2 * Copyright (c) 2009, 2010, 2011, 2012 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-provider.h"
26 #include "byte-order.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
36 #include "mac-learning.h"
37 #include "multipath.h"
44 #include "ofp-print.h"
45 #include "ofproto-dpif-sflow.h"
46 #include "poll-loop.h"
48 #include "unaligned.h"
50 #include "vlan-bitmap.h"
53 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
55 COVERAGE_DEFINE(ofproto_dpif_ctlr_action);
56 COVERAGE_DEFINE(ofproto_dpif_expired);
57 COVERAGE_DEFINE(ofproto_dpif_no_packet_in);
58 COVERAGE_DEFINE(ofproto_dpif_xlate);
59 COVERAGE_DEFINE(facet_changed_rule);
60 COVERAGE_DEFINE(facet_invalidated);
61 COVERAGE_DEFINE(facet_revalidate);
62 COVERAGE_DEFINE(facet_unexpected);
64 /* Maximum depth of flow table recursion (due to resubmit actions) in a
65 * flow translation. */
66 #define MAX_RESUBMIT_RECURSION 32
68 /* Number of implemented OpenFlow tables. */
69 enum { N_TABLES = 255 };
70 BUILD_ASSERT_DECL(N_TABLES >= 1 && N_TABLES <= 255);
78 long long int used; /* Time last used; time created if not used. */
82 * - Do include packets and bytes from facets that have been deleted or
83 * whose own statistics have been folded into the rule.
85 * - Do include packets and bytes sent "by hand" that were accounted to
86 * the rule without any facet being involved (this is a rare corner
87 * case in rule_execute()).
89 * - Do not include packet or bytes that can be obtained from any facet's
90 * packet_count or byte_count member or that can be obtained from the
91 * datapath by, e.g., dpif_flow_get() for any subfacet.
93 uint64_t packet_count; /* Number of packets received. */
94 uint64_t byte_count; /* Number of bytes received. */
96 tag_type tag; /* Caches rule_calculate_tag() result. */
98 struct list facets; /* List of "struct facet"s. */
101 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
103 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
106 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
107 const struct flow *, uint8_t table);
109 static void flow_push_stats(const struct rule_dpif *, const struct flow *,
110 uint64_t packets, uint64_t bytes,
113 static uint32_t rule_calculate_tag(const struct flow *,
114 const struct flow_wildcards *,
116 static void rule_invalidate(const struct rule_dpif *);
118 #define MAX_MIRRORS 32
119 typedef uint32_t mirror_mask_t;
120 #define MIRROR_MASK_C(X) UINT32_C(X)
121 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
123 struct ofproto_dpif *ofproto; /* Owning ofproto. */
124 size_t idx; /* In ofproto's "mirrors" array. */
125 void *aux; /* Key supplied by ofproto's client. */
126 char *name; /* Identifier for log messages. */
128 /* Selection criteria. */
129 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
130 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
131 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
133 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
134 struct ofbundle *out; /* Output port or NULL. */
135 int out_vlan; /* Output VLAN or -1. */
136 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
139 int64_t packet_count; /* Number of packets sent. */
140 int64_t byte_count; /* Number of bytes sent. */
143 static void mirror_destroy(struct ofmirror *);
144 static void update_mirror_stats(struct ofproto_dpif *ofproto,
145 mirror_mask_t mirrors,
146 uint64_t packets, uint64_t bytes);
149 struct ofproto_dpif *ofproto; /* Owning ofproto. */
150 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
151 void *aux; /* Key supplied by ofproto's client. */
152 char *name; /* Identifier for log messages. */
155 struct list ports; /* Contains "struct ofport"s. */
156 enum port_vlan_mode vlan_mode; /* VLAN mode */
157 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
158 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
159 * NULL if all VLANs are trunked. */
160 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
161 struct bond *bond; /* Nonnull iff more than one port. */
162 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
165 bool floodable; /* True if no port has OFPPC_NO_FLOOD set. */
167 /* Port mirroring info. */
168 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
169 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
170 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
173 static void bundle_remove(struct ofport *);
174 static void bundle_update(struct ofbundle *);
175 static void bundle_destroy(struct ofbundle *);
176 static void bundle_del_port(struct ofport_dpif *);
177 static void bundle_run(struct ofbundle *);
178 static void bundle_wait(struct ofbundle *);
179 static struct ofbundle *lookup_input_bundle(struct ofproto_dpif *,
180 uint16_t in_port, bool warn);
182 /* A controller may use OFPP_NONE as the ingress port to indicate that
183 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
184 * when an input bundle is needed for validation (e.g., mirroring or
185 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
186 * any 'port' structs, so care must be taken when dealing with it. */
187 static struct ofbundle ofpp_none_bundle = {
189 .vlan_mode = PORT_VLAN_TRUNK
192 static void stp_run(struct ofproto_dpif *ofproto);
193 static void stp_wait(struct ofproto_dpif *ofproto);
195 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
197 struct action_xlate_ctx {
198 /* action_xlate_ctx_init() initializes these members. */
201 struct ofproto_dpif *ofproto;
203 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
204 * this flow when actions change header fields. */
207 /* The packet corresponding to 'flow', or a null pointer if we are
208 * revalidating without a packet to refer to. */
209 const struct ofpbuf *packet;
211 /* Should OFPP_NORMAL MAC learning and NXAST_LEARN actions execute? We
212 * want to execute them if we are actually processing a packet, or if we
213 * are accounting for packets that the datapath has processed, but not if
214 * we are just revalidating. */
217 /* If nonnull, called just before executing a resubmit action.
219 * This is normally null so the client has to set it manually after
220 * calling action_xlate_ctx_init(). */
221 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *);
223 /* xlate_actions() initializes and uses these members. The client might want
224 * to look at them after it returns. */
226 struct ofpbuf *odp_actions; /* Datapath actions. */
227 tag_type tags; /* Tags associated with actions. */
228 bool may_set_up_flow; /* True ordinarily; false if the actions must
229 * be reassessed for every packet. */
230 bool has_learn; /* Actions include NXAST_LEARN? */
231 bool has_normal; /* Actions output to OFPP_NORMAL? */
232 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
233 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
235 /* xlate_actions() initializes and uses these members, but the client has no
236 * reason to look at them. */
238 int recurse; /* Recursion level, via xlate_table_action. */
239 struct flow base_flow; /* Flow at the last commit. */
240 uint32_t orig_skb_priority; /* Priority when packet arrived. */
241 uint8_t table_id; /* OpenFlow table ID where flow was found. */
242 uint32_t sflow_n_outputs; /* Number of output ports. */
243 uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
244 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
245 bool exit; /* No further actions should be processed. */
248 static void action_xlate_ctx_init(struct action_xlate_ctx *,
249 struct ofproto_dpif *, const struct flow *,
250 ovs_be16 initial_tci, const struct ofpbuf *);
251 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
252 const union ofp_action *in, size_t n_in);
254 /* An exact-match instantiation of an OpenFlow flow.
