2 * Copyright (c) 2009, 2010, 2011, 2012 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-provider.h"
26 #include "byte-order.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
36 #include "mac-learning.h"
37 #include "multipath.h"
44 #include "ofp-print.h"
45 #include "ofproto-dpif-sflow.h"
46 #include "poll-loop.h"
48 #include "unaligned.h"
50 #include "vlan-bitmap.h"
53 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
55 COVERAGE_DEFINE(ofproto_dpif_ctlr_action);
56 COVERAGE_DEFINE(ofproto_dpif_expired);
57 COVERAGE_DEFINE(ofproto_dpif_no_packet_in);
58 COVERAGE_DEFINE(ofproto_dpif_xlate);
59 COVERAGE_DEFINE(facet_changed_rule);
60 COVERAGE_DEFINE(facet_invalidated);
61 COVERAGE_DEFINE(facet_revalidate);
62 COVERAGE_DEFINE(facet_unexpected);
64 /* Maximum depth of flow table recursion (due to resubmit actions) in a
65 * flow translation. */
66 #define MAX_RESUBMIT_RECURSION 32
68 /* Number of implemented OpenFlow tables. */
69 enum { N_TABLES = 255 };
70 BUILD_ASSERT_DECL(N_TABLES >= 1 && N_TABLES <= 255);
78 long long int used; /* Time last used; time created if not used. */
82 * - Do include packets and bytes from facets that have been deleted or
83 * whose own statistics have been folded into the rule.
85 * - Do include packets and bytes sent "by hand" that were accounted to
86 * the rule without any facet being involved (this is a rare corner
87 * case in rule_execute()).
89 * - Do not include packet or bytes that can be obtained from any facet's
90 * packet_count or byte_count member or that can be obtained from the
91 * datapath by, e.g., dpif_flow_get() for any subfacet.
93 uint64_t packet_count; /* Number of packets received. */
94 uint64_t byte_count; /* Number of bytes received. */
96 tag_type tag; /* Caches rule_calculate_tag() result. */
98 struct list facets; /* List of "struct facet"s. */
101 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
103 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
106 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
107 const struct flow *, uint8_t table);
109 static void flow_push_stats(const struct rule_dpif *, const struct flow *,
110 uint64_t packets, uint64_t bytes,
113 static uint32_t rule_calculate_tag(const struct flow *,
114 const struct flow_wildcards *,
116 static void rule_invalidate(const struct rule_dpif *);
118 #define MAX_MIRRORS 32
119 typedef uint32_t mirror_mask_t;
120 #define MIRROR_MASK_C(X) UINT32_C(X)
121 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
123 struct ofproto_dpif *ofproto; /* Owning ofproto. */
124 size_t idx; /* In ofproto's "mirrors" array. */
125 void *aux; /* Key supplied by ofproto's client. */
126 char *name; /* Identifier for log messages. */
128 /* Selection criteria. */
129 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
130 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
131 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
133 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
134 struct ofbundle *out; /* Output port or NULL. */
135 int out_vlan; /* Output VLAN or -1. */
136 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
139 int64_t packet_count; /* Number of packets sent. */
140 int64_t byte_count; /* Number of bytes sent. */
143 static void mirror_destroy(struct ofmirror *);
144 static void update_mirror_stats(struct ofproto_dpif *ofproto,
145 mirror_mask_t mirrors,
146 uint64_t packets, uint64_t bytes);
149 struct ofproto_dpif *ofproto; /* Owning ofproto. */
150 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
151 void *aux; /* Key supplied by ofproto's client. */
152 char *name; /* Identifier for log messages. */
155 struct list ports; /* Contains "struct ofport"s. */
156 enum port_vlan_mode vlan_mode; /* VLAN mode */
157 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
158 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
159 * NULL if all VLANs are trunked. */
160 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
161 struct bond *bond; /* Nonnull iff more than one port. */
162 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
165 bool floodable; /* True if no port has OFPPC_NO_FLOOD set. */
167 /* Port mirroring info. */
168 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
169 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
170 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
173 static void bundle_remove(struct ofport *);
174 static void bundle_update(struct ofbundle *);
175 static void bundle_destroy(struct ofbundle *);
176 static void bundle_del_port(struct ofport_dpif *);
177 static void bundle_run(struct ofbundle *);
178 static void bundle_wait(struct ofbundle *);
179 static struct ofbundle *lookup_input_bundle(struct ofproto_dpif *,
180 uint16_t in_port, bool warn);
182 /* A controller may use OFPP_NONE as the ingress port to indicate that
183 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
184 * when an input bundle is needed for validation (e.g., mirroring or
185 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
186 * any 'port' structs, so care must be taken when dealing with it. */
187 static struct ofbundle ofpp_none_bundle = {
189 .vlan_mode = PORT_VLAN_TRUNK
192 static void stp_run(struct ofproto_dpif *ofproto);
193 static void stp_wait(struct ofproto_dpif *ofproto);
194 static int set_stp_port(struct ofport *,
195 const struct ofproto_port_stp_settings *);
197 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
199 struct action_xlate_ctx {
200 /* action_xlate_ctx_init() initializes these members. */
203 struct ofproto_dpif *ofproto;
205 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
206 * this flow when actions change header fields. */
209 /* The packet corresponding to 'flow', or a null pointer if we are
210 * revalidating without a packet to refer to. */
211 const struct ofpbuf *packet;
213 /* Should OFPP_NORMAL MAC learning and NXAST_LEARN actions execute? We
214 * want to execute them if we are actually processing a packet, or if we
215 * are accounting for packets that the datapath has processed, but not if
216 * we are just revalidating. */
219 /* Cookie of the currently matching rule, or 0. */
222 /* If nonnull, called just before executing a resubmit action.
224 * This is normally null so the client has to set it manually after
225 * calling action_xlate_ctx_init(). */
226 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *);
228 /* xlate_actions() initializes and uses these members. The client might want
229 * to look at them after it returns. */
231 struct ofpbuf *odp_actions; /* Datapath actions. */
232 tag_type tags; /* Tags associated with actions. */
233 bool may_set_up_flow; /* True ordinarily; false if the actions must
234 * be reassessed for every packet. */
235 bool has_learn; /* Actions include NXAST_LEARN? */
236 bool has_normal; /* Actions output to OFPP_NORMAL? */
237 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
238 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
240 /* xlate_actions() initializes and uses these members, but the client has no
241 * reason to look at them. */
243 int recurse; /* Recursion level, via xlate_table_action. */
244 struct flow base_flow; /* Flow at the last commit. */
245 uint32_t orig_skb_priority; /* Priority when packet arrived. */
246 uint8_t table_id; /* OpenFlow table ID where flow was found. */
247 uint32_t sflow_n_outputs; /* Number of output ports. */
248 uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
249 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
250 bool exit; /* No further actions should be processed. */
253 static void action_xlate_ctx_init(struct action_xlate_ctx *,
254 struct ofproto_dpif *, const struct flow *,
255 ovs_be16 initial_tci, ovs_be64 cookie,
256 const struct ofpbuf *);
257 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
258 const union ofp_action *in, size_t n_in);
260 /* An exact-match instantiation of an OpenFlow flow.
262 * A facet associates a "struct flow", which represents the Open vSwitch
263 * userspace idea of an exact-match flow, with one or more subfacets. Each
264 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
265 * the facet. When the kernel module (or other dpif implementation) and Open
266 * vSwitch userspace agree on the definition of a flow key, there is exactly
267 * one subfacet per facet. If the dpif implementation supports more-specific
268 * flow matching than userspace, however, a facet can have more than one
269 * subfacet, each of which corresponds to some distinction in flow that
270 * userspace simply doesn't understand.
272 * Flow expiration works in terms of subfacets, so a facet must have at least
273 * one subfacet or it will never expire, leaking memory. */
276 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
277 struct list list_node; /* In owning rule's 'facets' list. */
278 struct rule_dpif *rule; /* Owning rule. */
281 struct list subfacets;
282 long long int used; /* Time last used; time created if not used. */
289 * - Do include packets and bytes sent "by hand", e.g. with
292 * - Do include packets and bytes that were obtained from the datapath
293 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
294 * DPIF_FP_ZERO_STATS).
296 * - Do not include packets or bytes that can be obtained from the
297 * datapath for any existing subfacet.
299 uint64_t packet_count; /* Number of packets received. */
300 uint64_t byte_count; /* Number of bytes received. */
302 /* Resubmit statistics. */
303 uint64_t prev_packet_count; /* Number of packets from last stats push. */
304 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
305 long long int prev_used; /* Used time from last stats push. */
308 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
309 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
311 /* Properties of datapath actions.
313 * Every subfacet has its own actions because actions can differ slightly
314 * between splintered and non-splintered subfacets due to the VLAN tag
315 * being initially different (present vs. absent). All of them have these
316 * properties in common so we just store one copy of them here. */
317 bool may_install; /* Reassess actions for every packet? */
318 bool has_learn; /* Actions include NXAST_LEARN? */
319 bool has_normal; /* Actions output to OFPP_NORMAL? */
320 tag_type tags; /* Tags that would require revalidation. */
321 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
324 static struct facet *facet_create(struct rule_dpif *, const struct flow *);
325 static void facet_remove(struct ofproto_dpif *, struct facet *);
326 static void facet_free(struct facet *);
328 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
329 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
330 const struct flow *);
331 static bool facet_revalidate(struct ofproto_dpif *, struct facet *);
333 static void facet_flush_stats(struct ofproto_dpif *, struct facet *);
335 static void facet_update_time(struct ofproto_dpif *, struct facet *,
337 static void facet_reset_counters(struct facet *);
338 static void facet_push_stats(struct facet *);
339 static void facet_account(struct ofproto_dpif *, struct facet *);
341 static bool facet_is_controller_flow(struct facet *);
343 /* A dpif flow and actions associated with a facet.
345 * See also the large comment on struct facet. */
348 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
349 struct list list_node; /* In struct facet's 'facets' list. */
350 struct facet *facet; /* Owning facet. */
354 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
355 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
356 * regenerate the ODP flow key from ->facet->flow. */
357 enum odp_key_fitness key_fitness;
361 long long int used; /* Time last used; time created if not used. */
363 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
364 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
368 * These should be essentially identical for every subfacet in a facet, but
369 * may differ in trivial ways due to VLAN splinters. */
370 size_t actions_len; /* Number of bytes in actions[]. */
371 struct nlattr *actions; /* Datapath actions. */
373 bool installed; /* Installed in datapath? */
375 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
376 * splinters can cause it to differ. This value should be removed when
377 * the VLAN splinters feature is no longer needed. */
378 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
381 static struct subfacet *subfacet_create(struct ofproto_dpif *, struct facet *,
382 enum odp_key_fitness,
383 const struct nlattr *key,
384 size_t key_len, ovs_be16 initial_tci);
385 static struct subfacet *subfacet_find(struct ofproto_dpif *,
386 const struct nlattr *key, size_t key_len);
387 static void subfacet_destroy(struct ofproto_dpif *, struct subfacet *);
388 static void subfacet_destroy__(struct ofproto_dpif *, struct subfacet *);
389 static void subfacet_reset_dp_stats(struct subfacet *,
390 struct dpif_flow_stats *);
391 static void subfacet_update_time(struct ofproto_dpif *, struct subfacet *,
393 static void subfacet_update_stats(struct ofproto_dpif *, struct subfacet *,
394 const struct dpif_flow_stats *);
395 static void subfacet_make_actions(struct ofproto_dpif *, struct subfacet *,
396 const struct ofpbuf *packet);
397 static int subfacet_install(struct ofproto_dpif *, struct subfacet *,
398 const struct nlattr *actions, size_t actions_len,
399 struct dpif_flow_stats *);
400 static void subfacet_uninstall(struct ofproto_dpif *, struct subfacet *);
406 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
407 struct list bundle_node; /* In struct ofbundle's "ports" list. */
408 struct cfm *cfm; /* Connectivity Fault Management, if any. */
409 tag_type tag; /* Tag associated with this port. */
410 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
411 bool may_enable; /* May be enabled in bonds. */
414 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
415 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
416 long long int stp_state_entered;
418 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
420 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
422 * This is deprecated. It is only for compatibility with broken device
423 * drivers in old versions of Linux that do not properly support VLANs when
424 * VLAN devices are not used. When broken device drivers are no longer in
425 * widespread use, we will delete these interfaces. */
426 uint16_t realdev_ofp_port;
430 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
431 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
432 * traffic egressing the 'ofport' with that priority should be marked with. */
433 struct priority_to_dscp {
434 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
435 uint32_t priority; /* Priority of this queue (see struct flow). */
437 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
440 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
442 * This is deprecated. It is only for compatibility with broken device drivers
443 * in old versions of Linux that do not properly support VLANs when VLAN
444 * devices are not used. When broken device drivers are no longer in
445 * widespread use, we will delete these interfaces. */
446 struct vlan_splinter {
447 struct hmap_node realdev_vid_node;
448 struct hmap_node vlandev_node;
449 uint16_t realdev_ofp_port;
450 uint16_t vlandev_ofp_port;
454 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
455 uint32_t realdev, ovs_be16 vlan_tci);
456 static uint16_t vsp_vlandev_to_realdev(const struct ofproto_dpif *,
457 uint16_t vlandev, int *vid);
458 static void vsp_remove(struct ofport_dpif *);
459 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
461 static struct ofport_dpif *
462 ofport_dpif_cast(const struct ofport *ofport)
464 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
465 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
468 static void port_run(struct ofport_dpif *);
469 static void port_wait(struct ofport_dpif *);
470 static int set_cfm(struct ofport *, const struct cfm_settings *);
471 static void ofport_clear_priorities(struct ofport_dpif *);
473 struct dpif_completion {
474 struct list list_node;
475 struct ofoperation *op;
478 /* Extra information about a classifier table.
479 * Currently used just for optimized flow revalidation. */
481 /* If either of these is nonnull, then this table has a form that allows
482 * flows to be tagged to avoid revalidating most flows for the most common
483 * kinds of flow table changes. */
484 struct cls_table *catchall_table; /* Table that wildcards all fields. */
485 struct cls_table *other_table; /* Table with any other wildcard set. */
486 uint32_t basis; /* Keeps each table's tags separate. */
489 struct ofproto_dpif {
490 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
499 struct netflow *netflow;
500 struct dpif_sflow *sflow;
501 struct hmap bundles; /* Contains "struct ofbundle"s. */
502 struct mac_learning *ml;
503 struct ofmirror *mirrors[MAX_MIRRORS];
504 bool has_bonded_bundles;
507 struct timer next_expiration;
511 struct hmap subfacets;
514 struct table_dpif tables[N_TABLES];
515 bool need_revalidate;
516 struct tag_set revalidate_set;
518 /* Support for debugging async flow mods. */
519 struct list completions;
521 bool has_bundle_action; /* True when the first bundle action appears. */
522 struct netdev_stats stats; /* To account packets generated and consumed in
527 long long int stp_last_tick;
529 /* VLAN splinters. */
530 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
531 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
534 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
535 * for debugging the asynchronous flow_mod implementation.) */
538 /* All existing ofproto_dpif instances, indexed by ->up.name. */
539 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
541 static void ofproto_dpif_unixctl_init(void);
543 static struct ofproto_dpif *
544 ofproto_dpif_cast(const struct ofproto *ofproto)
546 assert(ofproto->ofproto_class == &ofproto_dpif_class);
547 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
550 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
552 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
555 /* Packet processing. */
556 static void update_learning_table(struct ofproto_dpif *,
557 const struct flow *, int vlan,
560 #define FLOW_MISS_MAX_BATCH 50
561 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
563 /* Flow expiration. */
564 static int expire(struct ofproto_dpif *);
567 static void send_netflow_active_timeouts(struct ofproto_dpif *);
570 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
572 compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions,
573 const struct flow *, uint32_t odp_port);
574 static void add_mirror_actions(struct action_xlate_ctx *ctx,
575 const struct flow *flow);
576 /* Global variables. */
577 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
579 /* Factory functions. */
582 enumerate_types(struct sset *types)
584 dp_enumerate_types(types);
588 enumerate_names(const char *type, struct sset *names)
590 return dp_enumerate_names(type, names);
594 del(const char *type, const char *name)
599 error = dpif_open(name, type, &dpif);
601 error = dpif_delete(dpif);
607 /* Basic life-cycle. */
609 static struct ofproto *
612 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
617 dealloc(struct ofproto *ofproto_)
619 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
624 construct(struct ofproto *ofproto_, int *n_tablesp)
626 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
627 const char *name = ofproto->up.name;
631 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
633 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
637 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
638 ofproto->n_matches = 0;
640 dpif_flow_flush(ofproto->dpif);
641 dpif_recv_purge(ofproto->dpif);
643 error = dpif_recv_set_mask(ofproto->dpif,
644 ((1u << DPIF_UC_MISS) |
645 (1u << DPIF_UC_ACTION)));
647 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
648 dpif_close(ofproto->dpif);
652 ofproto->netflow = NULL;
653 ofproto->sflow = NULL;
655 hmap_init(&ofproto->bundles);
656 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
657 for (i = 0; i < MAX_MIRRORS; i++) {
658 ofproto->mirrors[i] = NULL;
660 ofproto->has_bonded_bundles = false;
662 timer_set_duration(&ofproto->next_expiration, 1000);
664 hmap_init(&ofproto->facets);
665 hmap_init(&ofproto->subfacets);
667 for (i = 0; i < N_TABLES; i++) {
668 struct table_dpif *table = &ofproto->tables[i];
670 table->catchall_table = NULL;
671 table->other_table = NULL;
672 table->basis = random_uint32();
674 ofproto->need_revalidate = false;
675 tag_set_init(&ofproto->revalidate_set);
677 list_init(&ofproto->completions);
679 ofproto_dpif_unixctl_init();
681 ofproto->has_bundle_action = false;
683 hmap_init(&ofproto->vlandev_map);
684 hmap_init(&ofproto->realdev_vid_map);
686 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
687 hash_string(ofproto->up.name, 0));
689 *n_tablesp = N_TABLES;
690 memset(&ofproto->stats, 0, sizeof ofproto->stats);
695 complete_operations(struct ofproto_dpif *ofproto)
697 struct dpif_completion *c, *next;
699 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
700 ofoperation_complete(c->op, 0);
701 list_remove(&c->list_node);
707 destruct(struct ofproto *ofproto_)
709 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
710 struct rule_dpif *rule, *next_rule;
711 struct classifier *table;
714 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
715 complete_operations(ofproto);
717 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
718 struct cls_cursor cursor;
720 cls_cursor_init(&cursor, table, NULL);
721 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
722 ofproto_rule_destroy(&rule->up);
726 for (i = 0; i < MAX_MIRRORS; i++) {
727 mirror_destroy(ofproto->mirrors[i]);
730 netflow_destroy(ofproto->netflow);
731 dpif_sflow_destroy(ofproto->sflow);
732 hmap_destroy(&ofproto->bundles);
733 mac_learning_destroy(ofproto->ml);
735 hmap_destroy(&ofproto->facets);
736 hmap_destroy(&ofproto->subfacets);
738 hmap_destroy(&ofproto->vlandev_map);
739 hmap_destroy(&ofproto->realdev_vid_map);
741 dpif_close(ofproto->dpif);
745 run_fast(struct ofproto *ofproto_)
747 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
750 /* Handle one or more batches of upcalls, until there's nothing left to do
751 * or until we do a fixed total amount of work.
