2 * Copyright (c) 2009, 2010, 2011 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-provider.h"
26 #include "byte-order.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
36 #include "mac-learning.h"
37 #include "multipath.h"
44 #include "ofp-print.h"
45 #include "ofproto-dpif-sflow.h"
46 #include "poll-loop.h"
48 #include "unaligned.h"
50 #include "vlan-bitmap.h"
53 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
55 COVERAGE_DEFINE(ofproto_dpif_ctlr_action);
56 COVERAGE_DEFINE(ofproto_dpif_expired);
57 COVERAGE_DEFINE(ofproto_dpif_no_packet_in);
58 COVERAGE_DEFINE(ofproto_dpif_xlate);
59 COVERAGE_DEFINE(facet_changed_rule);
60 COVERAGE_DEFINE(facet_invalidated);
61 COVERAGE_DEFINE(facet_revalidate);
62 COVERAGE_DEFINE(facet_unexpected);
64 /* Maximum depth of flow table recursion (due to resubmit actions) in a
65 * flow translation. */
66 #define MAX_RESUBMIT_RECURSION 32
68 /* Number of implemented OpenFlow tables. */
69 enum { N_TABLES = 255 };
70 BUILD_ASSERT_DECL(N_TABLES >= 1 && N_TABLES <= 255);
78 long long int used; /* Time last used; time created if not used. */
82 * - Do include packets and bytes from facets that have been deleted or
83 * whose own statistics have been folded into the rule.
85 * - Do include packets and bytes sent "by hand" that were accounted to
86 * the rule without any facet being involved (this is a rare corner
87 * case in rule_execute()).
89 * - Do not include packet or bytes that can be obtained from any facet's
90 * packet_count or byte_count member or that can be obtained from the
91 * datapath by, e.g., dpif_flow_get() for any subfacet.
93 uint64_t packet_count; /* Number of packets received. */
94 uint64_t byte_count; /* Number of bytes received. */
96 tag_type tag; /* Caches rule_calculate_tag() result. */
98 struct list facets; /* List of "struct facet"s. */
101 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
103 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
106 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
107 const struct flow *, uint8_t table);
109 static void flow_push_stats(const struct rule_dpif *, const struct flow *,
110 uint64_t packets, uint64_t bytes,
113 static uint32_t rule_calculate_tag(const struct flow *,
114 const struct flow_wildcards *,
116 static void rule_invalidate(const struct rule_dpif *);
118 #define MAX_MIRRORS 32
119 typedef uint32_t mirror_mask_t;
120 #define MIRROR_MASK_C(X) UINT32_C(X)
121 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
123 struct ofproto_dpif *ofproto; /* Owning ofproto. */
124 size_t idx; /* In ofproto's "mirrors" array. */
125 void *aux; /* Key supplied by ofproto's client. */
126 char *name; /* Identifier for log messages. */
128 /* Selection criteria. */
129 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
130 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
131 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
133 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
134 struct ofbundle *out; /* Output port or NULL. */
135 int out_vlan; /* Output VLAN or -1. */
136 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
139 int64_t packet_count; /* Number of packets sent. */
140 int64_t byte_count; /* Number of bytes sent. */
143 static void mirror_destroy(struct ofmirror *);
144 static void update_mirror_stats(struct ofproto_dpif *ofproto,
145 mirror_mask_t mirrors,
146 uint64_t packets, uint64_t bytes);
149 struct ofproto_dpif *ofproto; /* Owning ofproto. */
150 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
151 void *aux; /* Key supplied by ofproto's client. */
152 char *name; /* Identifier for log messages. */
155 struct list ports; /* Contains "struct ofport"s. */
156 enum port_vlan_mode vlan_mode; /* VLAN mode */
157 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
158 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
159 * NULL if all VLANs are trunked. */
160 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
161 struct bond *bond; /* Nonnull iff more than one port. */
162 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
165 bool floodable; /* True if no port has OFPPC_NO_FLOOD set. */
167 /* Port mirroring info. */
168 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
169 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
170 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
173 static void bundle_remove(struct ofport *);
174 static void bundle_update(struct ofbundle *);
175 static void bundle_destroy(struct ofbundle *);
176 static void bundle_del_port(struct ofport_dpif *);
177 static void bundle_run(struct ofbundle *);
178 static void bundle_wait(struct ofbundle *);
179 static struct ofport_dpif *lookup_input_bundle(struct ofproto_dpif *,
180 uint16_t in_port, bool warn);
182 static void stp_run(struct ofproto_dpif *ofproto);
183 static void stp_wait(struct ofproto_dpif *ofproto);
185 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
187 struct action_xlate_ctx {
188 /* action_xlate_ctx_init() initializes these members. */
191 struct ofproto_dpif *ofproto;
193 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
194 * this flow when actions change header fields. */
197 /* The packet corresponding to 'flow', or a null pointer if we are
198 * revalidating without a packet to refer to. */
199 const struct ofpbuf *packet;
201 /* Should OFPP_NORMAL MAC learning and NXAST_LEARN actions execute? We
202 * want to execute them if we are actually processing a packet, or if we
203 * are accounting for packets that the datapath has processed, but not if
204 * we are just revalidating. */
207 /* If nonnull, called just before executing a resubmit action.
209 * This is normally null so the client has to set it manually after
210 * calling action_xlate_ctx_init(). */
211 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *);
213 /* xlate_actions() initializes and uses these members. The client might want
214 * to look at them after it returns. */
216 struct ofpbuf *odp_actions; /* Datapath actions. */
217 tag_type tags; /* Tags associated with actions. */
218 bool may_set_up_flow; /* True ordinarily; false if the actions must
219 * be reassessed for every packet. */
220 bool has_learn; /* Actions include NXAST_LEARN? */
221 bool has_normal; /* Actions output to OFPP_NORMAL? */
222 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
223 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
225 /* xlate_actions() initializes and uses these members, but the client has no
226 * reason to look at them. */
228 int recurse; /* Recursion level, via xlate_table_action. */
229 struct flow base_flow; /* Flow at the last commit. */
230 uint32_t original_priority; /* Priority when packet arrived. */
231 uint8_t table_id; /* OpenFlow table ID where flow was found. */
232 uint32_t sflow_n_outputs; /* Number of output ports. */
233 uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
234 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
235 bool exit; /* No further actions should be processed. */
238 static void action_xlate_ctx_init(struct action_xlate_ctx *,
239 struct ofproto_dpif *, const struct flow *,
240 ovs_be16 initial_tci, const struct ofpbuf *);
241 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
242 const union ofp_action *in, size_t n_in);
244 /* An exact-match instantiation of an OpenFlow flow.
246 * A facet associates a "struct flow", which represents the Open vSwitch
247 * userspace idea of an exact-match flow, with one or more subfacets. Each
248 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
249 * the facet. When the kernel module (or other dpif implementation) and Open
250 * vSwitch userspace agree on the definition of a flow key, there is exactly
251 * one subfacet per facet. If the dpif implementation supports more-specific
252 * flow matching than userspace, however, a facet can have more than one
253 * subfacet, each of which corresponds to some distinction in flow that
254 * userspace simply doesn't understand.
256 * Flow expiration works in terms of subfacets, so a facet must have at least
257 * one subfacet or it will never expire, leaking memory. */
260 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
261 struct list list_node; /* In owning rule's 'facets' list. */
262 struct rule_dpif *rule; /* Owning rule. */
265 struct list subfacets;
266 long long int used; /* Time last used; time created if not used. */
273 * - Do include packets and bytes sent "by hand", e.g. with
276 * - Do include packets and bytes that were obtained from the datapath
277 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
278 * DPIF_FP_ZERO_STATS).
280 * - Do not include packets or bytes that can be obtained from the
281 * datapath for any existing subfacet.
283 uint64_t packet_count; /* Number of packets received. */
284 uint64_t byte_count; /* Number of bytes received. */
286 /* Resubmit statistics. */
287 uint64_t prev_packet_count; /* Number of packets from last stats push. */
288 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
289 long long int prev_used; /* Used time from last stats push. */
292 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
293 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
295 /* Properties of datapath actions.
297 * Every subfacet has its own actions because actions can differ slightly
298 * between splintered and non-splintered subfacets due to the VLAN tag
299 * being initially different (present vs. absent). All of them have these
300 * properties in common so we just store one copy of them here. */
301 bool may_install; /* Reassess actions for every packet? */
302 bool has_learn; /* Actions include NXAST_LEARN? */
303 bool has_normal; /* Actions output to OFPP_NORMAL? */
304 tag_type tags; /* Tags that would require revalidation. */
305 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
308 static struct facet *facet_create(struct rule_dpif *, const struct flow *);
309 static void facet_remove(struct ofproto_dpif *, struct facet *);
310 static void facet_free(struct facet *);
312 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
313 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
314 const struct flow *);
315 static bool facet_revalidate(struct ofproto_dpif *, struct facet *);
317 static bool execute_controller_action(struct ofproto_dpif *,
319 const struct nlattr *odp_actions,
321 struct ofpbuf *packet, bool clone);
323 static void facet_flush_stats(struct ofproto_dpif *, struct facet *);
325 static void facet_update_time(struct ofproto_dpif *, struct facet *,
327 static void facet_reset_counters(struct facet *);
328 static void facet_push_stats(struct facet *);
329 static void facet_account(struct ofproto_dpif *, struct facet *);
331 static bool facet_is_controller_flow(struct facet *);
333 /* A dpif flow and actions associated with a facet.
335 * See also the large comment on struct facet. */
338 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
339 struct list list_node; /* In struct facet's 'facets' list. */
340 struct facet *facet; /* Owning facet. */
344 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
345 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
346 * regenerate the ODP flow key from ->facet->flow. */
347 enum odp_key_fitness key_fitness;
351 long long int used; /* Time last used; time created if not used. */
353 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
354 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
358 * These should be essentially identical for every subfacet in a facet, but
359 * may differ in trivial ways due to VLAN splinters. */
360 size_t actions_len; /* Number of bytes in actions[]. */
361 struct nlattr *actions; /* Datapath actions. */
363 bool installed; /* Installed in datapath? */
365 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
366 * splinters can cause it to differ. This value should be removed when
367 * the VLAN splinters feature is no longer needed. */
368 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
371 static struct subfacet *subfacet_create(struct ofproto_dpif *, struct facet *,
372 enum odp_key_fitness,
373 const struct nlattr *key,
374 size_t key_len, ovs_be16 initial_tci);
375 static struct subfacet *subfacet_find(struct ofproto_dpif *,
376 const struct nlattr *key, size_t key_len);
377 static void subfacet_destroy(struct ofproto_dpif *, struct subfacet *);
378 static void subfacet_destroy__(struct ofproto_dpif *, struct subfacet *);
379 static void subfacet_reset_dp_stats(struct subfacet *,
380 struct dpif_flow_stats *);
381 static void subfacet_update_time(struct ofproto_dpif *, struct subfacet *,
383 static void subfacet_update_stats(struct ofproto_dpif *, struct subfacet *,
384 const struct dpif_flow_stats *);
385 static void subfacet_make_actions(struct ofproto_dpif *, struct subfacet *,
386 const struct ofpbuf *packet);
387 static int subfacet_install(struct ofproto_dpif *, struct subfacet *,
388 const struct nlattr *actions, size_t actions_len,
389 struct dpif_flow_stats *);
390 static void subfacet_uninstall(struct ofproto_dpif *, struct subfacet *);
396 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
397 struct list bundle_node; /* In struct ofbundle's "ports" list. */
398 struct cfm *cfm; /* Connectivity Fault Management, if any. */
399 tag_type tag; /* Tag associated with this port. */
400 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
401 bool may_enable; /* May be enabled in bonds. */
404 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
405 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
406 long long int stp_state_entered;
408 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
410 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
412 * This is deprecated. It is only for compatibility with broken device
413 * drivers in old versions of Linux that do not properly support VLANs when
414 * VLAN devices are not used. When broken device drivers are no longer in
415 * widespread use, we will delete these interfaces. */
416 uint16_t realdev_ofp_port;
420 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
421 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
422 * traffic egressing the 'ofport' with that priority should be marked with. */
423 struct priority_to_dscp {
424 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
425 uint32_t priority; /* Priority of this queue (see struct flow). */
427 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
430 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
432 * This is deprecated. It is only for compatibility with broken device drivers
433 * in old versions of Linux that do not properly support VLANs when VLAN
434 * devices are not used. When broken device drivers are no longer in
435 * widespread use, we will delete these interfaces. */
436 struct vlan_splinter {
437 struct hmap_node realdev_vid_node;
438 struct hmap_node vlandev_node;
439 uint16_t realdev_ofp_port;
440 uint16_t vlandev_ofp_port;
444 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
445 uint32_t realdev, ovs_be16 vlan_tci);
446 static uint16_t vsp_vlandev_to_realdev(const struct ofproto_dpif *,
447 uint16_t vlandev, int *vid);
448 static void vsp_remove(struct ofport_dpif *);
449 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
451 static struct ofport_dpif *
452 ofport_dpif_cast(const struct ofport *ofport)
454 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
455 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
458 static void port_run(struct ofport_dpif *);
459 static void port_wait(struct ofport_dpif *);
460 static int set_cfm(struct ofport *, const struct cfm_settings *);
461 static void ofport_clear_priorities(struct ofport_dpif *);
463 struct dpif_completion {
464 struct list list_node;
465 struct ofoperation *op;
468 /* Extra information about a classifier table.
469 * Currently used just for optimized flow revalidation. */
471 /* If either of these is nonnull, then this table has a form that allows
472 * flows to be tagged to avoid revalidating most flows for the most common
473 * kinds of flow table changes. */
474 struct cls_table *catchall_table; /* Table that wildcards all fields. */
475 struct cls_table *other_table; /* Table with any other wildcard set. */
476 uint32_t basis; /* Keeps each table's tags separate. */
479 struct ofproto_dpif {
480 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
489 struct netflow *netflow;
490 struct dpif_sflow *sflow;
491 struct hmap bundles; /* Contains "struct ofbundle"s. */
492 struct mac_learning *ml;
493 struct ofmirror *mirrors[MAX_MIRRORS];
494 bool has_bonded_bundles;
497 struct timer next_expiration;
501 struct hmap subfacets;
504 struct table_dpif tables[N_TABLES];
505 bool need_revalidate;
506 struct tag_set revalidate_set;
508 /* Support for debugging async flow mods. */
509 struct list completions;
511 bool has_bundle_action; /* True when the first bundle action appears. */
512 struct netdev_stats stats; /* To account packets generated and consumed in
517 long long int stp_last_tick;
519 /* VLAN splinters. */
520 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
521 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
524 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
525 * for debugging the asynchronous flow_mod implementation.) */
528 /* All existing ofproto_dpif instances, indexed by ->up.name. */
529 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
531 static void ofproto_dpif_unixctl_init(void);
533 static struct ofproto_dpif *
534 ofproto_dpif_cast(const struct ofproto *ofproto)
536 assert(ofproto->ofproto_class == &ofproto_dpif_class);
537 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
540 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
542 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
545 /* Packet processing. */
546 static void update_learning_table(struct ofproto_dpif *,
547 const struct flow *, int vlan,
550 #define FLOW_MISS_MAX_BATCH 50
551 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
553 /* Flow expiration. */
554 static int expire(struct ofproto_dpif *);
557 static void send_netflow_active_timeouts(struct ofproto_dpif *);
560 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
562 compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions,
563 const struct flow *, uint32_t odp_port);
564 static void add_mirror_actions(struct action_xlate_ctx *ctx,
565 const struct flow *flow);
566 /* Global variables. */
567 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
569 /* Factory functions. */
572 enumerate_types(struct sset *types)
574 dp_enumerate_types(types);
578 enumerate_names(const char *type, struct sset *names)
580 return dp_enumerate_names(type, names);
584 del(const char *type, const char *name)
589 error = dpif_open(name, type, &dpif);
591 error = dpif_delete(dpif);
597 /* Basic life-cycle. */
599 static struct ofproto *
602 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
607 dealloc(struct ofproto *ofproto_)
609 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
614 construct(struct ofproto *ofproto_, int *n_tablesp)
616 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
617 const char *name = ofproto->up.name;
621 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
623 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
627 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
628 ofproto->n_matches = 0;
630 dpif_flow_flush(ofproto->dpif);
631 dpif_recv_purge(ofproto->dpif);
633 error = dpif_recv_set_mask(ofproto->dpif,
634 ((1u << DPIF_UC_MISS) |
635 (1u << DPIF_UC_ACTION)));
637 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
638 dpif_close(ofproto->dpif);
642 ofproto->netflow = NULL;
643 ofproto->sflow = NULL;
645 hmap_init(&ofproto->bundles);
646 ofproto->ml = mac_learning_create();
647 for (i = 0; i < MAX_MIRRORS; i++) {
648 ofproto->mirrors[i] = NULL;
650 ofproto->has_bonded_bundles = false;
652 timer_set_duration(&ofproto->next_expiration, 1000);
654 hmap_init(&ofproto->facets);
655 hmap_init(&ofproto->subfacets);
657 for (i = 0; i < N_TABLES; i++) {
658 struct table_dpif *table = &ofproto->tables[i];
660 table->catchall_table = NULL;
661 table->other_table = NULL;
662 table->basis = random_uint32();
664 ofproto->need_revalidate = false;
665 tag_set_init(&ofproto->revalidate_set);
667 list_init(&ofproto->completions);
669 ofproto_dpif_unixctl_init();
671 ofproto->has_bundle_action = false;
673 hmap_init(&ofproto->vlandev_map);
674 hmap_init(&ofproto->realdev_vid_map);
676 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
677 hash_string(ofproto->up.name, 0));
679 *n_tablesp = N_TABLES;
680 memset(&ofproto->stats, 0, sizeof ofproto->stats);
685 complete_operations(struct ofproto_dpif *ofproto)
687 struct dpif_completion *c, *next;
689 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
690 ofoperation_complete(c->op, 0);
691 list_remove(&c->list_node);
697 destruct(struct ofproto *ofproto_)
699 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
700 struct rule_dpif *rule, *next_rule;
701 struct classifier *table;
704 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
705 complete_operations(ofproto);
707 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
708 struct cls_cursor cursor;
710 cls_cursor_init(&cursor, table, NULL);
711 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
712 ofproto_rule_destroy(&rule->up);
716 for (i = 0; i < MAX_MIRRORS; i++) {
717 mirror_destroy(ofproto->mirrors[i]);
720 netflow_destroy(ofproto->netflow);
721 dpif_sflow_destroy(ofproto->sflow);
722 hmap_destroy(&ofproto->bundles);
723 mac_learning_destroy(ofproto->ml);
725 hmap_destroy(&ofproto->facets);
726 hmap_destroy(&ofproto->subfacets);
728 hmap_destroy(&ofproto->vlandev_map);
729 hmap_destroy(&ofproto->realdev_vid_map);
731 dpif_close(ofproto->dpif);
735 run_fast(struct ofproto *ofproto_)
737 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
740 /* Handle one or more batches of upcalls, until there's nothing left to do
741 * or until we do a fixed total amount of work.