256 * A facet associates a "struct flow", which represents the Open vSwitch
257 * userspace idea of an exact-match flow, with one or more subfacets. Each
258 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
259 * the facet. When the kernel module (or other dpif implementation) and Open
260 * vSwitch userspace agree on the definition of a flow key, there is exactly
261 * one subfacet per facet. If the dpif implementation supports more-specific
262 * flow matching than userspace, however, a facet can have more than one
263 * subfacet, each of which corresponds to some distinction in flow that
264 * userspace simply doesn't understand.
266 * Flow expiration works in terms of subfacets, so a facet must have at least
267 * one subfacet or it will never expire, leaking memory. */
270 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
271 struct list list_node; /* In owning rule's 'facets' list. */
272 struct rule_dpif *rule; /* Owning rule. */
275 struct list subfacets;
276 long long int used; /* Time last used; time created if not used. */
283 * - Do include packets and bytes sent "by hand", e.g. with
286 * - Do include packets and bytes that were obtained from the datapath
287 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
288 * DPIF_FP_ZERO_STATS).
290 * - Do not include packets or bytes that can be obtained from the
291 * datapath for any existing subfacet.
293 uint64_t packet_count; /* Number of packets received. */
294 uint64_t byte_count; /* Number of bytes received. */
296 /* Resubmit statistics. */
297 uint64_t prev_packet_count; /* Number of packets from last stats push. */
298 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
299 long long int prev_used; /* Used time from last stats push. */
302 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
303 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
305 /* Properties of datapath actions.
307 * Every subfacet has its own actions because actions can differ slightly
308 * between splintered and non-splintered subfacets due to the VLAN tag
309 * being initially different (present vs. absent). All of them have these
310 * properties in common so we just store one copy of them here. */
311 bool may_install; /* Reassess actions for every packet? */
312 bool has_learn; /* Actions include NXAST_LEARN? */
313 bool has_normal; /* Actions output to OFPP_NORMAL? */
314 tag_type tags; /* Tags that would require revalidation. */
315 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
318 static struct facet *facet_create(struct rule_dpif *, const struct flow *);
319 static void facet_remove(struct ofproto_dpif *, struct facet *);
320 static void facet_free(struct facet *);
322 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
323 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
324 const struct flow *);
325 static bool facet_revalidate(struct ofproto_dpif *, struct facet *);
327 static bool execute_controller_action(struct ofproto_dpif *,
329 const struct nlattr *odp_actions,
331 struct ofpbuf *packet, bool clone);
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();
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 stp_destroy(ofproto->stp);
1142 ofproto->stp = NULL;
1149 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1151 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1155 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1156 s->designated_root = stp_get_designated_root(ofproto->stp);
1157 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1166 update_stp_port_state(struct ofport_dpif *ofport)
1168 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1169 enum stp_state state;
1171 /* Figure out new state. */
1172 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1176 if (ofport->stp_state != state) {
1180 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1181 netdev_get_name(ofport->up.netdev),
1182 stp_state_name(ofport->stp_state),
1183 stp_state_name(state));
1184 if (stp_learn_in_state(ofport->stp_state)
1185 != stp_learn_in_state(state)) {
1186 /* xxx Learning action flows should also be flushed. */
1187 mac_learning_flush(ofproto->ml);
1189 fwd_change = stp_forward_in_state(ofport->stp_state)
1190 != stp_forward_in_state(state);
1192 ofproto->need_revalidate = true;
1193 ofport->stp_state = state;
1194 ofport->stp_state_entered = time_msec();
1196 if (fwd_change && ofport->bundle) {
1197 bundle_update(ofport->bundle);
1200 /* Update the STP state bits in the OpenFlow port description. */
1201 of_state = (ofport->up.opp.state & htonl(~OFPPS_STP_MASK))
1202 | htonl(state == STP_LISTENING ? OFPPS_STP_LISTEN
1203 : state == STP_LEARNING ? OFPPS_STP_LEARN
1204 : state == STP_FORWARDING ? OFPPS_STP_FORWARD
1205 : state == STP_BLOCKING ? OFPPS_STP_BLOCK
1207 ofproto_port_set_state(&ofport->up, of_state);
1211 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1212 * caller is responsible for assigning STP port numbers and ensuring
1213 * there are no duplicates. */
1215 set_stp_port(struct ofport *ofport_,
1216 const struct ofproto_port_stp_settings *s)
1218 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1219 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1220 struct stp_port *sp = ofport->stp_port;
1222 if (!s || !s->enable) {
1224 ofport->stp_port = NULL;
1225 stp_port_disable(sp);
1226 update_stp_port_state(ofport);
1229 } else if (sp && stp_port_no(sp) != s->port_num
1230 && ofport == stp_port_get_aux(sp)) {
1231 /* The port-id changed, so disable the old one if it's not
1232 * already in use by another port. */
1233 stp_port_disable(sp);
1236 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1237 stp_port_enable(sp);
1239 stp_port_set_aux(sp, ofport);
1240 stp_port_set_priority(sp, s->priority);
1241 stp_port_set_path_cost(sp, s->path_cost);
1243 update_stp_port_state(ofport);
1249 get_stp_port_status(struct ofport *ofport_,
1250 struct ofproto_port_stp_status *s)
1252 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1253 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1254 struct stp_port *sp = ofport->stp_port;
1256 if (!ofproto->stp || !sp) {
1262 s->port_id = stp_port_get_id(sp);
1263 s->state = stp_port_get_state(sp);
1264 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1265 s->role = stp_port_get_role(sp);
1266 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1272 stp_run(struct ofproto_dpif *ofproto)
1275 long long int now = time_msec();
1276 long long int elapsed = now - ofproto->stp_last_tick;
1277 struct stp_port *sp;
1280 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1281 ofproto->stp_last_tick = now;
1283 while (stp_get_changed_port(ofproto->stp, &sp)) {
1284 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1287 update_stp_port_state(ofport);
1294 stp_wait(struct ofproto_dpif *ofproto)