753 * We do work in batches because it can be much cheaper to set up a number
754 * of flows and fire off their patches all at once. We do multiple batches
755 * because in some cases handling a packet can cause another packet to be
756 * queued almost immediately as part of the return flow. Both
757 * optimizations can make major improvements on some benchmarks and
758 * presumably for real traffic as well. */
760 while (work < FLOW_MISS_MAX_BATCH) {
761 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
771 run(struct ofproto *ofproto_)
773 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
774 struct ofport_dpif *ofport;
775 struct ofbundle *bundle;
779 complete_operations(ofproto);
781 dpif_run(ofproto->dpif);
783 error = run_fast(ofproto_);
788 if (timer_expired(&ofproto->next_expiration)) {
789 int delay = expire(ofproto);
790 timer_set_duration(&ofproto->next_expiration, delay);
793 if (ofproto->netflow) {
794 if (netflow_run(ofproto->netflow)) {
795 send_netflow_active_timeouts(ofproto);
798 if (ofproto->sflow) {
799 dpif_sflow_run(ofproto->sflow);
802 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
805 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
810 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
812 /* Now revalidate if there's anything to do. */
813 if (ofproto->need_revalidate
814 || !tag_set_is_empty(&ofproto->revalidate_set)) {
815 struct tag_set revalidate_set = ofproto->revalidate_set;
816 bool revalidate_all = ofproto->need_revalidate;
817 struct facet *facet, *next;
819 /* Clear the revalidation flags. */
820 tag_set_init(&ofproto->revalidate_set);
821 ofproto->need_revalidate = false;
823 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
825 || tag_set_intersects(&revalidate_set, facet->tags)) {
826 facet_revalidate(ofproto, facet);
835 wait(struct ofproto *ofproto_)
837 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
838 struct ofport_dpif *ofport;
839 struct ofbundle *bundle;
841 if (!clogged && !list_is_empty(&ofproto->completions)) {
842 poll_immediate_wake();
845 dpif_wait(ofproto->dpif);
846 dpif_recv_wait(ofproto->dpif);
847 if (ofproto->sflow) {
848 dpif_sflow_wait(ofproto->sflow);
850 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
851 poll_immediate_wake();
853 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
856 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
859 if (ofproto->netflow) {
860 netflow_wait(ofproto->netflow);
862 mac_learning_wait(ofproto->ml);
864 if (ofproto->need_revalidate) {
865 /* Shouldn't happen, but if it does just go around again. */
866 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
867 poll_immediate_wake();
869 timer_wait(&ofproto->next_expiration);
874 flush(struct ofproto *ofproto_)
876 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
877 struct facet *facet, *next_facet;
879 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
880 /* Mark the facet as not installed so that facet_remove() doesn't
881 * bother trying to uninstall it. There is no point in uninstalling it
882 * individually since we are about to blow away all the facets with
883 * dpif_flow_flush(). */
884 struct subfacet *subfacet;
886 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
887 subfacet->installed = false;
888 subfacet->dp_packet_count = 0;
889 subfacet->dp_byte_count = 0;
891 facet_remove(ofproto, facet);
893 dpif_flow_flush(ofproto->dpif);
897 get_features(struct ofproto *ofproto_ OVS_UNUSED,
898 bool *arp_match_ip, uint32_t *actions)
900 *arp_match_ip = true;
901 *actions = ((1u << OFPAT_OUTPUT) |
902 (1u << OFPAT_SET_VLAN_VID) |
903 (1u << OFPAT_SET_VLAN_PCP) |
904 (1u << OFPAT_STRIP_VLAN) |
905 (1u << OFPAT_SET_DL_SRC) |
906 (1u << OFPAT_SET_DL_DST) |
907 (1u << OFPAT_SET_NW_SRC) |
908 (1u << OFPAT_SET_NW_DST) |
909 (1u << OFPAT_SET_NW_TOS) |
910 (1u << OFPAT_SET_TP_SRC) |
911 (1u << OFPAT_SET_TP_DST) |
912 (1u << OFPAT_ENQUEUE));
916 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
918 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
919 struct dpif_dp_stats s;
921 strcpy(ots->name, "classifier");
923 dpif_get_dp_stats(ofproto->dpif, &s);
924 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
925 put_32aligned_be64(&ots->matched_count,
926 htonll(s.n_hit + ofproto->n_matches));
929 static struct ofport *
932 struct ofport_dpif *port = xmalloc(sizeof *port);
937 port_dealloc(struct ofport *port_)
939 struct ofport_dpif *port = ofport_dpif_cast(port_);
944 port_construct(struct ofport *port_)
946 struct ofport_dpif *port = ofport_dpif_cast(port_);
947 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
949 ofproto->need_revalidate = true;
950 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
953 port->tag = tag_create_random();
954 port->may_enable = true;
955 port->stp_port = NULL;
956 port->stp_state = STP_DISABLED;
957 hmap_init(&port->priorities);
958 port->realdev_ofp_port = 0;
959 port->vlandev_vid = 0;
961 if (ofproto->sflow) {
962 dpif_sflow_add_port(ofproto->sflow, port_);
969 port_destruct(struct ofport *port_)
971 struct ofport_dpif *port = ofport_dpif_cast(port_);
972 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
974 ofproto->need_revalidate = true;
975 bundle_remove(port_);
976 set_cfm(port_, NULL);
977 if (ofproto->sflow) {
978 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
981 ofport_clear_priorities(port);
982 hmap_destroy(&port->priorities);
986 port_modified(struct ofport *port_)
988 struct ofport_dpif *port = ofport_dpif_cast(port_);
990 if (port->bundle && port->bundle->bond) {
991 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
996 port_reconfigured(struct ofport *port_, ovs_be32 old_config)
998 struct ofport_dpif *port = ofport_dpif_cast(port_);
999 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1000 ovs_be32 changed = old_config ^ port->up.opp.config;
1002 if (changed & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
1003 OFPPC_NO_FWD | OFPPC_NO_FLOOD)) {
1004 ofproto->need_revalidate = true;
1006 if (changed & htonl(OFPPC_NO_FLOOD) && port->bundle) {
1007 bundle_update(port->bundle);
1013 set_sflow(struct ofproto *ofproto_,
1014 const struct ofproto_sflow_options *sflow_options)
1016 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1017 struct dpif_sflow *ds = ofproto->sflow;
1019 if (sflow_options) {
1021 struct ofport_dpif *ofport;
1023 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1024 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1025 dpif_sflow_add_port(ds, &ofport->up);
1027 ofproto->need_revalidate = true;
1029 dpif_sflow_set_options(ds, sflow_options);
1032 dpif_sflow_destroy(ds);
1033 ofproto->need_revalidate = true;
1034 ofproto->sflow = NULL;
1041 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1043 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1050 struct ofproto_dpif *ofproto;
1052 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1053 ofproto->need_revalidate = true;
1054 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1057 if (cfm_configure(ofport->cfm, s)) {
1063 cfm_destroy(ofport->cfm);
1069 get_cfm_fault(const struct ofport *ofport_)
1071 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1073 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1077 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1080 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1083 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1090 /* Spanning Tree. */
1093 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1095 struct ofproto_dpif *ofproto = ofproto_;
1096 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1097 struct ofport_dpif *ofport;
1099 ofport = stp_port_get_aux(sp);
1101 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1102 ofproto->up.name, port_num);
1104 struct eth_header *eth = pkt->l2;
1106 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1107 if (eth_addr_is_zero(eth->eth_src)) {
1108 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1109 "with unknown MAC", ofproto->up.name, port_num);
1111 send_packet(ofport, pkt);
1117 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1119 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1121 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1123 /* Only revalidate flows if the configuration changed. */
1124 if (!s != !ofproto->stp) {
1125 ofproto->need_revalidate = true;
1129 if (!ofproto->stp) {
1130 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1131 send_bpdu_cb, ofproto);
1132 ofproto->stp_last_tick = time_msec();
1135 stp_set_bridge_id(ofproto->stp, s->system_id);
1136 stp_set_bridge_priority(ofproto->stp, s->priority);
1137 stp_set_hello_time(ofproto->stp, s->hello_time);
1138 stp_set_max_age(ofproto->stp, s->max_age);
1139 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1141 struct ofport *ofport;
1143 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1144 set_stp_port(ofport, NULL);
1147 stp_destroy(ofproto->stp);
1148 ofproto->stp = NULL;
1155 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1157 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1161 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1162 s->designated_root = stp_get_designated_root(ofproto->stp);
1163 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1172 update_stp_port_state(struct ofport_dpif *ofport)
1174 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1175 enum stp_state state;
1177 /* Figure out new state. */
1178 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1182 if (ofport->stp_state != state) {
1186 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1187 netdev_get_name(ofport->up.netdev),
1188 stp_state_name(ofport->stp_state),
1189 stp_state_name(state));
1190 if (stp_learn_in_state(ofport->stp_state)
1191 != stp_learn_in_state(state)) {
1192 /* xxx Learning action flows should also be flushed. */
1193 mac_learning_flush(ofproto->ml);
1195 fwd_change = stp_forward_in_state(ofport->stp_state)
1196 != stp_forward_in_state(state);
1198 ofproto->need_revalidate = true;
1199 ofport->stp_state = state;
1200 ofport->stp_state_entered = time_msec();
1202 if (fwd_change && ofport->bundle) {
1203 bundle_update(ofport->bundle);
1206 /* Update the STP state bits in the OpenFlow port description. */
1207 of_state = (ofport->up.opp.state & htonl(~OFPPS_STP_MASK))
1208 | htonl(state == STP_LISTENING ? OFPPS_STP_LISTEN
1209 : state == STP_LEARNING ? OFPPS_STP_LEARN
1210 : state == STP_FORWARDING ? OFPPS_STP_FORWARD
1211 : state == STP_BLOCKING ? OFPPS_STP_BLOCK
1213 ofproto_port_set_state(&ofport->up, of_state);
1217 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1218 * caller is responsible for assigning STP port numbers and ensuring
1219 * there are no duplicates. */
1221 set_stp_port(struct ofport *ofport_,
1222 const struct ofproto_port_stp_settings *s)
1224 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1225 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1226 struct stp_port *sp = ofport->stp_port;
1228 if (!s || !s->enable) {
1230 ofport->stp_port = NULL;
1231 stp_port_disable(sp);
1232 update_stp_port_state(ofport);
1235 } else if (sp && stp_port_no(sp) != s->port_num
1236 && ofport == stp_port_get_aux(sp)) {
1237 /* The port-id changed, so disable the old one if it's not
1238 * already in use by another port. */
1239 stp_port_disable(sp);
1242 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1243 stp_port_enable(sp);
1245 stp_port_set_aux(sp, ofport);
1246 stp_port_set_priority(sp, s->priority);
1247 stp_port_set_path_cost(sp, s->path_cost);
1249 update_stp_port_state(ofport);
1255 get_stp_port_status(struct ofport *ofport_,
1256 struct ofproto_port_stp_status *s)
1258 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1259 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1260 struct stp_port *sp = ofport->stp_port;
1262 if (!ofproto->stp || !sp) {
1268 s->port_id = stp_port_get_id(sp);
1269 s->state = stp_port_get_state(sp);
1270 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1271 s->role = stp_port_get_role(sp);
1272 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1278 stp_run(struct ofproto_dpif *ofproto)
1281 long long int now = time_msec();
1282 long long int elapsed = now - ofproto->stp_last_tick;
1283 struct stp_port *sp;
1286 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1287 ofproto->stp_last_tick = now;
1289 while (stp_get_changed_port(ofproto->stp, &sp)) {
1290 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1293 update_stp_port_state(ofport);
1300 stp_wait(struct ofproto_dpif *ofproto)
1303 poll_timer_wait(1000);
1307 /* Returns true if STP should process 'flow'. */
1309 stp_should_process_flow(const struct flow *flow)
1311 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1315 stp_process_packet(const struct ofport_dpif *ofport,
1316 const struct ofpbuf *packet)
1318 struct ofpbuf payload = *packet;
1319 struct eth_header *eth = payload.data;
1320 struct stp_port *sp = ofport->stp_port;
1322 /* Sink packets on ports that have STP disabled when the bridge has
1324 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1328 /* Trim off padding on payload. */
1329 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1330 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1333 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1334 stp_received_bpdu(sp, payload.data, payload.size);
1338 static struct priority_to_dscp *
1339 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1341 struct priority_to_dscp *pdscp;
1344 hash = hash_int(priority, 0);
1345 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1346 if (pdscp->priority == priority) {
1354 ofport_clear_priorities(struct ofport_dpif *ofport)
1356 struct priority_to_dscp *pdscp, *next;
1358 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1359 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1365 set_queues(struct ofport *ofport_,
1366 const struct ofproto_port_queue *qdscp_list,
1369 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1370 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1371 struct hmap new = HMAP_INITIALIZER(&new);
1374 for (i = 0; i < n_qdscp; i++) {
1375 struct priority_to_dscp *pdscp;
1379 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1380 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1385 pdscp = get_priority(ofport, priority);
1387 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1389 pdscp = xmalloc(sizeof *pdscp);
1390 pdscp->priority = priority;
1392 ofproto->need_revalidate = true;
1395 if (pdscp->dscp != dscp) {
1397 ofproto->need_revalidate = true;
1400 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1403 if (!hmap_is_empty(&ofport->priorities)) {
1404 ofport_clear_priorities(ofport);
1405 ofproto->need_revalidate = true;
1408 hmap_swap(&new, &ofport->priorities);
1416 /* Expires all MAC learning entries associated with 'bundle' and forces its
1417 * ofproto to revalidate every flow.
