)
2399 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2400 return (ofport->bundle && ofport->bundle->lacp
2401 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2405 /* Upcall handling. */
2407 /* Flow miss batching.
2409 * Some dpifs implement operations faster when you hand them off in a batch.
2410 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2411 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2412 * more packets, plus possibly installing the flow in the dpif.
2414 * So far we only batch the operations that affect flow setup time the most.
2415 * It's possible to batch more than that, but the benefit might be minimal. */
2417 struct hmap_node hmap_node;
2419 enum odp_key_fitness key_fitness;
2420 const struct nlattr *key;
2422 ovs_be16 initial_tci;
2423 struct list packets;
2426 struct flow_miss_op {
2427 union dpif_op dpif_op;
2428 struct subfacet *subfacet;
2431 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2432 * OpenFlow controller as necessary according to their individual
2433 * configurations. */
2435 send_packet_in_miss(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
2436 const struct flow *flow)
2438 struct ofputil_packet_in pin;
2440 pin.packet = packet->data;
2441 pin.packet_len = packet->size;
2442 pin.total_len = packet->size;
2443 pin.reason = OFPR_NO_MATCH;
2448 pin.buffer_id = 0; /* not yet known */
2449 pin.send_len = 0; /* not used for flow table misses */
2451 flow_get_metadata(flow, &pin.fmd);
2453 /* Registers aren't meaningful on a miss. */
2454 memset(pin.fmd.reg_masks, 0, sizeof pin.fmd.reg_masks);
2456 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow);
2460 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2461 const struct ofpbuf *packet)
2463 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2469 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2471 cfm_process_heartbeat(ofport->cfm, packet);
2474 } else if (ofport->bundle && ofport->bundle->lacp
2475 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2477 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2480 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2482 stp_process_packet(ofport, packet);
2489 static struct flow_miss *
2490 flow_miss_create(struct hmap *todo, const struct flow *flow,
2491 enum odp_key_fitness key_fitness,
2492 const struct nlattr *key, size_t key_len,
2493 ovs_be16 initial_tci)
2495 uint32_t hash = flow_hash(flow, 0);
2496 struct flow_miss *miss;
2498 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2499 if (flow_equal(&miss->flow, flow)) {
2504 miss = xmalloc(sizeof *miss);
2505 hmap_insert(todo, &miss->hmap_node, hash);
2507 miss->key_fitness = key_fitness;
2509 miss->key_len = key_len;
2510 miss->initial_tci = initial_tci;
2511 list_init(&miss->packets);
2516 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2517 struct flow_miss_op *ops, size_t *n_ops)
2519 const struct flow *flow = &miss->flow;
2520 struct ofpbuf *packet, *next_packet;
2521 struct subfacet *subfacet;
2522 struct facet *facet;
2524 facet = facet_lookup_valid(ofproto, flow);
2526 struct rule_dpif *rule;
2528 rule = rule_dpif_lookup(ofproto, flow, 0);
2530 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
2531 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
2533 if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) {
2534 COVERAGE_INC(ofproto_dpif_no_packet_in);
2535 /* XXX install 'drop' flow entry */
2539 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
2543 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2544 send_packet_in_miss(ofproto, packet, flow);
2550 facet = facet_create(rule, flow);
2553 subfacet = subfacet_create(ofproto, facet,
2554 miss->key_fitness, miss->key, miss->key_len,
2557 LIST_FOR_EACH_SAFE (packet, next_packet, list_node, &miss->packets) {
2558 struct dpif_flow_stats stats;
2559 struct flow_miss_op *op;
2560 struct dpif_execute *execute;
2562 ofproto->n_matches++;
2564 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2566 * Extra-special case for fail-open mode.
2568 * We are in fail-open mode and the packet matched the fail-open
2569 * rule, but we are connected to a controller too. We should send
2570 * the packet up to the controller in the hope that it will try to
2571 * set up a flow and thereby allow us to exit fail-open.
2573 * See the top-level comment in fail-open.c for more information.
2575 send_packet_in_miss(ofproto, packet, flow);
2578 if (!facet->may_install || !subfacet->actions) {
2579 subfacet_make_actions(ofproto, subfacet, packet);
2582 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2583 subfacet_update_stats(ofproto, subfacet, &stats);
2585 if (flow->vlan_tci != subfacet->initial_tci) {
2586 /* This packet was received on a VLAN splinter port. We added
2587 * a VLAN to the packet to make the packet resemble the flow,
2588 * but the actions were composed assuming that the packet
2589 * contained no VLAN. So, we must remove the VLAN header from
2590 * the packet before trying to execute the actions. */
2591 eth_pop_vlan(packet);
2594 op = &ops[(*n_ops)++];
2595 execute = &op->dpif_op.execute;
2596 op->subfacet = subfacet;
2597 execute->type = DPIF_OP_EXECUTE;
2598 execute->key = miss->key;
2599 execute->key_len = miss->key_len;
2600 execute->actions = (facet->may_install
2602 : xmemdup(subfacet->actions,
2603 subfacet->actions_len));
2604 execute->actions_len = subfacet->actions_len;
2605 execute->packet = packet;
2608 if (facet->may_install && subfacet->key_fitness != ODP_FIT_TOO_LITTLE) {
2609 struct flow_miss_op *op = &ops[(*n_ops)++];
2610 struct dpif_flow_put *put = &op->dpif_op.flow_put;
2612 op->subfacet = subfacet;
2613 put->type = DPIF_OP_FLOW_PUT;
2614 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2615 put->key = miss->key;
2616 put->key_len = miss->key_len;
2617 put->actions = subfacet->actions;
2618 put->actions_len = subfacet->actions_len;
2623 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
2624 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
2625 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
2626 * what a flow key should contain.
2628 * This function also includes some logic to help make VLAN splinters
2629 * transparent to the rest of the upcall processing logic. In particular, if
2630 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
2631 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
2632 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
2634 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
2635 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
2636 * (This differs from the value returned in flow->vlan_tci only for packets
2637 * received on VLAN splinters.)
2639 static enum odp_key_fitness
2640 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
2641 const struct nlattr *key, size_t key_len,
2642 struct flow *flow, ovs_be16 *initial_tci,
2643 struct ofpbuf *packet)
2645 enum odp_key_fitness fitness;
2649 fitness = odp_flow_key_to_flow(key, key_len, flow);
2650 if (fitness == ODP_FIT_ERROR) {
2653 *initial_tci = flow->vlan_tci;
2655 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
2657 /* Cause the flow to be processed as if it came in on the real device
2658 * with the VLAN device's VLAN ID. */
2659 flow->in_port = realdev;
2660 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
2662 /* Make the packet resemble the flow, so that it gets sent to an
2663 * OpenFlow controller properly, so that it looks correct for
2664 * sFlow, and so that flow_extract() will get the correct vlan_tci
2665 * if it is called on 'packet'.
2667 * The allocated space inside 'packet' probably also contains
2668 * 'key', that is, both 'packet' and 'key' are probably part of a
2669 * struct dpif_upcall (see the large comment on that structure
2670 * definition), so pushing data on 'packet' is in general not a
2671 * good idea since it could overwrite 'key' or free it as a side
2672 * effect. However, it's OK in this special case because we know
2673 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
2674 * will just overwrite the 4-byte "struct nlattr", which is fine
2675 * since we don't need that header anymore. */
2676 eth_push_vlan(packet, flow->vlan_tci);
2679 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
2680 if (fitness == ODP_FIT_PERFECT) {
2681 fitness = ODP_FIT_TOO_MUCH;
2689 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
2692 struct dpif_upcall *upcall;
2693 struct flow_miss *miss, *next_miss;
2694 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
2695 union dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
2704 /* Construct the to-do list.
2706 * This just amounts to extracting the flow from each packet and sticking
2707 * the packets that have the same flow in the same "flow_miss" structure so
2708 * that we can process them together. */
2710 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
2711 enum odp_key_fitness fitness;
2712 struct flow_miss *miss;
2713 ovs_be16 initial_tci;
2716 /* Obtain metadata and check userspace/kernel agreement on flow match,
2717 * then set 'flow''s header pointers. */
2718 fitness = ofproto_dpif_extract_flow_key(ofproto,
2719 upcall->key, upcall->key_len,
2720 &flow, &initial_tci,
2722 if (fitness == ODP_FIT_ERROR) {
2723 ofpbuf_delete(upcall->packet);
2726 flow_extract(upcall->packet, flow.skb_priority, flow.tun_id,
2727 flow.in_port, &flow);
2729 /* Handle 802.1ag, LACP, and STP specially. */
2730 if (process_special(ofproto, &flow, upcall->packet)) {
2731 ofproto_update_local_port_stats(&ofproto->up,
2732 0, upcall->packet->size);
2733 ofpbuf_delete(upcall->packet);
2734 ofproto->n_matches++;
2738 /* Add other packets to a to-do list. */
2739 miss = flow_miss_create(&todo, &flow, fitness,
2740 upcall->key, upcall->key_len, initial_tci);
2741 list_push_back(&miss->packets, &upcall->packet->list_node);
2744 /* Process each element in the to-do list, constructing the set of
2745 * operations to batch. */
2747 HMAP_FOR_EACH (miss, hmap_node, &todo) {
2748 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
2750 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
2752 /* Execute batch. */
2753 for (i = 0; i < n_ops; i++) {
2754 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
2756 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
2758 /* Free memory and update facets. */
2759 for (i = 0; i < n_ops; i++) {
2760 struct flow_miss_op *op = &flow_miss_ops[i];
2761 struct dpif_execute *execute;
2762 struct dpif_flow_put *put;
2764 switch (op->dpif_op.type) {
2765 case DPIF_OP_EXECUTE:
2766 execute = &op->dpif_op.execute;
2767 if (op->subfacet->actions != execute->actions) {
2768 free((struct nlattr *) execute->actions);
2772 case DPIF_OP_FLOW_PUT:
2773 put = &op->dpif_op.flow_put;
2775 op->subfacet->installed = true;
2780 HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) {
2781 ofpbuf_list_delete(&miss->packets);
2782 hmap_remove(&todo, &miss->hmap_node);
2785 hmap_destroy(&todo);
2789 handle_userspace_upcall(struct ofproto_dpif *ofproto,
2790 struct dpif_upcall *upcall)
2792 struct user_action_cookie cookie;
2793 enum odp_key_fitness fitness;
2794 ovs_be16 initial_tci;
2797 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
2799 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
2800 upcall->key_len, &flow,
2801 &initial_tci, upcall->packet);
2802 if (fitness == ODP_FIT_ERROR) {
2803 ofpbuf_delete(upcall->packet);
2807 if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
2808 if (ofproto->sflow) {
2809 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
2813 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
2815 ofpbuf_delete(upcall->packet);
2819 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
2821 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
2825 assert (max_batch <= FLOW_MISS_MAX_BATCH);
2828 for (i = 0; i < max_batch; i++) {
2829 struct dpif_upcall *upcall = &misses[n_misses];
2832 error = dpif_recv(ofproto->dpif, upcall);
2837 switch (upcall->type) {
2838 case DPIF_UC_ACTION:
2839 handle_userspace_upcall(ofproto, upcall);
2843 /* Handle it later. */
2847 case DPIF_N_UC_TYPES:
2849 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32,
2855 handle_miss_upcalls(ofproto, misses, n_misses);
2860 /* Flow expiration. */
2862 static int subfacet_max_idle(const struct ofproto_dpif *);
2863 static void update_stats(struct ofproto_dpif *);
2864 static void rule_expire(struct rule_dpif *);
2865 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
2867 /* This function is called periodically by run(). Its job is to collect
2868 * updates for the flows that have been installed into the datapath, most
2869 * importantly when they last were used, and then use that information to
2870 * expire flows that have not been used recently.