1297 poll_timer_wait(1000);
1301 /* Returns true if STP should process 'flow'. */
1303 stp_should_process_flow(const struct flow *flow)
1305 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1309 stp_process_packet(const struct ofport_dpif *ofport,
1310 const struct ofpbuf *packet)
1312 struct ofpbuf payload = *packet;
1313 struct eth_header *eth = payload.data;
1314 struct stp_port *sp = ofport->stp_port;
1316 /* Sink packets on ports that have STP disabled when the bridge has
1318 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1322 /* Trim off padding on payload. */
1323 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1324 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1327 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1328 stp_received_bpdu(sp, payload.data, payload.size);
1332 static struct priority_to_dscp *
1333 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1335 struct priority_to_dscp *pdscp;
1338 hash = hash_int(priority, 0);
1339 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1340 if (pdscp->priority == priority) {
1348 ofport_clear_priorities(struct ofport_dpif *ofport)
1350 struct priority_to_dscp *pdscp, *next;
1352 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1353 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1359 set_queues(struct ofport *ofport_,
1360 const struct ofproto_port_queue *qdscp_list,
1363 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1364 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1365 struct hmap new = HMAP_INITIALIZER(&new);
1368 for (i = 0; i < n_qdscp; i++) {
1369 struct priority_to_dscp *pdscp;
1373 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1374 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1379 pdscp = get_priority(ofport, priority);
1381 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1383 pdscp = xmalloc(sizeof *pdscp);
1384 pdscp->priority = priority;
1386 ofproto->need_revalidate = true;
1389 if (pdscp->dscp != dscp) {
1391 ofproto->need_revalidate = true;
1394 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1397 if (!hmap_is_empty(&ofport->priorities)) {
1398 ofport_clear_priorities(ofport);
1399 ofproto->need_revalidate = true;
1402 hmap_swap(&new, &ofport->priorities);
1410 /* Expires all MAC learning entries associated with 'bundle' and forces its
1411 * ofproto to revalidate every flow.
1413 * Normally MAC learning entries are removed only from the ofproto associated
1414 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1415 * are removed from every ofproto. When patch ports and SLB bonds are in use
1416 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1417 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1418 * with the host from which it migrated. */
1420 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1422 struct ofproto_dpif *ofproto = bundle->ofproto;
1423 struct mac_learning *ml = ofproto->ml;
1424 struct mac_entry *mac, *next_mac;
1426 ofproto->need_revalidate = true;
1427 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1428 if (mac->port.p == bundle) {
1430 struct ofproto_dpif *o;
1432 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1434 struct mac_entry *e;
1436 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1439 tag_set_add(&o->revalidate_set, e->tag);
1440 mac_learning_expire(o->ml, e);
1446 mac_learning_expire(ml, mac);
1451 static struct ofbundle *
1452 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1454 struct ofbundle *bundle;
1456 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1457 &ofproto->bundles) {
1458 if (bundle->aux == aux) {
1465 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1466 * ones that are found to 'bundles'. */
1468 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1469 void **auxes, size_t n_auxes,
1470 struct hmapx *bundles)
1474 hmapx_init(bundles);
1475 for (i = 0; i < n_auxes; i++) {
1476 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1478 hmapx_add(bundles, bundle);
1484 bundle_update(struct ofbundle *bundle)
1486 struct ofport_dpif *port;
1488 bundle->floodable = true;
1489 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1490 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
1491 bundle->floodable = false;
1498 bundle_del_port(struct ofport_dpif *port)
1500 struct ofbundle *bundle = port->bundle;
1502 bundle->ofproto->need_revalidate = true;
1504 list_remove(&port->bundle_node);
1505 port->bundle = NULL;
1508 lacp_slave_unregister(bundle->lacp, port);
1511 bond_slave_unregister(bundle->bond, port);
1514 bundle_update(bundle);
1518 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1519 struct lacp_slave_settings *lacp,
1520 uint32_t bond_stable_id)
1522 struct ofport_dpif *port;
1524 port = get_ofp_port(bundle->ofproto, ofp_port);
1529 if (port->bundle != bundle) {
1530 bundle->ofproto->need_revalidate = true;
1532 bundle_del_port(port);
1535 port->bundle = bundle;
1536 list_push_back(&bundle->ports, &port->bundle_node);
1537 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
1538 bundle->floodable = false;
1542 port->bundle->ofproto->need_revalidate = true;
1543 lacp_slave_register(bundle->lacp, port, lacp);
1546 port->bond_stable_id = bond_stable_id;
1552 bundle_destroy(struct ofbundle *bundle)
1554 struct ofproto_dpif *ofproto;
1555 struct ofport_dpif *port, *next_port;
1562 ofproto = bundle->ofproto;
1563 for (i = 0; i < MAX_MIRRORS; i++) {
1564 struct ofmirror *m = ofproto->mirrors[i];
1566 if (m->out == bundle) {
1568 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1569 || hmapx_find_and_delete(&m->dsts, bundle)) {
1570 ofproto->need_revalidate = true;
1575 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1576 bundle_del_port(port);
1579 bundle_flush_macs(bundle, true);
1580 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1582 free(bundle->trunks);
1583 lacp_destroy(bundle->lacp);
1584 bond_destroy(bundle->bond);
1589 bundle_set(struct ofproto *ofproto_, void *aux,
1590 const struct ofproto_bundle_settings *s)
1592 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1593 bool need_flush = false;
1594 struct ofport_dpif *port;
1595 struct ofbundle *bundle;
1596 unsigned long *trunks;
1602 bundle_destroy(bundle_lookup(ofproto, aux));
1606 assert(s->n_slaves == 1 || s->bond != NULL);
1607 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1609 bundle = bundle_lookup(ofproto, aux);
1611 bundle = xmalloc(sizeof *bundle);
1613 bundle->ofproto = ofproto;
1614 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1615 hash_pointer(aux, 0));
1617 bundle->name = NULL;
1619 list_init(&bundle->ports);
1620 bundle->vlan_mode = PORT_VLAN_TRUNK;
1622 bundle->trunks = NULL;
1623 bundle->use_priority_tags = s->use_priority_tags;
1624 bundle->lacp = NULL;
1625 bundle->bond = NULL;