1419 * Normally MAC learning entries are removed only from the ofproto associated
1420 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1421 * are removed from every ofproto. When patch ports and SLB bonds are in use
1422 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1423 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1424 * with the host from which it migrated. */
1426 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1428 struct ofproto_dpif *ofproto = bundle->ofproto;
1429 struct mac_learning *ml = ofproto->ml;
1430 struct mac_entry *mac, *next_mac;
1432 ofproto->need_revalidate = true;
1433 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1434 if (mac->port.p == bundle) {
1436 struct ofproto_dpif *o;
1438 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1440 struct mac_entry *e;
1442 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1445 tag_set_add(&o->revalidate_set, e->tag);
1446 mac_learning_expire(o->ml, e);
1452 mac_learning_expire(ml, mac);
1457 static struct ofbundle *
1458 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1460 struct ofbundle *bundle;
1462 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1463 &ofproto->bundles) {
1464 if (bundle->aux == aux) {
1471 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1472 * ones that are found to 'bundles'. */
1474 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1475 void **auxes, size_t n_auxes,
1476 struct hmapx *bundles)
1480 hmapx_init(bundles);
1481 for (i = 0; i < n_auxes; i++) {
1482 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1484 hmapx_add(bundles, bundle);
1490 bundle_update(struct ofbundle *bundle)
1492 struct ofport_dpif *port;
1494 bundle->floodable = true;
1495 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1496 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
1497 bundle->floodable = false;
1504 bundle_del_port(struct ofport_dpif *port)
1506 struct ofbundle *bundle = port->bundle;
1508 bundle->ofproto->need_revalidate = true;
1510 list_remove(&port->bundle_node);
1511 port->bundle = NULL;
1514 lacp_slave_unregister(bundle->lacp, port);
1517 bond_slave_unregister(bundle->bond, port);
1520 bundle_update(bundle);
1524 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1525 struct lacp_slave_settings *lacp,
1526 uint32_t bond_stable_id)
1528 struct ofport_dpif *port;
1530 port = get_ofp_port(bundle->ofproto, ofp_port);
1535 if (port->bundle != bundle) {
1536 bundle->ofproto->need_revalidate = true;
1538 bundle_del_port(port);
1541 port->bundle = bundle;
1542 list_push_back(&bundle->ports, &port->bundle_node);
1543 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
1544 bundle->floodable = false;
1548 port->bundle->ofproto->need_revalidate = true;
1549 lacp_slave_register(bundle->lacp, port, lacp);
1552 port->bond_stable_id = bond_stable_id;
1558 bundle_destroy(struct ofbundle *bundle)
1560 struct ofproto_dpif *ofproto;
1561 struct ofport_dpif *port, *next_port;
1568 ofproto = bundle->ofproto;
1569 for (i = 0; i < MAX_MIRRORS; i++) {
1570 struct ofmirror *m = ofproto->mirrors[i];
1572 if (m->out == bundle) {
1574 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1575 || hmapx_find_and_delete(&m->dsts, bundle)) {
1576 ofproto->need_revalidate = true;
1581 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1582 bundle_del_port(port);
1585 bundle_flush_macs(bundle, true);
1586 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1588 free(bundle->trunks);
1589 lacp_destroy(bundle->lacp);
1590 bond_destroy(bundle->bond);
1595 bundle_set(struct ofproto *ofproto_, void *aux,
1596 const struct ofproto_bundle_settings *s)
1598 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1599 bool need_flush = false;
1600 struct ofport_dpif *port;
1601 struct ofbundle *bundle;
1602 unsigned long *trunks;
1608 bundle_destroy(bundle_lookup(ofproto, aux));
1612 assert(s->n_slaves == 1 || s->bond != NULL);
1613 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1615 bundle = bundle_lookup(ofproto, aux);
1617 bundle = xmalloc(sizeof *bundle);
1619 bundle->ofproto = ofproto;
1620 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1621 hash_pointer(aux, 0));
1623 bundle->name = NULL;
1625 list_init(&bundle->ports);
1626 bundle->vlan_mode = PORT_VLAN_TRUNK;
1628 bundle->trunks = NULL;
1629 bundle->use_priority_tags = s->use_priority_tags;
1630 bundle->lacp = NULL;
1631 bundle->bond = NULL;
1633 bundle->floodable = true;
1635 bundle->src_mirrors = 0;
1636 bundle->dst_mirrors = 0;
1637 bundle->mirror_out = 0;
1640 if (!bundle->name || strcmp(s->name, bundle->name)) {
1642 bundle->name = xstrdup(s->name);
1647 if (!bundle->lacp) {
1648 ofproto->need_revalidate = true;
1649 bundle->lacp = lacp_create();
1651 lacp_configure(bundle->lacp, s->lacp);
1653 lacp_destroy(bundle->lacp);
1654 bundle->lacp = NULL;
1657 /* Update set of ports. */
1659 for (i = 0; i < s->n_slaves; i++) {
1660 if (!bundle_add_port(bundle, s->slaves[i],
1661 s->lacp ? &s->lacp_slaves[i] : NULL,
1662 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1666 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1667 struct ofport_dpif *next_port;
1669 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1670 for (i = 0; i < s->n_slaves; i++) {
1671 if (s->slaves[i] == port->up.ofp_port) {
1676 bundle_del_port(port);
1680 assert(list_size(&bundle->ports) <= s->n_slaves);
1682 if (list_is_empty(&bundle->ports)) {
1683 bundle_destroy(bundle);
1687 /* Set VLAN tagging mode */
1688 if (s->vlan_mode != bundle->vlan_mode
1689 || s->use_priority_tags != bundle->use_priority_tags) {
1690 bundle->vlan_mode = s->vlan_mode;
1691 bundle->use_priority_tags = s->use_priority_tags;
1696 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1697 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1699 if (vlan != bundle->vlan) {
1700 bundle->vlan = vlan;
1704 /* Get trunked VLANs. */
1705 switch (s->vlan_mode) {
1706 case PORT_VLAN_ACCESS:
1710 case PORT_VLAN_TRUNK:
1711 trunks = (unsigned long *) s->trunks;
1714 case PORT_VLAN_NATIVE_UNTAGGED:
1715 case PORT_VLAN_NATIVE_TAGGED:
1716 if (vlan != 0 && (!s->trunks
1717 || !bitmap_is_set(s->trunks, vlan)
1718 || bitmap_is_set(s->trunks, 0))) {
1719 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1721 trunks = bitmap_clone(s->trunks, 4096);
1723 trunks = bitmap_allocate1(4096);
1725 bitmap_set1(trunks, vlan);
1726 bitmap_set0(trunks, 0);
1728 trunks = (unsigned long *) s->trunks;
1735 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1736 free(bundle->trunks);
1737 if (trunks == s->trunks) {
1738 bundle->trunks = vlan_bitmap_clone(trunks);
1740 bundle->trunks = trunks;
1745 if (trunks != s->trunks) {
1750 if (!list_is_short(&bundle->ports)) {
1751 bundle->ofproto->has_bonded_bundles = true;
1753 if (bond_reconfigure(bundle->bond, s->bond)) {
1754 ofproto->need_revalidate = true;
1757 bundle->bond = bond_create(s->bond);
1758 ofproto->need_revalidate = true;
1761 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1762 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1766 bond_destroy(bundle->bond);
1767 bundle->bond = NULL;
1770 /* If we changed something that would affect MAC learning, un-learn
1771 * everything on this port and force flow revalidation. */
1773 bundle_flush_macs(bundle, false);
1780 bundle_remove(struct ofport *port_)
1782 struct ofport_dpif *port = ofport_dpif_cast(port_);
1783 struct ofbundle *bundle = port->bundle;
1786 bundle_del_port(port);
1787 if (list_is_empty(&bundle->ports)) {
1788 bundle_destroy(bundle);
1789 } else if (list_is_short(&bundle->ports)) {
1790 bond_destroy(bundle->bond);
1791 bundle->bond = NULL;
1797 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
1799 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1800 struct ofport_dpif *port = port_;
1801 uint8_t ea[ETH_ADDR_LEN];
1804 error = netdev_get_etheraddr(port->up.netdev, ea);
1806 struct ofpbuf packet;
1809 ofpbuf_init(&packet, 0);
1810 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1812 memcpy(packet_pdu, pdu, pdu_size);
1814 send_packet(port, &packet);
1815 ofpbuf_uninit(&packet);
1817 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1818 "%s (%s)", port->bundle->name,
1819 netdev_get_name(port->up.netdev), strerror(error));
1824 bundle_send_learning_packets(struct ofbundle *bundle)
1826 struct ofproto_dpif *ofproto = bundle->ofproto;
1827 int error, n_packets, n_errors;
1828 struct mac_entry *e;
1830 error = n_packets = n_errors = 0;
1831 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1832 if (e->port.p != bundle) {
1833 struct ofpbuf *learning_packet;
1834 struct ofport_dpif *port;
1838 /* The assignment to "port" is unnecessary but makes "grep"ing for
1839 * struct ofport_dpif more effective. */
1840 learning_packet = bond_compose_learning_packet(bundle->bond,
1844 ret = send_packet(port, learning_packet);
1845 ofpbuf_delete(learning_packet);
1855 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1856 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1857 "packets, last error was: %s",
1858 bundle->name, n_errors, n_packets, strerror(error));
1860 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1861 bundle->name, n_packets);
1866 bundle_run(struct ofbundle *bundle)
1869 lacp_run(bundle->lacp, send_pdu_cb);
1872 struct ofport_dpif *port;
1874 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1875 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1878 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1879 lacp_negotiated(bundle->lacp));
1880 if (bond_should_send_learning_packets(bundle->bond)) {
1881 bundle_send_learning_packets(bundle);
1887 bundle_wait(struct ofbundle *bundle)
1890 lacp_wait(bundle->lacp);
1893 bond_wait(bundle->bond);
1900 mirror_scan(struct ofproto_dpif *ofproto)
1904 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1905 if (!ofproto->mirrors[idx]) {
1912 static struct ofmirror *
1913 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1917 for (i = 0; i < MAX_MIRRORS; i++) {
1918 struct ofmirror *mirror = ofproto->mirrors[i];
1919 if (mirror && mirror->aux == aux) {
1927 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
1929 mirror_update_dups(struct ofproto_dpif *ofproto)
1933 for (i = 0; i < MAX_MIRRORS; i++) {
1934 struct ofmirror *m = ofproto->mirrors[i];
1937 m->dup_mirrors = MIRROR_MASK_C(1) << i;
1941 for (i = 0; i < MAX_MIRRORS; i++) {
1942 struct ofmirror *m1 = ofproto->mirrors[i];
1949 for (j = i + 1; j < MAX_MIRRORS; j++) {
1950 struct ofmirror *m2 = ofproto->mirrors[j];
1952 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
1953 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
1954 m2->dup_mirrors |= m1->dup_mirrors;
1961 mirror_set(struct ofproto *ofproto_, void *aux,
1962 const struct ofproto_mirror_settings *s)
1964 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1965 mirror_mask_t mirror_bit;
1966 struct ofbundle *bundle;
1967 struct ofmirror *mirror;
1968 struct ofbundle *out;
1969 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
1970 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
1973 mirror = mirror_lookup(ofproto, aux);
1975 mirror_destroy(mirror);
1981 idx = mirror_scan(ofproto);
1983 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
1985 ofproto->up.name, MAX_MIRRORS, s->name);
1989 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
1990 mirror->ofproto = ofproto;
1993 mirror->out_vlan = -1;
1994 mirror->name = NULL;
1997 if (!mirror->name || strcmp(s->name, mirror->name)) {
1999 mirror->name = xstrdup(s->name);
2002 /* Get the new configuration. */
2003 if (s->out_bundle) {
2004 out = bundle_lookup(ofproto, s->out_bundle);
2006 mirror_destroy(mirror);
2012 out_vlan = s->out_vlan;
2014 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2015 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2017 /* If the configuration has not changed, do nothing. */
2018 if (hmapx_equals(&srcs, &mirror->srcs)
2019 && hmapx_equals(&dsts, &mirror->dsts)
2020 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2021 && mirror->out == out
2022 && mirror->out_vlan == out_vlan)
2024 hmapx_destroy(&srcs);
2025 hmapx_destroy(&dsts);
2029 hmapx_swap(&srcs, &mirror->srcs);
2030 hmapx_destroy(&srcs);
2032 hmapx_swap(&dsts, &mirror->dsts);
2033 hmapx_destroy(&dsts);
2035 free(mirror->vlans);
2036 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2039 mirror->out_vlan = out_vlan;
2041 /* Update bundles. */
2042 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2043 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2044 if (hmapx_contains(&mirror->srcs, bundle)) {
2045 bundle->src_mirrors |= mirror_bit;
2047 bundle->src_mirrors &= ~mirror_bit;
2050 if (hmapx_contains(&mirror->dsts, bundle)) {
2051 bundle->dst_mirrors |= mirror_bit;
2053 bundle->dst_mirrors &= ~mirror_bit;
2056 if (mirror->out == bundle) {
2057 bundle->mirror_out |= mirror_bit;
2059 bundle->mirror_out &= ~mirror_bit;
2063 ofproto->need_revalidate = true;
2064 mac_learning_flush(ofproto->ml);
2065 mirror_update_dups(ofproto);
2071 mirror_destroy(struct ofmirror *mirror)
2073 struct ofproto_dpif *ofproto;
2074 mirror_mask_t mirror_bit;
2075 struct ofbundle *bundle;
2081 ofproto = mirror->ofproto;
2082 ofproto->need_revalidate = true;
2083 mac_learning_flush(ofproto->ml);
2085 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2086 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2087 bundle->src_mirrors &= ~mirror_bit;
2088 bundle->dst_mirrors &= ~mirror_bit;
2089 bundle->mirror_out &= ~mirror_bit;
2092 hmapx_destroy(&mirror->srcs);
2093 hmapx_destroy(&mirror->dsts);
2094 free(mirror->vlans);
2096 ofproto->mirrors[mirror->idx] = NULL;
2100 mirror_update_dups(ofproto);
2104 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2105 uint64_t *packets, uint64_t *bytes)
2107 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2108 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2111 *packets = *bytes = UINT64_MAX;
2115 *packets = mirror->packet_count;
2116 *bytes = mirror->byte_count;
2122 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2124 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2125 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2126 ofproto->need_revalidate = true;
2127 mac_learning_flush(ofproto->ml);
2133 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2135 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2136 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2137 return bundle && bundle->mirror_out != 0;
2141 forward_bpdu_changed(struct ofproto *ofproto_)
2143 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2144 /* Revalidate cached flows whenever forward_bpdu option changes. */
2145 ofproto->need_revalidate = true;
2149 set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time)
2151 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2152 mac_learning_set_idle_time(ofproto->ml, idle_time);
2157 static struct ofport_dpif *
2158 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
2160 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2161 return ofport ? ofport_dpif_cast(ofport) : NULL;
2164 static struct ofport_dpif *
2165 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
2167 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2171 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2172 struct dpif_port *dpif_port)
2174 ofproto_port->name = dpif_port->name;
2175 ofproto_port->type = dpif_port->type;
2176 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2180 port_run(struct ofport_dpif *ofport)
2182 bool enable = netdev_get_carrier(ofport->up.netdev);
2185 cfm_run(ofport->cfm);
2187 if (cfm_should_send_ccm(ofport->cfm)) {
2188 struct ofpbuf packet;
2190 ofpbuf_init(&packet, 0);
2191 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr);
2192 send_packet(ofport, &packet);
2193 ofpbuf_uninit(&packet);
2196 enable = enable && !cfm_get_fault(ofport->cfm)
2197 && cfm_get_opup(ofport->cfm);
2200 if (ofport->bundle) {
2201 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2204 if (ofport->may_enable != enable) {
2205 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2207 if (ofproto->has_bundle_action) {
2208 ofproto->need_revalidate = true;
2212 ofport->may_enable = enable;
2216 port_wait(struct ofport_dpif *ofport)
2219 cfm_wait(ofport->cfm);
2224 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2225 struct ofproto_port *ofproto_port)
2227 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2228 struct dpif_port dpif_port;
2231 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2233 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2239 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2241 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2245 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2247 *ofp_portp = odp_port_to_ofp_port(odp_port);
2253 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2255 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2258 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2260 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2262 /* The caller is going to close ofport->up.netdev. If this is a
2263 * bonded port, then the bond is using that netdev, so remove it
2264 * from the bond. The client will need to reconfigure everything
2265 * after deleting ports, so then the slave will get re-added. */
2266 bundle_remove(&ofport->up);
2273 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2275 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2278 error = netdev_get_stats(ofport->up.netdev, stats);
2280 if (!error && ofport->odp_port == OVSP_LOCAL) {
2281 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2283 /* ofproto->stats.tx_packets represents packets that we created
2284 * internally and sent to some port (e.g. packets sent with
2285 * send_packet()). Account for them as if they had come from
2286 * OFPP_LOCAL and got forwarded. */
2288 if (stats->rx_packets != UINT64_MAX) {
2289 stats->rx_packets += ofproto->stats.tx_packets;
2292 if (stats->rx_bytes != UINT64_MAX) {
2293 stats->rx_bytes += ofproto->stats.tx_bytes;
2296 /* ofproto->stats.rx_packets represents packets that were received on
2297 * some port and we processed internally and dropped (e.g. STP).
2298 * Account fro them as if they had been forwarded to OFPP_LOCAL. */
2300 if (stats->tx_packets != UINT64_MAX) {
2301 stats->tx_packets += ofproto->stats.rx_packets;
2304 if (stats->tx_bytes != UINT64_MAX) {
2305 stats->tx_bytes += ofproto->stats.rx_bytes;
2312 /* Account packets for LOCAL port. */
2314 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2315 size_t tx_size, size_t rx_size)
2317 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2320 ofproto->stats.rx_packets++;
2321 ofproto->stats.rx_bytes += rx_size;
2324 ofproto->stats.tx_packets++;
2325 ofproto->stats.tx_bytes += tx_size;
2329 struct port_dump_state {
2330 struct dpif_port_dump dump;
2335 port_dump_start(const struct ofproto *ofproto_, void **statep)
2337 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2338 struct port_dump_state *state;
2340 *statep = state = xmalloc(sizeof *state);
2341 dpif_port_dump_start(&state->dump, ofproto->dpif);
2342 state->done = false;
2347 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2348 struct ofproto_port *port)
2350 struct port_dump_state *state = state_;
2351 struct dpif_port dpif_port;
2353 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2354 ofproto_port_from_dpif_port(port, &dpif_port);
2357 int error = dpif_port_dump_done(&state->dump);
2359 return error ? error : EOF;
2364 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2366 struct port_dump_state *state = state_;
2369 dpif_port_dump_done(&state->dump);
2376 port_poll(const struct ofproto *ofproto_, char **devnamep)
2378 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2379 return dpif_port_poll(ofproto->dpif, devnamep);
2383 port_poll_wait(const struct ofproto *ofproto_)
2385 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2386 dpif_port_poll_wait(ofproto->dpif);
2390 port_is_lacp_current(const struct ofport *ofport_)
2392 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2393 return (ofport->bundle && ofport->bundle->lacp
2394 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2398 /* Upcall handling. */
2400 /* Flow miss batching.
2402 * Some dpifs implement operations faster when you hand them off in a batch.
2403 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2404 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2405 * more packets, plus possibly installing the flow in the dpif.
2407 * So far we only batch the operations that affect flow setup time the most.