743 * We do work in batches because it can be much cheaper to set up a number
744 * of flows and fire off their patches all at once. We do multiple batches
745 * because in some cases handling a packet can cause another packet to be
746 * queued almost immediately as part of the return flow. Both
747 * optimizations can make major improvements on some benchmarks and
748 * presumably for real traffic as well. */
750 while (work < FLOW_MISS_MAX_BATCH) {
751 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
761 run(struct ofproto *ofproto_)
763 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
764 struct ofport_dpif *ofport;
765 struct ofbundle *bundle;
769 complete_operations(ofproto);
771 dpif_run(ofproto->dpif);
773 error = run_fast(ofproto_);
778 if (timer_expired(&ofproto->next_expiration)) {
779 int delay = expire(ofproto);
780 timer_set_duration(&ofproto->next_expiration, delay);
783 if (ofproto->netflow) {
784 if (netflow_run(ofproto->netflow)) {
785 send_netflow_active_timeouts(ofproto);
788 if (ofproto->sflow) {
789 dpif_sflow_run(ofproto->sflow);
792 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
795 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
800 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
802 /* Now revalidate if there's anything to do. */
803 if (ofproto->need_revalidate
804 || !tag_set_is_empty(&ofproto->revalidate_set)) {
805 struct tag_set revalidate_set = ofproto->revalidate_set;
806 bool revalidate_all = ofproto->need_revalidate;
807 struct facet *facet, *next;
809 /* Clear the revalidation flags. */
810 tag_set_init(&ofproto->revalidate_set);
811 ofproto->need_revalidate = false;
813 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
815 || tag_set_intersects(&revalidate_set, facet->tags)) {
816 facet_revalidate(ofproto, facet);
825 wait(struct ofproto *ofproto_)
827 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
828 struct ofport_dpif *ofport;
829 struct ofbundle *bundle;
831 if (!clogged && !list_is_empty(&ofproto->completions)) {
832 poll_immediate_wake();
835 dpif_wait(ofproto->dpif);
836 dpif_recv_wait(ofproto->dpif);
837 if (ofproto->sflow) {
838 dpif_sflow_wait(ofproto->sflow);
840 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
841 poll_immediate_wake();
843 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
846 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
849 if (ofproto->netflow) {
850 netflow_wait(ofproto->netflow);
852 mac_learning_wait(ofproto->ml);
854 if (ofproto->need_revalidate) {
855 /* Shouldn't happen, but if it does just go around again. */
856 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
857 poll_immediate_wake();
859 timer_wait(&ofproto->next_expiration);
864 flush(struct ofproto *ofproto_)
866 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
867 struct facet *facet, *next_facet;
869 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
870 /* Mark the facet as not installed so that facet_remove() doesn't
871 * bother trying to uninstall it. There is no point in uninstalling it
872 * individually since we are about to blow away all the facets with
873 * dpif_flow_flush(). */
874 struct subfacet *subfacet;
876 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
877 subfacet->installed = false;
878 subfacet->dp_packet_count = 0;
879 subfacet->dp_byte_count = 0;
881 facet_remove(ofproto, facet);
883 dpif_flow_flush(ofproto->dpif);
887 get_features(struct ofproto *ofproto_ OVS_UNUSED,
888 bool *arp_match_ip, uint32_t *actions)
890 *arp_match_ip = true;
891 *actions = ((1u << OFPAT_OUTPUT) |
892 (1u << OFPAT_SET_VLAN_VID) |
893 (1u << OFPAT_SET_VLAN_PCP) |
894 (1u << OFPAT_STRIP_VLAN) |
895 (1u << OFPAT_SET_DL_SRC) |
896 (1u << OFPAT_SET_DL_DST) |
897 (1u << OFPAT_SET_NW_SRC) |
898 (1u << OFPAT_SET_NW_DST) |
899 (1u << OFPAT_SET_NW_TOS) |
900 (1u << OFPAT_SET_TP_SRC) |
901 (1u << OFPAT_SET_TP_DST) |
902 (1u << OFPAT_ENQUEUE));
906 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
908 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
909 struct dpif_dp_stats s;
911 strcpy(ots->name, "classifier");
913 dpif_get_dp_stats(ofproto->dpif, &s);
914 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
915 put_32aligned_be64(&ots->matched_count,
916 htonll(s.n_hit + ofproto->n_matches));
919 static struct ofport *
922 struct ofport_dpif *port = xmalloc(sizeof *port);
927 port_dealloc(struct ofport *port_)
929 struct ofport_dpif *port = ofport_dpif_cast(port_);
934 port_construct(struct ofport *port_)
936 struct ofport_dpif *port = ofport_dpif_cast(port_);
937 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
939 ofproto->need_revalidate = true;
940 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
943 port->tag = tag_create_random();
944 port->may_enable = true;
945 port->stp_port = NULL;
946 port->stp_state = STP_DISABLED;
947 hmap_init(&port->priorities);
948 port->realdev_ofp_port = 0;
949 port->vlandev_vid = 0;
951 if (ofproto->sflow) {
952 dpif_sflow_add_port(ofproto->sflow, port_);
959 port_destruct(struct ofport *port_)
961 struct ofport_dpif *port = ofport_dpif_cast(port_);
962 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
964 ofproto->need_revalidate = true;
965 bundle_remove(port_);
966 set_cfm(port_, NULL);
967 if (ofproto->sflow) {
968 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
971 ofport_clear_priorities(port);
972 hmap_destroy(&port->priorities);
976 port_modified(struct ofport *port_)
978 struct ofport_dpif *port = ofport_dpif_cast(port_);
980 if (port->bundle && port->bundle->bond) {
981 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
986 port_reconfigured(struct ofport *port_, ovs_be32 old_config)
988 struct ofport_dpif *port = ofport_dpif_cast(port_);
989 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
990 ovs_be32 changed = old_config ^ port->up.opp.config;
992 if (changed & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
993 OFPPC_NO_FWD | OFPPC_NO_FLOOD)) {
994 ofproto->need_revalidate = true;
996 if (changed & htonl(OFPPC_NO_FLOOD) && port->bundle) {
997 bundle_update(port->bundle);
1003 set_sflow(struct ofproto *ofproto_,
1004 const struct ofproto_sflow_options *sflow_options)
1006 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1007 struct dpif_sflow *ds = ofproto->sflow;
1009 if (sflow_options) {
1011 struct ofport_dpif *ofport;
1013 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1014 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1015 dpif_sflow_add_port(ds, &ofport->up);
1017 ofproto->need_revalidate = true;
1019 dpif_sflow_set_options(ds, sflow_options);
1022 dpif_sflow_destroy(ds);
1023 ofproto->need_revalidate = true;
1024 ofproto->sflow = NULL;
1031 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1033 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1040 struct ofproto_dpif *ofproto;
1042 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1043 ofproto->need_revalidate = true;
1044 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1047 if (cfm_configure(ofport->cfm, s)) {
1053 cfm_destroy(ofport->cfm);
1059 get_cfm_fault(const struct ofport *ofport_)
1061 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1063 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1067 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1070 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1073 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1080 /* Spanning Tree. */
1083 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1085 struct ofproto_dpif *ofproto = ofproto_;
1086 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1087 struct ofport_dpif *ofport;
1089 ofport = stp_port_get_aux(sp);
1091 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1092 ofproto->up.name, port_num);
1094 struct eth_header *eth = pkt->l2;
1096 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1097 if (eth_addr_is_zero(eth->eth_src)) {
1098 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1099 "with unknown MAC", ofproto->up.name, port_num);
1101 send_packet(ofport, pkt);
1107 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1109 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1111 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1113 /* Only revalidate flows if the configuration changed. */
1114 if (!s != !ofproto->stp) {
1115 ofproto->need_revalidate = true;
1119 if (!ofproto->stp) {
1120 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1121 send_bpdu_cb, ofproto);
1122 ofproto->stp_last_tick = time_msec();
1125 stp_set_bridge_id(ofproto->stp, s->system_id);
1126 stp_set_bridge_priority(ofproto->stp, s->priority);
1127 stp_set_hello_time(ofproto->stp, s->hello_time);
1128 stp_set_max_age(ofproto->stp, s->max_age);
1129 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1131 stp_destroy(ofproto->stp);
1132 ofproto->stp = NULL;
1139 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1141 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1145 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1146 s->designated_root = stp_get_designated_root(ofproto->stp);
1147 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1156 update_stp_port_state(struct ofport_dpif *ofport)
1158 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1159 enum stp_state state;
1161 /* Figure out new state. */
1162 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1166 if (ofport->stp_state != state) {
1170 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1171 netdev_get_name(ofport->up.netdev),
1172 stp_state_name(ofport->stp_state),
1173 stp_state_name(state));
1174 if (stp_learn_in_state(ofport->stp_state)
1175 != stp_learn_in_state(state)) {
1176 /* xxx Learning action flows should also be flushed. */
1177 mac_learning_flush(ofproto->ml);
1179 fwd_change = stp_forward_in_state(ofport->stp_state)
1180 != stp_forward_in_state(state);
1182 ofproto->need_revalidate = true;
1183 ofport->stp_state = state;
1184 ofport->stp_state_entered = time_msec();
1186 if (fwd_change && ofport->bundle) {
1187 bundle_update(ofport->bundle);
1190 /* Update the STP state bits in the OpenFlow port description. */
1191 of_state = (ofport->up.opp.state & htonl(~OFPPS_STP_MASK))
1192 | htonl(state == STP_LISTENING ? OFPPS_STP_LISTEN
1193 : state == STP_LEARNING ? OFPPS_STP_LEARN
1194 : state == STP_FORWARDING ? OFPPS_STP_FORWARD
1195 : state == STP_BLOCKING ? OFPPS_STP_BLOCK
1197 ofproto_port_set_state(&ofport->up, of_state);
1201 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1202 * caller is responsible for assigning STP port numbers and ensuring
1203 * there are no duplicates. */
1205 set_stp_port(struct ofport *ofport_,
1206 const struct ofproto_port_stp_settings *s)
1208 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1209 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1210 struct stp_port *sp = ofport->stp_port;
1212 if (!s || !s->enable) {
1214 ofport->stp_port = NULL;
1215 stp_port_disable(sp);
1216 update_stp_port_state(ofport);
1219 } else if (sp && stp_port_no(sp) != s->port_num
1220 && ofport == stp_port_get_aux(sp)) {
1221 /* The port-id changed, so disable the old one if it's not
1222 * already in use by another port. */
1223 stp_port_disable(sp);
1226 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1227 stp_port_enable(sp);
1229 stp_port_set_aux(sp, ofport);
1230 stp_port_set_priority(sp, s->priority);
1231 stp_port_set_path_cost(sp, s->path_cost);
1233 update_stp_port_state(ofport);
1239 get_stp_port_status(struct ofport *ofport_,
1240 struct ofproto_port_stp_status *s)
1242 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1243 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1244 struct stp_port *sp = ofport->stp_port;
1246 if (!ofproto->stp || !sp) {
1252 s->port_id = stp_port_get_id(sp);
1253 s->state = stp_port_get_state(sp);
1254 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1255 s->role = stp_port_get_role(sp);
1256 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1262 stp_run(struct ofproto_dpif *ofproto)
1265 long long int now = time_msec();
1266 long long int elapsed = now - ofproto->stp_last_tick;
1267 struct stp_port *sp;
1270 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1271 ofproto->stp_last_tick = now;
1273 while (stp_get_changed_port(ofproto->stp, &sp)) {
1274 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1277 update_stp_port_state(ofport);
1284 stp_wait(struct ofproto_dpif *ofproto)
1287 poll_timer_wait(1000);
1291 /* Returns true if STP should process 'flow'. */
1293 stp_should_process_flow(const struct flow *flow)
1295 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1299 stp_process_packet(const struct ofport_dpif *ofport,
1300 const struct ofpbuf *packet)
1302 struct ofpbuf payload = *packet;
1303 struct eth_header *eth = payload.data;
1304 struct stp_port *sp = ofport->stp_port;
1306 /* Sink packets on ports that have STP disabled when the bridge has
1308 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1312 /* Trim off padding on payload. */
1313 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1314 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1317 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1318 stp_received_bpdu(sp, payload.data, payload.size);
1322 static struct priority_to_dscp *
1323 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1325 struct priority_to_dscp *pdscp;
1328 hash = hash_int(priority, 0);
1329 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1330 if (pdscp->priority == priority) {
1338 ofport_clear_priorities(struct ofport_dpif *ofport)
1340 struct priority_to_dscp *pdscp, *next;
1342 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1343 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1349 set_queues(struct ofport *ofport_,
1350 const struct ofproto_port_queue *qdscp_list,
1353 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1354 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1355 struct hmap new = HMAP_INITIALIZER(&new);
1358 for (i = 0; i < n_qdscp; i++) {
1359 struct priority_to_dscp *pdscp;
1363 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1364 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1369 pdscp = get_priority(ofport, priority);
1371 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1373 pdscp = xmalloc(sizeof *pdscp);
1374 pdscp->priority = priority;
1376 ofproto->need_revalidate = true;
1379 if (pdscp->dscp != dscp) {
1381 ofproto->need_revalidate = true;
1384 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1387 if (!hmap_is_empty(&ofport->priorities)) {
1388 ofport_clear_priorities(ofport);
1389 ofproto->need_revalidate = true;
1392 hmap_swap(&new, &ofport->priorities);
1400 /* Expires all MAC learning entries associated with 'bundle' and forces its
1401 * ofproto to revalidate every flow.