1627 bundle->floodable = true;
1629 bundle->src_mirrors = 0;
1630 bundle->dst_mirrors = 0;
1631 bundle->mirror_out = 0;
1634 if (!bundle->name || strcmp(s->name, bundle->name)) {
1636 bundle->name = xstrdup(s->name);
1641 if (!bundle->lacp) {
1642 ofproto->need_revalidate = true;
1643 bundle->lacp = lacp_create();
1645 lacp_configure(bundle->lacp, s->lacp);
1647 lacp_destroy(bundle->lacp);
1648 bundle->lacp = NULL;
1651 /* Update set of ports. */
1653 for (i = 0; i < s->n_slaves; i++) {
1654 if (!bundle_add_port(bundle, s->slaves[i],
1655 s->lacp ? &s->lacp_slaves[i] : NULL,
1656 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1660 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1661 struct ofport_dpif *next_port;
1663 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1664 for (i = 0; i < s->n_slaves; i++) {
1665 if (s->slaves[i] == port->up.ofp_port) {
1670 bundle_del_port(port);
1674 assert(list_size(&bundle->ports) <= s->n_slaves);
1676 if (list_is_empty(&bundle->ports)) {
1677 bundle_destroy(bundle);
1681 /* Set VLAN tagging mode */
1682 if (s->vlan_mode != bundle->vlan_mode
1683 || s->use_priority_tags != bundle->use_priority_tags) {
1684 bundle->vlan_mode = s->vlan_mode;
1685 bundle->use_priority_tags = s->use_priority_tags;
1690 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1691 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1693 if (vlan != bundle->vlan) {
1694 bundle->vlan = vlan;
1698 /* Get trunked VLANs. */
1699 switch (s->vlan_mode) {
1700 case PORT_VLAN_ACCESS:
1704 case PORT_VLAN_TRUNK:
1705 trunks = (unsigned long *) s->trunks;
1708 case PORT_VLAN_NATIVE_UNTAGGED:
1709 case PORT_VLAN_NATIVE_TAGGED:
1710 if (vlan != 0 && (!s->trunks
1711 || !bitmap_is_set(s->trunks, vlan)
1712 || bitmap_is_set(s->trunks, 0))) {
1713 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1715 trunks = bitmap_clone(s->trunks, 4096);
1717 trunks = bitmap_allocate1(4096);
1719 bitmap_set1(trunks, vlan);
1720 bitmap_set0(trunks, 0);
1722 trunks = (unsigned long *) s->trunks;
1729 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1730 free(bundle->trunks);
1731 if (trunks == s->trunks) {
1732 bundle->trunks = vlan_bitmap_clone(trunks);
1734 bundle->trunks = trunks;
1739 if (trunks != s->trunks) {
1744 if (!list_is_short(&bundle->ports)) {
1745 bundle->ofproto->has_bonded_bundles = true;
1747 if (bond_reconfigure(bundle->bond, s->bond)) {
1748 ofproto->need_revalidate = true;
1751 bundle->bond = bond_create(s->bond);
1752 ofproto->need_revalidate = true;
1755 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1756 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1760 bond_destroy(bundle->bond);
1761 bundle->bond = NULL;
1764 /* If we changed something that would affect MAC learning, un-learn
1765 * everything on this port and force flow revalidation. */
1767 bundle_flush_macs(bundle, false);
1774 bundle_remove(struct ofport *port_)
1776 struct ofport_dpif *port = ofport_dpif_cast(port_);
1777 struct ofbundle *bundle = port->bundle;
1780 bundle_del_port(port);
1781 if (list_is_empty(&bundle->ports)) {
1782 bundle_destroy(bundle);
1783 } else if (list_is_short(&bundle->ports)) {
1784 bond_destroy(bundle->bond);
1785 bundle->bond = NULL;
1791 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
1793 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1794 struct ofport_dpif *port = port_;
1795 uint8_t ea[ETH_ADDR_LEN];
1798 error = netdev_get_etheraddr(port->up.netdev, ea);
1800 struct ofpbuf packet;
1803 ofpbuf_init(&packet, 0);
1804 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1806 memcpy(packet_pdu, pdu, pdu_size);
1808 send_packet(port, &packet);
1809 ofpbuf_uninit(&packet);
1811 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1812 "%s (%s)", port->bundle->name,
1813 netdev_get_name(port->up.netdev), strerror(error));
1818 bundle_send_learning_packets(struct ofbundle *bundle)
1820 struct ofproto_dpif *ofproto = bundle->ofproto;
1821 int error, n_packets, n_errors;
1822 struct mac_entry *e;
1824 error = n_packets = n_errors = 0;
1825 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1826 if (e->port.p != bundle) {
1827 struct ofpbuf *learning_packet;
1828 struct ofport_dpif *port;
1832 /* The assignment to "port" is unnecessary but makes "grep"ing for
1833 * struct ofport_dpif more effective. */
1834 learning_packet = bond_compose_learning_packet(bundle->bond,
1838 ret = send_packet(port, learning_packet);
1839 ofpbuf_delete(learning_packet);
1849 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1850 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1851 "packets, last error was: %s",
1852 bundle->name, n_errors, n_packets, strerror(error));
1854 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1855 bundle->name, n_packets);
1860 bundle_run(struct ofbundle *bundle)
1863 lacp_run(bundle->lacp, send_pdu_cb);
1866 struct ofport_dpif *port;
1868 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1869 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1872 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1873 lacp_negotiated(bundle->lacp));
1874 if (bond_should_send_learning_packets(bundle->bond)) {
1875 bundle_send_learning_packets(bundle);
1881 bundle_wait(struct ofbundle *bundle)
1884 lacp_wait(bundle->lacp);
1887 bond_wait(bundle->bond);
1894 mirror_scan(struct ofproto_dpif *ofproto)
1898 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1899 if (!ofproto->mirrors[idx]) {
1906 static struct ofmirror *
1907 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1911 for (i = 0; i < MAX_MIRRORS; i++) {
1912 struct ofmirror *mirror = ofproto->mirrors[i];
1913 if (mirror && mirror->aux == aux) {
1921 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
1923 mirror_update_dups(struct ofproto_dpif *ofproto)
1927 for (i = 0; i < MAX_MIRRORS; i++) {
1928 struct ofmirror *m = ofproto->mirrors[i];
1931 m->dup_mirrors = MIRROR_MASK_C(1) << i;
1935 for (i = 0; i < MAX_MIRRORS; i++) {
1936 struct ofmirror *m1 = ofproto->mirrors[i];
1943 for (j = i + 1; j < MAX_MIRRORS; j++) {
1944 struct ofmirror *m2 = ofproto->mirrors[j];
1946 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
1947 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
1948 m2->dup_mirrors |= m1->dup_mirrors;
1955 mirror_set(struct ofproto *ofproto_, void *aux,
1956 const struct ofproto_mirror_settings *s)
1958 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1959 mirror_mask_t mirror_bit;
1960 struct ofbundle *bundle;
1961 struct ofmirror *mirror;
1962 struct ofbundle *out;
1963 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
1964 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
1967 mirror = mirror_lookup(ofproto, aux);
1969 mirror_destroy(mirror);
1975 idx = mirror_scan(ofproto);
1977 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