2408 * It's possible to batch more than that, but the benefit might be minimal. */
2410 struct hmap_node hmap_node;
2412 enum odp_key_fitness key_fitness;
2413 const struct nlattr *key;
2415 ovs_be16 initial_tci;
2416 struct list packets;
2419 struct flow_miss_op {
2420 union dpif_op dpif_op;
2421 struct subfacet *subfacet;
2424 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2425 * OpenFlow controller as necessary according to their individual
2426 * configurations. */
2428 send_packet_in_miss(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
2429 const struct flow *flow)
2431 struct ofputil_packet_in pin;
2433 pin.packet = packet->data;
2434 pin.packet_len = packet->size;
2435 pin.total_len = packet->size;
2436 pin.reason = OFPR_NO_MATCH;
2441 pin.buffer_id = 0; /* not yet known */
2442 pin.send_len = 0; /* not used for flow table misses */
2444 flow_get_metadata(flow, &pin.fmd);
2446 /* Registers aren't meaningful on a miss. */
2447 memset(pin.fmd.reg_masks, 0, sizeof pin.fmd.reg_masks);
2449 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow);
2453 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2454 const struct ofpbuf *packet)
2456 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2462 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2464 cfm_process_heartbeat(ofport->cfm, packet);
2467 } else if (ofport->bundle && ofport->bundle->lacp
2468 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2470 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2473 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2475 stp_process_packet(ofport, packet);
2482 static struct flow_miss *
2483 flow_miss_create(struct hmap *todo, const struct flow *flow,
2484 enum odp_key_fitness key_fitness,
2485 const struct nlattr *key, size_t key_len,
2486 ovs_be16 initial_tci)
2488 uint32_t hash = flow_hash(flow, 0);
2489 struct flow_miss *miss;
2491 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2492 if (flow_equal(&miss->flow, flow)) {
2497 miss = xmalloc(sizeof *miss);
2498 hmap_insert(todo, &miss->hmap_node, hash);
2500 miss->key_fitness = key_fitness;
2502 miss->key_len = key_len;
2503 miss->initial_tci = initial_tci;
2504 list_init(&miss->packets);
2509 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2510 struct flow_miss_op *ops, size_t *n_ops)
2512 const struct flow *flow = &miss->flow;
2513 struct ofpbuf *packet, *next_packet;
2514 struct subfacet *subfacet;
2515 struct facet *facet;
2517 facet = facet_lookup_valid(ofproto, flow);
2519 struct rule_dpif *rule;
2521 rule = rule_dpif_lookup(ofproto, flow, 0);
2523 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
2524 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
2526 if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) {
2527 COVERAGE_INC(ofproto_dpif_no_packet_in);
2528 /* XXX install 'drop' flow entry */
2532 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
2536 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2537 send_packet_in_miss(ofproto, packet, flow);
2543 facet = facet_create(rule, flow);
2546 subfacet = subfacet_create(ofproto, facet,
2547 miss->key_fitness, miss->key, miss->key_len,
2550 LIST_FOR_EACH_SAFE (packet, next_packet, list_node, &miss->packets) {
2551 struct dpif_flow_stats stats;
2552 struct flow_miss_op *op;
2553 struct dpif_execute *execute;
2555 ofproto->n_matches++;
2557 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2559 * Extra-special case for fail-open mode.
2561 * We are in fail-open mode and the packet matched the fail-open
2562 * rule, but we are connected to a controller too. We should send
2563 * the packet up to the controller in the hope that it will try to
2564 * set up a flow and thereby allow us to exit fail-open.
2566 * See the top-level comment in fail-open.c for more information.
2568 send_packet_in_miss(ofproto, packet, flow);
2571 if (!facet->may_install || !subfacet->actions) {
2572 subfacet_make_actions(ofproto, subfacet, packet);
2575 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2576 subfacet_update_stats(ofproto, subfacet, &stats);
2578 if (flow->vlan_tci != subfacet->initial_tci) {
2579 /* This packet was received on a VLAN splinter port. We added
2580 * a VLAN to the packet to make the packet resemble the flow,
2581 * but the actions were composed assuming that the packet
2582 * contained no VLAN. So, we must remove the VLAN header from
2583 * the packet before trying to execute the actions. */
2584 eth_pop_vlan(packet);
2587 op = &ops[(*n_ops)++];
2588 execute = &op->dpif_op.execute;
2589 op->subfacet = subfacet;
2590 execute->type = DPIF_OP_EXECUTE;
2591 execute->key = miss->key;
2592 execute->key_len = miss->key_len;
2593 execute->actions = (facet->may_install
2595 : xmemdup(subfacet->actions,
2596 subfacet->actions_len));
2597 execute->actions_len = subfacet->actions_len;
2598 execute->packet = packet;
2601 if (facet->may_install && subfacet->key_fitness != ODP_FIT_TOO_LITTLE) {
2602 struct flow_miss_op *op = &ops[(*n_ops)++];
2603 struct dpif_flow_put *put = &op->dpif_op.flow_put;
2605 op->subfacet = subfacet;
2606 put->type = DPIF_OP_FLOW_PUT;
2607 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2608 put->key = miss->key;
2609 put->key_len = miss->key_len;
2610 put->actions = subfacet->actions;
2611 put->actions_len = subfacet->actions_len;
2616 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
2617 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
2618 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
2619 * what a flow key should contain.
2621 * This function also includes some logic to help make VLAN splinters
2622 * transparent to the rest of the upcall processing logic. In particular, if
2623 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
2624 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
2625 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
2627 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
2628 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
2629 * (This differs from the value returned in flow->vlan_tci only for packets
2630 * received on VLAN splinters.)
2632 static enum odp_key_fitness
2633 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
2634 const struct nlattr *key, size_t key_len,
2635 struct flow *flow, ovs_be16 *initial_tci,
2636 struct ofpbuf *packet)
2638 enum odp_key_fitness fitness;
2642 fitness = odp_flow_key_to_flow(key, key_len, flow);
2643 if (fitness == ODP_FIT_ERROR) {
2646 *initial_tci = flow->vlan_tci;
2648 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
2650 /* Cause the flow to be processed as if it came in on the real device
2651 * with the VLAN device's VLAN ID. */
2652 flow->in_port = realdev;
2653 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
2655 /* Make the packet resemble the flow, so that it gets sent to an
2656 * OpenFlow controller properly, so that it looks correct for
2657 * sFlow, and so that flow_extract() will get the correct vlan_tci
2658 * if it is called on 'packet'.
2660 * The allocated space inside 'packet' probably also contains
2661 * 'key', that is, both 'packet' and 'key' are probably part of a
2662 * struct dpif_upcall (see the large comment on that structure
2663 * definition), so pushing data on 'packet' is in general not a
2664 * good idea since it could overwrite 'key' or free it as a side
2665 * effect. However, it's OK in this special case because we know
2666 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
2667 * will just overwrite the 4-byte "struct nlattr", which is fine
2668 * since we don't need that header anymore. */
2669 eth_push_vlan(packet, flow->vlan_tci);
2672 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
2673 if (fitness == ODP_FIT_PERFECT) {
2674 fitness = ODP_FIT_TOO_MUCH;
2682 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
2685 struct dpif_upcall *upcall;
2686 struct flow_miss *miss, *next_miss;
2687 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
2688 union dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
2697 /* Construct the to-do list.
2699 * This just amounts to extracting the flow from each packet and sticking
2700 * the packets that have the same flow in the same "flow_miss" structure so
2701 * that we can process them together. */
2703 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
2704 enum odp_key_fitness fitness;
2705 struct flow_miss *miss;
2706 ovs_be16 initial_tci;
2709 /* Obtain metadata and check userspace/kernel agreement on flow match,
2710 * then set 'flow''s header pointers. */
2711 fitness = ofproto_dpif_extract_flow_key(ofproto,
2712 upcall->key, upcall->key_len,
2713 &flow, &initial_tci,
2715 if (fitness == ODP_FIT_ERROR) {
2716 ofpbuf_delete(upcall->packet);
2719 flow_extract(upcall->packet, flow.skb_priority, flow.tun_id,
2720 flow.in_port, &flow);
2722 /* Handle 802.1ag, LACP, and STP specially. */
2723 if (process_special(ofproto, &flow, upcall->packet)) {
2724 ofproto_update_local_port_stats(&ofproto->up,
2725 0, upcall->packet->size);
2726 ofpbuf_delete(upcall->packet);
2727 ofproto->n_matches++;
2731 /* Add other packets to a to-do list. */
2732 miss = flow_miss_create(&todo, &flow, fitness,
2733 upcall->key, upcall->key_len, initial_tci);
2734 list_push_back(&miss->packets, &upcall->packet->list_node);
2737 /* Process each element in the to-do list, constructing the set of
2738 * operations to batch. */
2740 HMAP_FOR_EACH (miss, hmap_node, &todo) {
2741 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
2743 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
2745 /* Execute batch. */
2746 for (i = 0; i < n_ops; i++) {
2747 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
2749 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
2751 /* Free memory and update facets. */
2752 for (i = 0; i < n_ops; i++) {
2753 struct flow_miss_op *op = &flow_miss_ops[i];
2754 struct dpif_execute *execute;
2755 struct dpif_flow_put *put;
2757 switch (op->dpif_op.type) {
2758 case DPIF_OP_EXECUTE:
2759 execute = &op->dpif_op.execute;
2760 if (op->subfacet->actions != execute->actions) {
2761 free((struct nlattr *) execute->actions);
2765 case DPIF_OP_FLOW_PUT:
2766 put = &op->dpif_op.flow_put;
2768 op->subfacet->installed = true;
2773 HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) {
2774 ofpbuf_list_delete(&miss->packets);
2775 hmap_remove(&todo, &miss->hmap_node);
2778 hmap_destroy(&todo);
2782 handle_userspace_upcall(struct ofproto_dpif *ofproto,
2783 struct dpif_upcall *upcall)
2785 struct user_action_cookie cookie;
2786 enum odp_key_fitness fitness;
2787 ovs_be16 initial_tci;
2790 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
2792 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
2793 upcall->key_len, &flow,
2794 &initial_tci, upcall->packet);
2795 if (fitness == ODP_FIT_ERROR) {
2796 ofpbuf_delete(upcall->packet);
2800 if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
2801 if (ofproto->sflow) {
2802 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
2806 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
2808 ofpbuf_delete(upcall->packet);
2812 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
2814 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
2818 assert (max_batch <= FLOW_MISS_MAX_BATCH);
2821 for (i = 0; i < max_batch; i++) {
2822 struct dpif_upcall *upcall = &misses[n_misses];
2825 error = dpif_recv(ofproto->dpif, upcall);
2830 switch (upcall->type) {
2831 case DPIF_UC_ACTION:
2832 handle_userspace_upcall(ofproto, upcall);
2836 /* Handle it later. */
2840 case DPIF_N_UC_TYPES:
2842 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32,
2848 handle_miss_upcalls(ofproto, misses, n_misses);
2853 /* Flow expiration. */
2855 static int subfacet_max_idle(const struct ofproto_dpif *);
2856 static void update_stats(struct ofproto_dpif *);
2857 static void rule_expire(struct rule_dpif *);
2858 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
2860 /* This function is called periodically by run(). Its job is to collect
2861 * updates for the flows that have been installed into the datapath, most
2862 * importantly when they last were used, and then use that information to
2863 * expire flows that have not been used recently.
2865 * Returns the number of milliseconds after which it should be called again. */
2867 expire(struct ofproto_dpif *ofproto)
2869 struct rule_dpif *rule, *next_rule;
2870 struct classifier *table;
2873 /* Update stats for each flow in the datapath. */
2874 update_stats(ofproto);
2876 /* Expire subfacets that have been idle too long. */
2877 dp_max_idle = subfacet_max_idle(ofproto);
2878 expire_subfacets(ofproto, dp_max_idle);
2880 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
2881 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
2882 struct cls_cursor cursor;
2884 cls_cursor_init(&cursor, table, NULL);
2885 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
2890 /* All outstanding data in existing flows has been accounted, so it's a
2891 * good time to do bond rebalancing. */
2892 if (ofproto->has_bonded_bundles) {
2893 struct ofbundle *bundle;
2895 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2897 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
2902 return MIN(dp_max_idle, 1000);
2905 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
2907 * This function also pushes statistics updates to rules which each facet
2908 * resubmits into. Generally these statistics will be accurate. However, if a
2909 * facet changes the rule it resubmits into at some time in between
2910 * update_stats() runs, it is possible that statistics accrued to the
2911 * old rule will be incorrectly attributed to the new rule. This could be
2912 * avoided by calling update_stats() whenever rules are created or
2913 * deleted. However, the performance impact of making so many calls to the
2914 * datapath do not justify the benefit of having perfectly accurate statistics.
2917 update_stats(struct ofproto_dpif *p)
2919 const struct dpif_flow_stats *stats;
2920 struct dpif_flow_dump dump;
2921 const struct nlattr *key;
2924 dpif_flow_dump_start(&dump, p->dpif);
2925 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
2926 struct subfacet *subfacet;
2928 subfacet = subfacet_find(p, key, key_len);
2929 if (subfacet && subfacet->installed) {
2930 struct facet *facet = subfacet->facet;
2932 if (stats->n_packets >= subfacet->dp_packet_count) {
2933 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
2934 facet->packet_count += extra;
2936 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
2939 if (stats->n_bytes >= subfacet->dp_byte_count) {
2940 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
2942 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
2945 subfacet->dp_packet_count = stats->n_packets;
2946 subfacet->dp_byte_count = stats->n_bytes;
2948 subfacet_update_time(p, subfacet, stats->used);
2949 facet_account(p, facet);
2950 facet_push_stats(facet);
2952 if (!VLOG_DROP_WARN(&rl)) {
2956 odp_flow_key_format(key, key_len, &s);
2957 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
2961 COVERAGE_INC(facet_unexpected);
2962 /* There's a flow in the datapath that we know nothing about, or a
2963 * flow that shouldn't be installed but was anyway. Delete it. */
2964 dpif_flow_del(p->dpif, key, key_len, NULL);
2967 dpif_flow_dump_done(&dump);
2970 /* Calculates and returns the number of milliseconds of idle time after which
2971 * subfacets should expire from the datapath. When a subfacet expires, we fold
2972 * its statistics into its facet, and when a facet's last subfacet expires, we
2973 * fold its statistic into its rule. */
2975 subfacet_max_idle(const struct ofproto_dpif *ofproto)
2978 * Idle time histogram.
2980 * Most of the time a switch has a relatively small number of subfacets.
2981 * When this is the case we might as well keep statistics for all of them
2982 * in userspace and to cache them in the kernel datapath for performance as
2985 * As the number of subfacets increases, the memory required to maintain
2986 * statistics about them in userspace and in the kernel becomes
2987 * significant. However, with a large number of subfacets it is likely
2988 * that only a few of them are "heavy hitters" that consume a large amount
2989 * of bandwidth. At this point, only heavy hitters are worth caching in
2990 * the kernel and maintaining in userspaces; other subfacets we can
2993 * The technique used to compute the idle time is to build a histogram with
2994 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
2995 * that is installed in the kernel gets dropped in the appropriate bucket.
2996 * After the histogram has been built, we compute the cutoff so that only
2997 * the most-recently-used 1% of subfacets (but at least
2998 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
2999 * the most-recently-used bucket of subfacets is kept, so actually an
3000 * arbitrary number of subfacets can be kept in any given expiration run
3001 * (though the next run will delete most of those unless they receive
3004 * This requires a second pass through the subfacets, in addition to the
3005 * pass made by update_stats(), because the former function never looks at
3006 * uninstallable subfacets.
3008 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3009 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3010 int buckets[N_BUCKETS] = { 0 };
3011 int total, subtotal, bucket;
3012 struct subfacet *subfacet;
3016 total = hmap_count(&ofproto->subfacets);
3017 if (total <= ofproto->up.flow_eviction_threshold) {
3018 return N_BUCKETS * BUCKET_WIDTH;
3021 /* Build histogram. */
3023 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3024 long long int idle = now - subfacet->used;
3025 int bucket = (idle <= 0 ? 0
3026 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3027 : (unsigned int) idle / BUCKET_WIDTH);
3031 /* Find the first bucket whose flows should be expired. */
3032 subtotal = bucket = 0;
3034 subtotal += buckets[bucket++];
3035 } while (bucket < N_BUCKETS &&
3036 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3038 if (VLOG_IS_DBG_ENABLED()) {
3042 ds_put_cstr(&s, "keep");
3043 for (i = 0; i < N_BUCKETS; i++) {
3045 ds_put_cstr(&s, ", drop");
3048 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3051 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3055 return bucket * BUCKET_WIDTH;
3059 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3061 long long int cutoff = time_msec() - dp_max_idle;
3062 struct subfacet *subfacet, *next_subfacet;
3064 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3065 &ofproto->subfacets) {
3066 if (subfacet->used < cutoff) {
3067 subfacet_destroy(ofproto, subfacet);
3072 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3073 * then delete it entirely. */
3075 rule_expire(struct rule_dpif *rule)
3077 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3078 struct facet *facet, *next_facet;
3082 /* Has 'rule' expired? */
3084 if (rule->up.hard_timeout
3085 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3086 reason = OFPRR_HARD_TIMEOUT;
3087 } else if (rule->up.idle_timeout && list_is_empty(&rule->facets)
3088 && now > rule->used + rule->up.idle_timeout * 1000) {
3089 reason = OFPRR_IDLE_TIMEOUT;
3094 COVERAGE_INC(ofproto_dpif_expired);
3096 /* Update stats. (This is a no-op if the rule expired due to an idle
3097 * timeout, because that only happens when the rule has no facets left.) */
3098 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3099 facet_remove(ofproto, facet);
3102 /* Get rid of the rule. */
3103 ofproto_rule_expire(&rule->up, reason);
3108 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3110 * The caller must already have determined that no facet with an identical
3111 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3112 * the ofproto's classifier table.
3114 * The facet will initially have no subfacets. The caller should create (at
3115 * least) one subfacet with subfacet_create(). */
3116 static struct facet *
3117 facet_create(struct rule_dpif *rule, const struct flow *flow)
3119 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3120 struct facet *facet;
3122 facet = xzalloc(sizeof *facet);
3123 facet->used = time_msec();
3124 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
3125 list_push_back(&rule->facets, &facet->list_node);
3127 facet->flow = *flow;
3128 list_init(&facet->subfacets);
3129 netflow_flow_init(&facet->nf_flow);
3130 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3136 facet_free(struct facet *facet)
3141 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3142 * 'packet', which arrived on 'in_port'.