2872 * Returns the number of milliseconds after which it should be called again. */
2874 expire(struct ofproto_dpif *ofproto)
2876 struct rule_dpif *rule, *next_rule;
2877 struct classifier *table;
2880 /* Update stats for each flow in the datapath. */
2881 update_stats(ofproto);
2883 /* Expire subfacets that have been idle too long. */
2884 dp_max_idle = subfacet_max_idle(ofproto);
2885 expire_subfacets(ofproto, dp_max_idle);
2887 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
2888 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
2889 struct cls_cursor cursor;
2891 cls_cursor_init(&cursor, table, NULL);
2892 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
2897 /* All outstanding data in existing flows has been accounted, so it's a
2898 * good time to do bond rebalancing. */
2899 if (ofproto->has_bonded_bundles) {
2900 struct ofbundle *bundle;
2902 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2904 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
2909 return MIN(dp_max_idle, 1000);
2912 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
2914 * This function also pushes statistics updates to rules which each facet
2915 * resubmits into. Generally these statistics will be accurate. However, if a
2916 * facet changes the rule it resubmits into at some time in between
2917 * update_stats() runs, it is possible that statistics accrued to the
2918 * old rule will be incorrectly attributed to the new rule. This could be
2919 * avoided by calling update_stats() whenever rules are created or
2920 * deleted. However, the performance impact of making so many calls to the
2921 * datapath do not justify the benefit of having perfectly accurate statistics.
2924 update_stats(struct ofproto_dpif *p)
2926 const struct dpif_flow_stats *stats;
2927 struct dpif_flow_dump dump;
2928 const struct nlattr *key;
2931 dpif_flow_dump_start(&dump, p->dpif);
2932 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
2933 struct subfacet *subfacet;
2935 subfacet = subfacet_find(p, key, key_len);
2936 if (subfacet && subfacet->installed) {
2937 struct facet *facet = subfacet->facet;
2939 if (stats->n_packets >= subfacet->dp_packet_count) {
2940 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
2941 facet->packet_count += extra;
2943 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
2946 if (stats->n_bytes >= subfacet->dp_byte_count) {
2947 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
2949 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
2952 subfacet->dp_packet_count = stats->n_packets;
2953 subfacet->dp_byte_count = stats->n_bytes;
2955 subfacet_update_time(p, subfacet, stats->used);
2956 facet_account(p, facet, true);
2957 facet_push_stats(facet);
2959 if (!VLOG_DROP_WARN(&rl)) {
2963 odp_flow_key_format(key, key_len, &s);
2964 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
2968 COVERAGE_INC(facet_unexpected);
2969 /* There's a flow in the datapath that we know nothing about, or a
2970 * flow that shouldn't be installed but was anyway. Delete it. */
2971 dpif_flow_del(p->dpif, key, key_len, NULL);
2974 dpif_flow_dump_done(&dump);
2977 /* Calculates and returns the number of milliseconds of idle time after which
2978 * subfacets should expire from the datapath. When a subfacet expires, we fold
2979 * its statistics into its facet, and when a facet's last subfacet expires, we
2980 * fold its statistic into its rule. */
2982 subfacet_max_idle(const struct ofproto_dpif *ofproto)
2985 * Idle time histogram.
2987 * Most of the time a switch has a relatively small number of subfacets.
2988 * When this is the case we might as well keep statistics for all of them
2989 * in userspace and to cache them in the kernel datapath for performance as
2992 * As the number of subfacets increases, the memory required to maintain
2993 * statistics about them in userspace and in the kernel becomes
2994 * significant. However, with a large number of subfacets it is likely
2995 * that only a few of them are "heavy hitters" that consume a large amount
2996 * of bandwidth. At this point, only heavy hitters are worth caching in
2997 * the kernel and maintaining in userspaces; other subfacets we can
3000 * The technique used to compute the idle time is to build a histogram with
3001 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3002 * that is installed in the kernel gets dropped in the appropriate bucket.
3003 * After the histogram has been built, we compute the cutoff so that only
3004 * the most-recently-used 1% of subfacets (but at least
3005 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3006 * the most-recently-used bucket of subfacets is kept, so actually an
3007 * arbitrary number of subfacets can be kept in any given expiration run
3008 * (though the next run will delete most of those unless they receive
3011 * This requires a second pass through the subfacets, in addition to the
3012 * pass made by update_stats(), because the former function never looks at
3013 * uninstallable subfacets.
3015 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3016 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3017 int buckets[N_BUCKETS] = { 0 };
3018 int total, subtotal, bucket;
3019 struct subfacet *subfacet;
3023 total = hmap_count(&ofproto->subfacets);
3024 if (total <= ofproto->up.flow_eviction_threshold) {
3025 return N_BUCKETS * BUCKET_WIDTH;
3028 /* Build histogram. */
3030 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3031 long long int idle = now - subfacet->used;
3032 int bucket = (idle <= 0 ? 0
3033 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3034 : (unsigned int) idle / BUCKET_WIDTH);
3038 /* Find the first bucket whose flows should be expired. */
3039 subtotal = bucket = 0;
3041 subtotal += buckets[bucket++];
3042 } while (bucket < N_BUCKETS &&
3043 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3045 if (VLOG_IS_DBG_ENABLED()) {
3049 ds_put_cstr(&s, "keep");
3050 for (i = 0; i < N_BUCKETS; i++) {
3052 ds_put_cstr(&s, ", drop");
3055 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3058 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3062 return bucket * BUCKET_WIDTH;
3066 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3068 long long int cutoff = time_msec() - dp_max_idle;
3069 struct subfacet *subfacet, *next_subfacet;
3071 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3072 &ofproto->subfacets) {
3073 if (subfacet->used < cutoff) {
3074 subfacet_destroy(ofproto, subfacet);
3079 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3080 * then delete it entirely. */
3082 rule_expire(struct rule_dpif *rule)
3084 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3085 struct facet *facet, *next_facet;
3089 /* Has 'rule' expired? */
3091 if (rule->up.hard_timeout
3092 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3093 reason = OFPRR_HARD_TIMEOUT;
3094 } else if (rule->up.idle_timeout && list_is_empty(&rule->facets)
3095 && now > rule->used + rule->up.idle_timeout * 1000) {
3096 reason = OFPRR_IDLE_TIMEOUT;
3101 COVERAGE_INC(ofproto_dpif_expired);
3103 /* Update stats. (This is a no-op if the rule expired due to an idle
3104 * timeout, because that only happens when the rule has no facets left.) */
3105 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3106 facet_remove(ofproto, facet);
3109 /* Get rid of the rule. */
3110 ofproto_rule_expire(&rule->up, reason);
3115 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3117 * The caller must already have determined that no facet with an identical
3118 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3119 * the ofproto's classifier table.
3121 * The facet will initially have no subfacets. The caller should create (at
3122 * least) one subfacet with subfacet_create(). */
3123 static struct facet *
3124 facet_create(struct rule_dpif *rule, const struct flow *flow)
3126 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3127 struct facet *facet;
3129 facet = xzalloc(sizeof *facet);
3130 facet->used = time_msec();
3131 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
3132 list_push_back(&rule->facets, &facet->list_node);
3134 facet->flow = *flow;
3135 list_init(&facet->subfacets);
3136 netflow_flow_init(&facet->nf_flow);
3137 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3143 facet_free(struct facet *facet)
3148 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3149 * 'packet', which arrived on 'in_port'.
3151 * Takes ownership of 'packet'. */
3153 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3154 const struct nlattr *odp_actions, size_t actions_len,
3155 struct ofpbuf *packet)
3157 struct odputil_keybuf keybuf;
3161 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3162 odp_flow_key_from_flow(&key, flow);
3164 error = dpif_execute(ofproto->dpif, key.data, key.size,
3165 odp_actions, actions_len, packet);
3167 ofpbuf_delete(packet);
3171 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3173 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3174 * rule's statistics, via subfacet_uninstall().
3176 * - Removes 'facet' from its rule and from ofproto->facets.
3179 facet_remove(struct ofproto_dpif *ofproto, struct facet *facet)
3181 struct subfacet *subfacet, *next_subfacet;
3183 assert(!list_is_empty(&facet->subfacets));
3185 /* First uninstall all of the subfacets to get final statistics. */
3186 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3187 subfacet_uninstall(ofproto, subfacet);
3190 /* Flush the final stats to the rule.
3192 * This might require us to have at least one subfacet around so that we
3193 * can use its actions for accounting in facet_account(), which is why we
3194 * have uninstalled but not yet destroyed the subfacets. */
3195 facet_flush_stats(ofproto, facet);
3197 /* Now we're really all done so destroy everything. */
3198 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3199 &facet->subfacets) {
3200 subfacet_destroy__(ofproto, subfacet);
3202 hmap_remove(&ofproto->facets, &facet->hmap_node);
3203 list_remove(&facet->list_node);
3208 facet_account(struct ofproto_dpif *ofproto, struct facet *facet,
3212 struct subfacet *subfacet;
3213 const struct nlattr *a;
3217 if (facet->byte_count <= facet->accounted_bytes) {
3220 n_bytes = facet->byte_count - facet->accounted_bytes;
3221 facet->accounted_bytes = facet->byte_count;
3223 /* Feed information from the active flows back into the learning table to
3224 * ensure that table is always in sync with what is actually flowing
3225 * through the datapath. */
3226 if (facet->has_learn || facet->has_normal) {
3227 struct action_xlate_ctx ctx;
3229 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3230 facet->flow.vlan_tci,
3231 facet->rule->up.flow_cookie, NULL);
3232 ctx.may_learn_macs = true;
3233 ctx.may_flow_mod = may_flow_mod;
3234 ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions,
3235 facet->rule->up.n_actions));
3238 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3242 /* This loop feeds byte counters to bond_account() for rebalancing to use
3243 * as a basis. We also need to track the actual VLAN on which the packet
3244 * is going to be sent to ensure that it matches the one passed to
3245 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3248 * We use the actions from an arbitrary subfacet because they should all
3249 * be equally valid for our purpose. */
3250 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3251 struct subfacet, list_node);
3252 vlan_tci = facet->flow.vlan_tci;
3253 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3254 subfacet->actions, subfacet->actions_len) {
3255 const struct ovs_action_push_vlan *vlan;
3256 struct ofport_dpif *port;
3258 switch (nl_attr_type(a)) {
3259 case OVS_ACTION_ATTR_OUTPUT:
3260 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3261 if (port && port->bundle && port->bundle->bond) {
3262 bond_account(port->bundle->bond, &facet->flow,
3263 vlan_tci_to_vid(vlan_tci), n_bytes);
3267 case OVS_ACTION_ATTR_POP_VLAN:
3268 vlan_tci = htons(0);
3271 case OVS_ACTION_ATTR_PUSH_VLAN:
3272 vlan = nl_attr_get(a);
3273 vlan_tci = vlan->vlan_tci;
3279 /* Returns true if the only action for 'facet' is to send to the controller.