1979 ofproto->up.name, MAX_MIRRORS, s->name);
1983 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
1984 mirror->ofproto = ofproto;
1987 mirror->out_vlan = -1;
1988 mirror->name = NULL;
1991 if (!mirror->name || strcmp(s->name, mirror->name)) {
1993 mirror->name = xstrdup(s->name);
1996 /* Get the new configuration. */
1997 if (s->out_bundle) {
1998 out = bundle_lookup(ofproto, s->out_bundle);
2000 mirror_destroy(mirror);
2006 out_vlan = s->out_vlan;
2008 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2009 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2011 /* If the configuration has not changed, do nothing. */
2012 if (hmapx_equals(&srcs, &mirror->srcs)
2013 && hmapx_equals(&dsts, &mirror->dsts)
2014 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2015 && mirror->out == out
2016 && mirror->out_vlan == out_vlan)
2018 hmapx_destroy(&srcs);
2019 hmapx_destroy(&dsts);
2023 hmapx_swap(&srcs, &mirror->srcs);
2024 hmapx_destroy(&srcs);
2026 hmapx_swap(&dsts, &mirror->dsts);
2027 hmapx_destroy(&dsts);
2029 free(mirror->vlans);
2030 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2033 mirror->out_vlan = out_vlan;
2035 /* Update bundles. */
2036 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2037 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2038 if (hmapx_contains(&mirror->srcs, bundle)) {
2039 bundle->src_mirrors |= mirror_bit;
2041 bundle->src_mirrors &= ~mirror_bit;
2044 if (hmapx_contains(&mirror->dsts, bundle)) {
2045 bundle->dst_mirrors |= mirror_bit;
2047 bundle->dst_mirrors &= ~mirror_bit;
2050 if (mirror->out == bundle) {
2051 bundle->mirror_out |= mirror_bit;
2053 bundle->mirror_out &= ~mirror_bit;
2057 ofproto->need_revalidate = true;
2058 mac_learning_flush(ofproto->ml);
2059 mirror_update_dups(ofproto);
2065 mirror_destroy(struct ofmirror *mirror)
2067 struct ofproto_dpif *ofproto;
2068 mirror_mask_t mirror_bit;
2069 struct ofbundle *bundle;
2075 ofproto = mirror->ofproto;
2076 ofproto->need_revalidate = true;
2077 mac_learning_flush(ofproto->ml);
2079 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2080 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2081 bundle->src_mirrors &= ~mirror_bit;
2082 bundle->dst_mirrors &= ~mirror_bit;
2083 bundle->mirror_out &= ~mirror_bit;
2086 hmapx_destroy(&mirror->srcs);
2087 hmapx_destroy(&mirror->dsts);
2088 free(mirror->vlans);
2090 ofproto->mirrors[mirror->idx] = NULL;
2094 mirror_update_dups(ofproto);
2098 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2099 uint64_t *packets, uint64_t *bytes)
2101 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2102 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2105 *packets = *bytes = UINT64_MAX;
2109 *packets = mirror->packet_count;
2110 *bytes = mirror->byte_count;
2116 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2118 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2119 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2120 ofproto->need_revalidate = true;
2121 mac_learning_flush(ofproto->ml);
2127 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2129 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2130 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2131 return bundle && bundle->mirror_out != 0;
2135 forward_bpdu_changed(struct ofproto *ofproto_)
2137 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2138 /* Revalidate cached flows whenever forward_bpdu option changes. */
2139 ofproto->need_revalidate = true;
2144 static struct ofport_dpif *
2145 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
2147 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2148 return ofport ? ofport_dpif_cast(ofport) : NULL;
2151 static struct ofport_dpif *
2152 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
2154 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2158 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2159 struct dpif_port *dpif_port)
2161 ofproto_port->name = dpif_port->name;
2162 ofproto_port->type = dpif_port->type;
2163 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2167 port_run(struct ofport_dpif *ofport)
2169 bool enable = netdev_get_carrier(ofport->up.netdev);
2172 cfm_run(ofport->cfm);
2174 if (cfm_should_send_ccm(ofport->cfm)) {
2175 struct ofpbuf packet;
2177 ofpbuf_init(&packet, 0);
2178 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr);
2179 send_packet(ofport, &packet);
2180 ofpbuf_uninit(&packet);
2183 enable = enable && !cfm_get_fault(ofport->cfm)
2184 && cfm_get_opup(ofport->cfm);
2187 if (ofport->bundle) {
2188 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2191 if (ofport->may_enable != enable) {
2192 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2194 if (ofproto->has_bundle_action) {
2195 ofproto->need_revalidate = true;
2199 ofport->may_enable = enable;
2203 port_wait(struct ofport_dpif *ofport)
2206 cfm_wait(ofport->cfm);
2211 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2212 struct ofproto_port *ofproto_port)
2214 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2215 struct dpif_port dpif_port;
2218 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2220 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2226 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2228 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2232 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2234 *ofp_portp = odp_port_to_ofp_port(odp_port);
2240 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2242 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2245 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2247 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2249 /* The caller is going to close ofport->up.netdev. If this is a
2250 * bonded port, then the bond is using that netdev, so remove it
2251 * from the bond. The client will need to reconfigure everything
2252 * after deleting ports, so then the slave will get re-added. */
2253 bundle_remove(&ofport->up);
2260 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2262 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2265 error = netdev_get_stats(ofport->up.netdev, stats);
2267 if (!error && ofport->odp_port == OVSP_LOCAL) {
2268 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2270 /* ofproto->stats.tx_packets represents packets that we created
2271 * internally and sent to some port (e.g. packets sent with
2272 * send_packet()). Account for them as if they had come from
2273 * OFPP_LOCAL and got forwarded. */
2275 if (stats->rx_packets != UINT64_MAX) {
2276 stats->rx_packets += ofproto->stats.tx_packets;
2279 if (stats->rx_bytes != UINT64_MAX) {
2280 stats->rx_bytes += ofproto->stats.tx_bytes;
2283 /* ofproto->stats.rx_packets represents packets that were received on
2284 * some port and we processed internally and dropped (e.g. STP).