3144 * Takes ownership of 'packet'. */
3146 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3147 const struct nlattr *odp_actions, size_t actions_len,
3148 struct ofpbuf *packet)
3150 struct odputil_keybuf keybuf;
3154 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3155 odp_flow_key_from_flow(&key, flow);
3157 error = dpif_execute(ofproto->dpif, key.data, key.size,
3158 odp_actions, actions_len, packet);
3160 ofpbuf_delete(packet);
3164 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3166 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3167 * rule's statistics, via subfacet_uninstall().
3169 * - Removes 'facet' from its rule and from ofproto->facets.
3172 facet_remove(struct ofproto_dpif *ofproto, struct facet *facet)
3174 struct subfacet *subfacet, *next_subfacet;
3176 assert(!list_is_empty(&facet->subfacets));
3178 /* First uninstall all of the subfacets to get final statistics. */
3179 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3180 subfacet_uninstall(ofproto, subfacet);
3183 /* Flush the final stats to the rule.
3185 * This might require us to have at least one subfacet around so that we
3186 * can use its actions for accounting in facet_account(), which is why we
3187 * have uninstalled but not yet destroyed the subfacets. */
3188 facet_flush_stats(ofproto, facet);
3190 /* Now we're really all done so destroy everything. */
3191 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3192 &facet->subfacets) {
3193 subfacet_destroy__(ofproto, subfacet);
3195 hmap_remove(&ofproto->facets, &facet->hmap_node);
3196 list_remove(&facet->list_node);
3201 facet_account(struct ofproto_dpif *ofproto, struct facet *facet)
3204 struct subfacet *subfacet;
3205 const struct nlattr *a;
3209 if (facet->byte_count <= facet->accounted_bytes) {
3212 n_bytes = facet->byte_count - facet->accounted_bytes;
3213 facet->accounted_bytes = facet->byte_count;
3215 /* Feed information from the active flows back into the learning table to
3216 * ensure that table is always in sync with what is actually flowing
3217 * through the datapath. */
3218 if (facet->has_learn || facet->has_normal) {
3219 struct action_xlate_ctx ctx;
3221 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3222 facet->flow.vlan_tci,
3223 facet->rule->up.flow_cookie, NULL);
3224 ctx.may_learn = true;
3225 ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions,
3226 facet->rule->up.n_actions));
3229 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3233 /* This loop feeds byte counters to bond_account() for rebalancing to use
3234 * as a basis. We also need to track the actual VLAN on which the packet
3235 * is going to be sent to ensure that it matches the one passed to
3236 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3239 * We use the actions from an arbitrary subfacet because they should all
3240 * be equally valid for our purpose. */
3241 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3242 struct subfacet, list_node);
3243 vlan_tci = facet->flow.vlan_tci;
3244 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3245 subfacet->actions, subfacet->actions_len) {
3246 const struct ovs_action_push_vlan *vlan;
3247 struct ofport_dpif *port;
3249 switch (nl_attr_type(a)) {
3250 case OVS_ACTION_ATTR_OUTPUT:
3251 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3252 if (port && port->bundle && port->bundle->bond) {
3253 bond_account(port->bundle->bond, &facet->flow,
3254 vlan_tci_to_vid(vlan_tci), n_bytes);
3258 case OVS_ACTION_ATTR_POP_VLAN:
3259 vlan_tci = htons(0);
3262 case OVS_ACTION_ATTR_PUSH_VLAN:
3263 vlan = nl_attr_get(a);
3264 vlan_tci = vlan->vlan_tci;
3270 /* Returns true if the only action for 'facet' is to send to the controller.
3271 * (We don't report NetFlow expiration messages for such facets because they
3272 * are just part of the control logic for the network, not real traffic). */
3274 facet_is_controller_flow(struct facet *facet)
3277 && facet->rule->up.n_actions == 1
3278 && action_outputs_to_port(&facet->rule->up.actions[0],
3279 htons(OFPP_CONTROLLER)));
3282 /* Folds all of 'facet''s statistics into its rule. Also updates the
3283 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3284 * 'facet''s statistics in the datapath should have been zeroed and folded into
3285 * its packet and byte counts before this function is called. */
3287 facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet)
3289 struct subfacet *subfacet;
3291 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3292 assert(!subfacet->dp_byte_count);
3293 assert(!subfacet->dp_packet_count);
3296 facet_push_stats(facet);
3297 facet_account(ofproto, facet);
3299 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3300 struct ofexpired expired;
3301 expired.flow = facet->flow;
3302 expired.packet_count = facet->packet_count;
3303 expired.byte_count = facet->byte_count;
3304 expired.used = facet->used;
3305 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3308 facet->rule->packet_count += facet->packet_count;
3309 facet->rule->byte_count += facet->byte_count;
3311 /* Reset counters to prevent double counting if 'facet' ever gets
3313 facet_reset_counters(facet);
3315 netflow_flow_clear(&facet->nf_flow);
3318 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3319 * Returns it if found, otherwise a null pointer.
3321 * The returned facet might need revalidation; use facet_lookup_valid()
3322 * instead if that is important. */
3323 static struct facet *
3324 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
3326 struct facet *facet;
3328 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
3330 if (flow_equal(flow, &facet->flow)) {
3338 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3339 * Returns it if found, otherwise a null pointer.
3341 * The returned facet is guaranteed to be valid. */
3342 static struct facet *
3343 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
3345 struct facet *facet = facet_find(ofproto, flow);
3347 /* The facet we found might not be valid, since we could be in need of
3348 * revalidation. If it is not valid, don't return it. */
3350 && (ofproto->need_revalidate
3351 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))
3352 && !facet_revalidate(ofproto, facet)) {
3353 COVERAGE_INC(facet_invalidated);
3360 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
3362 * - If the rule found is different from 'facet''s current rule, moves
3363 * 'facet' to the new rule and recompiles its actions.
3365 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3366 * where it is and recompiles its actions anyway.
3368 * - If there is none, destroys 'facet'.
3370 * Returns true if 'facet' still exists, false if it has been destroyed. */
3372 facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet)
3375 struct nlattr *odp_actions;
3378 struct actions *new_actions;
3380 struct action_xlate_ctx ctx;
3381 struct rule_dpif *new_rule;
3382 struct subfacet *subfacet;
3383 bool actions_changed;
3386 COVERAGE_INC(facet_revalidate);
3388 /* Determine the new rule. */
3389 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3391 /* No new rule, so delete the facet. */
3392 facet_remove(ofproto, facet);
3396 /* Calculate new datapath actions.
3398 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3399 * emit a NetFlow expiration and, if so, we need to have the old state
3400 * around to properly compose it. */
3402 /* If the datapath actions changed or the installability changed,
3403 * then we need to talk to the datapath. */
3406 memset(&ctx, 0, sizeof ctx);
3407 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3408 struct ofpbuf *odp_actions;
3409 bool should_install;
3411 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3412 subfacet->initial_tci, new_rule->up.flow_cookie,
3414 odp_actions = xlate_actions(&ctx, new_rule->up.actions,
3415 new_rule->up.n_actions);
3416 actions_changed = (subfacet->actions_len != odp_actions->size
3417 || memcmp(subfacet->actions, odp_actions->data,
3418 subfacet->actions_len));
3420 should_install = (ctx.may_set_up_flow
3421 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3422 if (actions_changed || should_install != subfacet->installed) {
3423 if (should_install) {
3424 struct dpif_flow_stats stats;
3426 subfacet_install(ofproto, subfacet,
3427 odp_actions->data, odp_actions->size, &stats);
3428 subfacet_update_stats(ofproto, subfacet, &stats);
3430 subfacet_uninstall(ofproto, subfacet);
3434 new_actions = xcalloc(list_size(&facet->subfacets),
3435 sizeof *new_actions);
3437 new_actions[i].odp_actions = xmemdup(odp_actions->data,
3439 new_actions[i].actions_len = odp_actions->size;
3442 ofpbuf_delete(odp_actions);
3446 facet_flush_stats(ofproto, facet);
3449 /* Update 'facet' now that we've taken care of all the old state. */
3450 facet->tags = ctx.tags;
3451 facet->nf_flow.output_iface = ctx.nf_output_iface;
3452 facet->may_install = ctx.may_set_up_flow;
3453 facet->has_learn = ctx.has_learn;
3454 facet->has_normal = ctx.has_normal;
3455 facet->mirrors = ctx.mirrors;
3458 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3459 if (new_actions[i].odp_actions) {
3460 free(subfacet->actions);
3461 subfacet->actions = new_actions[i].odp_actions;
3462 subfacet->actions_len = new_actions[i].actions_len;
3468 if (facet->rule != new_rule) {
3469 COVERAGE_INC(facet_changed_rule);
3470 list_remove(&facet->list_node);
3471 list_push_back(&new_rule->facets, &facet->list_node);
3472 facet->rule = new_rule;
3473 facet->used = new_rule->up.created;
3474 facet->prev_used = facet->used;
3480 /* Updates 'facet''s used time. Caller is responsible for calling
3481 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3483 facet_update_time(struct ofproto_dpif *ofproto, struct facet *facet,
3486 if (used > facet->used) {
3488 if (used > facet->rule->used) {
3489 facet->rule->used = used;
3491 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3496 facet_reset_counters(struct facet *facet)
3498 facet->packet_count = 0;
3499 facet->byte_count = 0;
3500 facet->prev_packet_count = 0;
3501 facet->prev_byte_count = 0;
3502 facet->accounted_bytes = 0;
3506 facet_push_stats(struct facet *facet)
3508 uint64_t new_packets, new_bytes;
3510 assert(facet->packet_count >= facet->prev_packet_count);
3511 assert(facet->byte_count >= facet->prev_byte_count);
3512 assert(facet->used >= facet->prev_used);
3514 new_packets = facet->packet_count - facet->prev_packet_count;
3515 new_bytes = facet->byte_count - facet->prev_byte_count;
3517 if (new_packets || new_bytes || facet->used > facet->prev_used) {
3518 facet->prev_packet_count = facet->packet_count;
3519 facet->prev_byte_count = facet->byte_count;
3520 facet->prev_used = facet->used;
3522 flow_push_stats(facet->rule, &facet->flow,
3523 new_packets, new_bytes, facet->used);
3525 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
3526 facet->mirrors, new_packets, new_bytes);
3530 struct ofproto_push {
3531 struct action_xlate_ctx ctx;
3538 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
3540 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3543 rule->packet_count += push->packets;
3544 rule->byte_count += push->bytes;
3545 rule->used = MAX(push->used, rule->used);
3549 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3550 * 'rule''s actions and mirrors. */
3552 flow_push_stats(const struct rule_dpif *rule,
3553 const struct flow *flow, uint64_t packets, uint64_t bytes,
3556 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3557 struct ofproto_push push;
3559 push.packets = packets;
3563 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci,
3564 rule->up.flow_cookie, NULL);
3565 push.ctx.resubmit_hook = push_resubmit;
3566 ofpbuf_delete(xlate_actions(&push.ctx,
3567 rule->up.actions, rule->up.n_actions));
3572 static struct subfacet *
3573 subfacet_find__(struct ofproto_dpif *ofproto,
3574 const struct nlattr *key, size_t key_len, uint32_t key_hash,
3575 const struct flow *flow)
3577 struct subfacet *subfacet;
3579 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
3580 &ofproto->subfacets) {
3582 ? (subfacet->key_len == key_len
3583 && !memcmp(key, subfacet->key, key_len))
3584 : flow_equal(flow, &subfacet->facet->flow)) {
3592 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
3593 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
3594 * there is one, otherwise creates and returns a new subfacet.
3596 * If the returned subfacet is new, then subfacet->actions will be NULL, in
3597 * which case the caller must populate the actions with
3598 * subfacet_make_actions(). */
3599 static struct subfacet *
3600 subfacet_create(struct ofproto_dpif *ofproto, struct facet *facet,
3601 enum odp_key_fitness key_fitness,
3602 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
3604 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3605 struct subfacet *subfacet;
3607 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
3609 if (subfacet->facet == facet) {
3613 /* This shouldn't happen. */
3614 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
3615 subfacet_destroy(ofproto, subfacet);
3618 subfacet = xzalloc(sizeof *subfacet);
3619 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
3620 list_push_back(&facet->subfacets, &subfacet->list_node);
3621 subfacet->facet = facet;
3622 subfacet->used = time_msec();
3623 subfacet->key_fitness = key_fitness;
3624 if (key_fitness != ODP_FIT_PERFECT) {
3625 subfacet->key = xmemdup(key, key_len);
3626 subfacet->key_len = key_len;
3628 subfacet->installed = false;
3629 subfacet->initial_tci = initial_tci;
3634 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
3635 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
3636 static struct subfacet *
3637 subfacet_find(struct ofproto_dpif *ofproto,
3638 const struct nlattr *key, size_t key_len)
3640 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3641 enum odp_key_fitness fitness;
3644 fitness = odp_flow_key_to_flow(key, key_len, &flow);
3645 if (fitness == ODP_FIT_ERROR) {
3649 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
3652 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
3653 * its facet within 'ofproto', and frees it. */
3655 subfacet_destroy__(struct ofproto_dpif *ofproto, struct subfacet *subfacet)
3657 subfacet_uninstall(ofproto, subfacet);
3658 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
3659 list_remove(&subfacet->list_node);
3660 free(subfacet->key);
3661 free(subfacet->actions);
3665 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
3666 * last remaining subfacet in its facet destroys the facet too. */
3668 subfacet_destroy(struct ofproto_dpif *ofproto, struct subfacet *subfacet)
3670 struct facet *facet = subfacet->facet;
3672 if (list_is_singleton(&facet->subfacets)) {
3673 /* facet_remove() needs at least one subfacet (it will remove it). */
3674 facet_remove(ofproto, facet);
3676 subfacet_destroy__(ofproto, subfacet);
3680 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
3681 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
3682 * for use as temporary storage. */
3684 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
3687 if (!subfacet->key) {
3688 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
3689 odp_flow_key_from_flow(key, &subfacet->facet->flow);
3691 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
3695 /* Composes the datapath actions for 'subfacet' based on its rule's actions. */
3697 subfacet_make_actions(struct ofproto_dpif *p, struct subfacet *subfacet,
3698 const struct ofpbuf *packet)
3700 struct facet *facet = subfacet->facet;
3701 const struct rule_dpif *rule = facet->rule;
3702 struct ofpbuf *odp_actions;
3703 struct action_xlate_ctx ctx;
3705 action_xlate_ctx_init(&ctx, p, &facet->flow, subfacet->initial_tci,
3706 rule->up.flow_cookie, packet);
3707 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3708 facet->tags = ctx.tags;
3709 facet->may_install = ctx.may_set_up_flow;
3710 facet->has_learn = ctx.has_learn;
3711 facet->has_normal = ctx.has_normal;
3712 facet->nf_flow.output_iface = ctx.nf_output_iface;
3713 facet->mirrors = ctx.mirrors;
3715 if (subfacet->actions_len != odp_actions->size
3716 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
3717 free(subfacet->actions);
3718 subfacet->actions_len = odp_actions->size;
3719 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
3722 ofpbuf_delete(odp_actions);
3725 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
3726 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
3727 * in the datapath will be zeroed and 'stats' will be updated with traffic new
3728 * since 'subfacet' was last updated.