1403 * Normally MAC learning entries are removed only from the ofproto associated
1404 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1405 * are removed from every ofproto. When patch ports and SLB bonds are in use
1406 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1407 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1408 * with the host from which it migrated. */
1410 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1412 struct ofproto_dpif *ofproto = bundle->ofproto;
1413 struct mac_learning *ml = ofproto->ml;
1414 struct mac_entry *mac, *next_mac;
1416 ofproto->need_revalidate = true;
1417 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1418 if (mac->port.p == bundle) {
1420 struct ofproto_dpif *o;
1422 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1424 struct mac_entry *e;
1426 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1429 tag_set_add(&o->revalidate_set, e->tag);
1430 mac_learning_expire(o->ml, e);
1436 mac_learning_expire(ml, mac);
1441 static struct ofbundle *
1442 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1444 struct ofbundle *bundle;
1446 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1447 &ofproto->bundles) {
1448 if (bundle->aux == aux) {
1455 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1456 * ones that are found to 'bundles'. */
1458 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1459 void **auxes, size_t n_auxes,
1460 struct hmapx *bundles)
1464 hmapx_init(bundles);
1465 for (i = 0; i < n_auxes; i++) {
1466 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1468 hmapx_add(bundles, bundle);
1474 bundle_update(struct ofbundle *bundle)
1476 struct ofport_dpif *port;
1478 bundle->floodable = true;
1479 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1480 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
1481 bundle->floodable = false;
1488 bundle_del_port(struct ofport_dpif *port)
1490 struct ofbundle *bundle = port->bundle;
1492 bundle->ofproto->need_revalidate = true;
1494 list_remove(&port->bundle_node);
1495 port->bundle = NULL;
1498 lacp_slave_unregister(bundle->lacp, port);
1501 bond_slave_unregister(bundle->bond, port);
1504 bundle_update(bundle);
1508 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1509 struct lacp_slave_settings *lacp,
1510 uint32_t bond_stable_id)
1512 struct ofport_dpif *port;
1514 port = get_ofp_port(bundle->ofproto, ofp_port);
1519 if (port->bundle != bundle) {
1520 bundle->ofproto->need_revalidate = true;
1522 bundle_del_port(port);
1525 port->bundle = bundle;
1526 list_push_back(&bundle->ports, &port->bundle_node);
1527 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
1528 bundle->floodable = false;
1532 port->bundle->ofproto->need_revalidate = true;
1533 lacp_slave_register(bundle->lacp, port, lacp);
1536 port->bond_stable_id = bond_stable_id;
1542 bundle_destroy(struct ofbundle *bundle)
1544 struct ofproto_dpif *ofproto;
1545 struct ofport_dpif *port, *next_port;
1552 ofproto = bundle->ofproto;
1553 for (i = 0; i < MAX_MIRRORS; i++) {
1554 struct ofmirror *m = ofproto->mirrors[i];
1556 if (m->out == bundle) {
1558 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1559 || hmapx_find_and_delete(&m->dsts, bundle)) {
1560 ofproto->need_revalidate = true;
1565 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1566 bundle_del_port(port);
1569 bundle_flush_macs(bundle, true);
1570 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1572 free(bundle->trunks);
1573 lacp_destroy(bundle->lacp);
1574 bond_destroy(bundle->bond);
1579 bundle_set(struct ofproto *ofproto_, void *aux,
1580 const struct ofproto_bundle_settings *s)
1582 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1583 bool need_flush = false;
1584 struct ofport_dpif *port;
1585 struct ofbundle *bundle;
1586 unsigned long *trunks;
1592 bundle_destroy(bundle_lookup(ofproto, aux));
1596 assert(s->n_slaves == 1 || s->bond != NULL);
1597 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1599 bundle = bundle_lookup(ofproto, aux);
1601 bundle = xmalloc(sizeof *bundle);
1603 bundle->ofproto = ofproto;
1604 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1605 hash_pointer(aux, 0));
1607 bundle->name = NULL;
1609 list_init(&bundle->ports);
1610 bundle->vlan_mode = PORT_VLAN_TRUNK;
1612 bundle->trunks = NULL;
1613 bundle->use_priority_tags = s->use_priority_tags;
1614 bundle->lacp = NULL;
1615 bundle->bond = NULL;
1617 bundle->floodable = true;
1619 bundle->src_mirrors = 0;
1620 bundle->dst_mirrors = 0;
1621 bundle->mirror_out = 0;
1624 if (!bundle->name || strcmp(s->name, bundle->name)) {
1626 bundle->name = xstrdup(s->name);
1631 if (!bundle->lacp) {
1632 ofproto->need_revalidate = true;
1633 bundle->lacp = lacp_create();
1635 lacp_configure(bundle->lacp, s->lacp);
1637 lacp_destroy(bundle->lacp);
1638 bundle->lacp = NULL;
1641 /* Update set of ports. */
1643 for (i = 0; i < s->n_slaves; i++) {
1644 if (!bundle_add_port(bundle, s->slaves[i],
1645 s->lacp ? &s->lacp_slaves[i] : NULL,
1646 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1650 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1651 struct ofport_dpif *next_port;
1653 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1654 for (i = 0; i < s->n_slaves; i++) {
1655 if (s->slaves[i] == port->up.ofp_port) {
1660 bundle_del_port(port);
1664 assert(list_size(&bundle->ports) <= s->n_slaves);
1666 if (list_is_empty(&bundle->ports)) {
1667 bundle_destroy(bundle);
1671 /* Set VLAN tagging mode */
1672 if (s->vlan_mode != bundle->vlan_mode
1673 || s->use_priority_tags != bundle->use_priority_tags) {
1674 bundle->vlan_mode = s->vlan_mode;
1675 bundle->use_priority_tags = s->use_priority_tags;
1680 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1681 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1683 if (vlan != bundle->vlan) {
1684 bundle->vlan = vlan;
1688 /* Get trunked VLANs. */
1689 switch (s->vlan_mode) {
1690 case PORT_VLAN_ACCESS:
1694 case PORT_VLAN_TRUNK:
1695 trunks = (unsigned long *) s->trunks;
1698 case PORT_VLAN_NATIVE_UNTAGGED:
1699 case PORT_VLAN_NATIVE_TAGGED:
1700 if (vlan != 0 && (!s->trunks
1701 || !bitmap_is_set(s->trunks, vlan)
1702 || bitmap_is_set(s->trunks, 0))) {
1703 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1705 trunks = bitmap_clone(s->trunks, 4096);
1707 trunks = bitmap_allocate1(4096);
1709 bitmap_set1(trunks, vlan);
1710 bitmap_set0(trunks, 0);
1712 trunks = (unsigned long *) s->trunks;
1719 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1720 free(bundle->trunks);
1721 if (trunks == s->trunks) {
1722 bundle->trunks = vlan_bitmap_clone(trunks);
1724 bundle->trunks = trunks;
1729 if (trunks != s->trunks) {
1734 if (!list_is_short(&bundle->ports)) {
1735 bundle->ofproto->has_bonded_bundles = true;
1737 if (bond_reconfigure(bundle->bond, s->bond)) {
1738 ofproto->need_revalidate = true;
1741 bundle->bond = bond_create(s->bond);
1742 ofproto->need_revalidate = true;
1745 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1746 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1750 bond_destroy(bundle->bond);
1751 bundle->bond = NULL;
1754 /* If we changed something that would affect MAC learning, un-learn
1755 * everything on this port and force flow revalidation. */
1757 bundle_flush_macs(bundle, false);
1764 bundle_remove(struct ofport *port_)
1766 struct ofport_dpif *port = ofport_dpif_cast(port_);
1767 struct ofbundle *bundle = port->bundle;
1770 bundle_del_port(port);
1771 if (list_is_empty(&bundle->ports)) {
1772 bundle_destroy(bundle);
1773 } else if (list_is_short(&bundle->ports)) {
1774 bond_destroy(bundle->bond);
1775 bundle->bond = NULL;
1781 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
1783 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1784 struct ofport_dpif *port = port_;
1785 uint8_t ea[ETH_ADDR_LEN];
1788 error = netdev_get_etheraddr(port->up.netdev, ea);
1790 struct ofpbuf packet;
1793 ofpbuf_init(&packet, 0);
1794 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1796 memcpy(packet_pdu, pdu, pdu_size);
1798 send_packet(port, &packet);
1799 ofpbuf_uninit(&packet);
1801 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1802 "%s (%s)", port->bundle->name,
1803 netdev_get_name(port->up.netdev), strerror(error));
1808 bundle_send_learning_packets(struct ofbundle *bundle)
1810 struct ofproto_dpif *ofproto = bundle->ofproto;
1811 int error, n_packets, n_errors;
1812 struct mac_entry *e;
1814 error = n_packets = n_errors = 0;
1815 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1816 if (e->port.p != bundle) {
1817 struct ofpbuf *learning_packet;
1818 struct ofport_dpif *port;
1821 learning_packet = bond_compose_learning_packet(bundle->bond, e->mac,
1824 ret = send_packet(port, learning_packet);
1825 ofpbuf_delete(learning_packet);
1835 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1836 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1837 "packets, last error was: %s",
1838 bundle->name, n_errors, n_packets, strerror(error));
1840 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1841 bundle->name, n_packets);
1846 bundle_run(struct ofbundle *bundle)
1849 lacp_run(bundle->lacp, send_pdu_cb);
1852 struct ofport_dpif *port;
1854 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1855 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1858 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1859 lacp_negotiated(bundle->lacp));
1860 if (bond_should_send_learning_packets(bundle->bond)) {
1861 bundle_send_learning_packets(bundle);
1867 bundle_wait(struct ofbundle *bundle)
1870 lacp_wait(bundle->lacp);
1873 bond_wait(bundle->bond);
1880 mirror_scan(struct ofproto_dpif *ofproto)
1884 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1885 if (!ofproto->mirrors[idx]) {
1892 static struct ofmirror *
1893 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1897 for (i = 0; i < MAX_MIRRORS; i++) {
1898 struct ofmirror *mirror = ofproto->mirrors[i];
1899 if (mirror && mirror->aux == aux) {
1907 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
1909 mirror_update_dups(struct ofproto_dpif *ofproto)
1913 for (i = 0; i < MAX_MIRRORS; i++) {
1914 struct ofmirror *m = ofproto->mirrors[i];
1917 m->dup_mirrors = MIRROR_MASK_C(1) << i;
1921 for (i = 0; i < MAX_MIRRORS; i++) {
1922 struct ofmirror *m1 = ofproto->mirrors[i];
1929 for (j = i + 1; j < MAX_MIRRORS; j++) {
1930 struct ofmirror *m2 = ofproto->mirrors[j];
1932 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
1933 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
1934 m2->dup_mirrors |= m1->dup_mirrors;
1941 mirror_set(struct ofproto *ofproto_, void *aux,
1942 const struct ofproto_mirror_settings *s)
1944 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1945 mirror_mask_t mirror_bit;
1946 struct ofbundle *bundle;
1947 struct ofmirror *mirror;
1948 struct ofbundle *out;
1949 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
1950 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
1953 mirror = mirror_lookup(ofproto, aux);
1955 mirror_destroy(mirror);
1961 idx = mirror_scan(ofproto);
1963 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
1965 ofproto->up.name, MAX_MIRRORS, s->name);
1969 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
1970 mirror->ofproto = ofproto;
1973 mirror->out_vlan = -1;
1974 mirror->name = NULL;
1977 if (!mirror->name || strcmp(s->name, mirror->name)) {
1979 mirror->name = xstrdup(s->name);
1982 /* Get the new configuration. */
1983 if (s->out_bundle) {
1984 out = bundle_lookup(ofproto, s->out_bundle);
1986 mirror_destroy(mirror);
1992 out_vlan = s->out_vlan;
1994 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
1995 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
1997 /* If the configuration has not changed, do nothing. */
1998 if (hmapx_equals(&srcs, &mirror->srcs)
1999 && hmapx_equals(&dsts, &mirror->dsts)
2000 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2001 && mirror->out == out
2002 && mirror->out_vlan == out_vlan)
2004 hmapx_destroy(&srcs);
2005 hmapx_destroy(&dsts);
2009 hmapx_swap(&srcs, &mirror->srcs);
2010 hmapx_destroy(&srcs);
2012 hmapx_swap(&dsts, &mirror->dsts);
2013 hmapx_destroy(&dsts);
2015 free(mirror->vlans);
2016 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2019 mirror->out_vlan = out_vlan;
2021 /* Update bundles. */
2022 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2023 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2024 if (hmapx_contains(&mirror->srcs, bundle)) {
2025 bundle->src_mirrors |= mirror_bit;
2027 bundle->src_mirrors &= ~mirror_bit;
2030 if (hmapx_contains(&mirror->dsts, bundle)) {
2031 bundle->dst_mirrors |= mirror_bit;
2033 bundle->dst_mirrors &= ~mirror_bit;
2036 if (mirror->out == bundle) {
2037 bundle->mirror_out |= mirror_bit;
2039 bundle->mirror_out &= ~mirror_bit;
2043 ofproto->need_revalidate = true;
2044 mac_learning_flush(ofproto->ml);
2045 mirror_update_dups(ofproto);
2051 mirror_destroy(struct ofmirror *mirror)
2053 struct ofproto_dpif *ofproto;
2054 mirror_mask_t mirror_bit;
2055 struct ofbundle *bundle;
2061 ofproto = mirror->ofproto;
2062 ofproto->need_revalidate = true;
2063 mac_learning_flush(ofproto->ml);
2065 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2066 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2067 bundle->src_mirrors &= ~mirror_bit;
2068 bundle->dst_mirrors &= ~mirror_bit;
2069 bundle->mirror_out &= ~mirror_bit;
2072 hmapx_destroy(&mirror->srcs);
2073 hmapx_destroy(&mirror->dsts);
2074 free(mirror->vlans);
2076 ofproto->mirrors[mirror->idx] = NULL;
2080 mirror_update_dups(ofproto);
2084 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2085 uint64_t *packets, uint64_t *bytes)
2087 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2088 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2091 *packets = *bytes = UINT64_MAX;
2095 *packets = mirror->packet_count;
2096 *bytes = mirror->byte_count;
2102 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2104 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2105 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2106 ofproto->need_revalidate = true;
2107 mac_learning_flush(ofproto->ml);
2113 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2115 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2116 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2117 return bundle && bundle->mirror_out != 0;
2121 forward_bpdu_changed(struct ofproto *ofproto_)
2123 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2124 /* Revalidate cached flows whenever forward_bpdu option changes. */
2125 ofproto->need_revalidate = true;
2130 static struct ofport_dpif *
2131 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
2133 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2134 return ofport ? ofport_dpif_cast(ofport) : NULL;
2137 static struct ofport_dpif *
2138 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
2140 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2144 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2145 struct dpif_port *dpif_port)
2147 ofproto_port->name = dpif_port->name;
2148 ofproto_port->type = dpif_port->type;
2149 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2153 port_run(struct ofport_dpif *ofport)
2155 bool enable = netdev_get_carrier(ofport->up.netdev);
2158 cfm_run(ofport->cfm);
2160 if (cfm_should_send_ccm(ofport->cfm)) {
2161 struct ofpbuf packet;
2163 ofpbuf_init(&packet, 0);
2164 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr);
2165 send_packet(ofport, &packet);
2166 ofpbuf_uninit(&packet);
2169 enable = enable && !cfm_get_fault(ofport->cfm)
2170 && cfm_get_opup(ofport->cfm);
2173 if (ofport->bundle) {
2174 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2177 if (ofport->may_enable != enable) {
2178 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2180 if (ofproto->has_bundle_action) {
2181 ofproto->need_revalidate = true;
2185 ofport->may_enable = enable;
2189 port_wait(struct ofport_dpif *ofport)
2192 cfm_wait(ofport->cfm);
2197 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2198 struct ofproto_port *ofproto_port)
2200 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2201 struct dpif_port dpif_port;
2204 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2206 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2212 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2214 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2218 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2220 *ofp_portp = odp_port_to_ofp_port(odp_port);
2226 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2228 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2231 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2233 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2235 /* The caller is going to close ofport->up.netdev. If this is a
2236 * bonded port, then the bond is using that netdev, so remove it
2237 * from the bond. The client will need to reconfigure everything
2238 * after deleting ports, so then the slave will get re-added. */
2239 bundle_remove(&ofport->up);
2246 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2248 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2251 error = netdev_get_stats(ofport->up.netdev, stats);
2253 if (!error && ofport->odp_port == OVSP_LOCAL) {
2254 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2256 /* ofproto->stats.tx_packets represents packets that we created
2257 * internally and sent to some port (e.g. packets sent with
2258 * send_packet()). Account for them as if they had come from
2259 * OFPP_LOCAL and got forwarded. */
2261 if (stats->rx_packets != UINT64_MAX) {
2262 stats->rx_packets += ofproto->stats.tx_packets;
2265 if (stats->rx_bytes != UINT64_MAX) {
2266 stats->rx_bytes += ofproto->stats.tx_bytes;
2269 /* ofproto->stats.rx_packets represents packets that were received on
2270 * some port and we processed internally and dropped (e.g. STP).
2271 * Account fro them as if they had been forwarded to OFPP_LOCAL. */
2273 if (stats->tx_packets != UINT64_MAX) {
2274 stats->tx_packets += ofproto->stats.rx_packets;
2277 if (stats->tx_bytes != UINT64_MAX) {
2278 stats->tx_bytes += ofproto->stats.rx_bytes;
2285 /* Account packets for LOCAL port. */
2287 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2288 size_t tx_size, size_t rx_size)
2290 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2293 ofproto->stats.rx_packets++;
2294 ofproto->stats.rx_bytes += rx_size;
2297 ofproto->stats.tx_packets++;
2298 ofproto->stats.tx_bytes += tx_size;
2302 struct port_dump_state {
2303 struct dpif_port_dump dump;
2308 port_dump_start(const struct ofproto *ofproto_, void **statep)
2310 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2311 struct port_dump_state *state;
2313 *statep = state = xmalloc(sizeof *state);
2314 dpif_port_dump_start(&state->dump, ofproto->dpif);
2315 state->done = false;
2320 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2321 struct ofproto_port *port)
2323 struct port_dump_state *state = state_;
2324 struct dpif_port dpif_port;
2326 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2327 ofproto_port_from_dpif_port(port, &dpif_port);
2330 int error = dpif_port_dump_done(&state->dump);
2332 return error ? error : EOF;
2337 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2339 struct port_dump_state *state = state_;
2342 dpif_port_dump_done(&state->dump);
2349 port_poll(const struct ofproto *ofproto_, char **devnamep)
2351 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2352 return dpif_port_poll(ofproto->dpif, devnamep);
2356 port_poll_wait(const struct ofproto *ofproto_)
2358 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2359 dpif_port_poll_wait(ofproto->dpif);
2363 port_is_lacp_current(const struct ofport *ofport_)
2365 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2366 return (ofport->bundle && ofport->bundle->lacp
2367 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2371 /* Upcall handling. */
2373 /* Flow miss batching.
2375 * Some dpifs implement operations faster when you hand them off in a batch.
2376 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2377 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2378 * more packets, plus possibly installing the flow in the dpif.
2380 * So far we only batch the operations that affect flow setup time the most.
2381 * It's possible to batch more than that, but the benefit might be minimal. */
2383 struct hmap_node hmap_node;
2385 enum odp_key_fitness key_fitness;
2386 const struct nlattr *key;
2388 ovs_be16 initial_tci;
2389 struct list packets;
2392 struct flow_miss_op {
2393 union dpif_op dpif_op;
2394 struct subfacet *subfacet;
2397 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2398 * OpenFlow controller as necessary according to their individual
2401 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
2402 * ownership is transferred to this function. */
2404 send_packet_in_miss(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
2405 const struct flow *flow, bool clone)
2407 struct ofputil_packet_in pin;
2409 pin.packet = packet;
2410 pin.in_port = flow->in_port;
2411 pin.reason = OFPR_NO_MATCH;
2412 pin.buffer_id = 0; /* not yet known */
2413 pin.send_len = 0; /* not used for flow table misses */
2414 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
2415 clone ? NULL : packet);
2418 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_ACTION to each
2419 * OpenFlow controller as necessary according to their individual
2422 * 'send_len' should be the number of bytes of 'packet' to send to the
2423 * controller, as specified in the action that caused the packet to be sent.