3280 * (We don't report NetFlow expiration messages for such facets because they
3281 * are just part of the control logic for the network, not real traffic). */
3283 facet_is_controller_flow(struct facet *facet)
3286 && facet->rule->up.n_actions == 1
3287 && action_outputs_to_port(&facet->rule->up.actions[0],
3288 htons(OFPP_CONTROLLER)));
3291 /* Folds all of 'facet''s statistics into its rule. Also updates the
3292 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3293 * 'facet''s statistics in the datapath should have been zeroed and folded into
3294 * its packet and byte counts before this function is called. */
3296 facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet)
3298 struct subfacet *subfacet;
3300 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3301 assert(!subfacet->dp_byte_count);
3302 assert(!subfacet->dp_packet_count);
3305 facet_push_stats(facet);
3306 facet_account(ofproto, facet, false);
3308 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3309 struct ofexpired expired;
3310 expired.flow = facet->flow;
3311 expired.packet_count = facet->packet_count;
3312 expired.byte_count = facet->byte_count;
3313 expired.used = facet->used;
3314 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3317 facet->rule->packet_count += facet->packet_count;
3318 facet->rule->byte_count += facet->byte_count;
3320 /* Reset counters to prevent double counting if 'facet' ever gets
3322 facet_reset_counters(facet);
3324 netflow_flow_clear(&facet->nf_flow);
3327 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3328 * Returns it if found, otherwise a null pointer.
3330 * The returned facet might need revalidation; use facet_lookup_valid()
3331 * instead if that is important. */
3332 static struct facet *
3333 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
3335 struct facet *facet;
3337 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
3339 if (flow_equal(flow, &facet->flow)) {
3347 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3348 * Returns it if found, otherwise a null pointer.
3350 * The returned facet is guaranteed to be valid. */
3351 static struct facet *
3352 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
3354 struct facet *facet = facet_find(ofproto, flow);
3356 /* The facet we found might not be valid, since we could be in need of
3357 * revalidation. If it is not valid, don't return it. */
3359 && (ofproto->need_revalidate
3360 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))
3361 && !facet_revalidate(ofproto, facet)) {
3362 COVERAGE_INC(facet_invalidated);
3369 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
3371 * - If the rule found is different from 'facet''s current rule, moves
3372 * 'facet' to the new rule and recompiles its actions.
3374 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3375 * where it is and recompiles its actions anyway.
3377 * - If there is none, destroys 'facet'.
3379 * Returns true if 'facet' still exists, false if it has been destroyed. */
3381 facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet)
3384 struct nlattr *odp_actions;
3387 struct actions *new_actions;
3389 struct action_xlate_ctx ctx;
3390 struct rule_dpif *new_rule;
3391 struct subfacet *subfacet;
3392 bool actions_changed;
3395 COVERAGE_INC(facet_revalidate);
3397 /* Determine the new rule. */
3398 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3400 /* No new rule, so delete the facet. */
3401 facet_remove(ofproto, facet);
3405 /* Calculate new datapath actions.
3407 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3408 * emit a NetFlow expiration and, if so, we need to have the old state
3409 * around to properly compose it. */
3411 /* If the datapath actions changed or the installability changed,
3412 * then we need to talk to the datapath. */
3415 memset(&ctx, 0, sizeof ctx);
3416 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3417 struct ofpbuf *odp_actions;
3418 bool should_install;
3420 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3421 subfacet->initial_tci, new_rule->up.flow_cookie,
3423 odp_actions = xlate_actions(&ctx, new_rule->up.actions,
3424 new_rule->up.n_actions);
3425 actions_changed = (subfacet->actions_len != odp_actions->size
3426 || memcmp(subfacet->actions, odp_actions->data,
3427 subfacet->actions_len));
3429 should_install = (ctx.may_set_up_flow
3430 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3431 if (actions_changed || should_install != subfacet->installed) {
3432 if (should_install) {
3433 struct dpif_flow_stats stats;
3435 subfacet_install(ofproto, subfacet,
3436 odp_actions->data, odp_actions->size, &stats);
3437 subfacet_update_stats(ofproto, subfacet, &stats);
3439 subfacet_uninstall(ofproto, subfacet);
3443 new_actions = xcalloc(list_size(&facet->subfacets),
3444 sizeof *new_actions);
3446 new_actions[i].odp_actions = xmemdup(odp_actions->data,
3448 new_actions[i].actions_len = odp_actions->size;
3451 ofpbuf_delete(odp_actions);
3455 facet_flush_stats(ofproto, facet);
3458 /* Update 'facet' now that we've taken care of all the old state. */
3459 facet->tags = ctx.tags;
3460 facet->nf_flow.output_iface = ctx.nf_output_iface;
3461 facet->may_install = ctx.may_set_up_flow;
3462 facet->has_learn = ctx.has_learn;
3463 facet->has_normal = ctx.has_normal;
3464 facet->mirrors = ctx.mirrors;
3467 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3468 if (new_actions[i].odp_actions) {
3469 free(subfacet->actions);
3470 subfacet->actions = new_actions[i].odp_actions;
3471 subfacet->actions_len = new_actions[i].actions_len;
3477 if (facet->rule != new_rule) {
3478 COVERAGE_INC(facet_changed_rule);
3479 list_remove(&facet->list_node);
3480 list_push_back(&new_rule->facets, &facet->list_node);
3481 facet->rule = new_rule;
3482 facet->used = new_rule->up.created;
3483 facet->prev_used = facet->used;
3489 /* Updates 'facet''s used time. Caller is responsible for calling
3490 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3492 facet_update_time(struct ofproto_dpif *ofproto, struct facet *facet,
3495 if (used > facet->used) {
3497 if (used > facet->rule->used) {
3498 facet->rule->used = used;
3500 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3505 facet_reset_counters(struct facet *facet)
3507 facet->packet_count = 0;
3508 facet->byte_count = 0;
3509 facet->prev_packet_count = 0;
3510 facet->prev_byte_count = 0;
3511 facet->accounted_bytes = 0;
3515 facet_push_stats(struct facet *facet)
3517 uint64_t new_packets, new_bytes;
3519 assert(facet->packet_count >= facet->prev_packet_count);
3520 assert(facet->byte_count >= facet->prev_byte_count);
3521 assert(facet->used >= facet->prev_used);
3523 new_packets = facet->packet_count - facet->prev_packet_count;
3524 new_bytes = facet->byte_count - facet->prev_byte_count;
3526 if (new_packets || new_bytes || facet->used > facet->prev_used) {
3527 facet->prev_packet_count = facet->packet_count;
3528 facet->prev_byte_count = facet->byte_count;
3529 facet->prev_used = facet->used;
3531 flow_push_stats(facet->rule, &facet->flow,
3532 new_packets, new_bytes, facet->used);
3534 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
3535 facet->mirrors, new_packets, new_bytes);
3539 struct ofproto_push {
3540 struct action_xlate_ctx ctx;
3547 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
3549 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3552 rule->packet_count += push->packets;
3553 rule->byte_count += push->bytes;
3554 rule->used = MAX(push->used, rule->used);
3558 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3559 * 'rule''s actions and mirrors. */
3561 flow_push_stats(const struct rule_dpif *rule,
3562 const struct flow *flow, uint64_t packets, uint64_t bytes,
3565 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3566 struct ofproto_push push;
3568 push.packets = packets;
3572 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci,
3573 rule->up.flow_cookie, NULL);
3574 push.ctx.resubmit_hook = push_resubmit;
3575 ofpbuf_delete(xlate_actions(&push.ctx,
3576 rule->up.actions, rule->up.n_actions));
3581 static struct subfacet *
3582 subfacet_find__(struct ofproto_dpif *ofproto,
3583 const struct nlattr *key, size_t key_len, uint32_t key_hash,
3584 const struct flow *flow)
3586 struct subfacet *subfacet;
3588 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
3589 &ofproto->subfacets) {
3591 ? (subfacet->key_len == key_len
3592 && !memcmp(key, subfacet->key, key_len))
3593 : flow_equal(flow, &subfacet->facet->flow)) {
3601 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
3602 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
3603 * there is one, otherwise creates and returns a new subfacet.
3605 * If the returned subfacet is new, then subfacet->actions will be NULL, in
3606 * which case the caller must populate the actions with
3607 * subfacet_make_actions(). */
3608 static struct subfacet *
3609 subfacet_create(struct ofproto_dpif *ofproto, struct facet *facet,
3610 enum odp_key_fitness key_fitness,
3611 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
3613 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3614 struct subfacet *subfacet;
3616 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
3618 if (subfacet->facet == facet) {
3622 /* This shouldn't happen. */
3623 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
3624 subfacet_destroy(ofproto, subfacet);
3627 subfacet = xzalloc(sizeof *subfacet);
3628 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
3629 list_push_back(&facet->subfacets, &subfacet->list_node);
3630 subfacet->facet = facet;
3631 subfacet->used = time_msec();
3632 subfacet->key_fitness = key_fitness;
3633 if (key_fitness != ODP_FIT_PERFECT) {
3634 subfacet->key = xmemdup(key, key_len);
3635 subfacet->key_len = key_len;
3637 subfacet->installed = false;
3638 subfacet->initial_tci = initial_tci;
3643 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
3644 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
3645 static struct subfacet *
3646 subfacet_find(struct ofproto_dpif *ofproto,
3647 const struct nlattr *key, size_t key_len)
3649 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3650 enum odp_key_fitness fitness;
3653 fitness = odp_flow_key_to_flow(key, key_len, &flow);
3654 if (fitness == ODP_FIT_ERROR) {
3658 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
3661 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
3662 * its facet within 'ofproto', and frees it. */
3664 subfacet_destroy__(struct ofproto_dpif *ofproto, struct subfacet *subfacet)
3666 subfacet_uninstall(ofproto, subfacet);
3667 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
3668 list_remove(&subfacet->list_node);
3669 free(subfacet->key);
3670 free(subfacet->actions);
3674 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
3675 * last remaining subfacet in its facet destroys the facet too. */
3677 subfacet_destroy(struct ofproto_dpif *ofproto, struct subfacet *subfacet)
3679 struct facet *facet = subfacet->facet;
3681 if (list_is_singleton(&facet->subfacets)) {
3682 /* facet_remove() needs at least one subfacet (it will remove it). */
3683 facet_remove(ofproto, facet);
3685 subfacet_destroy__(ofproto, subfacet);
3689 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
3690 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
3691 * for use as temporary storage. */
3693 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
3696 if (!subfacet->key) {
3697 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
3698 odp_flow_key_from_flow(key, &subfacet->facet->flow);
3700 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
3704 /* Composes the datapath actions for 'subfacet' based on its rule's actions. */
3706 subfacet_make_actions(struct ofproto_dpif *p, struct subfacet *subfacet,
3707 const struct ofpbuf *packet)
3709 struct facet *facet = subfacet->facet;
3710 const struct rule_dpif *rule = facet->rule;
3711 struct ofpbuf *odp_actions;
3712 struct action_xlate_ctx ctx;
3714 action_xlate_ctx_init(&ctx, p, &facet->flow, subfacet->initial_tci,
3715 rule->up.flow_cookie, packet);
3716 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3717 facet->tags = ctx.tags;
3718 facet->may_install = ctx.may_set_up_flow;
3719 facet->has_learn = ctx.has_learn;
3720 facet->has_normal = ctx.has_normal;
3721 facet->nf_flow.output_iface = ctx.nf_output_iface;
3722 facet->mirrors = ctx.mirrors;
3724 if (subfacet->actions_len != odp_actions->size
3725 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
3726 free(subfacet->actions);
3727 subfacet->actions_len = odp_actions->size;
3728 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
3731 ofpbuf_delete(odp_actions);
3734 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
3735 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
3736 * in the datapath will be zeroed and 'stats' will be updated with traffic new
3737 * since 'subfacet' was last updated.