3730 * Returns 0 if successful, otherwise a positive errno value. */
3732 subfacet_install(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3733 const struct nlattr *actions, size_t actions_len,
3734 struct dpif_flow_stats *stats)
3736 struct odputil_keybuf keybuf;
3737 enum dpif_flow_put_flags flags;
3741 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3743 flags |= DPIF_FP_ZERO_STATS;
3746 subfacet_get_key(subfacet, &keybuf, &key);
3747 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
3748 actions, actions_len, stats);
3751 subfacet_reset_dp_stats(subfacet, stats);
3757 /* If 'subfacet' is installed in the datapath, uninstalls it. */
3759 subfacet_uninstall(struct ofproto_dpif *p, struct subfacet *subfacet)
3761 if (subfacet->installed) {
3762 struct odputil_keybuf keybuf;
3763 struct dpif_flow_stats stats;
3767 subfacet_get_key(subfacet, &keybuf, &key);
3768 error = dpif_flow_del(p->dpif, key.data, key.size, &stats);
3769 subfacet_reset_dp_stats(subfacet, &stats);
3771 subfacet_update_stats(p, subfacet, &stats);
3773 subfacet->installed = false;
3775 assert(subfacet->dp_packet_count == 0);
3776 assert(subfacet->dp_byte_count == 0);
3780 /* Resets 'subfacet''s datapath statistics counters. This should be called
3781 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
3782 * non-null, it should contain the statistics returned by dpif when 'subfacet'
3783 * was reset in the datapath. 'stats' will be modified to include only
3784 * statistics new since 'subfacet' was last updated. */
3786 subfacet_reset_dp_stats(struct subfacet *subfacet,
3787 struct dpif_flow_stats *stats)
3790 && subfacet->dp_packet_count <= stats->n_packets
3791 && subfacet->dp_byte_count <= stats->n_bytes) {
3792 stats->n_packets -= subfacet->dp_packet_count;
3793 stats->n_bytes -= subfacet->dp_byte_count;
3796 subfacet->dp_packet_count = 0;
3797 subfacet->dp_byte_count = 0;
3800 /* Updates 'subfacet''s used time. The caller is responsible for calling
3801 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
3803 subfacet_update_time(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3806 if (used > subfacet->used) {
3807 subfacet->used = used;
3808 facet_update_time(ofproto, subfacet->facet, used);
3812 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
3814 * Because of the meaning of a subfacet's counters, it only makes sense to do
3815 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
3816 * represents a packet that was sent by hand or if it represents statistics
3817 * that have been cleared out of the datapath. */
3819 subfacet_update_stats(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3820 const struct dpif_flow_stats *stats)
3822 if (stats->n_packets || stats->used > subfacet->used) {
3823 struct facet *facet = subfacet->facet;
3825 subfacet_update_time(ofproto, subfacet, stats->used);
3826 facet->packet_count += stats->n_packets;
3827 facet->byte_count += stats->n_bytes;
3828 facet_push_stats(facet);
3829 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3835 static struct rule_dpif *
3836 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
3839 struct cls_rule *cls_rule;
3840 struct classifier *cls;
3842 if (table_id >= N_TABLES) {
3846 cls = &ofproto->up.tables[table_id];
3847 if (flow->nw_frag & FLOW_NW_FRAG_ANY
3848 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3849 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
3850 * are unavailable. */
3851 struct flow ofpc_normal_flow = *flow;
3852 ofpc_normal_flow.tp_src = htons(0);
3853 ofpc_normal_flow.tp_dst = htons(0);
3854 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
3856 cls_rule = classifier_lookup(cls, flow);
3858 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
3862 complete_operation(struct rule_dpif *rule)
3864 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3866 rule_invalidate(rule);
3868 struct dpif_completion *c = xmalloc(sizeof *c);
3869 c->op = rule->up.pending;
3870 list_push_back(&ofproto->completions, &c->list_node);
3872 ofoperation_complete(rule->up.pending, 0);
3876 static struct rule *
3879 struct rule_dpif *rule = xmalloc(sizeof *rule);
3884 rule_dealloc(struct rule *rule_)
3886 struct rule_dpif *rule = rule_dpif_cast(rule_);
3891 rule_construct(struct rule *rule_)
3893 struct rule_dpif *rule = rule_dpif_cast(rule_);
3894 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3895 struct rule_dpif *victim;
3899 error = validate_actions(rule->up.actions, rule->up.n_actions,
3900 &rule->up.cr.flow, ofproto->max_ports);
3905 rule->used = rule->up.created;
3906 rule->packet_count = 0;
3907 rule->byte_count = 0;
3909 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
3910 if (victim && !list_is_empty(&victim->facets)) {
3911 struct facet *facet;
3913 rule->facets = victim->facets;
3914 list_moved(&rule->facets);
3915 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3916 /* XXX: We're only clearing our local counters here. It's possible
3917 * that quite a few packets are unaccounted for in the datapath
3918 * statistics. These will be accounted to the new rule instead of
3919 * cleared as required. This could be fixed by clearing out the
3920 * datapath statistics for this facet, but currently it doesn't
3922 facet_reset_counters(facet);
3926 /* Must avoid list_moved() in this case. */
3927 list_init(&rule->facets);
3930 table_id = rule->up.table_id;
3931 rule->tag = (victim ? victim->tag
3933 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
3934 ofproto->tables[table_id].basis));
3936 complete_operation(rule);
3941 rule_destruct(struct rule *rule_)
3943 struct rule_dpif *rule = rule_dpif_cast(rule_);
3944 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3945 struct facet *facet, *next_facet;
3947 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3948 facet_revalidate(ofproto, facet);
3951 complete_operation(rule);
3955 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
3957 struct rule_dpif *rule = rule_dpif_cast(rule_);
3958 struct facet *facet;
3960 /* Start from historical data for 'rule' itself that are no longer tracked
3961 * in facets. This counts, for example, facets that have expired. */
3962 *packets = rule->packet_count;
3963 *bytes = rule->byte_count;
3965 /* Add any statistics that are tracked by facets. This includes
3966 * statistical data recently updated by ofproto_update_stats() as well as
3967 * stats for packets that were executed "by hand" via dpif_execute(). */
3968 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3969 *packets += facet->packet_count;
3970 *bytes += facet->byte_count;
3975 rule_execute(struct rule *rule_, const struct flow *flow,
3976 struct ofpbuf *packet)
3978 struct rule_dpif *rule = rule_dpif_cast(rule_);
3979 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3980 struct action_xlate_ctx ctx;
3981 struct ofpbuf *odp_actions;
3984 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
3985 rule->up.flow_cookie, packet);
3986 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3987 size = packet->size;
3988 if (execute_odp_actions(ofproto, flow, odp_actions->data,
3989 odp_actions->size, packet)) {
3990 rule->used = time_msec();
3991 rule->packet_count++;
3992 rule->byte_count += size;
3993 flow_push_stats(rule, flow, 1, size, rule->used);
3995 ofpbuf_delete(odp_actions);
4001 rule_modify_actions(struct rule *rule_)
4003 struct rule_dpif *rule = rule_dpif_cast(rule_);
4004 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4007 error = validate_actions(rule->up.actions, rule->up.n_actions,
4008 &rule->up.cr.flow, ofproto->max_ports);
4010 ofoperation_complete(rule->up.pending, error);
4014 complete_operation(rule);
4017 /* Sends 'packet' out 'ofport'.
4018 * May modify 'packet'.
4019 * Returns 0 if successful, otherwise a positive errno value. */
4021 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4023 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4024 struct ofpbuf key, odp_actions;
4025 struct odputil_keybuf keybuf;
4030 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
4031 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4033 if (odp_port != ofport->odp_port) {
4034 eth_pop_vlan(packet);
4035 flow.vlan_tci = htons(0);
4038 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4039 odp_flow_key_from_flow(&key, &flow);
4041 ofpbuf_init(&odp_actions, 32);
4042 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4044 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4045 error = dpif_execute(ofproto->dpif,
4047 odp_actions.data, odp_actions.size,
4049 ofpbuf_uninit(&odp_actions);
4052 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4053 ofproto->up.name, odp_port, strerror(error));
4055 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
4059 /* OpenFlow to datapath action translation. */
4061 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
4062 struct action_xlate_ctx *ctx);
4063 static void xlate_normal(struct action_xlate_ctx *);
4066 put_userspace_action(const struct ofproto_dpif *ofproto,
4067 struct ofpbuf *odp_actions,
4068 const struct flow *flow,
4069 const struct user_action_cookie *cookie)
4073 pid = dpif_port_get_pid(ofproto->dpif,
4074 ofp_port_to_odp_port(flow->in_port));
4076 return odp_put_userspace_action(pid, cookie, odp_actions);
4079 /* Compose SAMPLE action for sFlow. */
4081 compose_sflow_action(const struct ofproto_dpif *ofproto,
4082 struct ofpbuf *odp_actions,
4083 const struct flow *flow,
4086 uint32_t port_ifindex;
4087 uint32_t probability;
4088 struct user_action_cookie cookie;
4089 size_t sample_offset, actions_offset;
4090 int cookie_offset, n_output;
4092 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4096 if (odp_port == OVSP_NONE) {
4100 port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4104 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4106 /* Number of packets out of UINT_MAX to sample. */
4107 probability = dpif_sflow_get_probability(ofproto->sflow);
4108 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4110 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4112 cookie.type = USER_ACTION_COOKIE_SFLOW;
4113 cookie.data = port_ifindex;
4114 cookie.n_output = n_output;
4115 cookie.vlan_tci = 0;
4116 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4118 nl_msg_end_nested(odp_actions, actions_offset);
4119 nl_msg_end_nested(odp_actions, sample_offset);
4120 return cookie_offset;
4123 /* SAMPLE action must be first action in any given list of actions.
4124 * At this point we do not have all information required to build it. So try to
4125 * build sample action as complete as possible. */
4127 add_sflow_action(struct action_xlate_ctx *ctx)
4129 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4131 &ctx->flow, OVSP_NONE);
4132 ctx->sflow_odp_port = 0;
4133 ctx->sflow_n_outputs = 0;
4136 /* Fix SAMPLE action according to data collected while composing ODP actions.
4137 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4138 * USERSPACE action's user-cookie which is required for sflow. */
4140 fix_sflow_action(struct action_xlate_ctx *ctx)
4142 const struct flow *base = &ctx->base_flow;
4143 struct user_action_cookie *cookie;
4145 if (!ctx->user_cookie_offset) {
4149 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4151 assert(cookie != NULL);
4152 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4154 if (ctx->sflow_n_outputs) {
4155 cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow,
4156 ctx->sflow_odp_port);
4158 if (ctx->sflow_n_outputs >= 255) {
4159 cookie->n_output = 255;
4161 cookie->n_output = ctx->sflow_n_outputs;
4163 cookie->vlan_tci = base->vlan_tci;
4167 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4170 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4171 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4172 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4173 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4177 struct priority_to_dscp *pdscp;
4179 if (ofport->up.opp.config & htonl(OFPPC_NO_FWD)
4180 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4184 pdscp = get_priority(ofport, ctx->flow.skb_priority);
4186 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4187 ctx->flow.nw_tos |= pdscp->dscp;
4190 /* We may not have an ofport record for this port, but it doesn't hurt
4191 * to allow forwarding to it anyhow. Maybe such a port will appear
4192 * later and we're pre-populating the flow table. */
4195 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4196 ctx->flow.vlan_tci);
4197 if (out_port != odp_port) {
4198 ctx->flow.vlan_tci = htons(0);
4200 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4201 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4203 ctx->sflow_odp_port = odp_port;
4204 ctx->sflow_n_outputs++;
4205 ctx->nf_output_iface = ofp_port;
4206 ctx->flow.vlan_tci = flow_vlan_tci;
4207 ctx->flow.nw_tos = flow_nw_tos;
4211 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4213 compose_output_action__(ctx, ofp_port, true);
4217 xlate_table_action(struct action_xlate_ctx *ctx,
4218 uint16_t in_port, uint8_t table_id)
4220 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4221 struct ofproto_dpif *ofproto = ctx->ofproto;
4222 struct rule_dpif *rule;
4223 uint16_t old_in_port;
4224 uint8_t old_table_id;
4226 old_table_id = ctx->table_id;
4227 ctx->table_id = table_id;
4229 /* Look up a flow with 'in_port' as the input port. */
4230 old_in_port = ctx->flow.in_port;
4231 ctx->flow.in_port = in_port;
4232 rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
4235 if (table_id > 0 && table_id < N_TABLES) {
4236 struct table_dpif *table = &ofproto->tables[table_id];
4237 if (table->other_table) {
4240 : rule_calculate_tag(&ctx->flow,
4241 &table->other_table->wc,
4246 /* Restore the original input port. Otherwise OFPP_NORMAL and
4247 * OFPP_IN_PORT will have surprising behavior. */
4248 ctx->flow.in_port = old_in_port;
4250 if (ctx->resubmit_hook) {
4251 ctx->resubmit_hook(ctx, rule);
4255 ovs_be64 old_cookie = ctx->cookie;
4258 ctx->cookie = rule->up.flow_cookie;
4259 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
4260 ctx->cookie = old_cookie;
4264 ctx->table_id = old_table_id;
4266 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4268 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
4269 MAX_RESUBMIT_RECURSION);
4274 xlate_resubmit_table(struct action_xlate_ctx *ctx,
4275 const struct nx_action_resubmit *nar)
4280 in_port = (nar->in_port == htons(OFPP_IN_PORT)
4282 : ntohs(nar->in_port));
4283 table_id = nar->table == 255 ? ctx->table_id : nar->table;
4285 xlate_table_action(ctx, in_port, table_id);
4289 flood_packets(struct action_xlate_ctx *ctx, bool all)
4291 struct ofport_dpif *ofport;
4293 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
4294 uint16_t ofp_port = ofport->up.ofp_port;
4296 if (ofp_port == ctx->flow.in_port) {
4301 compose_output_action__(ctx, ofp_port, false);
4302 } else if (!(ofport->up.opp.config & htonl(OFPPC_NO_FLOOD))) {
4303 compose_output_action(ctx, ofp_port);
4307 ctx->nf_output_iface = NF_OUT_FLOOD;
4311 execute_controller_action(struct action_xlate_ctx *ctx, int len)
4313 struct ofputil_packet_in pin;
4314 struct ofpbuf *packet;
4316 ctx->may_set_up_flow = false;
4321 packet = ofpbuf_clone(ctx->packet);
4323 if (packet->l2 && packet->l3) {
4324 struct eth_header *eh;
4326 eth_pop_vlan(packet);
4328 assert(eh->eth_type == ctx->flow.dl_type);
4329 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
4330 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
4332 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
4333 eth_push_vlan(packet, ctx->flow.vlan_tci);
4337 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4338 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
4339 ctx->flow.nw_tos, ctx->flow.nw_ttl);
4343 if (ctx->flow.nw_proto == IPPROTO_TCP) {
4344 packet_set_tcp_port(packet, ctx->flow.tp_src,
4346 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
4347 packet_set_udp_port(packet, ctx->flow.tp_src,
4354 pin.packet = packet->data;
4355 pin.packet_len = packet->size;
4356 pin.reason = OFPR_ACTION;
4357 pin.table_id = ctx->table_id;
4358 pin.cookie = ctx->cookie;
4362 pin.total_len = packet->size;
4363 flow_get_metadata(&ctx->flow, &pin.fmd);
4365 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin, &ctx->flow);
4366 ofpbuf_delete(packet);
4370 xlate_output_action__(struct action_xlate_ctx *ctx,
4371 uint16_t port, uint16_t max_len)
4373 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
4375 ctx->nf_output_iface = NF_OUT_DROP;
4379 compose_output_action(ctx, ctx->flow.in_port);
4382 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
4388 flood_packets(ctx, false);
4391 flood_packets(ctx, true);
4393 case OFPP_CONTROLLER:
4394 execute_controller_action(ctx, max_len);
4400 if (port != ctx->flow.in_port) {
4401 compose_output_action(ctx, port);
4406 if (prev_nf_output_iface == NF_OUT_FLOOD) {
4407 ctx->nf_output_iface = NF_OUT_FLOOD;
4408 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
4409 ctx->nf_output_iface = prev_nf_output_iface;
4410 } else if (prev_nf_output_iface != NF_OUT_DROP &&
4411 ctx->nf_output_iface != NF_OUT_FLOOD) {
4412 ctx->nf_output_iface = NF_OUT_MULTI;
4417 xlate_output_reg_action(struct action_xlate_ctx *ctx,
4418 const struct nx_action_output_reg *naor)
4422 ofp_port = nxm_read_field_bits(naor->src, naor->ofs_nbits, &ctx->flow);
4424 if (ofp_port <= UINT16_MAX) {
4425 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
4430 xlate_output_action(struct action_xlate_ctx *ctx,
4431 const struct ofp_action_output *oao)
4433 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
4437 xlate_enqueue_action(struct action_xlate_ctx *ctx,
4438 const struct ofp_action_enqueue *oae)
4441 uint32_t flow_priority, priority;
4444 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
4447 /* Fall back to ordinary output action. */
4448 xlate_output_action__(ctx, ntohs(oae->port), 0);
4452 /* Figure out datapath output port. */
4453 ofp_port = ntohs(oae->port);
4454 if (ofp_port == OFPP_IN_PORT) {
4455 ofp_port = ctx->flow.in_port;
4456 } else if (ofp_port == ctx->flow.in_port) {
4460 /* Add datapath actions. */
4461 flow_priority = ctx->flow.skb_priority;
4462 ctx->flow.skb_priority = priority;
4463 compose_output_action(ctx, ofp_port);
4464 ctx->flow.skb_priority = flow_priority;
4466 /* Update NetFlow output port. */
4467 if (ctx->nf_output_iface == NF_OUT_DROP) {
4468 ctx->nf_output_iface = ofp_port;
4469 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
4470 ctx->nf_output_iface = NF_OUT_MULTI;
4475 xlate_set_queue_action(struct action_xlate_ctx *ctx,
4476 const struct nx_action_set_queue *nasq)
4481 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
4484 /* Couldn't translate queue to a priority, so ignore. A warning
4485 * has already been logged. */
4489 ctx->flow.skb_priority = priority;
4492 struct xlate_reg_state {