2425 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
2426 * Otherwise, ownership is transferred to this function. */
2428 send_packet_in_action(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
2429 uint64_t userdata, const struct flow *flow, bool clone)
2431 struct ofputil_packet_in pin;
2432 struct user_action_cookie cookie;
2434 memcpy(&cookie, &userdata, sizeof(cookie));
2436 pin.packet = packet;
2437 pin.in_port = flow->in_port;
2438 pin.reason = OFPR_ACTION;
2439 pin.buffer_id = 0; /* not yet known */
2440 pin.send_len = cookie.data;
2441 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
2442 clone ? NULL : packet);
2446 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2447 const struct ofpbuf *packet)
2449 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2455 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2457 cfm_process_heartbeat(ofport->cfm, packet);
2460 } else if (ofport->bundle && ofport->bundle->lacp
2461 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2463 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2466 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2468 stp_process_packet(ofport, packet);
2475 static struct flow_miss *
2476 flow_miss_create(struct hmap *todo, const struct flow *flow,
2477 enum odp_key_fitness key_fitness,
2478 const struct nlattr *key, size_t key_len,
2479 ovs_be16 initial_tci)
2481 uint32_t hash = flow_hash(flow, 0);
2482 struct flow_miss *miss;
2484 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2485 if (flow_equal(&miss->flow, flow)) {
2490 miss = xmalloc(sizeof *miss);
2491 hmap_insert(todo, &miss->hmap_node, hash);
2493 miss->key_fitness = key_fitness;
2495 miss->key_len = key_len;
2496 miss->initial_tci = initial_tci;
2497 list_init(&miss->packets);
2502 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2503 struct flow_miss_op *ops, size_t *n_ops)
2505 const struct flow *flow = &miss->flow;
2506 struct ofpbuf *packet, *next_packet;
2507 struct subfacet *subfacet;
2508 struct facet *facet;
2510 facet = facet_lookup_valid(ofproto, flow);
2512 struct rule_dpif *rule;
2514 rule = rule_dpif_lookup(ofproto, flow, 0);
2516 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
2517 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
2519 if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) {
2520 COVERAGE_INC(ofproto_dpif_no_packet_in);
2521 /* XXX install 'drop' flow entry */
2525 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
2529 LIST_FOR_EACH_SAFE (packet, next_packet, list_node,
2531 list_remove(&packet->list_node);
2532 send_packet_in_miss(ofproto, packet, flow, false);
2538 facet = facet_create(rule, flow);
2541 subfacet = subfacet_create(ofproto, facet,
2542 miss->key_fitness, miss->key, miss->key_len,
2545 LIST_FOR_EACH_SAFE (packet, next_packet, list_node, &miss->packets) {
2546 struct dpif_flow_stats stats;
2548 list_remove(&packet->list_node);
2549 ofproto->n_matches++;
2551 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2553 * Extra-special case for fail-open mode.
2555 * We are in fail-open mode and the packet matched the fail-open
2556 * rule, but we are connected to a controller too. We should send
2557 * the packet up to the controller in the hope that it will try to
2558 * set up a flow and thereby allow us to exit fail-open.
2560 * See the top-level comment in fail-open.c for more information.
2562 send_packet_in_miss(ofproto, packet, flow, true);
2565 if (!facet->may_install || !subfacet->actions) {
2566 subfacet_make_actions(ofproto, subfacet, packet);
2569 /* Credit statistics to subfacet for this packet. We must do this now
2570 * because execute_controller_action() below may destroy 'packet'. */
2571 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2572 subfacet_update_stats(ofproto, subfacet, &stats);
2574 if (!execute_controller_action(ofproto, &facet->flow,
2576 subfacet->actions_len, packet, true)) {
2577 struct flow_miss_op *op = &ops[(*n_ops)++];
2578 struct dpif_execute *execute = &op->dpif_op.execute;
2580 op->subfacet = subfacet;
2581 execute->type = DPIF_OP_EXECUTE;
2582 execute->key = miss->key;
2583 execute->key_len = miss->key_len;
2585 = (facet->may_install
2587 : xmemdup(subfacet->actions, subfacet->actions_len));
2588 execute->actions_len = subfacet->actions_len;
2589 execute->packet = packet;
2593 if (facet->may_install && subfacet->key_fitness != ODP_FIT_TOO_LITTLE) {
2594 struct flow_miss_op *op = &ops[(*n_ops)++];
2595 struct dpif_flow_put *put = &op->dpif_op.flow_put;
2597 op->subfacet = subfacet;
2598 put->type = DPIF_OP_FLOW_PUT;
2599 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2600 put->key = miss->key;
2601 put->key_len = miss->key_len;
2602 put->actions = subfacet->actions;
2603 put->actions_len = subfacet->actions_len;
2608 static enum odp_key_fitness
2609 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
2610 const struct nlattr *key, size_t key_len,
2611 struct flow *flow, ovs_be16 *initial_tci)
2613 enum odp_key_fitness fitness;
2617 fitness = odp_flow_key_to_flow(key, key_len, flow);
2618 if (fitness == ODP_FIT_ERROR) {
2621 *initial_tci = flow->vlan_tci;
2623 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
2625 /* Cause the flow to be processed as if it came in on the real device
2626 * with the VLAN device's VLAN ID. */
2627 flow->in_port = realdev;
2628 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
2630 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
2631 if (fitness == ODP_FIT_PERFECT) {
2632 fitness = ODP_FIT_TOO_MUCH;
2640 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
2643 struct dpif_upcall *upcall;
2644 struct flow_miss *miss, *next_miss;
2645 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
2646 union dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
2655 /* Construct the to-do list.
2657 * This just amounts to extracting the flow from each packet and sticking
2658 * the packets that have the same flow in the same "flow_miss" structure so
2659 * that we can process them together. */
2661 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
2662 enum odp_key_fitness fitness;
2663 struct flow_miss *miss;
2664 ovs_be16 initial_tci;
2667 /* Obtain metadata and check userspace/kernel agreement on flow match,
2668 * then set 'flow''s header pointers. */
2669 fitness = ofproto_dpif_extract_flow_key(ofproto,
2670 upcall->key, upcall->key_len,
2671 &flow, &initial_tci);
2672 if (fitness == ODP_FIT_ERROR) {
2673 ofpbuf_delete(upcall->packet);
2676 flow_extract(upcall->packet, flow.priority, flow.tun_id,
2677 flow.in_port, &flow);
2679 /* Handle 802.1ag, LACP, and STP specially. */
2680 if (process_special(ofproto, &flow, upcall->packet)) {
2681 ofproto_update_local_port_stats(&ofproto->up,
2682 0, upcall->packet->size);
2683 ofpbuf_delete(upcall->packet);
2684 ofproto->n_matches++;
2688 /* Add other packets to a to-do list. */
2689 miss = flow_miss_create(&todo, &flow, fitness,
2690 upcall->key, upcall->key_len, initial_tci);
2691 list_push_back(&miss->packets, &upcall->packet->list_node);
2694 /* Process each element in the to-do list, constructing the set of
2695 * operations to batch. */
2697 HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) {
2698 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
2699 ofpbuf_list_delete(&miss->packets);
2700 hmap_remove(&todo, &miss->hmap_node);
2703 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
2704 hmap_destroy(&todo);
2706 /* Execute batch. */
2707 for (i = 0; i < n_ops; i++) {
2708 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
2710 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
2712 /* Free memory and update facets. */
2713 for (i = 0; i < n_ops; i++) {
2714 struct flow_miss_op *op = &flow_miss_ops[i];
2715 struct dpif_execute *execute;
2716 struct dpif_flow_put *put;
2718 switch (op->dpif_op.type) {
2719 case DPIF_OP_EXECUTE:
2720 execute = &op->dpif_op.execute;
2721 if (op->subfacet->actions != execute->actions) {
2722 free((struct nlattr *) execute->actions);
2724 ofpbuf_delete((struct ofpbuf *) execute->packet);
2727 case DPIF_OP_FLOW_PUT:
2728 put = &op->dpif_op.flow_put;
2730 op->subfacet->installed = true;
2738 handle_userspace_upcall(struct ofproto_dpif *ofproto,
2739 struct dpif_upcall *upcall)
2741 struct user_action_cookie cookie;
2742 enum odp_key_fitness fitness;
2743 ovs_be16 initial_tci;
2746 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
2748 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
2749 upcall->key_len, &flow,
2751 if (fitness == ODP_FIT_ERROR) {
2752 ofpbuf_delete(upcall->packet);
2756 if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
2757 if (ofproto->sflow) {
2758 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
2761 ofpbuf_delete(upcall->packet);
2762 } else if (cookie.type == USER_ACTION_COOKIE_CONTROLLER) {
2763 COVERAGE_INC(ofproto_dpif_ctlr_action);
2764 send_packet_in_action(ofproto, upcall->packet, upcall->userdata,
2767 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
2768 ofpbuf_delete(upcall->packet);
2773 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
2775 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
2779 assert (max_batch <= FLOW_MISS_MAX_BATCH);
2782 for (i = 0; i < max_batch; i++) {
2783 struct dpif_upcall *upcall = &misses[n_misses];
2786 error = dpif_recv(ofproto->dpif, upcall);
2791 switch (upcall->type) {
2792 case DPIF_UC_ACTION:
2793 handle_userspace_upcall(ofproto, upcall);
2797 /* Handle it later. */
2801 case DPIF_N_UC_TYPES:
2803 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32,
2809 handle_miss_upcalls(ofproto, misses, n_misses);
2814 /* Flow expiration. */
2816 static int subfacet_max_idle(const struct ofproto_dpif *);
2817 static void update_stats(struct ofproto_dpif *);
2818 static void rule_expire(struct rule_dpif *);
2819 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
2821 /* This function is called periodically by run(). Its job is to collect
2822 * updates for the flows that have been installed into the datapath, most
2823 * importantly when they last were used, and then use that information to
2824 * expire flows that have not been used recently.
2826 * Returns the number of milliseconds after which it should be called again. */
2828 expire(struct ofproto_dpif *ofproto)
2830 struct rule_dpif *rule, *next_rule;
2831 struct classifier *table;
2834 /* Update stats for each flow in the datapath. */
2835 update_stats(ofproto);
2837 /* Expire subfacets that have been idle too long. */
2838 dp_max_idle = subfacet_max_idle(ofproto);
2839 expire_subfacets(ofproto, dp_max_idle);
2841 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
2842 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
2843 struct cls_cursor cursor;
2845 cls_cursor_init(&cursor, table, NULL);
2846 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
2851 /* All outstanding data in existing flows has been accounted, so it's a
2852 * good time to do bond rebalancing. */
2853 if (ofproto->has_bonded_bundles) {
2854 struct ofbundle *bundle;
2856 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2858 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
2863 return MIN(dp_max_idle, 1000);
2866 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
2868 * This function also pushes statistics updates to rules which each facet
2869 * resubmits into. Generally these statistics will be accurate. However, if a
2870 * facet changes the rule it resubmits into at some time in between
2871 * update_stats() runs, it is possible that statistics accrued to the
2872 * old rule will be incorrectly attributed to the new rule. This could be
2873 * avoided by calling update_stats() whenever rules are created or
2874 * deleted. However, the performance impact of making so many calls to the
2875 * datapath do not justify the benefit of having perfectly accurate statistics.
2878 update_stats(struct ofproto_dpif *p)
2880 const struct dpif_flow_stats *stats;
2881 struct dpif_flow_dump dump;
2882 const struct nlattr *key;
2885 dpif_flow_dump_start(&dump, p->dpif);
2886 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
2887 struct subfacet *subfacet;
2889 subfacet = subfacet_find(p, key, key_len);
2890 if (subfacet && subfacet->installed) {
2891 struct facet *facet = subfacet->facet;
2893 if (stats->n_packets >= subfacet->dp_packet_count) {
2894 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
2895 facet->packet_count += extra;
2897 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
2900 if (stats->n_bytes >= subfacet->dp_byte_count) {
2901 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
2903 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
2906 subfacet->dp_packet_count = stats->n_packets;
2907 subfacet->dp_byte_count = stats->n_bytes;
2909 subfacet_update_time(p, subfacet, stats->used);
2910 facet_account(p, facet);
2911 facet_push_stats(facet);
2913 if (!VLOG_DROP_WARN(&rl)) {
2917 odp_flow_key_format(key, key_len, &s);
2918 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
2922 COVERAGE_INC(facet_unexpected);
2923 /* There's a flow in the datapath that we know nothing about, or a
2924 * flow that shouldn't be installed but was anyway. Delete it. */
2925 dpif_flow_del(p->dpif, key, key_len, NULL);
2928 dpif_flow_dump_done(&dump);
2931 /* Calculates and returns the number of milliseconds of idle time after which
2932 * subfacets should expire from the datapath. When a subfacet expires, we fold
2933 * its statistics into its facet, and when a facet's last subfacet expires, we
2934 * fold its statistic into its rule. */
2936 subfacet_max_idle(const struct ofproto_dpif *ofproto)
2939 * Idle time histogram.
2941 * Most of the time a switch has a relatively small number of subfacets.
2942 * When this is the case we might as well keep statistics for all of them
2943 * in userspace and to cache them in the kernel datapath for performance as
2946 * As the number of subfacets increases, the memory required to maintain
2947 * statistics about them in userspace and in the kernel becomes
2948 * significant. However, with a large number of subfacets it is likely
2949 * that only a few of them are "heavy hitters" that consume a large amount
2950 * of bandwidth. At this point, only heavy hitters are worth caching in
2951 * the kernel and maintaining in userspaces; other subfacets we can
2954 * The technique used to compute the idle time is to build a histogram with
2955 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
2956 * that is installed in the kernel gets dropped in the appropriate bucket.
2957 * After the histogram has been built, we compute the cutoff so that only
2958 * the most-recently-used 1% of subfacets (but at least
2959 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
2960 * the most-recently-used bucket of subfacets is kept, so actually an
2961 * arbitrary number of subfacets can be kept in any given expiration run
2962 * (though the next run will delete most of those unless they receive
2965 * This requires a second pass through the subfacets, in addition to the
2966 * pass made by update_stats(), because the former function never looks at
2967 * uninstallable subfacets.
2969 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
2970 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
2971 int buckets[N_BUCKETS] = { 0 };
2972 int total, subtotal, bucket;
2973 struct subfacet *subfacet;
2977 total = hmap_count(&ofproto->subfacets);
2978 if (total <= ofproto->up.flow_eviction_threshold) {
2979 return N_BUCKETS * BUCKET_WIDTH;
2982 /* Build histogram. */
2984 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
2985 long long int idle = now - subfacet->used;
2986 int bucket = (idle <= 0 ? 0
2987 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
2988 : (unsigned int) idle / BUCKET_WIDTH);
2992 /* Find the first bucket whose flows should be expired. */
2993 subtotal = bucket = 0;
2995 subtotal += buckets[bucket++];
2996 } while (bucket < N_BUCKETS &&
2997 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
2999 if (VLOG_IS_DBG_ENABLED()) {
3003 ds_put_cstr(&s, "keep");
3004 for (i = 0; i < N_BUCKETS; i++) {
3006 ds_put_cstr(&s, ", drop");
3009 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3012 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3016 return bucket * BUCKET_WIDTH;
3020 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3022 long long int cutoff = time_msec() - dp_max_idle;
3023 struct subfacet *subfacet, *next_subfacet;
3025 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3026 &ofproto->subfacets) {
3027 if (subfacet->used < cutoff) {
3028 subfacet_destroy(ofproto, subfacet);
3033 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3034 * then delete it entirely. */
3036 rule_expire(struct rule_dpif *rule)
3038 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3039 struct facet *facet, *next_facet;
3043 /* Has 'rule' expired? */
3045 if (rule->up.hard_timeout
3046 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3047 reason = OFPRR_HARD_TIMEOUT;
3048 } else if (rule->up.idle_timeout && list_is_empty(&rule->facets)
3049 && now > rule->used + rule->up.idle_timeout * 1000) {
3050 reason = OFPRR_IDLE_TIMEOUT;
3055 COVERAGE_INC(ofproto_dpif_expired);
3057 /* Update stats. (This is a no-op if the rule expired due to an idle
3058 * timeout, because that only happens when the rule has no facets left.) */
3059 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3060 facet_remove(ofproto, facet);
3063 /* Get rid of the rule. */
3064 ofproto_rule_expire(&rule->up, reason);
3069 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3071 * The caller must already have determined that no facet with an identical
3072 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3073 * the ofproto's classifier table.
3075 * The facet will initially have no subfacets. The caller should create (at
3076 * least) one subfacet with subfacet_create(). */
3077 static struct facet *
3078 facet_create(struct rule_dpif *rule, const struct flow *flow)
3080 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3081 struct facet *facet;
3083 facet = xzalloc(sizeof *facet);
3084 facet->used = time_msec();
3085 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
3086 list_push_back(&rule->facets, &facet->list_node);
3088 facet->flow = *flow;
3089 list_init(&facet->subfacets);
3090 netflow_flow_init(&facet->nf_flow);
3091 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3097 facet_free(struct facet *facet)
3102 /* If the 'actions_len' bytes of actions in 'odp_actions' are just a single
3103 * OVS_ACTION_ATTR_USERSPACE action, executes it internally and returns true.
3104 * Otherwise, returns false without doing anything.
3106 * If 'clone' is true, the caller always retains ownership of 'packet'.