3739 * Returns 0 if successful, otherwise a positive errno value. */
3741 subfacet_install(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3742 const struct nlattr *actions, size_t actions_len,
3743 struct dpif_flow_stats *stats)
3745 struct odputil_keybuf keybuf;
3746 enum dpif_flow_put_flags flags;
3750 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3752 flags |= DPIF_FP_ZERO_STATS;
3755 subfacet_get_key(subfacet, &keybuf, &key);
3756 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
3757 actions, actions_len, stats);
3760 subfacet_reset_dp_stats(subfacet, stats);
3766 /* If 'subfacet' is installed in the datapath, uninstalls it. */
3768 subfacet_uninstall(struct ofproto_dpif *p, struct subfacet *subfacet)
3770 if (subfacet->installed) {
3771 struct odputil_keybuf keybuf;
3772 struct dpif_flow_stats stats;
3776 subfacet_get_key(subfacet, &keybuf, &key);
3777 error = dpif_flow_del(p->dpif, key.data, key.size, &stats);
3778 subfacet_reset_dp_stats(subfacet, &stats);
3780 subfacet_update_stats(p, subfacet, &stats);
3782 subfacet->installed = false;
3784 assert(subfacet->dp_packet_count == 0);
3785 assert(subfacet->dp_byte_count == 0);
3789 /* Resets 'subfacet''s datapath statistics counters. This should be called
3790 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
3791 * non-null, it should contain the statistics returned by dpif when 'subfacet'
3792 * was reset in the datapath. 'stats' will be modified to include only
3793 * statistics new since 'subfacet' was last updated. */
3795 subfacet_reset_dp_stats(struct subfacet *subfacet,
3796 struct dpif_flow_stats *stats)
3799 && subfacet->dp_packet_count <= stats->n_packets
3800 && subfacet->dp_byte_count <= stats->n_bytes) {
3801 stats->n_packets -= subfacet->dp_packet_count;
3802 stats->n_bytes -= subfacet->dp_byte_count;
3805 subfacet->dp_packet_count = 0;
3806 subfacet->dp_byte_count = 0;
3809 /* Updates 'subfacet''s used time. The caller is responsible for calling
3810 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
3812 subfacet_update_time(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3815 if (used > subfacet->used) {
3816 subfacet->used = used;
3817 facet_update_time(ofproto, subfacet->facet, used);
3821 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
3823 * Because of the meaning of a subfacet's counters, it only makes sense to do
3824 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
3825 * represents a packet that was sent by hand or if it represents statistics
3826 * that have been cleared out of the datapath. */
3828 subfacet_update_stats(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3829 const struct dpif_flow_stats *stats)
3831 if (stats->n_packets || stats->used > subfacet->used) {
3832 struct facet *facet = subfacet->facet;
3834 subfacet_update_time(ofproto, subfacet, stats->used);
3835 facet->packet_count += stats->n_packets;
3836 facet->byte_count += stats->n_bytes;
3837 facet_push_stats(facet);
3838 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3844 static struct rule_dpif *
3845 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
3848 struct cls_rule *cls_rule;
3849 struct classifier *cls;
3851 if (table_id >= N_TABLES) {
3855 cls = &ofproto->up.tables[table_id];
3856 if (flow->nw_frag & FLOW_NW_FRAG_ANY
3857 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3858 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
3859 * are unavailable. */
3860 struct flow ofpc_normal_flow = *flow;
3861 ofpc_normal_flow.tp_src = htons(0);
3862 ofpc_normal_flow.tp_dst = htons(0);
3863 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
3865 cls_rule = classifier_lookup(cls, flow);
3867 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
3871 complete_operation(struct rule_dpif *rule)
3873 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3875 rule_invalidate(rule);
3877 struct dpif_completion *c = xmalloc(sizeof *c);
3878 c->op = rule->up.pending;
3879 list_push_back(&ofproto->completions, &c->list_node);
3881 ofoperation_complete(rule->up.pending, 0);
3885 static struct rule *
3888 struct rule_dpif *rule = xmalloc(sizeof *rule);
3893 rule_dealloc(struct rule *rule_)
3895 struct rule_dpif *rule = rule_dpif_cast(rule_);
3900 rule_construct(struct rule *rule_)
3902 struct rule_dpif *rule = rule_dpif_cast(rule_);
3903 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3904 struct rule_dpif *victim;
3908 error = validate_actions(rule->up.actions, rule->up.n_actions,
3909 &rule->up.cr.flow, ofproto->max_ports);
3914 rule->used = rule->up.created;
3915 rule->packet_count = 0;
3916 rule->byte_count = 0;
3918 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
3919 if (victim && !list_is_empty(&victim->facets)) {
3920 struct facet *facet;
3922 rule->facets = victim->facets;
3923 list_moved(&rule->facets);
3924 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3925 /* XXX: We're only clearing our local counters here. It's possible
3926 * that quite a few packets are unaccounted for in the datapath
3927 * statistics. These will be accounted to the new rule instead of
3928 * cleared as required. This could be fixed by clearing out the
3929 * datapath statistics for this facet, but currently it doesn't
3931 facet_reset_counters(facet);
3935 /* Must avoid list_moved() in this case. */
3936 list_init(&rule->facets);
3939 table_id = rule->up.table_id;
3940 rule->tag = (victim ? victim->tag
3942 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
3943 ofproto->tables[table_id].basis));
3945 complete_operation(rule);
3950 rule_destruct(struct rule *rule_)
3952 struct rule_dpif *rule = rule_dpif_cast(rule_);
3953 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3954 struct facet *facet, *next_facet;
3956 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3957 facet_revalidate(ofproto, facet);
3960 complete_operation(rule);
3964 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
3966 struct rule_dpif *rule = rule_dpif_cast(rule_);
3967 struct facet *facet;
3969 /* Start from historical data for 'rule' itself that are no longer tracked
3970 * in facets. This counts, for example, facets that have expired. */
3971 *packets = rule->packet_count;
3972 *bytes = rule->byte_count;
3974 /* Add any statistics that are tracked by facets. This includes
3975 * statistical data recently updated by ofproto_update_stats() as well as
3976 * stats for packets that were executed "by hand" via dpif_execute(). */
3977 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3978 *packets += facet->packet_count;
3979 *bytes += facet->byte_count;
3984 rule_execute(struct rule *rule_, const struct flow *flow,
3985 struct ofpbuf *packet)
3987 struct rule_dpif *rule = rule_dpif_cast(rule_);
3988 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3989 struct action_xlate_ctx ctx;
3990 struct ofpbuf *odp_actions;
3993 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
3994 rule->up.flow_cookie, packet);
3995 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3996 size = packet->size;
3997 if (execute_odp_actions(ofproto, flow, odp_actions->data,
3998 odp_actions->size, packet)) {
3999 rule->used = time_msec();
4000 rule->packet_count++;
4001 rule->byte_count += size;
4002 flow_push_stats(rule, flow, 1, size, rule->used);
4004 ofpbuf_delete(odp_actions);
4010 rule_modify_actions(struct rule *rule_)
4012 struct rule_dpif *rule = rule_dpif_cast(rule_);
4013 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4016 error = validate_actions(rule->up.actions, rule->up.n_actions,
4017 &rule->up.cr.flow, ofproto->max_ports);
4019 ofoperation_complete(rule->up.pending, error);
4023 complete_operation(rule);
4026 /* Sends 'packet' out 'ofport'.
4027 * May modify 'packet'.
4028 * Returns 0 if successful, otherwise a positive errno value. */
4030 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4032 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4033 struct ofpbuf key, odp_actions;
4034 struct odputil_keybuf keybuf;
4039 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
4040 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4042 if (odp_port != ofport->odp_port) {
4043 eth_pop_vlan(packet);
4044 flow.vlan_tci = htons(0);
4047 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4048 odp_flow_key_from_flow(&key, &flow);
4050 ofpbuf_init(&odp_actions, 32);
4051 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4053 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4054 error = dpif_execute(ofproto->dpif,
4056 odp_actions.data, odp_actions.size,
4058 ofpbuf_uninit(&odp_actions);
4061 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4062 ofproto->up.name, odp_port, strerror(error));
4064 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
4068 /* OpenFlow to datapath action translation. */
4070 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
4071 struct action_xlate_ctx *ctx);
4072 static void xlate_normal(struct action_xlate_ctx *);
4075 put_userspace_action(const struct ofproto_dpif *ofproto,
4076 struct ofpbuf *odp_actions,
4077 const struct flow *flow,
4078 const struct user_action_cookie *cookie)
4082 pid = dpif_port_get_pid(ofproto->dpif,
4083 ofp_port_to_odp_port(flow->in_port));
4085 return odp_put_userspace_action(pid, cookie, odp_actions);
4088 /* Compose SAMPLE action for sFlow. */
4090 compose_sflow_action(const struct ofproto_dpif *ofproto,
4091 struct ofpbuf *odp_actions,
4092 const struct flow *flow,
4095 uint32_t port_ifindex;
4096 uint32_t probability;
4097 struct user_action_cookie cookie;
4098 size_t sample_offset, actions_offset;
4099 int cookie_offset, n_output;
4101 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4105 if (odp_port == OVSP_NONE) {
4109 port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4113 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4115 /* Number of packets out of UINT_MAX to sample. */
4116 probability = dpif_sflow_get_probability(ofproto->sflow);
4117 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4119 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4121 cookie.type = USER_ACTION_COOKIE_SFLOW;
4122 cookie.data = port_ifindex;
4123 cookie.n_output = n_output;
4124 cookie.vlan_tci = 0;
4125 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4127 nl_msg_end_nested(odp_actions, actions_offset);
4128 nl_msg_end_nested(odp_actions, sample_offset);
4129 return cookie_offset;
4132 /* SAMPLE action must be first action in any given list of actions.
4133 * At this point we do not have all information required to build it. So try to
4134 * build sample action as complete as possible. */
4136 add_sflow_action(struct action_xlate_ctx *ctx)
4138 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4140 &ctx->flow, OVSP_NONE);
4141 ctx->sflow_odp_port = 0;
4142 ctx->sflow_n_outputs = 0;
4145 /* Fix SAMPLE action according to data collected while composing ODP actions.