4498 xlate_autopath(struct action_xlate_ctx *ctx,
4499 const struct nx_action_autopath *naa)
4501 uint16_t ofp_port = ntohl(naa->id);
4502 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
4504 if (!port || !port->bundle) {
4505 ofp_port = OFPP_NONE;
4506 } else if (port->bundle->bond) {
4507 /* Autopath does not support VLAN hashing. */
4508 struct ofport_dpif *slave = bond_choose_output_slave(
4509 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
4511 ofp_port = slave->up.ofp_port;
4514 autopath_execute(naa, &ctx->flow, ofp_port);
4518 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
4520 struct ofproto_dpif *ofproto = ofproto_;
4521 struct ofport_dpif *port;
4531 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
4534 port = get_ofp_port(ofproto, ofp_port);
4535 return port ? port->may_enable : false;
4540 xlate_learn_action(struct action_xlate_ctx *ctx,
4541 const struct nx_action_learn *learn)
4543 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
4544 struct ofputil_flow_mod fm;
4547 learn_execute(learn, &ctx->flow, &fm);
4549 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
4550 if (error && !VLOG_DROP_WARN(&rl)) {
4551 char *msg = ofputil_error_to_string(error);
4552 VLOG_WARN("learning action failed to modify flow table (%s)", msg);
4560 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
4562 if (port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
4563 ? htonl(OFPPC_NO_RECV_STP)
4564 : htonl(OFPPC_NO_RECV))) {
4568 /* Only drop packets here if both forwarding and learning are
4569 * disabled. If just learning is enabled, we need to have
4570 * OFPP_NORMAL and the learning action have a look at the packet
4571 * before we can drop it. */
4572 if (!stp_forward_in_state(port->stp_state)
4573 && !stp_learn_in_state(port->stp_state)) {
4581 do_xlate_actions(const union ofp_action *in, size_t n_in,
4582 struct action_xlate_ctx *ctx)
4584 const struct ofport_dpif *port;
4585 const union ofp_action *ia;
4588 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
4589 if (port && !may_receive(port, ctx)) {
4590 /* Drop this flow. */
4594 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
4595 const struct ofp_action_dl_addr *oada;
4596 const struct nx_action_resubmit *nar;
4597 const struct nx_action_set_tunnel *nast;
4598 const struct nx_action_set_queue *nasq;
4599 const struct nx_action_multipath *nam;
4600 const struct nx_action_autopath *naa;
4601 const struct nx_action_bundle *nab;
4602 const struct nx_action_output_reg *naor;
4603 enum ofputil_action_code code;
4610 code = ofputil_decode_action_unsafe(ia);
4612 case OFPUTIL_OFPAT_OUTPUT:
4613 xlate_output_action(ctx, &ia->output);
4616 case OFPUTIL_OFPAT_SET_VLAN_VID:
4617 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
4618 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
4621 case OFPUTIL_OFPAT_SET_VLAN_PCP:
4622 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
4623 ctx->flow.vlan_tci |= htons(
4624 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
4627 case OFPUTIL_OFPAT_STRIP_VLAN:
4628 ctx->flow.vlan_tci = htons(0);
4631 case OFPUTIL_OFPAT_SET_DL_SRC:
4632 oada = ((struct ofp_action_dl_addr *) ia);
4633 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
4636 case OFPUTIL_OFPAT_SET_DL_DST:
4637 oada = ((struct ofp_action_dl_addr *) ia);
4638 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
4641 case OFPUTIL_OFPAT_SET_NW_SRC:
4642 ctx->flow.nw_src = ia->nw_addr.nw_addr;
4645 case OFPUTIL_OFPAT_SET_NW_DST:
4646 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
4649 case OFPUTIL_OFPAT_SET_NW_TOS:
4650 /* OpenFlow 1.0 only supports IPv4. */
4651 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4652 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4653 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
4657 case OFPUTIL_OFPAT_SET_TP_SRC:
4658 ctx->flow.tp_src = ia->tp_port.tp_port;
4661 case OFPUTIL_OFPAT_SET_TP_DST:
4662 ctx->flow.tp_dst = ia->tp_port.tp_port;
4665 case OFPUTIL_OFPAT_ENQUEUE:
4666 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
4669 case OFPUTIL_NXAST_RESUBMIT:
4670 nar = (const struct nx_action_resubmit *) ia;
4671 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
4674 case OFPUTIL_NXAST_RESUBMIT_TABLE:
4675 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
4678 case OFPUTIL_NXAST_SET_TUNNEL:
4679 nast = (const struct nx_action_set_tunnel *) ia;
4680 tun_id = htonll(ntohl(nast->tun_id));
4681 ctx->flow.tun_id = tun_id;
4684 case OFPUTIL_NXAST_SET_QUEUE:
4685 nasq = (const struct nx_action_set_queue *) ia;
4686 xlate_set_queue_action(ctx, nasq);
4689 case OFPUTIL_NXAST_POP_QUEUE:
4690 ctx->flow.skb_priority = ctx->orig_skb_priority;
4693 case OFPUTIL_NXAST_REG_MOVE:
4694 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
4698 case OFPUTIL_NXAST_REG_LOAD:
4699 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
4703 case OFPUTIL_NXAST_NOTE:
4704 /* Nothing to do. */
4707 case OFPUTIL_NXAST_SET_TUNNEL64:
4708 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
4709 ctx->flow.tun_id = tun_id;
4712 case OFPUTIL_NXAST_MULTIPATH:
4713 nam = (const struct nx_action_multipath *) ia;
4714 multipath_execute(nam, &ctx->flow);
4717 case OFPUTIL_NXAST_AUTOPATH:
4718 naa = (const struct nx_action_autopath *) ia;
4719 xlate_autopath(ctx, naa);
4722 case OFPUTIL_NXAST_BUNDLE:
4723 ctx->ofproto->has_bundle_action = true;
4724 nab = (const struct nx_action_bundle *) ia;
4725 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
4730 case OFPUTIL_NXAST_BUNDLE_LOAD:
4731 ctx->ofproto->has_bundle_action = true;
4732 nab = (const struct nx_action_bundle *) ia;
4733 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
4737 case OFPUTIL_NXAST_OUTPUT_REG:
4738 naor = (const struct nx_action_output_reg *) ia;
4739 xlate_output_reg_action(ctx, naor);
4742 case OFPUTIL_NXAST_LEARN:
4743 ctx->has_learn = true;
4744 if (ctx->may_learn) {
4745 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
4749 case OFPUTIL_NXAST_EXIT:
4755 /* We've let OFPP_NORMAL and the learning action look at the packet,
4756 * so drop it now if forwarding is disabled. */
4757 if (port && !stp_forward_in_state(port->stp_state)) {
4758 ofpbuf_clear(ctx->odp_actions);
4759 add_sflow_action(ctx);
4764 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
4765 struct ofproto_dpif *ofproto, const struct flow *flow,
4766 ovs_be16 initial_tci, ovs_be64 cookie,
4767 const struct ofpbuf *packet)
4769 ctx->ofproto = ofproto;
4771 ctx->base_flow = ctx->flow;
4772 ctx->base_flow.tun_id = 0;
4773 ctx->base_flow.vlan_tci = initial_tci;
4774 ctx->cookie = cookie;
4775 ctx->packet = packet;
4776 ctx->may_learn = packet != NULL;
4777 ctx->resubmit_hook = NULL;
4780 static struct ofpbuf *
4781 xlate_actions(struct action_xlate_ctx *ctx,
4782 const union ofp_action *in, size_t n_in)
4784 struct flow orig_flow = ctx->flow;
4786 COVERAGE_INC(ofproto_dpif_xlate);
4788 ctx->odp_actions = ofpbuf_new(512);
4789 ofpbuf_reserve(ctx->odp_actions, NL_A_U32_SIZE);
4791 ctx->may_set_up_flow = true;
4792 ctx->has_learn = false;
4793 ctx->has_normal = false;
4794 ctx->nf_output_iface = NF_OUT_DROP;
4797 ctx->orig_skb_priority = ctx->flow.skb_priority;
4801 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
4802 switch (ctx->ofproto->up.frag_handling) {
4803 case OFPC_FRAG_NORMAL:
4804 /* We must pretend that transport ports are unavailable. */
4805 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
4806 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
4809 case OFPC_FRAG_DROP:
4810 return ctx->odp_actions;
4812 case OFPC_FRAG_REASM:
4815 case OFPC_FRAG_NX_MATCH:
4816 /* Nothing to do. */
4821 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
4822 ctx->may_set_up_flow = false;
4823 return ctx->odp_actions;
4825 add_sflow_action(ctx);
4826 do_xlate_actions(in, n_in, ctx);
4828 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
4829 ctx->odp_actions->data,
4830 ctx->odp_actions->size)) {
4831 ctx->may_set_up_flow = false;
4833 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
4835 compose_output_action(ctx, OFPP_LOCAL);
4838 add_mirror_actions(ctx, &orig_flow);
4839 fix_sflow_action(ctx);
4842 return ctx->odp_actions;
4845 /* OFPP_NORMAL implementation. */
4847 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
4849 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
4850 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
4851 * the bundle on which the packet was received, returns the VLAN to which the
4854 * Both 'vid' and the return value are in the range 0...4095. */
4856 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
4858 switch (in_bundle->vlan_mode) {
4859 case PORT_VLAN_ACCESS:
4860 return in_bundle->vlan;
4863 case PORT_VLAN_TRUNK:
4866 case PORT_VLAN_NATIVE_UNTAGGED:
4867 case PORT_VLAN_NATIVE_TAGGED:
4868 return vid ? vid : in_bundle->vlan;
4875 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
4876 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
4879 * 'vid' should be the VID obtained from the 802.1Q header that was received as
4880 * part of a packet (specify 0 if there was no 802.1Q header), in the range
4883 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
4885 /* Allow any VID on the OFPP_NONE port. */
4886 if (in_bundle == &ofpp_none_bundle) {
4890 switch (in_bundle->vlan_mode) {
4891 case PORT_VLAN_ACCESS:
4894 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4895 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
4896 "packet received on port %s configured as VLAN "
4897 "%"PRIu16" access port",
4898 in_bundle->ofproto->up.name, vid,
4899 in_bundle->name, in_bundle->vlan);
4905 case PORT_VLAN_NATIVE_UNTAGGED:
4906 case PORT_VLAN_NATIVE_TAGGED:
4908 /* Port must always carry its native VLAN. */
4912 case PORT_VLAN_TRUNK:
4913 if (!ofbundle_includes_vlan(in_bundle, vid)) {
4915 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4916 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
4917 "received on port %s not configured for trunking "
4919 in_bundle->ofproto->up.name, vid,
4920 in_bundle->name, vid);
4932 /* Given 'vlan', the VLAN that a packet belongs to, and
4933 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
4934 * that should be included in the 802.1Q header. (If the return value is 0,
4935 * then the 802.1Q header should only be included in the packet if there is a
4938 * Both 'vlan' and the return value are in the range 0...4095. */
4940 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
4942 switch (out_bundle->vlan_mode) {
4943 case PORT_VLAN_ACCESS:
4946 case PORT_VLAN_TRUNK:
4947 case PORT_VLAN_NATIVE_TAGGED:
4950 case PORT_VLAN_NATIVE_UNTAGGED:
4951 return vlan == out_bundle->vlan ? 0 : vlan;
4959 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
4962 struct ofport_dpif *port;
4964 ovs_be16 tci, old_tci;
4966 vid = output_vlan_to_vid(out_bundle, vlan);
4967 if (!out_bundle->bond) {
4968 port = ofbundle_get_a_port(out_bundle);
4970 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
4973 /* No slaves enabled, so drop packet. */
4978 old_tci = ctx->flow.vlan_tci;
4980 if (tci || out_bundle->use_priority_tags) {
4981 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
4983 tci |= htons(VLAN_CFI);
4986 ctx->flow.vlan_tci = tci;
4988 compose_output_action(ctx, port->up.ofp_port);
4989 ctx->flow.vlan_tci = old_tci;
4993 mirror_mask_ffs(mirror_mask_t mask)
4995 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
5000 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
5002 return (bundle->vlan_mode != PORT_VLAN_ACCESS
5003 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
5007 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
5009 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
5012 /* Returns an arbitrary interface within 'bundle'. */
5013 static struct ofport_dpif *
5014 ofbundle_get_a_port(const struct ofbundle *bundle)
5016 return CONTAINER_OF(list_front(&bundle->ports),
5017 struct ofport_dpif, bundle_node);
5021 vlan_is_mirrored(const struct ofmirror *m, int vlan)
5023 return !m->vlans || bitmap_is_set(m->vlans, vlan);
5026 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
5027 * to a VLAN. In general most packets may be mirrored but we want to drop
5028 * protocols that may confuse switches. */
5030 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
5032 /* If you change this function's behavior, please update corresponding
5033 * documentation in vswitch.xml at the same time. */
5034 if (dst[0] != 0x01) {
5035 /* All the currently banned MACs happen to start with 01 currently, so
5036 * this is a quick way to eliminate most of the good ones. */
5038 if (eth_addr_is_reserved(dst)) {
5039 /* Drop STP, IEEE pause frames, and other reserved protocols
5040 * (01-80-c2-00-00-0x). */
5044 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
5046 if ((dst[3] & 0xfe) == 0xcc &&
5047 (dst[4] & 0xfe) == 0xcc &&
5048 (dst[5] & 0xfe) == 0xcc) {
5049 /* Drop the following protocols plus others following the same
5052 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
5053 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
5054 STP Uplink Fast (01-00-0c-cd-cd-cd) */
5058 if (!(dst[3] | dst[4] | dst[5])) {
5059 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
5068 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5070 struct ofproto_dpif *ofproto = ctx->ofproto;
5071 mirror_mask_t mirrors;
5072 struct ofbundle *in_bundle;
5075 const struct nlattr *a;
5078 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5079 ctx->packet != NULL);
5083 mirrors = in_bundle->src_mirrors;
5085 /* Drop frames on bundles reserved for mirroring. */
5086 if (in_bundle->mirror_out) {
5087 if (ctx->packet != NULL) {
5088 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5089 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5090 "%s, which is reserved exclusively for mirroring",
5091 ctx->ofproto->up.name, in_bundle->name);
5097 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5098 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5101 vlan = input_vid_to_vlan(in_bundle, vid);
5103 /* Look at the output ports to check for destination selections. */
5105 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5106 ctx->odp_actions->size) {
5107 enum ovs_action_attr type = nl_attr_type(a);
5108 struct ofport_dpif *ofport;
5110 if (type != OVS_ACTION_ATTR_OUTPUT) {
5114 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5115 if (ofport && ofport->bundle) {
5116 mirrors |= ofport->bundle->dst_mirrors;
5124 /* Restore the original packet before adding the mirror actions. */
5125 ctx->flow = *orig_flow;
5130 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5132 if (!vlan_is_mirrored(m, vlan)) {
5133 mirrors &= mirrors - 1;
5137 mirrors &= ~m->dup_mirrors;
5138 ctx->mirrors |= m->dup_mirrors;
5140 output_normal(ctx, m->out, vlan);
5141 } else if (eth_dst_may_rspan(orig_flow->dl_dst)
5142 && vlan != m->out_vlan) {
5143 struct ofbundle *bundle;
5145 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5146 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5147 && !bundle->mirror_out) {
5148 output_normal(ctx, bundle, m->out_vlan);
5156 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5157 uint64_t packets, uint64_t bytes)
5163 for (; mirrors; mirrors &= mirrors - 1) {
5166 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5169 /* In normal circumstances 'm' will not be NULL. However,
5170 * if mirrors are reconfigured, we can temporarily get out
5171 * of sync in facet_revalidate(). We could "correct" the
5172 * mirror list before reaching here, but doing that would
5173 * not properly account the traffic stats we've currently
5174 * accumulated for previous mirror configuration. */
5178 m->packet_count += packets;
5179 m->byte_count += bytes;
5183 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
5184 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
5185 * indicate this; newer upstream kernels use gratuitous ARP requests. */
5187 is_gratuitous_arp(const struct flow *flow)
5189 return (flow->dl_type == htons(ETH_TYPE_ARP)
5190 && eth_addr_is_broadcast(flow->dl_dst)
5191 && (flow->nw_proto == ARP_OP_REPLY
5192 || (flow->nw_proto == ARP_OP_REQUEST
5193 && flow->nw_src == flow->nw_dst)));
5197 update_learning_table(struct ofproto_dpif *ofproto,
5198 const struct flow *flow, int vlan,
5199 struct ofbundle *in_bundle)
5201 struct mac_entry *mac;
5203 /* Don't learn the OFPP_NONE port. */
5204 if (in_bundle == &ofpp_none_bundle) {
5208 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
5212 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
5213 if (is_gratuitous_arp(flow)) {
5214 /* We don't want to learn from gratuitous ARP packets that are
5215 * reflected back over bond slaves so we lock the learning table. */
5216 if (!in_bundle->bond) {
5217 mac_entry_set_grat_arp_lock(mac);
5218 } else if (mac_entry_is_grat_arp_locked(mac)) {
5223 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
5224 /* The log messages here could actually be useful in debugging,
5225 * so keep the rate limit relatively high. */
5226 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5227 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
5228 "on port %s in VLAN %d",
5229 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
5230 in_bundle->name, vlan);
5232 mac->port.p = in_bundle;
5233 tag_set_add(&ofproto->revalidate_set,
5234 mac_learning_changed(ofproto->ml, mac));
5238 static struct ofbundle *
5239 lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn)
5241 struct ofport_dpif *ofport;
5243 /* Special-case OFPP_NONE, which a controller may use as the ingress
5244 * port for traffic that it is sourcing. */
5245 if (in_port == OFPP_NONE) {
5246 return &ofpp_none_bundle;
5249 /* Find the port and bundle for the received packet. */
5250 ofport = get_ofp_port(ofproto, in_port);
5251 if (ofport && ofport->bundle) {
5252 return ofport->bundle;
5255 /* Odd. A few possible reasons here:
5257 * - We deleted a port but there are still a few packets queued up
5260 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
5261 * we don't know about.
5263 * - The ofproto client didn't configure the port as part of a bundle.
5266 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5268 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
5269 "port %"PRIu16, ofproto->up.name, in_port);
5274 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
5275 * dropped. Returns true if they may be forwarded, false if they should be
5278 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
5279 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
5281 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
5282 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
5283 * checked by input_vid_is_valid().
5285 * May also add tags to '*tags', although the current implementation only does
5286 * so in one special case.