3107 * Otherwise, ownership is transferred to this function if it returns true. */
3109 execute_controller_action(struct ofproto_dpif *ofproto,
3110 const struct flow *flow,
3111 const struct nlattr *odp_actions, size_t actions_len,
3112 struct ofpbuf *packet, bool clone)
3115 && odp_actions->nla_type == OVS_ACTION_ATTR_USERSPACE
3116 && NLA_ALIGN(odp_actions->nla_len) == actions_len) {
3117 /* As an optimization, avoid a round-trip from userspace to kernel to
3118 * userspace. This also avoids possibly filling up kernel packet
3119 * buffers along the way.
3121 * This optimization will not accidentally catch sFlow
3122 * OVS_ACTION_ATTR_USERSPACE actions, since those are encapsulated
3123 * inside OVS_ACTION_ATTR_SAMPLE. */
3124 const struct nlattr *nla;
3126 nla = nl_attr_find_nested(odp_actions, OVS_USERSPACE_ATTR_USERDATA);
3127 send_packet_in_action(ofproto, packet, nl_attr_get_u64(nla), flow,
3135 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3136 * 'packet', which arrived on 'in_port'.
3138 * Takes ownership of 'packet'. */
3140 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3141 const struct nlattr *odp_actions, size_t actions_len,
3142 struct ofpbuf *packet)
3144 struct odputil_keybuf keybuf;
3148 if (execute_controller_action(ofproto, flow, odp_actions, actions_len,
3153 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3154 odp_flow_key_from_flow(&key, flow);
3156 error = dpif_execute(ofproto->dpif, key.data, key.size,
3157 odp_actions, actions_len, packet);
3159 ofpbuf_delete(packet);
3163 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3165 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3166 * rule's statistics, via subfacet_uninstall().
3168 * - Removes 'facet' from its rule and from ofproto->facets.
3171 facet_remove(struct ofproto_dpif *ofproto, struct facet *facet)
3173 struct subfacet *subfacet, *next_subfacet;
3175 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3176 &facet->subfacets) {
3177 subfacet_destroy__(ofproto, subfacet);
3180 facet_flush_stats(ofproto, facet);
3181 hmap_remove(&ofproto->facets, &facet->hmap_node);
3182 list_remove(&facet->list_node);
3187 facet_account(struct ofproto_dpif *ofproto, struct facet *facet)
3190 struct subfacet *subfacet;
3191 const struct nlattr *a;
3195 if (facet->byte_count <= facet->accounted_bytes) {
3198 n_bytes = facet->byte_count - facet->accounted_bytes;
3199 facet->accounted_bytes = facet->byte_count;
3201 /* Feed information from the active flows back into the learning table to
3202 * ensure that table is always in sync with what is actually flowing
3203 * through the datapath. */
3204 if (facet->has_learn || facet->has_normal) {
3205 struct action_xlate_ctx ctx;
3207 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3208 facet->flow.vlan_tci, NULL);
3209 ctx.may_learn = true;
3210 ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions,
3211 facet->rule->up.n_actions));
3214 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3218 /* This loop feeds byte counters to bond_account() for rebalancing to use
3219 * as a basis. We also need to track the actual VLAN on which the packet
3220 * is going to be sent to ensure that it matches the one passed to
3221 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3224 * We use the actions from an arbitrary subfacet because they should all
3225 * be equally valid for our purpose. */
3226 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3227 struct subfacet, list_node);
3228 vlan_tci = facet->flow.vlan_tci;
3229 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3230 subfacet->actions, subfacet->actions_len) {
3231 const struct ovs_action_push_vlan *vlan;
3232 struct ofport_dpif *port;
3234 switch (nl_attr_type(a)) {
3235 case OVS_ACTION_ATTR_OUTPUT:
3236 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3237 if (port && port->bundle && port->bundle->bond) {
3238 bond_account(port->bundle->bond, &facet->flow,
3239 vlan_tci_to_vid(vlan_tci), n_bytes);
3243 case OVS_ACTION_ATTR_POP_VLAN:
3244 vlan_tci = htons(0);
3247 case OVS_ACTION_ATTR_PUSH_VLAN:
3248 vlan = nl_attr_get(a);
3249 vlan_tci = vlan->vlan_tci;
3255 /* Returns true if the only action for 'facet' is to send to the controller.
3256 * (We don't report NetFlow expiration messages for such facets because they
3257 * are just part of the control logic for the network, not real traffic). */
3259 facet_is_controller_flow(struct facet *facet)
3262 && facet->rule->up.n_actions == 1
3263 && action_outputs_to_port(&facet->rule->up.actions[0],
3264 htons(OFPP_CONTROLLER)));
3267 /* Folds all of 'facet''s statistics into its rule. Also updates the
3268 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3269 * 'facet''s statistics in the datapath should have been zeroed and folded into
3270 * its packet and byte counts before this function is called. */
3272 facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet)
3274 struct subfacet *subfacet;
3276 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3277 assert(!subfacet->dp_byte_count);
3278 assert(!subfacet->dp_packet_count);
3281 facet_push_stats(facet);
3282 facet_account(ofproto, facet);
3284 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3285 struct ofexpired expired;
3286 expired.flow = facet->flow;
3287 expired.packet_count = facet->packet_count;
3288 expired.byte_count = facet->byte_count;
3289 expired.used = facet->used;
3290 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3293 facet->rule->packet_count += facet->packet_count;
3294 facet->rule->byte_count += facet->byte_count;
3296 /* Reset counters to prevent double counting if 'facet' ever gets
3298 facet_reset_counters(facet);
3300 netflow_flow_clear(&facet->nf_flow);
3303 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3304 * Returns it if found, otherwise a null pointer.
3306 * The returned facet might need revalidation; use facet_lookup_valid()
3307 * instead if that is important. */
3308 static struct facet *
3309 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
3311 struct facet *facet;
3313 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
3315 if (flow_equal(flow, &facet->flow)) {
3323 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3324 * Returns it if found, otherwise a null pointer.
3326 * The returned facet is guaranteed to be valid. */
3327 static struct facet *
3328 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
3330 struct facet *facet = facet_find(ofproto, flow);
3332 /* The facet we found might not be valid, since we could be in need of
3333 * revalidation. If it is not valid, don't return it. */
3335 && (ofproto->need_revalidate
3336 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))
3337 && !facet_revalidate(ofproto, facet)) {
3338 COVERAGE_INC(facet_invalidated);
3345 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
3347 * - If the rule found is different from 'facet''s current rule, moves
3348 * 'facet' to the new rule and recompiles its actions.
3350 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3351 * where it is and recompiles its actions anyway.
3353 * - If there is none, destroys 'facet'.
3355 * Returns true if 'facet' still exists, false if it has been destroyed. */
3357 facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet)
3360 struct nlattr *odp_actions;
3363 struct actions *new_actions;
3365 struct action_xlate_ctx ctx;
3366 struct rule_dpif *new_rule;
3367 struct subfacet *subfacet;
3368 bool actions_changed;
3371 COVERAGE_INC(facet_revalidate);
3373 /* Determine the new rule. */
3374 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3376 /* No new rule, so delete the facet. */
3377 facet_remove(ofproto, facet);
3381 /* Calculate new datapath actions.
3383 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3384 * emit a NetFlow expiration and, if so, we need to have the old state
3385 * around to properly compose it. */
3387 /* If the datapath actions changed or the installability changed,
3388 * then we need to talk to the datapath. */
3391 memset(&ctx, 0, sizeof ctx);
3392 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3393 struct ofpbuf *odp_actions;
3394 bool should_install;
3396 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3397 subfacet->initial_tci, NULL);
3398 odp_actions = xlate_actions(&ctx, new_rule->up.actions,
3399 new_rule->up.n_actions);
3400 actions_changed = (subfacet->actions_len != odp_actions->size
3401 || memcmp(subfacet->actions, odp_actions->data,
3402 subfacet->actions_len));
3404 should_install = (ctx.may_set_up_flow
3405 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3406 if (actions_changed || should_install != subfacet->installed) {
3407 if (should_install) {
3408 struct dpif_flow_stats stats;
3410 subfacet_install(ofproto, subfacet,
3411 odp_actions->data, odp_actions->size, &stats);
3412 subfacet_update_stats(ofproto, subfacet, &stats);
3414 subfacet_uninstall(ofproto, subfacet);
3418 new_actions = xcalloc(list_size(&facet->subfacets),
3419 sizeof *new_actions);
3421 new_actions[i].odp_actions = xmemdup(odp_actions->data,
3423 new_actions[i].actions_len = odp_actions->size;
3426 ofpbuf_delete(odp_actions);
3430 facet_flush_stats(ofproto, facet);
3433 /* Update 'facet' now that we've taken care of all the old state. */
3434 facet->tags = ctx.tags;
3435 facet->nf_flow.output_iface = ctx.nf_output_iface;
3436 facet->may_install = ctx.may_set_up_flow;
3437 facet->has_learn = ctx.has_learn;
3438 facet->has_normal = ctx.has_normal;
3439 facet->mirrors = ctx.mirrors;
3442 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3443 if (new_actions[i].odp_actions) {
3444 free(subfacet->actions);
3445 subfacet->actions = new_actions[i].odp_actions;
3446 subfacet->actions_len = new_actions[i].actions_len;
3452 if (facet->rule != new_rule) {
3453 COVERAGE_INC(facet_changed_rule);
3454 list_remove(&facet->list_node);
3455 list_push_back(&new_rule->facets, &facet->list_node);
3456 facet->rule = new_rule;
3457 facet->used = new_rule->up.created;
3458 facet->prev_used = facet->used;
3464 /* Updates 'facet''s used time. Caller is responsible for calling
3465 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3467 facet_update_time(struct ofproto_dpif *ofproto, struct facet *facet,
3470 if (used > facet->used) {
3472 if (used > facet->rule->used) {
3473 facet->rule->used = used;
3475 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3480 facet_reset_counters(struct facet *facet)
3482 facet->packet_count = 0;
3483 facet->byte_count = 0;
3484 facet->prev_packet_count = 0;
3485 facet->prev_byte_count = 0;
3486 facet->accounted_bytes = 0;
3490 facet_push_stats(struct facet *facet)
3492 uint64_t new_packets, new_bytes;
3494 assert(facet->packet_count >= facet->prev_packet_count);
3495 assert(facet->byte_count >= facet->prev_byte_count);
3496 assert(facet->used >= facet->prev_used);
3498 new_packets = facet->packet_count - facet->prev_packet_count;
3499 new_bytes = facet->byte_count - facet->prev_byte_count;
3501 if (new_packets || new_bytes || facet->used > facet->prev_used) {
3502 facet->prev_packet_count = facet->packet_count;
3503 facet->prev_byte_count = facet->byte_count;
3504 facet->prev_used = facet->used;
3506 flow_push_stats(facet->rule, &facet->flow,
3507 new_packets, new_bytes, facet->used);
3509 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
3510 facet->mirrors, new_packets, new_bytes);
3514 struct ofproto_push {
3515 struct action_xlate_ctx ctx;
3522 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
3524 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3527 rule->packet_count += push->packets;
3528 rule->byte_count += push->bytes;
3529 rule->used = MAX(push->used, rule->used);
3533 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3534 * 'rule''s actions and mirrors. */
3536 flow_push_stats(const struct rule_dpif *rule,
3537 const struct flow *flow, uint64_t packets, uint64_t bytes,
3540 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3541 struct ofproto_push push;
3543 push.packets = packets;
3547 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, NULL);
3548 push.ctx.resubmit_hook = push_resubmit;
3549 ofpbuf_delete(xlate_actions(&push.ctx,
3550 rule->up.actions, rule->up.n_actions));
3555 static struct subfacet *
3556 subfacet_find__(struct ofproto_dpif *ofproto,
3557 const struct nlattr *key, size_t key_len, uint32_t key_hash,
3558 const struct flow *flow)
3560 struct subfacet *subfacet;
3562 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
3563 &ofproto->subfacets) {
3565 ? (subfacet->key_len == key_len
3566 && !memcmp(key, subfacet->key, key_len))
3567 : flow_equal(flow, &subfacet->facet->flow)) {
3575 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
3576 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
3577 * there is one, otherwise creates and returns a new subfacet.
3579 * If the returned subfacet is new, then subfacet->actions will be NULL, in
3580 * which case the caller must populate the actions with
3581 * subfacet_make_actions(). */
3582 static struct subfacet *
3583 subfacet_create(struct ofproto_dpif *ofproto, struct facet *facet,
3584 enum odp_key_fitness key_fitness,
3585 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
3587 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3588 struct subfacet *subfacet;
3590 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
3592 if (subfacet->facet == facet) {
3596 /* This shouldn't happen. */
3597 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
3598 subfacet_destroy(ofproto, subfacet);
3601 subfacet = xzalloc(sizeof *subfacet);
3602 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
3603 list_push_back(&facet->subfacets, &subfacet->list_node);
3604 subfacet->facet = facet;
3605 subfacet->used = time_msec();
3606 subfacet->key_fitness = key_fitness;
3607 if (key_fitness != ODP_FIT_PERFECT) {
3608 subfacet->key = xmemdup(key, key_len);
3609 subfacet->key_len = key_len;
3611 subfacet->installed = false;
3612 subfacet->initial_tci = initial_tci;
3617 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
3618 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
3619 static struct subfacet *
3620 subfacet_find(struct ofproto_dpif *ofproto,
3621 const struct nlattr *key, size_t key_len)
3623 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3624 enum odp_key_fitness fitness;
3627 fitness = odp_flow_key_to_flow(key, key_len, &flow);
3628 if (fitness == ODP_FIT_ERROR) {
3632 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
3635 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
3636 * its facet within 'ofproto', and frees it. */
3638 subfacet_destroy__(struct ofproto_dpif *ofproto, struct subfacet *subfacet)
3640 subfacet_uninstall(ofproto, subfacet);
3641 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
3642 list_remove(&subfacet->list_node);
3643 free(subfacet->key);
3644 free(subfacet->actions);
3648 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
3649 * last remaining subfacet in its facet destroys the facet too. */
3651 subfacet_destroy(struct ofproto_dpif *ofproto, struct subfacet *subfacet)
3653 struct facet *facet = subfacet->facet;
3655 subfacet_destroy__(ofproto, subfacet);
3656 if (list_is_empty(&facet->subfacets)) {
3657 facet_remove(ofproto, facet);
3661 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
3662 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
3663 * for use as temporary storage. */
3665 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
3668 if (!subfacet->key) {
3669 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
3670 odp_flow_key_from_flow(key, &subfacet->facet->flow);
3672 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
3676 /* Composes the datapath actions for 'subfacet' based on its rule's actions. */
3678 subfacet_make_actions(struct ofproto_dpif *p, struct subfacet *subfacet,
3679 const struct ofpbuf *packet)
3681 struct facet *facet = subfacet->facet;
3682 const struct rule_dpif *rule = facet->rule;
3683 struct ofpbuf *odp_actions;
3684 struct action_xlate_ctx ctx;
3686 action_xlate_ctx_init(&ctx, p, &facet->flow, subfacet->initial_tci,
3688 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3689 facet->tags = ctx.tags;
3690 facet->may_install = ctx.may_set_up_flow;
3691 facet->has_learn = ctx.has_learn;
3692 facet->has_normal = ctx.has_normal;
3693 facet->nf_flow.output_iface = ctx.nf_output_iface;
3694 facet->mirrors = ctx.mirrors;
3696 if (subfacet->actions_len != odp_actions->size
3697 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
3698 free(subfacet->actions);
3699 subfacet->actions_len = odp_actions->size;
3700 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
3703 ofpbuf_delete(odp_actions);
3706 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
3707 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
3708 * in the datapath will be zeroed and 'stats' will be updated with traffic new
3709 * since 'subfacet' was last updated.