2285 * Account fro them as if they had been forwarded to OFPP_LOCAL. */
2287 if (stats->tx_packets != UINT64_MAX) {
2288 stats->tx_packets += ofproto->stats.rx_packets;
2291 if (stats->tx_bytes != UINT64_MAX) {
2292 stats->tx_bytes += ofproto->stats.rx_bytes;
2299 /* Account packets for LOCAL port. */
2301 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2302 size_t tx_size, size_t rx_size)
2304 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2307 ofproto->stats.rx_packets++;
2308 ofproto->stats.rx_bytes += rx_size;
2311 ofproto->stats.tx_packets++;
2312 ofproto->stats.tx_bytes += tx_size;
2316 struct port_dump_state {
2317 struct dpif_port_dump dump;
2322 port_dump_start(const struct ofproto *ofproto_, void **statep)
2324 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2325 struct port_dump_state *state;
2327 *statep = state = xmalloc(sizeof *state);
2328 dpif_port_dump_start(&state->dump, ofproto->dpif);
2329 state->done = false;
2334 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2335 struct ofproto_port *port)
2337 struct port_dump_state *state = state_;
2338 struct dpif_port dpif_port;
2340 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2341 ofproto_port_from_dpif_port(port, &dpif_port);
2344 int error = dpif_port_dump_done(&state->dump);
2346 return error ? error : EOF;
2351 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2353 struct port_dump_state *state = state_;
2356 dpif_port_dump_done(&state->dump);
2363 port_poll(const struct ofproto *ofproto_, char **devnamep)
2365 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2366 return dpif_port_poll(ofproto->dpif, devnamep);
2370 port_poll_wait(const struct ofproto *ofproto_)
2372 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2373 dpif_port_poll_wait(ofproto->dpif);
2377 port_is_lacp_current(const struct ofport *ofport_)
2379 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2380 return (ofport->bundle && ofport->bundle->lacp
2381 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2385 /* Upcall handling. */
2387 /* Flow miss batching.
2389 * Some dpifs implement operations faster when you hand them off in a batch.
2390 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2391 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2392 * more packets, plus possibly installing the flow in the dpif.
2394 * So far we only batch the operations that affect flow setup time the most.
2395 * It's possible to batch more than that, but the benefit might be minimal. */
2397 struct hmap_node hmap_node;
2399 enum odp_key_fitness key_fitness;
2400 const struct nlattr *key;
2402 ovs_be16 initial_tci;
2403 struct list packets;
2406 struct flow_miss_op {
2407 union dpif_op dpif_op;
2408 struct subfacet *subfacet;
2411 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2412 * OpenFlow controller as necessary according to their individual
2415 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
2416 * ownership is transferred to this function. */
2418 send_packet_in_miss(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
2419 const struct flow *flow, bool clone)
2421 struct ofputil_packet_in pin;
2423 pin.packet = packet;
2424 pin.in_port = flow->in_port;
2425 pin.reason = OFPR_NO_MATCH;
2426 pin.buffer_id = 0; /* not yet known */
2427 pin.send_len = 0; /* not used for flow table misses */
2428 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
2429 clone ? NULL : packet);
2432 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_ACTION to each
2433 * OpenFlow controller as necessary according to their individual
2436 * 'send_len' should be the number of bytes of 'packet' to send to the
2437 * controller, as specified in the action that caused the packet to be sent.
2439 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
2440 * Otherwise, ownership is transferred to this function. */
2442 send_packet_in_action(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
2443 uint64_t userdata, const struct flow *flow, bool clone)
2445 struct ofputil_packet_in pin;
2446 struct user_action_cookie cookie;
2448 memcpy(&cookie, &userdata, sizeof(cookie));
2450 pin.packet = packet;
2451 pin.in_port = flow->in_port;
2452 pin.reason = OFPR_ACTION;
2453 pin.buffer_id = 0; /* not yet known */
2454 pin.send_len = cookie.data;
2455 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
2456 clone ? NULL : packet);
2460 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2461 const struct ofpbuf *packet)
2463 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2469 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2471 cfm_process_heartbeat(ofport->cfm, packet);
2474 } else if (ofport->bundle && ofport->bundle->lacp
2475 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2477 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2480 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2482 stp_process_packet(ofport, packet);
2489 static struct flow_miss *
2490 flow_miss_create(struct hmap *todo, const struct flow *flow,
2491 enum odp_key_fitness key_fitness,
2492 const struct nlattr *key, size_t key_len,
2493 ovs_be16 initial_tci)
2495 uint32_t hash = flow_hash(flow, 0);
2496 struct flow_miss *miss;
2498 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2499 if (flow_equal(&miss->flow, flow)) {
2504 miss = xmalloc(sizeof *miss);
2505 hmap_insert(todo, &miss->hmap_node, hash);
2507 miss->key_fitness = key_fitness;
2509 miss->key_len = key_len;
2510 miss->initial_tci = initial_tci;
2511 list_init(&miss->packets);
2516 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2517 struct flow_miss_op *ops, size_t *n_ops)
2519 const struct flow *flow = &miss->flow;
2520 struct ofpbuf *packet, *next_packet;
2521 struct subfacet *subfacet;
2522 struct facet *facet;
2524 facet = facet_lookup_valid(ofproto, flow);
2526 struct rule_dpif *rule;
2528 rule = rule_dpif_lookup(ofproto, flow, 0);
2530 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
2531 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
2533 if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) {
2534 COVERAGE_INC(ofproto_dpif_no_packet_in);
2535 /* XXX install 'drop' flow entry */
2539 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
2543 LIST_FOR_EACH_SAFE (packet, next_packet, list_node,
2545 list_remove(&packet->list_node);
2546 send_packet_in_miss(ofproto, packet, flow, false);
2552 facet = facet_create(rule, flow);
2555 subfacet = subfacet_create(ofproto, facet,
2556 miss->key_fitness, miss->key, miss->key_len,
2559 LIST_FOR_EACH_SAFE (packet, next_packet, list_node, &miss->packets) {
2560 struct dpif_flow_stats stats;
2562 list_remove(&packet->list_node);
2563 ofproto->n_matches++;
2565 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2567 * Extra-special case for fail-open mode.
2569 * We are in fail-open mode and the packet matched the fail-open
2570 * rule, but we are connected to a controller too. We should send
2571 * the packet up to the controller in the hope that it will try to
2572 * set up a flow and thereby allow us to exit fail-open.