4146 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4147 * USERSPACE action's user-cookie which is required for sflow. */
4149 fix_sflow_action(struct action_xlate_ctx *ctx)
4151 const struct flow *base = &ctx->base_flow;
4152 struct user_action_cookie *cookie;
4154 if (!ctx->user_cookie_offset) {
4158 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4160 assert(cookie != NULL);
4161 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4163 if (ctx->sflow_n_outputs) {
4164 cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow,
4165 ctx->sflow_odp_port);
4167 if (ctx->sflow_n_outputs >= 255) {
4168 cookie->n_output = 255;
4170 cookie->n_output = ctx->sflow_n_outputs;
4172 cookie->vlan_tci = base->vlan_tci;
4176 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4179 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4180 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4181 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4182 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4186 struct priority_to_dscp *pdscp;
4188 if (ofport->up.opp.config & htonl(OFPPC_NO_FWD)
4189 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4193 pdscp = get_priority(ofport, ctx->flow.skb_priority);
4195 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4196 ctx->flow.nw_tos |= pdscp->dscp;
4199 /* We may not have an ofport record for this port, but it doesn't hurt
4200 * to allow forwarding to it anyhow. Maybe such a port will appear
4201 * later and we're pre-populating the flow table. */
4204 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4205 ctx->flow.vlan_tci);
4206 if (out_port != odp_port) {
4207 ctx->flow.vlan_tci = htons(0);
4209 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4210 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4212 ctx->sflow_odp_port = odp_port;
4213 ctx->sflow_n_outputs++;
4214 ctx->nf_output_iface = ofp_port;
4215 ctx->flow.vlan_tci = flow_vlan_tci;
4216 ctx->flow.nw_tos = flow_nw_tos;
4220 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4222 compose_output_action__(ctx, ofp_port, true);
4226 xlate_table_action(struct action_xlate_ctx *ctx,
4227 uint16_t in_port, uint8_t table_id)
4229 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4230 struct ofproto_dpif *ofproto = ctx->ofproto;
4231 struct rule_dpif *rule;
4232 uint16_t old_in_port;
4233 uint8_t old_table_id;
4235 old_table_id = ctx->table_id;
4236 ctx->table_id = table_id;
4238 /* Look up a flow with 'in_port' as the input port. */
4239 old_in_port = ctx->flow.in_port;
4240 ctx->flow.in_port = in_port;
4241 rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
4244 if (table_id > 0 && table_id < N_TABLES) {
4245 struct table_dpif *table = &ofproto->tables[table_id];
4246 if (table->other_table) {
4247 ctx->tags |= (rule && rule->tag
4249 : rule_calculate_tag(&ctx->flow,
4250 &table->other_table->wc,
4255 /* Restore the original input port. Otherwise OFPP_NORMAL and
4256 * OFPP_IN_PORT will have surprising behavior. */
4257 ctx->flow.in_port = old_in_port;
4259 if (ctx->resubmit_hook) {
4260 ctx->resubmit_hook(ctx, rule);
4264 ovs_be64 old_cookie = ctx->cookie;
4267 ctx->cookie = rule->up.flow_cookie;
4268 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
4269 ctx->cookie = old_cookie;
4273 ctx->table_id = old_table_id;
4275 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4277 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
4278 MAX_RESUBMIT_RECURSION);
4283 xlate_resubmit_table(struct action_xlate_ctx *ctx,
4284 const struct nx_action_resubmit *nar)
4289 in_port = (nar->in_port == htons(OFPP_IN_PORT)
4291 : ntohs(nar->in_port));
4292 table_id = nar->table == 255 ? ctx->table_id : nar->table;
4294 xlate_table_action(ctx, in_port, table_id);
4298 flood_packets(struct action_xlate_ctx *ctx, bool all)
4300 struct ofport_dpif *ofport;
4302 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
4303 uint16_t ofp_port = ofport->up.ofp_port;
4305 if (ofp_port == ctx->flow.in_port) {
4310 compose_output_action__(ctx, ofp_port, false);
4311 } else if (!(ofport->up.opp.config & htonl(OFPPC_NO_FLOOD))) {
4312 compose_output_action(ctx, ofp_port);
4316 ctx->nf_output_iface = NF_OUT_FLOOD;
4320 execute_controller_action(struct action_xlate_ctx *ctx, int len)
4322 struct ofputil_packet_in pin;
4323 struct ofpbuf *packet;
4325 ctx->may_set_up_flow = false;
4330 packet = ofpbuf_clone(ctx->packet);
4332 if (packet->l2 && packet->l3) {
4333 struct eth_header *eh;
4335 eth_pop_vlan(packet);
4337 assert(eh->eth_type == ctx->flow.dl_type);
4338 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
4339 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
4341 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
4342 eth_push_vlan(packet, ctx->flow.vlan_tci);
4346 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4347 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
4348 ctx->flow.nw_tos, ctx->flow.nw_ttl);
4352 if (ctx->flow.nw_proto == IPPROTO_TCP) {
4353 packet_set_tcp_port(packet, ctx->flow.tp_src,
4355 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
4356 packet_set_udp_port(packet, ctx->flow.tp_src,
4363 pin.packet = packet->data;
4364 pin.packet_len = packet->size;
4365 pin.reason = OFPR_ACTION;
4366 pin.table_id = ctx->table_id;
4367 pin.cookie = ctx->cookie;
4371 pin.total_len = packet->size;
4372 flow_get_metadata(&ctx->flow, &pin.fmd);
4374 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin, &ctx->flow);
4375 ofpbuf_delete(packet);
4379 xlate_output_action__(struct action_xlate_ctx *ctx,
4380 uint16_t port, uint16_t max_len)
4382 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
4384 ctx->nf_output_iface = NF_OUT_DROP;
4388 compose_output_action(ctx, ctx->flow.in_port);
4391 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
4397 flood_packets(ctx, false);
4400 flood_packets(ctx, true);
4402 case OFPP_CONTROLLER:
4403 execute_controller_action(ctx, max_len);
4409 if (port != ctx->flow.in_port) {
4410 compose_output_action(ctx, port);
4415 if (prev_nf_output_iface == NF_OUT_FLOOD) {
4416 ctx->nf_output_iface = NF_OUT_FLOOD;
4417 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
4418 ctx->nf_output_iface = prev_nf_output_iface;
4419 } else if (prev_nf_output_iface != NF_OUT_DROP &&
4420 ctx->nf_output_iface != NF_OUT_FLOOD) {
4421 ctx->nf_output_iface = NF_OUT_MULTI;
4426 xlate_output_reg_action(struct action_xlate_ctx *ctx,
4427 const struct nx_action_output_reg *naor)
4431 ofp_port = nxm_read_field_bits(naor->src, naor->ofs_nbits, &ctx->flow);
4433 if (ofp_port <= UINT16_MAX) {
4434 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
4439 xlate_output_action(struct action_xlate_ctx *ctx,
4440 const struct ofp_action_output *oao)
4442 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
4446 xlate_enqueue_action(struct action_xlate_ctx *ctx,
4447 const struct ofp_action_enqueue *oae)
4450 uint32_t flow_priority, priority;
4453 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
4456 /* Fall back to ordinary output action. */
4457 xlate_output_action__(ctx, ntohs(oae->port), 0);
4461 /* Figure out datapath output port. */
4462 ofp_port = ntohs(oae->port);
4463 if (ofp_port == OFPP_IN_PORT) {
4464 ofp_port = ctx->flow.in_port;
4465 } else if (ofp_port == ctx->flow.in_port) {
4469 /* Add datapath actions. */
4470 flow_priority = ctx->flow.skb_priority;
4471 ctx->flow.skb_priority = priority;
4472 compose_output_action(ctx, ofp_port);
4473 ctx->flow.skb_priority = flow_priority;
4475 /* Update NetFlow output port. */
4476 if (ctx->nf_output_iface == NF_OUT_DROP) {
4477 ctx->nf_output_iface = ofp_port;
4478 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
4479 ctx->nf_output_iface = NF_OUT_MULTI;
4484 xlate_set_queue_action(struct action_xlate_ctx *ctx,
4485 const struct nx_action_set_queue *nasq)
4490 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
4493 /* Couldn't translate queue to a priority, so ignore. A warning
4494 * has already been logged. */
4498 ctx->flow.skb_priority = priority;
4501 struct xlate_reg_state {
4507 xlate_autopath(struct action_xlate_ctx *ctx,
4508 const struct nx_action_autopath *naa)
4510 uint16_t ofp_port = ntohl(naa->id);
4511 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
4513 if (!port || !port->bundle) {
4514 ofp_port = OFPP_NONE;
4515 } else if (port->bundle->bond) {
4516 /* Autopath does not support VLAN hashing. */
4517 struct ofport_dpif *slave = bond_choose_output_slave(
4518 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
4520 ofp_port = slave->up.ofp_port;
4523 autopath_execute(naa, &ctx->flow, ofp_port);
4527 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
4529 struct ofproto_dpif *ofproto = ofproto_;
4530 struct ofport_dpif *port;
4540 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
4543 port = get_ofp_port(ofproto, ofp_port);
4544 return port ? port->may_enable : false;
4549 xlate_learn_action(struct action_xlate_ctx *ctx,
4550 const struct nx_action_learn *learn)
4552 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
4553 struct ofputil_flow_mod fm;
4556 learn_execute(learn, &ctx->flow, &fm);
4558 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
4559 if (error && !VLOG_DROP_WARN(&rl)) {
4560 char *msg = ofputil_error_to_string(error);
4561 VLOG_WARN("learning action failed to modify flow table (%s)", msg);
4569 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
4571 if (port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
4572 ? htonl(OFPPC_NO_RECV_STP)
4573 : htonl(OFPPC_NO_RECV))) {
4577 /* Only drop packets here if both forwarding and learning are
4578 * disabled. If just learning is enabled, we need to have
4579 * OFPP_NORMAL and the learning action have a look at the packet
4580 * before we can drop it. */
4581 if (!stp_forward_in_state(port->stp_state)
4582 && !stp_learn_in_state(port->stp_state)) {
4590 do_xlate_actions(const union ofp_action *in, size_t n_in,
4591 struct action_xlate_ctx *ctx)
4593 const struct ofport_dpif *port;
4594 const union ofp_action *ia;
4597 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
4598 if (port && !may_receive(port, ctx)) {
4599 /* Drop this flow. */
4603 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
4604 const struct ofp_action_dl_addr *oada;
4605 const struct nx_action_resubmit *nar;
4606 const struct nx_action_set_tunnel *nast;
4607 const struct nx_action_set_queue *nasq;
4608 const struct nx_action_multipath *nam;
4609 const struct nx_action_autopath *naa;
4610 const struct nx_action_bundle *nab;
4611 const struct nx_action_output_reg *naor;
4612 enum ofputil_action_code code;
4619 code = ofputil_decode_action_unsafe(ia);
4621 case OFPUTIL_OFPAT_OUTPUT:
4622 xlate_output_action(ctx, &ia->output);
4625 case OFPUTIL_OFPAT_SET_VLAN_VID:
4626 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
4627 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
4630 case OFPUTIL_OFPAT_SET_VLAN_PCP:
4631 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
4632 ctx->flow.vlan_tci |= htons(