3711 * Returns 0 if successful, otherwise a positive errno value. */
3713 subfacet_install(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3714 const struct nlattr *actions, size_t actions_len,
3715 struct dpif_flow_stats *stats)
3717 struct odputil_keybuf keybuf;
3718 enum dpif_flow_put_flags flags;
3722 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3724 flags |= DPIF_FP_ZERO_STATS;
3727 subfacet_get_key(subfacet, &keybuf, &key);
3728 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
3729 actions, actions_len, stats);
3732 subfacet_reset_dp_stats(subfacet, stats);
3738 /* If 'subfacet' is installed in the datapath, uninstalls it. */
3740 subfacet_uninstall(struct ofproto_dpif *p, struct subfacet *subfacet)
3742 if (subfacet->installed) {
3743 struct odputil_keybuf keybuf;
3744 struct dpif_flow_stats stats;
3748 subfacet_get_key(subfacet, &keybuf, &key);
3749 error = dpif_flow_del(p->dpif, key.data, key.size, &stats);
3750 subfacet_reset_dp_stats(subfacet, &stats);
3752 subfacet_update_stats(p, subfacet, &stats);
3754 subfacet->installed = false;
3756 assert(subfacet->dp_packet_count == 0);
3757 assert(subfacet->dp_byte_count == 0);
3761 /* Resets 'subfacet''s datapath statistics counters. This should be called
3762 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
3763 * non-null, it should contain the statistics returned by dpif when 'subfacet'
3764 * was reset in the datapath. 'stats' will be modified to include only
3765 * statistics new since 'subfacet' was last updated. */
3767 subfacet_reset_dp_stats(struct subfacet *subfacet,
3768 struct dpif_flow_stats *stats)
3771 && subfacet->dp_packet_count <= stats->n_packets
3772 && subfacet->dp_byte_count <= stats->n_bytes) {
3773 stats->n_packets -= subfacet->dp_packet_count;
3774 stats->n_bytes -= subfacet->dp_byte_count;
3777 subfacet->dp_packet_count = 0;
3778 subfacet->dp_byte_count = 0;
3781 /* Updates 'subfacet''s used time. The caller is responsible for calling
3782 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
3784 subfacet_update_time(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3787 if (used > subfacet->used) {
3788 subfacet->used = used;
3789 facet_update_time(ofproto, subfacet->facet, used);
3793 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
3795 * Because of the meaning of a subfacet's counters, it only makes sense to do
3796 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
3797 * represents a packet that was sent by hand or if it represents statistics
3798 * that have been cleared out of the datapath. */
3800 subfacet_update_stats(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3801 const struct dpif_flow_stats *stats)
3803 if (stats->n_packets || stats->used > subfacet->used) {
3804 struct facet *facet = subfacet->facet;
3806 subfacet_update_time(ofproto, subfacet, stats->used);
3807 facet->packet_count += stats->n_packets;
3808 facet->byte_count += stats->n_bytes;
3809 facet_push_stats(facet);
3810 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3816 static struct rule_dpif *
3817 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
3820 struct cls_rule *cls_rule;
3821 struct classifier *cls;
3823 if (table_id >= N_TABLES) {
3827 cls = &ofproto->up.tables[table_id];
3828 if (flow->nw_frag & FLOW_NW_FRAG_ANY
3829 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3830 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
3831 * are unavailable. */
3832 struct flow ofpc_normal_flow = *flow;
3833 ofpc_normal_flow.tp_src = htons(0);
3834 ofpc_normal_flow.tp_dst = htons(0);
3835 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
3837 cls_rule = classifier_lookup(cls, flow);
3839 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
3843 complete_operation(struct rule_dpif *rule)
3845 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3847 rule_invalidate(rule);
3849 struct dpif_completion *c = xmalloc(sizeof *c);
3850 c->op = rule->up.pending;
3851 list_push_back(&ofproto->completions, &c->list_node);
3853 ofoperation_complete(rule->up.pending, 0);
3857 static struct rule *
3860 struct rule_dpif *rule = xmalloc(sizeof *rule);
3865 rule_dealloc(struct rule *rule_)
3867 struct rule_dpif *rule = rule_dpif_cast(rule_);
3872 rule_construct(struct rule *rule_)
3874 struct rule_dpif *rule = rule_dpif_cast(rule_);
3875 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3876 struct rule_dpif *victim;
3880 error = validate_actions(rule->up.actions, rule->up.n_actions,
3881 &rule->up.cr.flow, ofproto->max_ports);
3886 rule->used = rule->up.created;
3887 rule->packet_count = 0;
3888 rule->byte_count = 0;
3890 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
3891 if (victim && !list_is_empty(&victim->facets)) {
3892 struct facet *facet;
3894 rule->facets = victim->facets;
3895 list_moved(&rule->facets);
3896 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3897 /* XXX: We're only clearing our local counters here. It's possible
3898 * that quite a few packets are unaccounted for in the datapath
3899 * statistics. These will be accounted to the new rule instead of
3900 * cleared as required. This could be fixed by clearing out the
3901 * datapath statistics for this facet, but currently it doesn't
3903 facet_reset_counters(facet);
3907 /* Must avoid list_moved() in this case. */
3908 list_init(&rule->facets);
3911 table_id = rule->up.table_id;
3912 rule->tag = (victim ? victim->tag
3914 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
3915 ofproto->tables[table_id].basis));
3917 complete_operation(rule);
3922 rule_destruct(struct rule *rule_)
3924 struct rule_dpif *rule = rule_dpif_cast(rule_);
3925 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3926 struct facet *facet, *next_facet;
3928 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3929 facet_revalidate(ofproto, facet);
3932 complete_operation(rule);
3936 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
3938 struct rule_dpif *rule = rule_dpif_cast(rule_);
3939 struct facet *facet;
3941 /* Start from historical data for 'rule' itself that are no longer tracked
3942 * in facets. This counts, for example, facets that have expired. */
3943 *packets = rule->packet_count;
3944 *bytes = rule->byte_count;
3946 /* Add any statistics that are tracked by facets. This includes
3947 * statistical data recently updated by ofproto_update_stats() as well as
3948 * stats for packets that were executed "by hand" via dpif_execute(). */
3949 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3950 *packets += facet->packet_count;
3951 *bytes += facet->byte_count;
3956 rule_execute(struct rule *rule_, const struct flow *flow,
3957 struct ofpbuf *packet)
3959 struct rule_dpif *rule = rule_dpif_cast(rule_);
3960 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3961 struct action_xlate_ctx ctx;
3962 struct ofpbuf *odp_actions;
3965 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, packet);
3966 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3967 size = packet->size;
3968 if (execute_odp_actions(ofproto, flow, odp_actions->data,
3969 odp_actions->size, packet)) {
3970 rule->used = time_msec();
3971 rule->packet_count++;
3972 rule->byte_count += size;
3973 flow_push_stats(rule, flow, 1, size, rule->used);
3975 ofpbuf_delete(odp_actions);
3981 rule_modify_actions(struct rule *rule_)
3983 struct rule_dpif *rule = rule_dpif_cast(rule_);
3984 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3987 error = validate_actions(rule->up.actions, rule->up.n_actions,
3988 &rule->up.cr.flow, ofproto->max_ports);
3990 ofoperation_complete(rule->up.pending, error);
3994 complete_operation(rule);
3997 /* Sends 'packet' out 'ofport'.
3998 * May modify 'packet'.
3999 * Returns 0 if successful, otherwise a positive errno value. */
4001 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4003 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4004 struct ofpbuf key, odp_actions;
4005 struct odputil_keybuf keybuf;
4010 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
4011 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4013 if (odp_port != ofport->odp_port) {
4014 eth_pop_vlan(packet);
4015 flow.vlan_tci = htons(0);
4018 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4019 odp_flow_key_from_flow(&key, &flow);
4021 ofpbuf_init(&odp_actions, 32);
4022 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4024 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4025 error = dpif_execute(ofproto->dpif,
4027 odp_actions.data, odp_actions.size,
4029 ofpbuf_uninit(&odp_actions);
4032 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4033 ofproto->up.name, odp_port, strerror(error));
4035 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
4039 /* OpenFlow to datapath action translation. */
4041 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
4042 struct action_xlate_ctx *ctx);
4043 static void xlate_normal(struct action_xlate_ctx *);
4046 put_userspace_action(const struct ofproto_dpif *ofproto,
4047 struct ofpbuf *odp_actions,
4048 const struct flow *flow,
4049 const struct user_action_cookie *cookie)
4053 pid = dpif_port_get_pid(ofproto->dpif,
4054 ofp_port_to_odp_port(flow->in_port));
4056 return odp_put_userspace_action(pid, cookie, odp_actions);
4059 /* Compose SAMPLE action for sFlow. */
4061 compose_sflow_action(const struct ofproto_dpif *ofproto,
4062 struct ofpbuf *odp_actions,
4063 const struct flow *flow,
4066 uint32_t port_ifindex;
4067 uint32_t probability;
4068 struct user_action_cookie cookie;
4069 size_t sample_offset, actions_offset;
4070 int cookie_offset, n_output;
4072 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4076 if (odp_port == OVSP_NONE) {
4080 port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4084 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4086 /* Number of packets out of UINT_MAX to sample. */
4087 probability = dpif_sflow_get_probability(ofproto->sflow);
4088 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4090 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4092 cookie.type = USER_ACTION_COOKIE_SFLOW;
4093 cookie.data = port_ifindex;
4094 cookie.n_output = n_output;
4095 cookie.vlan_tci = 0;
4096 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4098 nl_msg_end_nested(odp_actions, actions_offset);
4099 nl_msg_end_nested(odp_actions, sample_offset);
4100 return cookie_offset;
4103 /* SAMPLE action must be first action in any given list of actions.
4104 * At this point we do not have all information required to build it. So try to
4105 * build sample action as complete as possible. */
4107 add_sflow_action(struct action_xlate_ctx *ctx)
4109 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4111 &ctx->flow, OVSP_NONE);
4112 ctx->sflow_odp_port = 0;
4113 ctx->sflow_n_outputs = 0;
4116 /* Fix SAMPLE action according to data collected while composing ODP actions.
4117 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4118 * USERSPACE action's user-cookie which is required for sflow. */
4120 fix_sflow_action(struct action_xlate_ctx *ctx)
4122 const struct flow *base = &ctx->base_flow;
4123 struct user_action_cookie *cookie;
4125 if (!ctx->user_cookie_offset) {
4129 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4131 assert(cookie != NULL);
4132 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4134 if (ctx->sflow_n_outputs) {
4135 cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow,
4136 ctx->sflow_odp_port);
4138 if (ctx->sflow_n_outputs >= 255) {
4139 cookie->n_output = 255;
4141 cookie->n_output = ctx->sflow_n_outputs;
4143 cookie->vlan_tci = base->vlan_tci;
4147 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4150 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4151 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4152 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4153 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4157 struct priority_to_dscp *pdscp;
4159 if (ofport->up.opp.config & htonl(OFPPC_NO_FWD)
4160 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4164 pdscp = get_priority(ofport, ctx->flow.priority);
4166 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4167 ctx->flow.nw_tos |= pdscp->dscp;
4170 /* We may not have an ofport record for this port, but it doesn't hurt
4171 * to allow forwarding to it anyhow. Maybe such a port will appear
4172 * later and we're pre-populating the flow table. */
4175 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4176 ctx->flow.vlan_tci);
4177 if (out_port != odp_port) {
4178 ctx->flow.vlan_tci = htons(0);
4180 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4181 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4183 ctx->sflow_odp_port = odp_port;
4184 ctx->sflow_n_outputs++;
4185 ctx->nf_output_iface = ofp_port;
4186 ctx->flow.vlan_tci = flow_vlan_tci;
4187 ctx->flow.nw_tos = flow_nw_tos;
4191 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4193 compose_output_action__(ctx, ofp_port, true);
4197 xlate_table_action(struct action_xlate_ctx *ctx,
4198 uint16_t in_port, uint8_t table_id)
4200 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4201 struct ofproto_dpif *ofproto = ctx->ofproto;
4202 struct rule_dpif *rule;
4203 uint16_t old_in_port;
4204 uint8_t old_table_id;
4206 old_table_id = ctx->table_id;
4207 ctx->table_id = table_id;
4209 /* Look up a flow with 'in_port' as the input port. */
4210 old_in_port = ctx->flow.in_port;
4211 ctx->flow.in_port = in_port;
4212 rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
4215 if (table_id > 0 && table_id < N_TABLES) {
4216 struct table_dpif *table = &ofproto->tables[table_id];
4217 if (table->other_table) {
4220 : rule_calculate_tag(&ctx->flow,
4221 &table->other_table->wc,
4226 /* Restore the original input port. Otherwise OFPP_NORMAL and
4227 * OFPP_IN_PORT will have surprising behavior. */
4228 ctx->flow.in_port = old_in_port;
4230 if (ctx->resubmit_hook) {
4231 ctx->resubmit_hook(ctx, rule);
4236 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
4240 ctx->table_id = old_table_id;
4242 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4244 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
4245 MAX_RESUBMIT_RECURSION);
4250 xlate_resubmit_table(struct action_xlate_ctx *ctx,
4251 const struct nx_action_resubmit *nar)
4256 in_port = (nar->in_port == htons(OFPP_IN_PORT)
4258 : ntohs(nar->in_port));
4259 table_id = nar->table == 255 ? ctx->table_id : nar->table;
4261 xlate_table_action(ctx, in_port, table_id);
4265 flood_packets(struct action_xlate_ctx *ctx, bool all)
4267 struct ofport_dpif *ofport;
4269 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
4270 uint16_t ofp_port = ofport->up.ofp_port;
4272 if (ofp_port == ctx->flow.in_port) {
4277 compose_output_action__(ctx, ofp_port, false);
4278 } else if (!(ofport->up.opp.config & htonl(OFPPC_NO_FLOOD))) {
4279 compose_output_action(ctx, ofp_port);
4283 ctx->nf_output_iface = NF_OUT_FLOOD;
4287 compose_controller_action(struct action_xlate_ctx *ctx, int len)
4289 struct user_action_cookie cookie;
4291 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4292 cookie.type = USER_ACTION_COOKIE_CONTROLLER;
4294 cookie.n_output = 0;
4295 cookie.vlan_tci = 0;
4296 put_userspace_action(ctx->ofproto, ctx->odp_actions, &ctx->flow, &cookie);
4300 xlate_output_action__(struct action_xlate_ctx *ctx,
4301 uint16_t port, uint16_t max_len)
4303 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
4305 ctx->nf_output_iface = NF_OUT_DROP;
4309 compose_output_action(ctx, ctx->flow.in_port);
4312 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
4318 flood_packets(ctx, false);
4321 flood_packets(ctx, true);
4323 case OFPP_CONTROLLER:
4324 compose_controller_action(ctx, max_len);
4327 compose_output_action(ctx, OFPP_LOCAL);
4332 if (port != ctx->flow.in_port) {
4333 compose_output_action(ctx, port);
4338 if (prev_nf_output_iface == NF_OUT_FLOOD) {
4339 ctx->nf_output_iface = NF_OUT_FLOOD;
4340 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
4341 ctx->nf_output_iface = prev_nf_output_iface;
4342 } else if (prev_nf_output_iface != NF_OUT_DROP &&
4343 ctx->nf_output_iface != NF_OUT_FLOOD) {
4344 ctx->nf_output_iface = NF_OUT_MULTI;
4349 xlate_output_reg_action(struct action_xlate_ctx *ctx,
4350 const struct nx_action_output_reg *naor)
4354 ofp_port = nxm_read_field_bits(naor->src, naor->ofs_nbits, &ctx->flow);
4356 if (ofp_port <= UINT16_MAX) {
4357 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
4362 xlate_output_action(struct action_xlate_ctx *ctx,
4363 const struct ofp_action_output *oao)
4365 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
4369 xlate_enqueue_action(struct action_xlate_ctx *ctx,
4370 const struct ofp_action_enqueue *oae)
4373 uint32_t flow_priority, priority;
4376 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
4379 /* Fall back to ordinary output action. */
4380 xlate_output_action__(ctx, ntohs(oae->port), 0);
4384 /* Figure out datapath output port. */
4385 ofp_port = ntohs(oae->port);
4386 if (ofp_port == OFPP_IN_PORT) {
4387 ofp_port = ctx->flow.in_port;
4388 } else if (ofp_port == ctx->flow.in_port) {
4392 /* Add datapath actions. */
4393 flow_priority = ctx->flow.priority;
4394 ctx->flow.priority = priority;
4395 compose_output_action(ctx, ofp_port);
4396 ctx->flow.priority = flow_priority;
4398 /* Update NetFlow output port. */
4399 if (ctx->nf_output_iface == NF_OUT_DROP) {
4400 ctx->nf_output_iface = ofp_port;
4401 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
4402 ctx->nf_output_iface = NF_OUT_MULTI;
4407 xlate_set_queue_action(struct action_xlate_ctx *ctx,
4408 const struct nx_action_set_queue *nasq)
4413 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
4416 /* Couldn't translate queue to a priority, so ignore. A warning
4417 * has already been logged. */
4421 ctx->flow.priority = priority;
4424 struct xlate_reg_state {
4430 xlate_autopath(struct action_xlate_ctx *ctx,
4431 const struct nx_action_autopath *naa)
4433 uint16_t ofp_port = ntohl(naa->id);
4434 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
4436 if (!port || !port->bundle) {
4437 ofp_port = OFPP_NONE;
4438 } else if (port->bundle->bond) {
4439 /* Autopath does not support VLAN hashing. */
4440 struct ofport_dpif *slave = bond_choose_output_slave(
4441 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
4443 ofp_port = slave->up.ofp_port;
4446 autopath_execute(naa, &ctx->flow, ofp_port);
4450 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
4452 struct ofproto_dpif *ofproto = ofproto_;
4453 struct ofport_dpif *port;
4463 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
4466 port = get_ofp_port(ofproto, ofp_port);
4467 return port ? port->may_enable : false;