2574 * See the top-level comment in fail-open.c for more information.
2576 send_packet_in_miss(ofproto, packet, flow, true);
2579 if (!facet->may_install || !subfacet->actions) {
2580 subfacet_make_actions(ofproto, subfacet, packet);
2583 /* Credit statistics to subfacet for this packet. We must do this now
2584 * because execute_controller_action() below may destroy 'packet'. */
2585 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2586 subfacet_update_stats(ofproto, subfacet, &stats);
2588 if (!execute_controller_action(ofproto, &facet->flow,
2590 subfacet->actions_len, packet, true)
2591 && subfacet->actions_len > 0) {
2592 struct flow_miss_op *op = &ops[(*n_ops)++];
2593 struct dpif_execute *execute = &op->dpif_op.execute;
2595 if (flow->vlan_tci != subfacet->initial_tci) {
2596 /* This packet was received on a VLAN splinter port. We added
2597 * a VLAN to the packet to make the packet resemble the flow,
2598 * but the actions were composed assuming that the packet
2599 * contained no VLAN. So, we must remove the VLAN header from
2600 * the packet before trying to execute the actions. */
2601 eth_pop_vlan(packet);
2604 op->subfacet = subfacet;
2605 execute->type = DPIF_OP_EXECUTE;
2606 execute->key = miss->key;
2607 execute->key_len = miss->key_len;
2609 = (facet->may_install
2611 : xmemdup(subfacet->actions, subfacet->actions_len));
2612 execute->actions_len = subfacet->actions_len;
2613 execute->packet = packet;
2617 if (facet->may_install && subfacet->key_fitness != ODP_FIT_TOO_LITTLE) {
2618 struct flow_miss_op *op = &ops[(*n_ops)++];
2619 struct dpif_flow_put *put = &op->dpif_op.flow_put;
2621 op->subfacet = subfacet;
2622 put->type = DPIF_OP_FLOW_PUT;
2623 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2624 put->key = miss->key;
2625 put->key_len = miss->key_len;
2626 put->actions = subfacet->actions;
2627 put->actions_len = subfacet->actions_len;
2632 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
2633 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
2634 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
2635 * what a flow key should contain.
2637 * This function also includes some logic to help make VLAN splinters
2638 * transparent to the rest of the upcall processing logic. In particular, if
2639 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
2640 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
2641 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
2643 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
2644 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
2645 * (This differs from the value returned in flow->vlan_tci only for packets
2646 * received on VLAN splinters.)
2648 static enum odp_key_fitness
2649 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
2650 const struct nlattr *key, size_t key_len,
2651 struct flow *flow, ovs_be16 *initial_tci,
2652 struct ofpbuf *packet)
2654 enum odp_key_fitness fitness;
2658 fitness = odp_flow_key_to_flow(key, key_len, flow);
2659 if (fitness == ODP_FIT_ERROR) {
2662 *initial_tci = flow->vlan_tci;
2664 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
2666 /* Cause the flow to be processed as if it came in on the real device
2667 * with the VLAN device's VLAN ID. */
2668 flow->in_port = realdev;
2669 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
2671 /* Make the packet resemble the flow, so that it gets sent to an
2672 * OpenFlow controller properly, so that it looks correct for
2673 * sFlow, and so that flow_extract() will get the correct vlan_tci
2674 * if it is called on 'packet'.
2676 * The allocated space inside 'packet' probably also contains
2677 * 'key', that is, both 'packet' and 'key' are probably part of a
2678 * struct dpif_upcall (see the large comment on that structure
2679 * definition), so pushing data on 'packet' is in general not a
2680 * good idea since it could overwrite 'key' or free it as a side
2681 * effect. However, it's OK in this special case because we know
2682 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
2683 * will just overwrite the 4-byte "struct nlattr", which is fine
2684 * since we don't need that header anymore. */
2685 eth_push_vlan(packet, flow->vlan_tci);
2688 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
2689 if (fitness == ODP_FIT_PERFECT) {
2690 fitness = ODP_FIT_TOO_MUCH;
2698 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
2701 struct dpif_upcall *upcall;
2702 struct flow_miss *miss, *next_miss;
2703 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
2704 union dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
2713 /* Construct the to-do list.
2715 * This just amounts to extracting the flow from each packet and sticking
2716 * the packets that have the same flow in the same "flow_miss" structure so
2717 * that we can process them together. */
2719 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
2720 enum odp_key_fitness fitness;
2721 struct flow_miss *miss;
2722 ovs_be16 initial_tci;
2725 /* Obtain metadata and check userspace/kernel agreement on flow match,
2726 * then set 'flow''s header pointers. */
2727 fitness = ofproto_dpif_extract_flow_key(ofproto,
2728 upcall->key, upcall->key_len,
2729 &flow, &initial_tci,
2731 if (fitness == ODP_FIT_ERROR) {
2732 ofpbuf_delete(upcall->packet);
2735 flow_extract(upcall->packet, flow.skb_priority, flow.tun_id,
2736 flow.in_port, &flow);
2738 /* Handle 802.1ag, LACP, and STP specially. */
2739 if (process_special(ofproto, &flow, upcall->packet)) {
2740 ofproto_update_local_port_stats(&ofproto->up,
2741 0, upcall->packet->size);
2742 ofpbuf_delete(upcall->packet);
2743 ofproto->n_matches++;
2747 /* Add other packets to a to-do list. */
2748 miss = flow_miss_create(&todo, &flow, fitness,
2749 upcall->key, upcall->key_len, initial_tci);
2750 list_push_back(&miss->packets, &upcall->packet->list_node);
2753 /* Process each element in the to-do list, constructing the set of
2754 * operations to batch. */
2756 HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) {
2757 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
2758 ofpbuf_list_delete(&miss->packets);
2759 hmap_remove(&todo, &miss->hmap_node);
2762 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
2763 hmap_destroy(&todo);
2765 /* Execute batch. */
2766 for (i = 0; i < n_ops; i++) {
2767 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
2769 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
2771 /* Free memory and update facets. */
2772 for (i = 0; i < n_ops; i++) {
2773 struct flow_miss_op *op = &flow_miss_ops[i];
2774 struct dpif_execute *execute;
2775 struct dpif_flow_put *put;
2777 switch (op->dpif_op.type) {
2778 case DPIF_OP_EXECUTE:
2779 execute = &op->dpif_op.execute;
2780 if (op->subfacet->actions != execute->actions) {
2781 free((struct nlattr *) execute->actions);
2783 ofpbuf_delete((struct ofpbuf *) execute->packet);
2786 case DPIF_OP_FLOW_PUT:
2787 put = &op->dpif_op.flow_put;