4633 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
4636 case OFPUTIL_OFPAT_STRIP_VLAN:
4637 ctx->flow.vlan_tci = htons(0);
4640 case OFPUTIL_OFPAT_SET_DL_SRC:
4641 oada = ((struct ofp_action_dl_addr *) ia);
4642 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
4645 case OFPUTIL_OFPAT_SET_DL_DST:
4646 oada = ((struct ofp_action_dl_addr *) ia);
4647 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
4650 case OFPUTIL_OFPAT_SET_NW_SRC:
4651 ctx->flow.nw_src = ia->nw_addr.nw_addr;
4654 case OFPUTIL_OFPAT_SET_NW_DST:
4655 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
4658 case OFPUTIL_OFPAT_SET_NW_TOS:
4659 /* OpenFlow 1.0 only supports IPv4. */
4660 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4661 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4662 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
4666 case OFPUTIL_OFPAT_SET_TP_SRC:
4667 ctx->flow.tp_src = ia->tp_port.tp_port;
4670 case OFPUTIL_OFPAT_SET_TP_DST:
4671 ctx->flow.tp_dst = ia->tp_port.tp_port;
4674 case OFPUTIL_OFPAT_ENQUEUE:
4675 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
4678 case OFPUTIL_NXAST_RESUBMIT:
4679 nar = (const struct nx_action_resubmit *) ia;
4680 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
4683 case OFPUTIL_NXAST_RESUBMIT_TABLE:
4684 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
4687 case OFPUTIL_NXAST_SET_TUNNEL:
4688 nast = (const struct nx_action_set_tunnel *) ia;
4689 tun_id = htonll(ntohl(nast->tun_id));
4690 ctx->flow.tun_id = tun_id;
4693 case OFPUTIL_NXAST_SET_QUEUE:
4694 nasq = (const struct nx_action_set_queue *) ia;
4695 xlate_set_queue_action(ctx, nasq);
4698 case OFPUTIL_NXAST_POP_QUEUE:
4699 ctx->flow.skb_priority = ctx->orig_skb_priority;
4702 case OFPUTIL_NXAST_REG_MOVE:
4703 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
4707 case OFPUTIL_NXAST_REG_LOAD:
4708 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
4712 case OFPUTIL_NXAST_NOTE:
4713 /* Nothing to do. */
4716 case OFPUTIL_NXAST_SET_TUNNEL64:
4717 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
4718 ctx->flow.tun_id = tun_id;
4721 case OFPUTIL_NXAST_MULTIPATH:
4722 nam = (const struct nx_action_multipath *) ia;
4723 multipath_execute(nam, &ctx->flow);
4726 case OFPUTIL_NXAST_AUTOPATH:
4727 naa = (const struct nx_action_autopath *) ia;
4728 xlate_autopath(ctx, naa);
4731 case OFPUTIL_NXAST_BUNDLE:
4732 ctx->ofproto->has_bundle_action = true;
4733 nab = (const struct nx_action_bundle *) ia;
4734 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
4739 case OFPUTIL_NXAST_BUNDLE_LOAD:
4740 ctx->ofproto->has_bundle_action = true;
4741 nab = (const struct nx_action_bundle *) ia;
4742 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
4746 case OFPUTIL_NXAST_OUTPUT_REG:
4747 naor = (const struct nx_action_output_reg *) ia;
4748 xlate_output_reg_action(ctx, naor);
4751 case OFPUTIL_NXAST_LEARN:
4752 ctx->has_learn = true;
4753 if (ctx->may_flow_mod) {
4754 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
4758 case OFPUTIL_NXAST_EXIT:
4764 /* We've let OFPP_NORMAL and the learning action look at the packet,
4765 * so drop it now if forwarding is disabled. */
4766 if (port && !stp_forward_in_state(port->stp_state)) {
4767 ofpbuf_clear(ctx->odp_actions);
4768 add_sflow_action(ctx);
4773 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
4774 struct ofproto_dpif *ofproto, const struct flow *flow,
4775 ovs_be16 initial_tci, ovs_be64 cookie,
4776 const struct ofpbuf *packet)
4778 ctx->ofproto = ofproto;
4780 ctx->base_flow = ctx->flow;
4781 ctx->base_flow.tun_id = 0;
4782 ctx->base_flow.vlan_tci = initial_tci;
4783 ctx->cookie = cookie;
4784 ctx->packet = packet;
4785 ctx->may_learn_macs = packet != NULL;
4786 ctx->may_flow_mod = packet != NULL;
4787 ctx->resubmit_hook = NULL;
4790 static struct ofpbuf *
4791 xlate_actions(struct action_xlate_ctx *ctx,
4792 const union ofp_action *in, size_t n_in)
4794 struct flow orig_flow = ctx->flow;
4796 COVERAGE_INC(ofproto_dpif_xlate);
4798 ctx->odp_actions = ofpbuf_new(512);
4799 ofpbuf_reserve(ctx->odp_actions, NL_A_U32_SIZE);
4801 ctx->may_set_up_flow = true;
4802 ctx->has_learn = false;
4803 ctx->has_normal = false;
4804 ctx->nf_output_iface = NF_OUT_DROP;
4807 ctx->orig_skb_priority = ctx->flow.skb_priority;
4811 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
4812 switch (ctx->ofproto->up.frag_handling) {
4813 case OFPC_FRAG_NORMAL:
4814 /* We must pretend that transport ports are unavailable. */
4815 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
4816 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
4819 case OFPC_FRAG_DROP:
4820 return ctx->odp_actions;
4822 case OFPC_FRAG_REASM:
4825 case OFPC_FRAG_NX_MATCH:
4826 /* Nothing to do. */
4831 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
4832 ctx->may_set_up_flow = false;
4833 return ctx->odp_actions;
4835 add_sflow_action(ctx);
4836 do_xlate_actions(in, n_in, ctx);
4838 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
4839 ctx->odp_actions->data,
4840 ctx->odp_actions->size)) {
4841 ctx->may_set_up_flow = false;
4843 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
4845 compose_output_action(ctx, OFPP_LOCAL);
4848 add_mirror_actions(ctx, &orig_flow);
4849 fix_sflow_action(ctx);
4852 return ctx->odp_actions;
4855 /* OFPP_NORMAL implementation. */
4857 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
4859 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
4860 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
4861 * the bundle on which the packet was received, returns the VLAN to which the
4864 * Both 'vid' and the return value are in the range 0...4095. */
4866 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
4868 switch (in_bundle->vlan_mode) {
4869 case PORT_VLAN_ACCESS:
4870 return in_bundle->vlan;
4873 case PORT_VLAN_TRUNK:
4876 case PORT_VLAN_NATIVE_UNTAGGED:
4877 case PORT_VLAN_NATIVE_TAGGED:
4878 return vid ? vid : in_bundle->vlan;
4885 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
4886 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
4889 * 'vid' should be the VID obtained from the 802.1Q header that was received as
4890 * part of a packet (specify 0 if there was no 802.1Q header), in the range
4893 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
4895 /* Allow any VID on the OFPP_NONE port. */
4896 if (in_bundle == &ofpp_none_bundle) {
4900 switch (in_bundle->vlan_mode) {
4901 case PORT_VLAN_ACCESS:
4904 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4905 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
4906 "packet received on port %s configured as VLAN "
4907 "%"PRIu16" access port",
4908 in_bundle->ofproto->up.name, vid,
4909 in_bundle->name, in_bundle->vlan);
4915 case PORT_VLAN_NATIVE_UNTAGGED:
4916 case PORT_VLAN_NATIVE_TAGGED:
4918 /* Port must always carry its native VLAN. */
4922 case PORT_VLAN_TRUNK:
4923 if (!ofbundle_includes_vlan(in_bundle, vid)) {
4925 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4926 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
4927 "received on port %s not configured for trunking "
4929 in_bundle->ofproto->up.name, vid,
4930 in_bundle->name, vid);
4942 /* Given 'vlan', the VLAN that a packet belongs to, and
4943 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
4944 * that should be included in the 802.1Q header. (If the return value is 0,
4945 * then the 802.1Q header should only be included in the packet if there is a
4948 * Both 'vlan' and the return value are in the range 0...4095. */
4950 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
4952 switch (out_bundle->vlan_mode) {
4953 case PORT_VLAN_ACCESS:
4956 case PORT_VLAN_TRUNK:
4957 case PORT_VLAN_NATIVE_TAGGED:
4960 case PORT_VLAN_NATIVE_UNTAGGED:
4961 return vlan == out_bundle->vlan ? 0 : vlan;
4969 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
4972 struct ofport_dpif *port;
4974 ovs_be16 tci, old_tci;
4976 vid = output_vlan_to_vid(out_bundle, vlan);
4977 if (!out_bundle->bond) {
4978 port = ofbundle_get_a_port(out_bundle);
4980 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
4983 /* No slaves enabled, so drop packet. */
4988 old_tci = ctx->flow.vlan_tci;
4990 if (tci || out_bundle->use_priority_tags) {
4991 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
4993 tci |= htons(VLAN_CFI);
4996 ctx->flow.vlan_tci = tci;
4998 compose_output_action(ctx, port->up.ofp_port);
4999 ctx->flow.vlan_tci = old_tci;
5003 mirror_mask_ffs(mirror_mask_t mask)
5005 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
5010 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
5012 return (bundle->vlan_mode != PORT_VLAN_ACCESS
5013 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
5017 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
5019 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
5022 /* Returns an arbitrary interface within 'bundle'. */
5023 static struct ofport_dpif *
5024 ofbundle_get_a_port(const struct ofbundle *bundle)
5026 return CONTAINER_OF(list_front(&bundle->ports),
5027 struct ofport_dpif, bundle_node);
5031 vlan_is_mirrored(const struct ofmirror *m, int vlan)
5033 return !m->vlans || bitmap_is_set(m->vlans, vlan);
5036 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
5037 * to a VLAN. In general most packets may be mirrored but we want to drop
5038 * protocols that may confuse switches. */
5040 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
5042 /* If you change this function's behavior, please update corresponding
5043 * documentation in vswitch.xml at the same time. */
5044 if (dst[0] != 0x01) {
5045 /* All the currently banned MACs happen to start with 01 currently, so
5046 * this is a quick way to eliminate most of the good ones. */
5048 if (eth_addr_is_reserved(dst)) {
5049 /* Drop STP, IEEE pause frames, and other reserved protocols
5050 * (01-80-c2-00-00-0x). */
5054 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
5056 if ((dst[3] & 0xfe) == 0xcc &&
5057 (dst[4] & 0xfe) == 0xcc &&
5058 (dst[5] & 0xfe) == 0xcc) {
5059 /* Drop the following protocols plus others following the same
5062 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
5063 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
5064 STP Uplink Fast (01-00-0c-cd-cd-cd) */
5068 if (!(dst[3] | dst[4] | dst[5])) {
5069 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
5078 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5080 struct ofproto_dpif *ofproto = ctx->ofproto;
5081 mirror_mask_t mirrors;
5082 struct ofbundle *in_bundle;
5085 const struct nlattr *a;
5088 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5089 ctx->packet != NULL);
5093 mirrors = in_bundle->src_mirrors;
5095 /* Drop frames on bundles reserved for mirroring. */
5096 if (in_bundle->mirror_out) {
5097 if (ctx->packet != NULL) {
5098 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5099 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5100 "%s, which is reserved exclusively for mirroring",
5101 ctx->ofproto->up.name, in_bundle->name);