5289 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
5290 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
5292 struct ofbundle *in_bundle = in_port->bundle;
5294 /* Drop frames for reserved multicast addresses
5295 * only if forward_bpdu option is absent. */
5296 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
5300 if (in_bundle->bond) {
5301 struct mac_entry *mac;
5303 switch (bond_check_admissibility(in_bundle->bond, in_port,
5304 flow->dl_dst, tags)) {
5311 case BV_DROP_IF_MOVED:
5312 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
5313 if (mac && mac->port.p != in_bundle &&
5314 (!is_gratuitous_arp(flow)
5315 || mac_entry_is_grat_arp_locked(mac))) {
5326 xlate_normal(struct action_xlate_ctx *ctx)
5328 struct ofport_dpif *in_port;
5329 struct ofbundle *in_bundle;
5330 struct mac_entry *mac;
5334 ctx->has_normal = true;
5336 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
5337 ctx->packet != NULL);
5342 /* We know 'in_port' exists unless it is "ofpp_none_bundle",
5343 * since lookup_input_bundle() succeeded. */
5344 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5346 /* Drop malformed frames. */
5347 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
5348 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
5349 if (ctx->packet != NULL) {
5350 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5351 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
5352 "VLAN tag received on port %s",
5353 ctx->ofproto->up.name, in_bundle->name);
5358 /* Drop frames on bundles reserved for mirroring. */
5359 if (in_bundle->mirror_out) {
5360 if (ctx->packet != NULL) {
5361 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5362 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5363 "%s, which is reserved exclusively for mirroring",
5364 ctx->ofproto->up.name, in_bundle->name);
5370 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
5371 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5374 vlan = input_vid_to_vlan(in_bundle, vid);
5376 /* Check other admissibility requirements. */
5378 !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
5382 /* Learn source MAC. */
5383 if (ctx->may_learn) {
5384 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
5387 /* Determine output bundle. */
5388 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
5391 if (mac->port.p != in_bundle) {
5392 output_normal(ctx, mac->port.p, vlan);
5395 struct ofbundle *bundle;
5397 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
5398 if (bundle != in_bundle
5399 && ofbundle_includes_vlan(bundle, vlan)
5400 && bundle->floodable
5401 && !bundle->mirror_out) {
5402 output_normal(ctx, bundle, vlan);
5405 ctx->nf_output_iface = NF_OUT_FLOOD;
5409 /* Optimized flow revalidation.
5411 * It's a difficult problem, in general, to tell which facets need to have
5412 * their actions recalculated whenever the OpenFlow flow table changes. We
5413 * don't try to solve that general problem: for most kinds of OpenFlow flow
5414 * table changes, we recalculate the actions for every facet. This is
5415 * relatively expensive, but it's good enough if the OpenFlow flow table
5416 * doesn't change very often.
5418 * However, we can expect one particular kind of OpenFlow flow table change to
5419 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5420 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5421 * table, we add a special case that applies to flow tables in which every rule
5422 * has the same form (that is, the same wildcards), except that the table is
5423 * also allowed to have a single "catch-all" flow that matches all packets. We
5424 * optimize this case by tagging all of the facets that resubmit into the table
5425 * and invalidating the same tag whenever a flow changes in that table. The
5426 * end result is that we revalidate just the facets that need it (and sometimes
5427 * a few more, but not all of the facets or even all of the facets that
5428 * resubmit to the table modified by MAC learning). */
5430 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
5431 * into an OpenFlow table with the given 'basis'. */
5433 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
5436 if (flow_wildcards_is_catchall(wc)) {
5439 struct flow tag_flow = *flow;
5440 flow_zero_wildcards(&tag_flow, wc);
5441 return tag_create_deterministic(flow_hash(&tag_flow, secret));
5445 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5446 * taggability of that table.
5448 * This function must be called after *each* change to a flow table. If you
5449 * skip calling it on some changes then the pointer comparisons at the end can
5450 * be invalid if you get unlucky. For example, if a flow removal causes a
5451 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5452 * different wildcards to be created with the same address, then this function
5453 * will incorrectly skip revalidation. */
5455 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5457 struct table_dpif *table = &ofproto->tables[table_id];
5458 const struct classifier *cls = &ofproto->up.tables[table_id];
5459 struct cls_table *catchall, *other;
5460 struct cls_table *t;
5462 catchall = other = NULL;
5464 switch (hmap_count(&cls->tables)) {
5466 /* We could tag this OpenFlow table but it would make the logic a
5467 * little harder and it's a corner case that doesn't seem worth it
5473 HMAP_FOR_EACH (t, hmap_node, &cls->tables) {
5474 if (cls_table_is_catchall(t)) {
5476 } else if (!other) {
5479 /* Indicate that we can't tag this by setting both tables to
5480 * NULL. (We know that 'catchall' is already NULL.) */
5487 /* Can't tag this table. */
5491 if (table->catchall_table != catchall || table->other_table != other) {
5492 table->catchall_table = catchall;
5493 table->other_table = other;
5494 ofproto->need_revalidate = true;
5498 /* Given 'rule' that has changed in some way (either it is a rule being
5499 * inserted, a rule being deleted, or a rule whose actions are being
5500 * modified), marks facets for revalidation to ensure that packets will be
5501 * forwarded correctly according to the new state of the flow table.
5503 * This function must be called after *each* change to a flow table. See
5504 * the comment on table_update_taggable() for more information. */
5506 rule_invalidate(const struct rule_dpif *rule)
5508 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5510 table_update_taggable(ofproto, rule->up.table_id);
5512 if (!ofproto->need_revalidate) {
5513 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5515 if (table->other_table && rule->tag) {
5516 tag_set_add(&ofproto->revalidate_set, rule->tag);
5518 ofproto->need_revalidate = true;
5524 set_frag_handling(struct ofproto *ofproto_,
5525 enum ofp_config_flags frag_handling)
5527 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5529 if (frag_handling != OFPC_FRAG_REASM) {
5530 ofproto->need_revalidate = true;
5538 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5539 const struct flow *flow,
5540 const union ofp_action *ofp_actions, size_t n_ofp_actions)
5542 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5545 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
5546 return ofp_mkerr_nicira(OFPET_BAD_REQUEST, NXBRC_BAD_IN_PORT);
5549 error = validate_actions(ofp_actions, n_ofp_actions, flow,
5550 ofproto->max_ports);
5552 struct odputil_keybuf keybuf;
5553 struct ofpbuf *odp_actions;
5554 struct ofproto_push push;
5557 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5558 odp_flow_key_from_flow(&key, flow);
5560 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, 0,
5563 /* Ensure that resubmits in 'ofp_actions' get accounted to their
5564 * matching rules. */
5566 push.bytes = packet->size;
5567 push.used = time_msec();
5568 push.ctx.resubmit_hook = push_resubmit;
5570 odp_actions = xlate_actions(&push.ctx, ofp_actions, n_ofp_actions);
5571 dpif_execute(ofproto->dpif, key.data, key.size,
5572 odp_actions->data, odp_actions->size, packet);
5573 ofpbuf_delete(odp_actions);
5581 set_netflow(struct ofproto *ofproto_,
5582 const struct netflow_options *netflow_options)
5584 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5586 if (netflow_options) {
5587 if (!ofproto->netflow) {
5588 ofproto->netflow = netflow_create();
5590 return netflow_set_options(ofproto->netflow, netflow_options);
5592 netflow_destroy(ofproto->netflow);
5593 ofproto->netflow = NULL;
5599 get_netflow_ids(const struct ofproto *ofproto_,
5600 uint8_t *engine_type, uint8_t *engine_id)
5602 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5604 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
5608 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5610 if (!facet_is_controller_flow(facet) &&
5611 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5612 struct subfacet *subfacet;
5613 struct ofexpired expired;
5615 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5616 if (subfacet->installed) {
5617 struct dpif_flow_stats stats;
5619 subfacet_install(ofproto, subfacet, subfacet->actions,
5620 subfacet->actions_len, &stats);
5621 subfacet_update_stats(ofproto, subfacet, &stats);
5625 expired.flow = facet->flow;
5626 expired.packet_count = facet->packet_count;
5627 expired.byte_count = facet->byte_count;
5628 expired.used = facet->used;
5629 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5634 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5636 struct facet *facet;
5638 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5639 send_active_timeout(ofproto, facet);
5643 static struct ofproto_dpif *
5644 ofproto_dpif_lookup(const char *name)
5646 struct ofproto_dpif *ofproto;
5648 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5649 hash_string(name, 0), &all_ofproto_dpifs) {
5650 if (!strcmp(ofproto->up.name, name)) {
5658 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5659 const char *argv[], void *aux OVS_UNUSED)
5661 struct ofproto_dpif *ofproto;
5664 ofproto = ofproto_dpif_lookup(argv[1]);
5666 unixctl_command_reply(conn, 501, "no such bridge");
5669 mac_learning_flush(ofproto->ml);
5670 ofproto->need_revalidate = true;
5672 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5673 mac_learning_flush(ofproto->ml);
5674 ofproto->need_revalidate = true;
5678 unixctl_command_reply(conn, 200, "table successfully flushed");
5682 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5683 const char *argv[], void *aux OVS_UNUSED)
5685 struct ds ds = DS_EMPTY_INITIALIZER;
5686 const struct ofproto_dpif *ofproto;
5687 const struct mac_entry *e;
5689 ofproto = ofproto_dpif_lookup(argv[1]);
5691 unixctl_command_reply(conn, 501, "no such bridge");
5695 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5696 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5697 struct ofbundle *bundle = e->port.p;
5698 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
5699 ofbundle_get_a_port(bundle)->odp_port,
5700 e->vlan, ETH_ADDR_ARGS(e->mac),
5701 mac_entry_age(ofproto->ml, e));
5703 unixctl_command_reply(conn, 200, ds_cstr(&ds));
5707 struct ofproto_trace {
5708 struct action_xlate_ctx ctx;
5714 trace_format_rule(struct ds *result, uint8_t table_id, int level,
5715 const struct rule_dpif *rule)
5717 ds_put_char_multiple(result, '\t', level);
5719 ds_put_cstr(result, "No match\n");
5723 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5724 table_id, ntohll(rule->up.flow_cookie));
5725 cls_rule_format(&rule->up.cr, result);
5726 ds_put_char(result, '\n');
5728 ds_put_char_multiple(result, '\t', level);
5729 ds_put_cstr(result, "OpenFlow ");
5730 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
5731 ds_put_char(result, '\n');
5735 trace_format_flow(struct ds *result, int level, const char *title,
5736 struct ofproto_trace *trace)
5738 ds_put_char_multiple(result, '\t', level);
5739 ds_put_format(result, "%s: ", title);
5740 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
5741 ds_put_cstr(result, "unchanged");
5743 flow_format(result, &trace->ctx.flow);
5744 trace->flow = trace->ctx.flow;
5746 ds_put_char(result, '\n');
5750 trace_format_regs(struct ds *result, int level, const char *title,
5751 struct ofproto_trace *trace)
5755 ds_put_char_multiple(result, '\t', level);
5756 ds_put_format(result, "%s:", title);
5757 for (i = 0; i < FLOW_N_REGS; i++) {
5758 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5760 ds_put_char(result, '\n');
5764 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
5766 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
5767 struct ds *result = trace->result;
5769 ds_put_char(result, '\n');
5770 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
5771 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
5772 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
5776 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5777 void *aux OVS_UNUSED)
5779 const char *dpname = argv[1];
5780 struct ofproto_dpif *ofproto;
5781 struct ofpbuf odp_key;
5782 struct ofpbuf *packet;
5783 struct rule_dpif *rule;
5784 ovs_be16 initial_tci;
5790 ofpbuf_init(&odp_key, 0);
5793 ofproto = ofproto_dpif_lookup(dpname);
5795 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
5799 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
5800 /* ofproto/trace dpname flow [-generate] */
5801 const char *flow_s = argv[2];
5802 const char *generate_s = argv[3];
5805 /* Convert string to datapath key. */
5806 ofpbuf_init(&odp_key, 0);
5807 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
5809 unixctl_command_reply(conn, 501, "Bad flow syntax");
5813 /* Convert odp_key to flow. */
5814 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
5815 odp_key.size, &flow,
5816 &initial_tci, NULL);
5817 if (error == ODP_FIT_ERROR) {
5818 unixctl_command_reply(conn, 501, "Invalid flow");
5822 /* Generate a packet, if requested. */
5824 packet = ofpbuf_new(0);
5825 flow_compose(packet, &flow);
5827 } else if (argc == 6) {
5828 /* ofproto/trace dpname priority tun_id in_port packet */
5829 const char *priority_s = argv[2];
5830 const char *tun_id_s = argv[3];
5831 const char *in_port_s = argv[4];
5832 const char *packet_s = argv[5];
5833 uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s));
5834 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
5835 uint32_t priority = atoi(priority_s);
5838 msg = eth_from_hex(packet_s, &packet);
5840 unixctl_command_reply(conn, 501, msg);
5844 ds_put_cstr(&result, "Packet: ");
5845 s = ofp_packet_to_string(packet->data, packet->size);
5846 ds_put_cstr(&result, s);
5849 flow_extract(packet, priority, tun_id, in_port, &flow);
5850 initial_tci = flow.vlan_tci;
5852 unixctl_command_reply(conn, 501, "Bad command syntax");
5856 ds_put_cstr(&result, "Flow: ");
5857 flow_format(&result, &flow);
5858 ds_put_char(&result, '\n');
5860 rule = rule_dpif_lookup(ofproto, &flow, 0);
5861 trace_format_rule(&result, 0, 0, rule);
5863 struct ofproto_trace trace;
5864 struct ofpbuf *odp_actions;
5866 trace.result = &result;
5868 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, initial_tci,
5869 rule->up.flow_cookie, packet);
5870 trace.ctx.resubmit_hook = trace_resubmit;
5871 odp_actions = xlate_actions(&trace.ctx,
5872 rule->up.actions, rule->up.n_actions);
5874 ds_put_char(&result, '\n');
5875 trace_format_flow(&result, 0, "Final flow", &trace);
5876 ds_put_cstr(&result, "Datapath actions: ");
5877 format_odp_actions(&result, odp_actions->data, odp_actions->size);
5878 ofpbuf_delete(odp_actions);
5880 if (!trace.ctx.may_set_up_flow) {
5882 ds_put_cstr(&result, "\nThis flow is not cachable.");
5884 ds_put_cstr(&result, "\nThe datapath actions are incomplete--"
5885 "for complete actions, please supply a packet.");
5890 unixctl_command_reply(conn, 200, ds_cstr(&result));
5893 ds_destroy(&result);
5894 ofpbuf_delete(packet);
5895 ofpbuf_uninit(&odp_key);
5899 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5900 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5903 unixctl_command_reply(conn, 200, NULL);
5907 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5908 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5911 unixctl_command_reply(conn, 200, NULL);
5915 ofproto_dpif_unixctl_init(void)
5917 static bool registered;
5923 unixctl_command_register(
5925 "bridge {tun_id in_port packet | odp_flow [-generate]}",
5926 2, 5, ofproto_unixctl_trace, NULL);
5927 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
5928 ofproto_unixctl_fdb_flush, NULL);
5929 unixctl_command_register("fdb/show", "bridge", 1, 1,
5930 ofproto_unixctl_fdb_show, NULL);
5931 unixctl_command_register("ofproto/clog", "", 0, 0,
5932 ofproto_dpif_clog, NULL);
5933 unixctl_command_register("ofproto/unclog", "", 0, 0,
5934 ofproto_dpif_unclog, NULL);
5937 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5939 * This is deprecated. It is only for compatibility with broken device drivers
5940 * in old versions of Linux that do not properly support VLANs when VLAN
5941 * devices are not used. When broken device drivers are no longer in
5942 * widespread use, we will delete these interfaces. */
5945 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
5947 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
5948 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
5950 if (realdev_ofp_port == ofport->realdev_ofp_port
5951 && vid == ofport->vlandev_vid) {
5955 ofproto->need_revalidate = true;
5957 if (ofport->realdev_ofp_port) {
5960 if (realdev_ofp_port && ofport->bundle) {
5961 /* vlandevs are enslaved to their realdevs, so they are not allowed to
5962 * themselves be part of a bundle. */
5963 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
5966 ofport->realdev_ofp_port = realdev_ofp_port;
5967 ofport->vlandev_vid = vid;
5969 if (realdev_ofp_port) {
5970 vsp_add(ofport, realdev_ofp_port, vid);
5977 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
5979 return hash_2words(realdev_ofp_port, vid);
5983 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
5984 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
5986 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
5987 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
5988 int vid = vlan_tci_to_vid(vlan_tci);
5989 const struct vlan_splinter *vsp;
5991 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
5992 hash_realdev_vid(realdev_ofp_port, vid),
5993 &ofproto->realdev_vid_map) {
5994 if (vsp->realdev_ofp_port == realdev_ofp_port
5995 && vsp->vid == vid) {
5996 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
6000 return realdev_odp_port;
6003 static struct vlan_splinter *
6004 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
6006 struct vlan_splinter *vsp;
6008 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
6009 &ofproto->vlandev_map) {
6010 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6019 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6020 uint16_t vlandev_ofp_port, int *vid)
6022 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6023 const struct vlan_splinter *vsp;
6025 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6030 return vsp->realdev_ofp_port;
6037 vsp_remove(struct ofport_dpif *port)
6039 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6040 struct vlan_splinter *vsp;
6042 vsp = vlandev_find(ofproto, port->up.ofp_port);
6044 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6045 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6048 port->realdev_ofp_port = 0;
6050 VLOG_ERR("missing vlan device record");
6055 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
6057 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6059 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6060 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6061 == realdev_ofp_port)) {
6062 struct vlan_splinter *vsp;
6064 vsp = xmalloc(sizeof *vsp);
6065 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6066 hash_int(port->up.ofp_port, 0));
6067 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6068 hash_realdev_vid(realdev_ofp_port, vid));
6069 vsp->realdev_ofp_port = realdev_ofp_port;
6070 vsp->vlandev_ofp_port = port->up.ofp_port;
6073 port->realdev_ofp_port = realdev_ofp_port;
6075 VLOG_ERR("duplicate vlan device record");
6079 const struct ofproto_class ofproto_dpif_class = {
6108 port_is_lacp_current,
6109 NULL, /* rule_choose_table */
6116 rule_modify_actions,
6124 get_cfm_remote_mpids,
6128 get_stp_port_status,
6135 is_mirror_output_bundle,
6136 forward_bpdu_changed,