4472 xlate_learn_action(struct action_xlate_ctx *ctx,
4473 const struct nx_action_learn *learn)
4475 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
4476 struct ofputil_flow_mod fm;
4479 learn_execute(learn, &ctx->flow, &fm);
4481 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
4482 if (error && !VLOG_DROP_WARN(&rl)) {
4483 char *msg = ofputil_error_to_string(error);
4484 VLOG_WARN("learning action failed to modify flow table (%s)", msg);
4492 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
4494 if (port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
4495 ? htonl(OFPPC_NO_RECV_STP)
4496 : htonl(OFPPC_NO_RECV))) {
4500 /* Only drop packets here if both forwarding and learning are
4501 * disabled. If just learning is enabled, we need to have
4502 * OFPP_NORMAL and the learning action have a look at the packet
4503 * before we can drop it. */
4504 if (!stp_forward_in_state(port->stp_state)
4505 && !stp_learn_in_state(port->stp_state)) {
4513 do_xlate_actions(const union ofp_action *in, size_t n_in,
4514 struct action_xlate_ctx *ctx)
4516 const struct ofport_dpif *port;
4517 const union ofp_action *ia;
4520 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
4521 if (port && !may_receive(port, ctx)) {
4522 /* Drop this flow. */
4526 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
4527 const struct ofp_action_dl_addr *oada;
4528 const struct nx_action_resubmit *nar;
4529 const struct nx_action_set_tunnel *nast;
4530 const struct nx_action_set_queue *nasq;
4531 const struct nx_action_multipath *nam;
4532 const struct nx_action_autopath *naa;
4533 const struct nx_action_bundle *nab;
4534 const struct nx_action_output_reg *naor;
4535 enum ofputil_action_code code;
4542 code = ofputil_decode_action_unsafe(ia);
4544 case OFPUTIL_OFPAT_OUTPUT:
4545 xlate_output_action(ctx, &ia->output);
4548 case OFPUTIL_OFPAT_SET_VLAN_VID:
4549 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
4550 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
4553 case OFPUTIL_OFPAT_SET_VLAN_PCP:
4554 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
4555 ctx->flow.vlan_tci |= htons(
4556 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
4559 case OFPUTIL_OFPAT_STRIP_VLAN:
4560 ctx->flow.vlan_tci = htons(0);
4563 case OFPUTIL_OFPAT_SET_DL_SRC:
4564 oada = ((struct ofp_action_dl_addr *) ia);
4565 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
4568 case OFPUTIL_OFPAT_SET_DL_DST:
4569 oada = ((struct ofp_action_dl_addr *) ia);
4570 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
4573 case OFPUTIL_OFPAT_SET_NW_SRC:
4574 ctx->flow.nw_src = ia->nw_addr.nw_addr;
4577 case OFPUTIL_OFPAT_SET_NW_DST:
4578 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
4581 case OFPUTIL_OFPAT_SET_NW_TOS:
4582 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4583 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
4586 case OFPUTIL_OFPAT_SET_TP_SRC:
4587 ctx->flow.tp_src = ia->tp_port.tp_port;
4590 case OFPUTIL_OFPAT_SET_TP_DST:
4591 ctx->flow.tp_dst = ia->tp_port.tp_port;
4594 case OFPUTIL_OFPAT_ENQUEUE:
4595 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
4598 case OFPUTIL_NXAST_RESUBMIT:
4599 nar = (const struct nx_action_resubmit *) ia;
4600 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
4603 case OFPUTIL_NXAST_RESUBMIT_TABLE:
4604 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
4607 case OFPUTIL_NXAST_SET_TUNNEL:
4608 nast = (const struct nx_action_set_tunnel *) ia;
4609 tun_id = htonll(ntohl(nast->tun_id));
4610 ctx->flow.tun_id = tun_id;
4613 case OFPUTIL_NXAST_SET_QUEUE:
4614 nasq = (const struct nx_action_set_queue *) ia;
4615 xlate_set_queue_action(ctx, nasq);
4618 case OFPUTIL_NXAST_POP_QUEUE:
4619 ctx->flow.priority = ctx->original_priority;
4622 case OFPUTIL_NXAST_REG_MOVE:
4623 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
4627 case OFPUTIL_NXAST_REG_LOAD:
4628 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
4632 case OFPUTIL_NXAST_NOTE:
4633 /* Nothing to do. */
4636 case OFPUTIL_NXAST_SET_TUNNEL64:
4637 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
4638 ctx->flow.tun_id = tun_id;
4641 case OFPUTIL_NXAST_MULTIPATH:
4642 nam = (const struct nx_action_multipath *) ia;
4643 multipath_execute(nam, &ctx->flow);
4646 case OFPUTIL_NXAST_AUTOPATH:
4647 naa = (const struct nx_action_autopath *) ia;
4648 xlate_autopath(ctx, naa);
4651 case OFPUTIL_NXAST_BUNDLE:
4652 ctx->ofproto->has_bundle_action = true;
4653 nab = (const struct nx_action_bundle *) ia;
4654 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
4659 case OFPUTIL_NXAST_BUNDLE_LOAD:
4660 ctx->ofproto->has_bundle_action = true;
4661 nab = (const struct nx_action_bundle *) ia;
4662 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
4666 case OFPUTIL_NXAST_OUTPUT_REG:
4667 naor = (const struct nx_action_output_reg *) ia;
4668 xlate_output_reg_action(ctx, naor);
4671 case OFPUTIL_NXAST_LEARN:
4672 ctx->has_learn = true;
4673 if (ctx->may_learn) {
4674 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
4678 case OFPUTIL_NXAST_EXIT:
4684 /* We've let OFPP_NORMAL and the learning action look at the packet,
4685 * so drop it now if forwarding is disabled. */
4686 if (port && !stp_forward_in_state(port->stp_state)) {
4687 ofpbuf_clear(ctx->odp_actions);
4688 add_sflow_action(ctx);
4693 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
4694 struct ofproto_dpif *ofproto, const struct flow *flow,
4695 ovs_be16 initial_tci, const struct ofpbuf *packet)
4697 ctx->ofproto = ofproto;
4699 ctx->base_flow = ctx->flow;
4700 ctx->base_flow.tun_id = 0;
4701 ctx->base_flow.vlan_tci = initial_tci;
4702 ctx->packet = packet;
4703 ctx->may_learn = packet != NULL;
4704 ctx->resubmit_hook = NULL;
4707 static struct ofpbuf *
4708 xlate_actions(struct action_xlate_ctx *ctx,
4709 const union ofp_action *in, size_t n_in)
4711 struct flow orig_flow = ctx->flow;
4713 COVERAGE_INC(ofproto_dpif_xlate);
4715 ctx->odp_actions = ofpbuf_new(512);
4716 ofpbuf_reserve(ctx->odp_actions, NL_A_U32_SIZE);
4718 ctx->may_set_up_flow = true;
4719 ctx->has_learn = false;
4720 ctx->has_normal = false;
4721 ctx->nf_output_iface = NF_OUT_DROP;
4724 ctx->original_priority = ctx->flow.priority;
4728 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
4729 switch (ctx->ofproto->up.frag_handling) {
4730 case OFPC_FRAG_NORMAL:
4731 /* We must pretend that transport ports are unavailable. */
4732 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
4733 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
4736 case OFPC_FRAG_DROP:
4737 return ctx->odp_actions;
4739 case OFPC_FRAG_REASM:
4742 case OFPC_FRAG_NX_MATCH:
4743 /* Nothing to do. */
4748 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
4749 ctx->may_set_up_flow = false;
4750 return ctx->odp_actions;
4752 add_sflow_action(ctx);
4753 do_xlate_actions(in, n_in, ctx);
4755 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
4756 ctx->odp_actions->data,
4757 ctx->odp_actions->size)) {
4758 ctx->may_set_up_flow = false;
4760 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
4762 compose_output_action(ctx, OFPP_LOCAL);
4765 add_mirror_actions(ctx, &orig_flow);
4766 fix_sflow_action(ctx);
4769 return ctx->odp_actions;
4772 /* OFPP_NORMAL implementation. */
4774 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
4776 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
4777 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
4778 * the bundle on which the packet was received, returns the VLAN to which the
4781 * Both 'vid' and the return value are in the range 0...4095. */
4783 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
4785 switch (in_bundle->vlan_mode) {
4786 case PORT_VLAN_ACCESS:
4787 return in_bundle->vlan;
4790 case PORT_VLAN_TRUNK:
4793 case PORT_VLAN_NATIVE_UNTAGGED:
4794 case PORT_VLAN_NATIVE_TAGGED:
4795 return vid ? vid : in_bundle->vlan;
4802 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
4803 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
4806 * 'vid' should be the VID obtained from the 802.1Q header that was received as
4807 * part of a packet (specify 0 if there was no 802.1Q header), in the range
4810 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
4812 switch (in_bundle->vlan_mode) {
4813 case PORT_VLAN_ACCESS:
4816 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4817 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
4818 "packet received on port %s configured as VLAN "
4819 "%"PRIu16" access port",
4820 in_bundle->ofproto->up.name, vid,
4821 in_bundle->name, in_bundle->vlan);
4827 case PORT_VLAN_NATIVE_UNTAGGED:
4828 case PORT_VLAN_NATIVE_TAGGED:
4830 /* Port must always carry its native VLAN. */
4834 case PORT_VLAN_TRUNK:
4835 if (!ofbundle_includes_vlan(in_bundle, vid)) {
4837 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4838 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
4839 "received on port %s not configured for trunking "
4841 in_bundle->ofproto->up.name, vid,
4842 in_bundle->name, vid);
4854 /* Given 'vlan', the VLAN that a packet belongs to, and
4855 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
4856 * that should be included in the 802.1Q header. (If the return value is 0,
4857 * then the 802.1Q header should only be included in the packet if there is a
4860 * Both 'vlan' and the return value are in the range 0...4095. */
4862 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
4864 switch (out_bundle->vlan_mode) {
4865 case PORT_VLAN_ACCESS:
4868 case PORT_VLAN_TRUNK:
4869 case PORT_VLAN_NATIVE_TAGGED:
4872 case PORT_VLAN_NATIVE_UNTAGGED:
4873 return vlan == out_bundle->vlan ? 0 : vlan;
4881 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
4884 struct ofport_dpif *port;
4886 ovs_be16 tci, old_tci;
4888 vid = output_vlan_to_vid(out_bundle, vlan);
4889 if (!out_bundle->bond) {
4890 port = ofbundle_get_a_port(out_bundle);
4892 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
4895 /* No slaves enabled, so drop packet. */
4900 old_tci = ctx->flow.vlan_tci;
4902 if (tci || out_bundle->use_priority_tags) {
4903 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
4905 tci |= htons(VLAN_CFI);
4908 ctx->flow.vlan_tci = tci;
4910 compose_output_action(ctx, port->up.ofp_port);
4911 ctx->flow.vlan_tci = old_tci;
4915 mirror_mask_ffs(mirror_mask_t mask)
4917 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
4922 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
4924 return (bundle->vlan_mode != PORT_VLAN_ACCESS
4925 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
4929 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
4931 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
4934 /* Returns an arbitrary interface within 'bundle'. */
4935 static struct ofport_dpif *
4936 ofbundle_get_a_port(const struct ofbundle *bundle)
4938 return CONTAINER_OF(list_front(&bundle->ports),
4939 struct ofport_dpif, bundle_node);
4943 vlan_is_mirrored(const struct ofmirror *m, int vlan)
4945 return !m->vlans || bitmap_is_set(m->vlans, vlan);
4948 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
4949 * to a VLAN. In general most packets may be mirrored but we want to drop
4950 * protocols that may confuse switches. */
4952 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
4954 /* If you change this function's behavior, please update corresponding
4955 * documentation in vswitch.xml at the same time. */
4956 if (dst[0] != 0x01) {
4957 /* All the currently banned MACs happen to start with 01 currently, so
4958 * this is a quick way to eliminate most of the good ones. */
4960 if (eth_addr_is_reserved(dst)) {
4961 /* Drop STP, IEEE pause frames, and other reserved protocols
4962 * (01-80-c2-00-00-0x). */
4966 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
4968 if ((dst[3] & 0xfe) == 0xcc &&
4969 (dst[4] & 0xfe) == 0xcc &&
4970 (dst[5] & 0xfe) == 0xcc) {
4971 /* Drop the following protocols plus others following the same
4974 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
4975 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
4976 STP Uplink Fast (01-00-0c-cd-cd-cd) */
4980 if (!(dst[3] | dst[4] | dst[5])) {
4981 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
4990 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
4992 struct ofproto_dpif *ofproto = ctx->ofproto;
4993 mirror_mask_t mirrors;
4994 struct ofport_dpif *in_port;
4995 struct ofbundle *in_bundle;
4998 const struct nlattr *a;
5001 /* Obtain in_port from orig_flow.in_port.
5003 * lookup_input_bundle() also ensures that in_port belongs to a bundle. */
5004 in_port = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5005 ctx->packet != NULL);
5009 in_bundle = in_port->bundle;
5010 mirrors = in_bundle->src_mirrors;
5012 /* Drop frames on bundles reserved for mirroring. */
5013 if (in_bundle->mirror_out) {
5014 if (ctx->packet != NULL) {
5015 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5016 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5017 "%s, which is reserved exclusively for mirroring",
5018 ctx->ofproto->up.name, in_bundle->name);
5024 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5025 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5028 vlan = input_vid_to_vlan(in_bundle, vid);
5030 /* Look at the output ports to check for destination selections. */
5032 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5033 ctx->odp_actions->size) {
5034 enum ovs_action_attr type = nl_attr_type(a);
5035 struct ofport_dpif *ofport;
5037 if (type != OVS_ACTION_ATTR_OUTPUT) {
5041 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5042 if (ofport && ofport->bundle) {
5043 mirrors |= ofport->bundle->dst_mirrors;
5051 /* Restore the original packet before adding the mirror actions. */
5052 ctx->flow = *orig_flow;
5057 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5059 if (!vlan_is_mirrored(m, vlan)) {
5060 mirrors &= mirrors - 1;
5064 mirrors &= ~m->dup_mirrors;
5065 ctx->mirrors |= m->dup_mirrors;
5067 output_normal(ctx, m->out, vlan);
5068 } else if (eth_dst_may_rspan(orig_flow->dl_dst)
5069 && vlan != m->out_vlan) {
5070 struct ofbundle *bundle;
5072 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5073 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5074 && !bundle->mirror_out) {
5075 output_normal(ctx, bundle, m->out_vlan);
5083 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5084 uint64_t packets, uint64_t bytes)
5090 for (; mirrors; mirrors &= mirrors - 1) {
5093 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5096 /* In normal circumstances 'm' will not be NULL. However,
5097 * if mirrors are reconfigured, we can temporarily get out
5098 * of sync in facet_revalidate(). We could "correct" the
5099 * mirror list before reaching here, but doing that would
5100 * not properly account the traffic stats we've currently
5101 * accumulated for previous mirror configuration. */
5105 m->packet_count += packets;
5106 m->byte_count += bytes;
5110 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
5111 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
5112 * indicate this; newer upstream kernels use gratuitous ARP requests. */
5114 is_gratuitous_arp(const struct flow *flow)
5116 return (flow->dl_type == htons(ETH_TYPE_ARP)
5117 && eth_addr_is_broadcast(flow->dl_dst)
5118 && (flow->nw_proto == ARP_OP_REPLY
5119 || (flow->nw_proto == ARP_OP_REQUEST
5120 && flow->nw_src == flow->nw_dst)));
5124 update_learning_table(struct ofproto_dpif *ofproto,
5125 const struct flow *flow, int vlan,
5126 struct ofbundle *in_bundle)
5128 struct mac_entry *mac;
5130 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
5134 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
5135 if (is_gratuitous_arp(flow)) {
5136 /* We don't want to learn from gratuitous ARP packets that are
5137 * reflected back over bond slaves so we lock the learning table. */
5138 if (!in_bundle->bond) {
5139 mac_entry_set_grat_arp_lock(mac);
5140 } else if (mac_entry_is_grat_arp_locked(mac)) {
5145 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
5146 /* The log messages here could actually be useful in debugging,
5147 * so keep the rate limit relatively high. */
5148 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5149 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
5150 "on port %s in VLAN %d",
5151 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
5152 in_bundle->name, vlan);
5154 mac->port.p = in_bundle;
5155 tag_set_add(&ofproto->revalidate_set,
5156 mac_learning_changed(ofproto->ml, mac));
5160 static struct ofport_dpif *
5161 lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn)
5163 struct ofport_dpif *ofport;
5165 /* Find the port and bundle for the received packet. */
5166 ofport = get_ofp_port(ofproto, in_port);
5167 if (ofport && ofport->bundle) {
5171 /* Odd. A few possible reasons here:
5173 * - We deleted a port but there are still a few packets queued up
5176 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
5177 * we don't know about.
5179 * - The ofproto client didn't configure the port as part of a bundle.
5182 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5184 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
5185 "port %"PRIu16, ofproto->up.name, in_port);
5190 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
5191 * dropped. Returns true if they may be forwarded, false if they should be
5194 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
5195 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
5197 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
5198 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
5199 * checked by input_vid_is_valid().
5201 * May also add tags to '*tags', although the current implementation only does
5202 * so in one special case.
5205 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
5206 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
5208 struct ofbundle *in_bundle = in_port->bundle;
5210 /* Drop frames for reserved multicast addresses
5211 * only if forward_bpdu option is absent. */
5212 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
5216 if (in_bundle->bond) {
5217 struct mac_entry *mac;
5219 switch (bond_check_admissibility(in_bundle->bond, in_port,
5220 flow->dl_dst, tags)) {
5227 case BV_DROP_IF_MOVED:
5228 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
5229 if (mac && mac->port.p != in_bundle &&
5230 (!is_gratuitous_arp(flow)
5231 || mac_entry_is_grat_arp_locked(mac))) {
5242 xlate_normal(struct action_xlate_ctx *ctx)
5244 struct ofport_dpif *in_port;
5245 struct ofbundle *in_bundle;
5246 struct mac_entry *mac;
5250 ctx->has_normal = true;
5252 /* Obtain in_port from ctx->flow.in_port.