5107 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5108 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5111 vlan = input_vid_to_vlan(in_bundle, vid);
5113 /* Look at the output ports to check for destination selections. */
5115 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5116 ctx->odp_actions->size) {
5117 enum ovs_action_attr type = nl_attr_type(a);
5118 struct ofport_dpif *ofport;
5120 if (type != OVS_ACTION_ATTR_OUTPUT) {
5124 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5125 if (ofport && ofport->bundle) {
5126 mirrors |= ofport->bundle->dst_mirrors;
5134 /* Restore the original packet before adding the mirror actions. */
5135 ctx->flow = *orig_flow;
5140 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5142 if (!vlan_is_mirrored(m, vlan)) {
5143 mirrors &= mirrors - 1;
5147 mirrors &= ~m->dup_mirrors;
5148 ctx->mirrors |= m->dup_mirrors;
5150 output_normal(ctx, m->out, vlan);
5151 } else if (eth_dst_may_rspan(orig_flow->dl_dst)
5152 && vlan != m->out_vlan) {
5153 struct ofbundle *bundle;
5155 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5156 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5157 && !bundle->mirror_out) {
5158 output_normal(ctx, bundle, m->out_vlan);
5166 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5167 uint64_t packets, uint64_t bytes)
5173 for (; mirrors; mirrors &= mirrors - 1) {
5176 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5179 /* In normal circumstances 'm' will not be NULL. However,
5180 * if mirrors are reconfigured, we can temporarily get out
5181 * of sync in facet_revalidate(). We could "correct" the
5182 * mirror list before reaching here, but doing that would
5183 * not properly account the traffic stats we've currently
5184 * accumulated for previous mirror configuration. */
5188 m->packet_count += packets;
5189 m->byte_count += bytes;
5193 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
5194 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
5195 * indicate this; newer upstream kernels use gratuitous ARP requests. */
5197 is_gratuitous_arp(const struct flow *flow)
5199 return (flow->dl_type == htons(ETH_TYPE_ARP)
5200 && eth_addr_is_broadcast(flow->dl_dst)
5201 && (flow->nw_proto == ARP_OP_REPLY
5202 || (flow->nw_proto == ARP_OP_REQUEST
5203 && flow->nw_src == flow->nw_dst)));
5207 update_learning_table(struct ofproto_dpif *ofproto,
5208 const struct flow *flow, int vlan,
5209 struct ofbundle *in_bundle)
5211 struct mac_entry *mac;
5213 /* Don't learn the OFPP_NONE port. */
5214 if (in_bundle == &ofpp_none_bundle) {
5218 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
5222 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
5223 if (is_gratuitous_arp(flow)) {
5224 /* We don't want to learn from gratuitous ARP packets that are
5225 * reflected back over bond slaves so we lock the learning table. */
5226 if (!in_bundle->bond) {
5227 mac_entry_set_grat_arp_lock(mac);
5228 } else if (mac_entry_is_grat_arp_locked(mac)) {
5233 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
5234 /* The log messages here could actually be useful in debugging,
5235 * so keep the rate limit relatively high. */
5236 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5237 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
5238 "on port %s in VLAN %d",
5239 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
5240 in_bundle->name, vlan);
5242 mac->port.p = in_bundle;
5243 tag_set_add(&ofproto->revalidate_set,
5244 mac_learning_changed(ofproto->ml, mac));
5248 static struct ofbundle *
5249 lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn)
5251 struct ofport_dpif *ofport;
5253 /* Special-case OFPP_NONE, which a controller may use as the ingress
5254 * port for traffic that it is sourcing. */
5255 if (in_port == OFPP_NONE) {
5256 return &ofpp_none_bundle;
5259 /* Find the port and bundle for the received packet. */
5260 ofport = get_ofp_port(ofproto, in_port);
5261 if (ofport && ofport->bundle) {
5262 return ofport->bundle;
5265 /* Odd. A few possible reasons here:
5267 * - We deleted a port but there are still a few packets queued up
5270 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
5271 * we don't know about.
5273 * - The ofproto client didn't configure the port as part of a bundle.
5276 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5278 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
5279 "port %"PRIu16, ofproto->up.name, in_port);
5284 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
5285 * dropped. Returns true if they may be forwarded, false if they should be
5288 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
5289 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
5291 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
5292 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
5293 * checked by input_vid_is_valid().
5295 * May also add tags to '*tags', although the current implementation only does
5296 * so in one special case.
5299 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
5300 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
5302 struct ofbundle *in_bundle = in_port->bundle;
5304 /* Drop frames for reserved multicast addresses
5305 * only if forward_bpdu option is absent. */
5306 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
5310 if (in_bundle->bond) {
5311 struct mac_entry *mac;
5313 switch (bond_check_admissibility(in_bundle->bond, in_port,
5314 flow->dl_dst, tags)) {
5321 case BV_DROP_IF_MOVED:
5322 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
5323 if (mac && mac->port.p != in_bundle &&
5324 (!is_gratuitous_arp(flow)
5325 || mac_entry_is_grat_arp_locked(mac))) {
5336 xlate_normal(struct action_xlate_ctx *ctx)
5338 struct ofport_dpif *in_port;
5339 struct ofbundle *in_bundle;
5340 struct mac_entry *mac;
5344 ctx->has_normal = true;
5346 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
5347 ctx->packet != NULL);
5352 /* We know 'in_port' exists unless it is "ofpp_none_bundle",
5353 * since lookup_input_bundle() succeeded. */
5354 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5356 /* Drop malformed frames. */
5357 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
5358 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
5359 if (ctx->packet != NULL) {
5360 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5361 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
5362 "VLAN tag received on port %s",
5363 ctx->ofproto->up.name, in_bundle->name);
5368 /* Drop frames on bundles reserved for mirroring. */
5369 if (in_bundle->mirror_out) {
5370 if (ctx->packet != NULL) {
5371 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5372 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5373 "%s, which is reserved exclusively for mirroring",
5374 ctx->ofproto->up.name, in_bundle->name);
5380 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
5381 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5384 vlan = input_vid_to_vlan(in_bundle, vid);
5386 /* Check other admissibility requirements. */
5388 !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
5392 /* Learn source MAC. */
5393 if (ctx->may_learn_macs) {
5394 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
5397 /* Determine output bundle. */
5398 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
5401 if (mac->port.p != in_bundle) {
5402 output_normal(ctx, mac->port.p, vlan);
5405 struct ofbundle *bundle;
5407 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
5408 if (bundle != in_bundle
5409 && ofbundle_includes_vlan(bundle, vlan)
5410 && bundle->floodable
5411 && !bundle->mirror_out) {
5412 output_normal(ctx, bundle, vlan);
5415 ctx->nf_output_iface = NF_OUT_FLOOD;
5419 /* Optimized flow revalidation.
5421 * It's a difficult problem, in general, to tell which facets need to have
5422 * their actions recalculated whenever the OpenFlow flow table changes. We
5423 * don't try to solve that general problem: for most kinds of OpenFlow flow
5424 * table changes, we recalculate the actions for every facet. This is
5425 * relatively expensive, but it's good enough if the OpenFlow flow table
5426 * doesn't change very often.
5428 * However, we can expect one particular kind of OpenFlow flow table change to
5429 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5430 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5431 * table, we add a special case that applies to flow tables in which every rule
5432 * has the same form (that is, the same wildcards), except that the table is
5433 * also allowed to have a single "catch-all" flow that matches all packets. We
5434 * optimize this case by tagging all of the facets that resubmit into the table
5435 * and invalidating the same tag whenever a flow changes in that table. The
5436 * end result is that we revalidate just the facets that need it (and sometimes
5437 * a few more, but not all of the facets or even all of the facets that
5438 * resubmit to the table modified by MAC learning). */
5440 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
5441 * into an OpenFlow table with the given 'basis'. */
5443 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
5446 if (flow_wildcards_is_catchall(wc)) {
5449 struct flow tag_flow = *flow;
5450 flow_zero_wildcards(&tag_flow, wc);
5451 return tag_create_deterministic(flow_hash(&tag_flow, secret));
5455 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5456 * taggability of that table.
5458 * This function must be called after *each* change to a flow table. If you
5459 * skip calling it on some changes then the pointer comparisons at the end can
5460 * be invalid if you get unlucky. For example, if a flow removal causes a
5461 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5462 * different wildcards to be created with the same address, then this function
5463 * will incorrectly skip revalidation. */
5465 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5467 struct table_dpif *table = &ofproto->tables[table_id];
5468 const struct classifier *cls = &ofproto->up.tables[table_id];
5469 struct cls_table *catchall, *other;
5470 struct cls_table *t;
5472 catchall = other = NULL;
5474 switch (hmap_count(&cls->tables)) {
5476 /* We could tag this OpenFlow table but it would make the logic a
5477 * little harder and it's a corner case that doesn't seem worth it
5483 HMAP_FOR_EACH (t, hmap_node, &cls->tables) {
5484 if (cls_table_is_catchall(t)) {
5486 } else if (!other) {
5489 /* Indicate that we can't tag this by setting both tables to
5490 * NULL. (We know that 'catchall' is already NULL.) */
5497 /* Can't tag this table. */
5501 if (table->catchall_table != catchall || table->other_table != other) {
5502 table->catchall_table = catchall;
5503 table->other_table = other;
5504 ofproto->need_revalidate = true;
5508 /* Given 'rule' that has changed in some way (either it is a rule being
5509 * inserted, a rule being deleted, or a rule whose actions are being
5510 * modified), marks facets for revalidation to ensure that packets will be
5511 * forwarded correctly according to the new state of the flow table.
5513 * This function must be called after *each* change to a flow table. See
5514 * the comment on table_update_taggable() for more information. */
5516 rule_invalidate(const struct rule_dpif *rule)
5518 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5520 table_update_taggable(ofproto, rule->up.table_id);
5522 if (!ofproto->need_revalidate) {
5523 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5525 if (table->other_table && rule->tag) {
5526 tag_set_add(&ofproto->revalidate_set, rule->tag);
5528 ofproto->need_revalidate = true;
5534 set_frag_handling(struct ofproto *ofproto_,
5535 enum ofp_config_flags frag_handling)
5537 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5539 if (frag_handling != OFPC_FRAG_REASM) {
5540 ofproto->need_revalidate = true;
5548 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5549 const struct flow *flow,
5550 const union ofp_action *ofp_actions, size_t n_ofp_actions)
5552 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5555 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
5556 return ofp_mkerr_nicira(OFPET_BAD_REQUEST, NXBRC_BAD_IN_PORT);
5559 error = validate_actions(ofp_actions, n_ofp_actions, flow,
5560 ofproto->max_ports);
5562 struct odputil_keybuf keybuf;
5563 struct ofpbuf *odp_actions;
5564 struct ofproto_push push;
5567 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5568 odp_flow_key_from_flow(&key, flow);
5570 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, 0,
5573 /* Ensure that resubmits in 'ofp_actions' get accounted to their
5574 * matching rules. */
5576 push.bytes = packet->size;
5577 push.used = time_msec();
5578 push.ctx.resubmit_hook = push_resubmit;
5580 odp_actions = xlate_actions(&push.ctx, ofp_actions, n_ofp_actions);
5581 dpif_execute(ofproto->dpif, key.data, key.size,
5582 odp_actions->data, odp_actions->size, packet);
5583 ofpbuf_delete(odp_actions);
5591 set_netflow(struct ofproto *ofproto_,
5592 const struct netflow_options *netflow_options)
5594 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5596 if (netflow_options) {
5597 if (!ofproto->netflow) {
5598 ofproto->netflow = netflow_create();
5600 return netflow_set_options(ofproto->netflow, netflow_options);
5602 netflow_destroy(ofproto->netflow);
5603 ofproto->netflow = NULL;
5609 get_netflow_ids(const struct ofproto *ofproto_,
5610 uint8_t *engine_type, uint8_t *engine_id)
5612 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5614 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
5618 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5620 if (!facet_is_controller_flow(facet) &&
5621 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5622 struct subfacet *subfacet;
5623 struct ofexpired expired;
5625 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5626 if (subfacet->installed) {
5627 struct dpif_flow_stats stats;
5629 subfacet_install(ofproto, subfacet, subfacet->actions,
5630 subfacet->actions_len, &stats);
5631 subfacet_update_stats(ofproto, subfacet, &stats);
5635 expired.flow = facet->flow;
5636 expired.packet_count = facet->packet_count;
5637 expired.byte_count = facet->byte_count;
5638 expired.used = facet->used;
5639 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5644 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5646 struct facet *facet;
5648 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5649 send_active_timeout(ofproto, facet);
5653 static struct ofproto_dpif *
5654 ofproto_dpif_lookup(const char *name)
5656 struct ofproto_dpif *ofproto;
5658 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5659 hash_string(name, 0), &all_ofproto_dpifs) {
5660 if (!strcmp(ofproto->up.name, name)) {
5668 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5669 const char *argv[], void *aux OVS_UNUSED)
5671 struct ofproto_dpif *ofproto;
5674 ofproto = ofproto_dpif_lookup(argv[1]);
5676 unixctl_command_reply(conn, 501, "no such bridge");
5679 mac_learning_flush(ofproto->ml);
5680 ofproto->need_revalidate = true;
5682 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5683 mac_learning_flush(ofproto->ml);
5684 ofproto->need_revalidate = true;
5688 unixctl_command_reply(conn, 200, "table successfully flushed");
5692 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5693 const char *argv[], void *aux OVS_UNUSED)
5695 struct ds ds = DS_EMPTY_INITIALIZER;
5696 const struct ofproto_dpif *ofproto;
5697 const struct mac_entry *e;
5699 ofproto = ofproto_dpif_lookup(argv[1]);
5701 unixctl_command_reply(conn, 501, "no such bridge");
5705 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5706 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5707 struct ofbundle *bundle = e->port.p;
5708 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
5709 ofbundle_get_a_port(bundle)->odp_port,
5710 e->vlan, ETH_ADDR_ARGS(e->mac),
5711 mac_entry_age(ofproto->ml, e));
5713 unixctl_command_reply(conn, 200, ds_cstr(&ds));
5717 struct ofproto_trace {
5718 struct action_xlate_ctx ctx;
5724 trace_format_rule(struct ds *result, uint8_t table_id, int level,
5725 const struct rule_dpif *rule)
5727 ds_put_char_multiple(result, '\t', level);
5729 ds_put_cstr(result, "No match\n");
5733 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5734 table_id, ntohll(rule->up.flow_cookie));
5735 cls_rule_format(&rule->up.cr, result);
5736 ds_put_char(result, '\n');
5738 ds_put_char_multiple(result, '\t', level);
5739 ds_put_cstr(result, "OpenFlow ");
5740 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
5741 ds_put_char(result, '\n');
5745 trace_format_flow(struct ds *result, int level, const char *title,
5746 struct ofproto_trace *trace)
5748 ds_put_char_multiple(result, '\t', level);
5749 ds_put_format(result, "%s: ", title);
5750 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
5751 ds_put_cstr(result, "unchanged");
5753 flow_format(result, &trace->ctx.flow);
5754 trace->flow = trace->ctx.flow;
5756 ds_put_char(result, '\n');
5760 trace_format_regs(struct ds *result, int level, const char *title,
5761 struct ofproto_trace *trace)
5765 ds_put_char_multiple(result, '\t', level);
5766 ds_put_format(result, "%s:", title);
5767 for (i = 0; i < FLOW_N_REGS; i++) {
5768 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5770 ds_put_char(result, '\n');
5774 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
5776 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
5777 struct ds *result = trace->result;
5779 ds_put_char(result, '\n');
5780 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
5781 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
5782 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
5786 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5787 void *aux OVS_UNUSED)
5789 const char *dpname = argv[1];
5790 struct ofproto_dpif *ofproto;
5791 struct ofpbuf odp_key;
5792 struct ofpbuf *packet;
5793 struct rule_dpif *rule;
5794 ovs_be16 initial_tci;
5800 ofpbuf_init(&odp_key, 0);
5803 ofproto = ofproto_dpif_lookup(dpname);
5805 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
5809 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
5810 /* ofproto/trace dpname flow [-generate] */
5811 const char *flow_s = argv[2];
5812 const char *generate_s = argv[3];
5815 /* Convert string to datapath key. */
5816 ofpbuf_init(&odp_key, 0);
5817 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
5819 unixctl_command_reply(conn, 501, "Bad flow syntax");
5823 /* Convert odp_key to flow. */
5824 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
5825 odp_key.size, &flow,
5826 &initial_tci, NULL);
5827 if (error == ODP_FIT_ERROR) {
5828 unixctl_command_reply(conn, 501, "Invalid flow");
5832 /* Generate a packet, if requested. */
5834 packet = ofpbuf_new(0);
5835 flow_compose(packet, &flow);
5837 } else if (argc == 6) {
5838 /* ofproto/trace dpname priority tun_id in_port packet */
5839 const char *priority_s = argv[2];
5840 const char *tun_id_s = argv[3];
5841 const char *in_port_s = argv[4];
5842 const char *packet_s = argv[5];
5843 uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s));
5844 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
5845 uint32_t priority = atoi(priority_s);
5848 msg = eth_from_hex(packet_s, &packet);
5850 unixctl_command_reply(conn, 501, msg);
5854 ds_put_cstr(&result, "Packet: ");
5855 s = ofp_packet_to_string(packet->data, packet->size);
5856 ds_put_cstr(&result, s);
5859 flow_extract(packet, priority, tun_id, in_port, &flow);
5860 initial_tci = flow.vlan_tci;
5862 unixctl_command_reply(conn, 501, "Bad command syntax");
5866 ds_put_cstr(&result, "Flow: ");
5867 flow_format(&result, &flow);
5868 ds_put_char(&result, '\n');
5870 rule = rule_dpif_lookup(ofproto, &flow, 0);
5871 trace_format_rule(&result, 0, 0, rule);
5873 struct ofproto_trace trace;
5874 struct ofpbuf *odp_actions;
5876 trace.result = &result;
5878 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, initial_tci,
5879 rule->up.flow_cookie, packet);
5880 trace.ctx.resubmit_hook = trace_resubmit;
5881 odp_actions = xlate_actions(&trace.ctx,
5882 rule->up.actions, rule->up.n_actions);
5884 ds_put_char(&result, '\n');
5885 trace_format_flow(&result, 0, "Final flow", &trace);
5886 ds_put_cstr(&result, "Datapath actions: ");
5887 format_odp_actions(&result, odp_actions->data, odp_actions->size);
5888 ofpbuf_delete(odp_actions);
5890 if (!trace.ctx.may_set_up_flow) {
5892 ds_put_cstr(&result, "\nThis flow is not cachable.");
5894 ds_put_cstr(&result, "\nThe datapath actions are incomplete--"
5895 "for complete actions, please supply a packet.");
5900 unixctl_command_reply(conn, 200, ds_cstr(&result));
5903 ds_destroy(&result);
5904 ofpbuf_delete(packet);
5905 ofpbuf_uninit(&odp_key);
5909 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5910 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5913 unixctl_command_reply(conn, 200, NULL);
5917 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5918 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5921 unixctl_command_reply(conn, 200, NULL);
5925 ofproto_dpif_unixctl_init(void)
5927 static bool registered;
5933 unixctl_command_register(
5935 "bridge {tun_id in_port packet | odp_flow [-generate]}",
5936 2, 5, ofproto_unixctl_trace, NULL);
5937 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
5938 ofproto_unixctl_fdb_flush, NULL);
5939 unixctl_command_register("fdb/show", "bridge", 1, 1,
5940 ofproto_unixctl_fdb_show, NULL);
5941 unixctl_command_register("ofproto/clog", "", 0, 0,
5942 ofproto_dpif_clog, NULL);
5943 unixctl_command_register("ofproto/unclog", "", 0, 0,
5944 ofproto_dpif_unclog, NULL);
5947 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5949 * This is deprecated. It is only for compatibility with broken device drivers
5950 * in old versions of Linux that do not properly support VLANs when VLAN
5951 * devices are not used. When broken device drivers are no longer in
5952 * widespread use, we will delete these interfaces. */
5955 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
5957 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
5958 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
5960 if (realdev_ofp_port == ofport->realdev_ofp_port
5961 && vid == ofport->vlandev_vid) {
5965 ofproto->need_revalidate = true;
5967 if (ofport->realdev_ofp_port) {
5970 if (realdev_ofp_port && ofport->bundle) {
5971 /* vlandevs are enslaved to their realdevs, so they are not allowed to
5972 * themselves be part of a bundle. */
5973 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
5976 ofport->realdev_ofp_port = realdev_ofp_port;
5977 ofport->vlandev_vid = vid;
5979 if (realdev_ofp_port) {
5980 vsp_add(ofport, realdev_ofp_port, vid);
5987 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
5989 return hash_2words(realdev_ofp_port, vid);
5993 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
5994 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
5996 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
5997 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
5998 int vid = vlan_tci_to_vid(vlan_tci);
5999 const struct vlan_splinter *vsp;
6001 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6002 hash_realdev_vid(realdev_ofp_port, vid),
6003 &ofproto->realdev_vid_map) {
6004 if (vsp->realdev_ofp_port == realdev_ofp_port
6005 && vsp->vid == vid) {
6006 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
6010 return realdev_odp_port;
6013 static struct vlan_splinter *
6014 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
6016 struct vlan_splinter *vsp;
6018 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
6019 &ofproto->vlandev_map) {
6020 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6029 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6030 uint16_t vlandev_ofp_port, int *vid)
6032 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6033 const struct vlan_splinter *vsp;
6035 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6040 return vsp->realdev_ofp_port;
6047 vsp_remove(struct ofport_dpif *port)
6049 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6050 struct vlan_splinter *vsp;
6052 vsp = vlandev_find(ofproto, port->up.ofp_port);
6054 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6055 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6058 port->realdev_ofp_port = 0;
6060 VLOG_ERR("missing vlan device record");
6065 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
6067 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6069 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6070 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6071 == realdev_ofp_port)) {
6072 struct vlan_splinter *vsp;
6074 vsp = xmalloc(sizeof *vsp);
6075 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6076 hash_int(port->up.ofp_port, 0));
6077 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6078 hash_realdev_vid(realdev_ofp_port, vid));
6079 vsp->realdev_ofp_port = realdev_ofp_port;
6080 vsp->vlandev_ofp_port = port->up.ofp_port;
6083 port->realdev_ofp_port = realdev_ofp_port;
6085 VLOG_ERR("duplicate vlan device record");
6089 const struct ofproto_class ofproto_dpif_class = {
6118 port_is_lacp_current,
6119 NULL, /* rule_choose_table */
6126 rule_modify_actions,
6134 get_cfm_remote_mpids,
6138 get_stp_port_status,
6145 is_mirror_output_bundle,
6146 forward_bpdu_changed,