5254 * lookup_input_bundle() also ensures that in_port belongs to a bundle. */
5255 in_port = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
5256 ctx->packet != NULL);
5260 in_bundle = in_port->bundle;
5262 /* Drop malformed frames. */
5263 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
5264 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
5265 if (ctx->packet != NULL) {
5266 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5267 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
5268 "VLAN tag received on port %s",
5269 ctx->ofproto->up.name, in_bundle->name);
5274 /* Drop frames on bundles reserved for mirroring. */
5275 if (in_bundle->mirror_out) {
5276 if (ctx->packet != NULL) {
5277 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5278 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5279 "%s, which is reserved exclusively for mirroring",
5280 ctx->ofproto->up.name, in_bundle->name);
5286 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
5287 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5290 vlan = input_vid_to_vlan(in_bundle, vid);
5292 /* Check other admissibility requirements. */
5293 if (!is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
5297 /* Learn source MAC. */
5298 if (ctx->may_learn) {
5299 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
5302 /* Determine output bundle. */
5303 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
5306 if (mac->port.p != in_bundle) {
5307 output_normal(ctx, mac->port.p, vlan);
5309 } else if (!ctx->packet && !eth_addr_is_multicast(ctx->flow.dl_dst)) {
5310 /* If we are revalidating but don't have a learning entry then eject
5311 * the flow. Installing a flow that floods packets opens up a window
5312 * of time where we could learn from a packet reflected on a bond and
5313 * blackhole packets before the learning table is updated to reflect
5314 * the correct port. */
5315 ctx->may_set_up_flow = false;
5318 struct ofbundle *bundle;
5320 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
5321 if (bundle != in_bundle
5322 && ofbundle_includes_vlan(bundle, vlan)
5323 && bundle->floodable
5324 && !bundle->mirror_out) {
5325 output_normal(ctx, bundle, vlan);
5328 ctx->nf_output_iface = NF_OUT_FLOOD;
5332 /* Optimized flow revalidation.
5334 * It's a difficult problem, in general, to tell which facets need to have
5335 * their actions recalculated whenever the OpenFlow flow table changes. We
5336 * don't try to solve that general problem: for most kinds of OpenFlow flow
5337 * table changes, we recalculate the actions for every facet. This is
5338 * relatively expensive, but it's good enough if the OpenFlow flow table
5339 * doesn't change very often.
5341 * However, we can expect one particular kind of OpenFlow flow table change to
5342 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5343 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5344 * table, we add a special case that applies to flow tables in which every rule
5345 * has the same form (that is, the same wildcards), except that the table is
5346 * also allowed to have a single "catch-all" flow that matches all packets. We
5347 * optimize this case by tagging all of the facets that resubmit into the table
5348 * and invalidating the same tag whenever a flow changes in that table. The
5349 * end result is that we revalidate just the facets that need it (and sometimes
5350 * a few more, but not all of the facets or even all of the facets that
5351 * resubmit to the table modified by MAC learning). */
5353 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
5354 * into an OpenFlow table with the given 'basis'. */
5356 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
5359 if (flow_wildcards_is_catchall(wc)) {
5362 struct flow tag_flow = *flow;
5363 flow_zero_wildcards(&tag_flow, wc);
5364 return tag_create_deterministic(flow_hash(&tag_flow, secret));
5368 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5369 * taggability of that table.
5371 * This function must be called after *each* change to a flow table. If you
5372 * skip calling it on some changes then the pointer comparisons at the end can
5373 * be invalid if you get unlucky. For example, if a flow removal causes a
5374 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5375 * different wildcards to be created with the same address, then this function
5376 * will incorrectly skip revalidation. */
5378 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5380 struct table_dpif *table = &ofproto->tables[table_id];
5381 const struct classifier *cls = &ofproto->up.tables[table_id];
5382 struct cls_table *catchall, *other;
5383 struct cls_table *t;
5385 catchall = other = NULL;
5387 switch (hmap_count(&cls->tables)) {
5389 /* We could tag this OpenFlow table but it would make the logic a
5390 * little harder and it's a corner case that doesn't seem worth it
5396 HMAP_FOR_EACH (t, hmap_node, &cls->tables) {
5397 if (cls_table_is_catchall(t)) {
5399 } else if (!other) {
5402 /* Indicate that we can't tag this by setting both tables to
5403 * NULL. (We know that 'catchall' is already NULL.) */
5410 /* Can't tag this table. */
5414 if (table->catchall_table != catchall || table->other_table != other) {
5415 table->catchall_table = catchall;
5416 table->other_table = other;
5417 ofproto->need_revalidate = true;
5421 /* Given 'rule' that has changed in some way (either it is a rule being
5422 * inserted, a rule being deleted, or a rule whose actions are being
5423 * modified), marks facets for revalidation to ensure that packets will be
5424 * forwarded correctly according to the new state of the flow table.
5426 * This function must be called after *each* change to a flow table. See
5427 * the comment on table_update_taggable() for more information. */
5429 rule_invalidate(const struct rule_dpif *rule)
5431 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5433 table_update_taggable(ofproto, rule->up.table_id);
5435 if (!ofproto->need_revalidate) {
5436 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5438 if (table->other_table && rule->tag) {
5439 tag_set_add(&ofproto->revalidate_set, rule->tag);
5441 ofproto->need_revalidate = true;
5447 set_frag_handling(struct ofproto *ofproto_,
5448 enum ofp_config_flags frag_handling)
5450 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5452 if (frag_handling != OFPC_FRAG_REASM) {
5453 ofproto->need_revalidate = true;
5461 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5462 const struct flow *flow,
5463 const union ofp_action *ofp_actions, size_t n_ofp_actions)
5465 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5468 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
5469 return ofp_mkerr_nicira(OFPET_BAD_REQUEST, NXBRC_BAD_IN_PORT);
5472 error = validate_actions(ofp_actions, n_ofp_actions, flow,
5473 ofproto->max_ports);
5475 struct odputil_keybuf keybuf;
5476 struct action_xlate_ctx ctx;
5477 struct ofpbuf *odp_actions;
5480 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5481 odp_flow_key_from_flow(&key, flow);
5483 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, packet);
5484 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
5485 dpif_execute(ofproto->dpif, key.data, key.size,
5486 odp_actions->data, odp_actions->size, packet);
5487 ofpbuf_delete(odp_actions);
5495 set_netflow(struct ofproto *ofproto_,
5496 const struct netflow_options *netflow_options)
5498 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5500 if (netflow_options) {
5501 if (!ofproto->netflow) {
5502 ofproto->netflow = netflow_create();
5504 return netflow_set_options(ofproto->netflow, netflow_options);
5506 netflow_destroy(ofproto->netflow);
5507 ofproto->netflow = NULL;
5513 get_netflow_ids(const struct ofproto *ofproto_,
5514 uint8_t *engine_type, uint8_t *engine_id)
5516 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5518 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
5522 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5524 if (!facet_is_controller_flow(facet) &&
5525 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5526 struct subfacet *subfacet;
5527 struct ofexpired expired;
5529 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5530 if (subfacet->installed) {
5531 struct dpif_flow_stats stats;
5533 subfacet_install(ofproto, subfacet, subfacet->actions,
5534 subfacet->actions_len, &stats);
5535 subfacet_update_stats(ofproto, subfacet, &stats);
5539 expired.flow = facet->flow;
5540 expired.packet_count = facet->packet_count;
5541 expired.byte_count = facet->byte_count;
5542 expired.used = facet->used;
5543 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5548 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5550 struct facet *facet;
5552 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5553 send_active_timeout(ofproto, facet);
5557 static struct ofproto_dpif *
5558 ofproto_dpif_lookup(const char *name)
5560 struct ofproto_dpif *ofproto;
5562 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5563 hash_string(name, 0), &all_ofproto_dpifs) {
5564 if (!strcmp(ofproto->up.name, name)) {
5572 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn,
5573 const char *args, void *aux OVS_UNUSED)
5575 const struct ofproto_dpif *ofproto;
5577 ofproto = ofproto_dpif_lookup(args);
5579 unixctl_command_reply(conn, 501, "no such bridge");
5582 mac_learning_flush(ofproto->ml);
5584 unixctl_command_reply(conn, 200, "table successfully flushed");
5588 ofproto_unixctl_fdb_show(struct unixctl_conn *conn,
5589 const char *args, void *aux OVS_UNUSED)
5591 struct ds ds = DS_EMPTY_INITIALIZER;
5592 const struct ofproto_dpif *ofproto;
5593 const struct mac_entry *e;
5595 ofproto = ofproto_dpif_lookup(args);
5597 unixctl_command_reply(conn, 501, "no such bridge");
5601 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5602 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5603 struct ofbundle *bundle = e->port.p;
5604 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
5605 ofbundle_get_a_port(bundle)->odp_port,
5606 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
5608 unixctl_command_reply(conn, 200, ds_cstr(&ds));
5612 struct ofproto_trace {
5613 struct action_xlate_ctx ctx;
5619 trace_format_rule(struct ds *result, uint8_t table_id, int level,
5620 const struct rule_dpif *rule)
5622 ds_put_char_multiple(result, '\t', level);
5624 ds_put_cstr(result, "No match\n");
5628 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5629 table_id, ntohll(rule->up.flow_cookie));
5630 cls_rule_format(&rule->up.cr, result);
5631 ds_put_char(result, '\n');
5633 ds_put_char_multiple(result, '\t', level);
5634 ds_put_cstr(result, "OpenFlow ");
5635 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
5636 ds_put_char(result, '\n');
5640 trace_format_flow(struct ds *result, int level, const char *title,
5641 struct ofproto_trace *trace)
5643 ds_put_char_multiple(result, '\t', level);
5644 ds_put_format(result, "%s: ", title);
5645 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
5646 ds_put_cstr(result, "unchanged");
5648 flow_format(result, &trace->ctx.flow);
5649 trace->flow = trace->ctx.flow;
5651 ds_put_char(result, '\n');
5655 trace_format_regs(struct ds *result, int level, const char *title,
5656 struct ofproto_trace *trace)
5660 ds_put_char_multiple(result, '\t', level);
5661 ds_put_format(result, "%s:", title);
5662 for (i = 0; i < FLOW_N_REGS; i++) {
5663 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5665 ds_put_char(result, '\n');
5669 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
5671 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
5672 struct ds *result = trace->result;
5674 ds_put_char(result, '\n');
5675 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
5676 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
5677 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
5681 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
5682 void *aux OVS_UNUSED)
5684 char *dpname, *arg1, *arg2, *arg3, *arg4;
5685 char *args = xstrdup(args_);
5686 char *save_ptr = NULL;
5687 struct ofproto_dpif *ofproto;
5688 struct ofpbuf odp_key;
5689 struct ofpbuf *packet;
5690 struct rule_dpif *rule;
5691 ovs_be16 initial_tci;
5697 ofpbuf_init(&odp_key, 0);
5700 dpname = strtok_r(args, " ", &save_ptr);
5702 unixctl_command_reply(conn, 501, "Bad command syntax");
5706 ofproto = ofproto_dpif_lookup(dpname);
5708 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
5712 arg1 = strtok_r(NULL, " ", &save_ptr);
5713 arg2 = strtok_r(NULL, " ", &save_ptr);
5714 arg3 = strtok_r(NULL, " ", &save_ptr);
5715 arg4 = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
5716 if (dpname && arg1 && (!arg2 || !strcmp(arg2, "-generate")) && !arg3) {
5717 /* ofproto/trace dpname flow [-generate] */
5720 /* Convert string to datapath key. */
5721 ofpbuf_init(&odp_key, 0);
5722 error = odp_flow_key_from_string(arg1, NULL, &odp_key);
5724 unixctl_command_reply(conn, 501, "Bad flow syntax");
5728 /* Convert odp_key to flow. */
5729 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
5730 odp_key.size, &flow,
5732 if (error == ODP_FIT_ERROR) {
5733 unixctl_command_reply(conn, 501, "Invalid flow");
5737 /* Generate a packet, if requested. */
5739 packet = ofpbuf_new(0);
5740 flow_compose(packet, &flow);
5742 } else if (dpname && arg1 && arg2 && arg3 && arg4) {
5743 /* ofproto/trace dpname priority tun_id in_port packet */
5748 priority = atoi(arg1);
5749 tun_id = htonll(strtoull(arg2, NULL, 0));
5750 in_port = ofp_port_to_odp_port(atoi(arg3));
5752 packet = ofpbuf_new(strlen(args) / 2);
5753 arg4 = ofpbuf_put_hex(packet, arg4, NULL);
5754 arg4 += strspn(arg4, " ");
5755 if (*arg4 != '\0') {
5756 unixctl_command_reply(conn, 501, "Trailing garbage in command");
5759 if (packet->size < ETH_HEADER_LEN) {
5760 unixctl_command_reply(conn, 501,
5761 "Packet data too short for Ethernet");
5765 ds_put_cstr(&result, "Packet: ");
5766 s = ofp_packet_to_string(packet->data, packet->size, packet->size);
5767 ds_put_cstr(&result, s);
5770 flow_extract(packet, priority, tun_id, in_port, &flow);
5771 initial_tci = flow.vlan_tci;
5773 unixctl_command_reply(conn, 501, "Bad command syntax");
5777 ds_put_cstr(&result, "Flow: ");
5778 flow_format(&result, &flow);
5779 ds_put_char(&result, '\n');
5781 rule = rule_dpif_lookup(ofproto, &flow, 0);
5782 trace_format_rule(&result, 0, 0, rule);
5784 struct ofproto_trace trace;
5785 struct ofpbuf *odp_actions;
5787 trace.result = &result;
5789 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, initial_tci, packet);
5790 trace.ctx.resubmit_hook = trace_resubmit;
5791 odp_actions = xlate_actions(&trace.ctx,
5792 rule->up.actions, rule->up.n_actions);
5794 ds_put_char(&result, '\n');
5795 trace_format_flow(&result, 0, "Final flow", &trace);
5796 ds_put_cstr(&result, "Datapath actions: ");
5797 format_odp_actions(&result, odp_actions->data, odp_actions->size);
5798 ofpbuf_delete(odp_actions);
5800 if (!trace.ctx.may_set_up_flow) {
5802 ds_put_cstr(&result, "\nThis flow is not cachable.");
5804 ds_put_cstr(&result, "\nThe datapath actions are incomplete--"
5805 "for complete actions, please supply a packet.");
5810 unixctl_command_reply(conn, 200, ds_cstr(&result));
5813 ds_destroy(&result);
5814 ofpbuf_delete(packet);
5815 ofpbuf_uninit(&odp_key);
5820 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED,
5821 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
5824 unixctl_command_reply(conn, 200, NULL);
5828 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED,
5829 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
5832 unixctl_command_reply(conn, 200, NULL);
5836 ofproto_dpif_unixctl_init(void)
5838 static bool registered;
5844 unixctl_command_register("ofproto/trace",
5845 "bridge {tun_id in_port packet | odp_flow [-generate]}",
5846 ofproto_unixctl_trace, NULL);
5847 unixctl_command_register("fdb/flush", "bridge", ofproto_unixctl_fdb_flush,
5849 unixctl_command_register("fdb/show", "bridge", ofproto_unixctl_fdb_show,
5851 unixctl_command_register("ofproto/clog", "", ofproto_dpif_clog, NULL);
5852 unixctl_command_register("ofproto/unclog", "", ofproto_dpif_unclog, NULL);
5855 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5857 * This is deprecated. It is only for compatibility with broken device drivers
5858 * in old versions of Linux that do not properly support VLANs when VLAN
5859 * devices are not used. When broken device drivers are no longer in
5860 * widespread use, we will delete these interfaces. */
5863 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
5865 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
5866 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
5868 if (realdev_ofp_port == ofport->realdev_ofp_port
5869 && vid == ofport->vlandev_vid) {
5873 ofproto->need_revalidate = true;
5875 if (ofport->realdev_ofp_port) {
5878 if (realdev_ofp_port && ofport->bundle) {
5879 /* vlandevs are enslaved to their realdevs, so they are not allowed to
5880 * themselves be part of a bundle. */
5881 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
5884 ofport->realdev_ofp_port = realdev_ofp_port;
5885 ofport->vlandev_vid = vid;
5887 if (realdev_ofp_port) {
5888 vsp_add(ofport, realdev_ofp_port, vid);
5895 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
5897 return hash_2words(realdev_ofp_port, vid);
5901 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
5902 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
5904 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
5905 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
5906 int vid = vlan_tci_to_vid(vlan_tci);
5907 const struct vlan_splinter *vsp;
5909 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
5910 hash_realdev_vid(realdev_ofp_port, vid),
5911 &ofproto->realdev_vid_map) {
5912 if (vsp->realdev_ofp_port == realdev_ofp_port
5913 && vsp->vid == vid) {
5914 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
5918 return realdev_odp_port;
5921 static struct vlan_splinter *
5922 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
5924 struct vlan_splinter *vsp;
5926 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
5927 &ofproto->vlandev_map) {
5928 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
5937 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
5938 uint16_t vlandev_ofp_port, int *vid)
5940 if (!hmap_is_empty(&ofproto->vlandev_map)) {
5941 const struct vlan_splinter *vsp;
5943 vsp = vlandev_find(ofproto, vlandev_ofp_port);
5948 return vsp->realdev_ofp_port;
5955 vsp_remove(struct ofport_dpif *port)
5957 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
5958 struct vlan_splinter *vsp;
5960 vsp = vlandev_find(ofproto, port->up.ofp_port);
5962 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
5963 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
5966 port->realdev_ofp_port = 0;
5968 VLOG_ERR("missing vlan device record");
5973 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
5975 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
5977 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
5978 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
5979 == realdev_ofp_port)) {
5980 struct vlan_splinter *vsp;
5982 vsp = xmalloc(sizeof *vsp);
5983 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
5984 hash_int(port->up.ofp_port, 0));
5985 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
5986 hash_realdev_vid(realdev_ofp_port, vid));
5987 vsp->realdev_ofp_port = realdev_ofp_port;
5988 vsp->vlandev_ofp_port = port->up.ofp_port;
5991 port->realdev_ofp_port = realdev_ofp_port;
5993 VLOG_ERR("duplicate vlan device record");
5997 const struct ofproto_class ofproto_dpif_class = {
6026 port_is_lacp_current,
6027 NULL, /* rule_choose_table */
6034 rule_modify_actions,
6042 get_cfm_remote_mpids,
6046 get_stp_port_status,
6053 is_mirror_output_bundle,
6054 forward_bpdu_changed,