2789 op->subfacet->installed = true;
2797 handle_userspace_upcall(struct ofproto_dpif *ofproto,
2798 struct dpif_upcall *upcall)
2800 struct user_action_cookie cookie;
2801 enum odp_key_fitness fitness;
2802 ovs_be16 initial_tci;
2805 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
2807 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
2808 upcall->key_len, &flow,
2809 &initial_tci, upcall->packet);
2810 if (fitness == ODP_FIT_ERROR) {
2811 ofpbuf_delete(upcall->packet);
2815 if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
2816 if (ofproto->sflow) {
2817 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
2820 ofpbuf_delete(upcall->packet);
2821 } else if (cookie.type == USER_ACTION_COOKIE_CONTROLLER) {
2822 COVERAGE_INC(ofproto_dpif_ctlr_action);
2823 send_packet_in_action(ofproto, upcall->packet, upcall->userdata,
2826 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
2827 ofpbuf_delete(upcall->packet);
2832 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
2834 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
2838 assert (max_batch <= FLOW_MISS_MAX_BATCH);
2841 for (i = 0; i < max_batch; i++) {
2842 struct dpif_upcall *upcall = &misses[n_misses];
2845 error = dpif_recv(ofproto->dpif, upcall);
2850 switch (upcall->type) {
2851 case DPIF_UC_ACTION:
2852 handle_userspace_upcall(ofproto, upcall);
2856 /* Handle it later. */
2860 case DPIF_N_UC_TYPES:
2862 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32,
2868 handle_miss_upcalls(ofproto, misses, n_misses);
2873 /* Flow expiration. */
2875 static int subfacet_max_idle(const struct ofproto_dpif *);
2876 static void update_stats(struct ofproto_dpif *);
2877 static void rule_expire(struct rule_dpif *);
2878 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
2880 /* This function is called periodically by run(). Its job is to collect
2881 * updates for the flows that have been installed into the datapath, most
2882 * importantly when they last were used, and then use that information to
2883 * expire flows that have not been used recently.
2885 * Returns the number of milliseconds after which it should be called again. */
2887 expire(struct ofproto_dpif *ofproto)
2889 struct rule_dpif *rule, *next_rule;
2890 struct classifier *table;
2893 /* Update stats for each flow in the datapath. */
2894 update_stats(ofproto);
2896 /* Expire subfacets that have been idle too long. */
2897 dp_max_idle = subfacet_max_idle(ofproto);
2898 expire_subfacets(ofproto, dp_max_idle);
2900 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
2901 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
2902 struct cls_cursor cursor;
2904 cls_cursor_init(&cursor, table, NULL);
2905 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
2910 /* All outstanding data in existing flows has been accounted, so it's a
2911 * good time to do bond rebalancing. */
2912 if (ofproto->has_bonded_bundles) {
2913 struct ofbundle *bundle;
2915 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2917 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
2922 return MIN(dp_max_idle, 1000);
2925 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
2927 * This function also pushes statistics updates to rules which each facet
2928 * resubmits into. Generally these statistics will be accurate. However, if a
2929 * facet changes the rule it resubmits into at some time in between
2930 * update_stats() runs, it is possible that statistics accrued to the
2931 * old rule will be incorrectly attributed to the new rule. This could be
2932 * avoided by calling update_stats() whenever rules are created or
2933 * deleted. However, the performance impact of making so many calls to the
2934 * datapath do not justify the benefit of having perfectly accurate statistics.
2937 update_stats(struct ofproto_dpif *p)
2939 const struct dpif_flow_stats *stats;
2940 struct dpif_flow_dump dump;
2941 const struct nlattr *key;
2944 dpif_flow_dump_start(&dump, p->dpif);
2945 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
2946 struct subfacet *subfacet;
2948 subfacet = subfacet_find(p, key, key_len);
2949 if (subfacet && subfacet->installed) {
2950 struct facet *facet = subfacet->facet;
2952 if (stats->n_packets >= subfacet->dp_packet_count) {
2953 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
2954 facet->packet_count += extra;
2956 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
2959 if (stats->n_bytes >= subfacet->dp_byte_count) {
2960 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
2962 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
2965 subfacet->dp_packet_count = stats->n_packets;
2966 subfacet->dp_byte_count = stats->n_bytes;
2968 subfacet_update_time(p, subfacet, stats->used);
2969 facet_account(p, facet);
2970 facet_push_stats(facet);
2972 if (!VLOG_DROP_WARN(&rl)) {
2976 odp_flow_key_format(key, key_len, &s);
2977 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
2981 COVERAGE_INC(facet_unexpected);
2982 /* There's a flow in the datapath that we know nothing about, or a
2983 * flow that shouldn't be installed but was anyway. Delete it. */
2984 dpif_flow_del(p->dpif, key, key_len, NULL);
2987 dpif_flow_dump_done(&dump);
2990 /* Calculates and returns the number of milliseconds of idle time after which
2991 * subfacets should expire from the datapath. When a subfacet expires, we fold
2992 * its statistics into its facet, and when a facet's last subfacet expires, we
2993 * fold its statistic into its rule. */
2995 subfacet_max_idle(const struct ofproto_dpif *ofproto)
2998 * Idle time histogram.
3000 * Most of the time a switch has a relatively small number of subfacets.
3001 * When this is the case we might as well keep statistics for all of them
3002 * in userspace and to cache them in the kernel datapath for performance as
3005 * As the number of subfacets increases, the memory required to maintain
3006 * statistics about them in userspace and in the kernel becomes
3007 * significant. However, with a large number of subfacets it is likely
3008 * that only a few of them are "heavy hitters" that consume a large amount
3009 * of bandwidth. At this point, only heavy hitters are worth caching in
3010 * the kernel and maintaining in userspaces; other subfacets we can
3013 * The technique used to compute the idle time is to build a histogram with
3014 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3015 * that is installed in the kernel gets dropped in the appropriate bucket.
3016 * After the histogram has been built, we compute the cutoff so that only
3017 * the most-recently-used 1% of subfacets (but at least
3018 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3019 * the most-recently-used bucket of subfacets is kept, so actually an
3020 * arbitrary number of subfacets can be kept in any given expiration run
3021 * (though the next run will delete most of those unless they receive
3024 * This requires a second pass through the subfacets, in addition to the
3025 * pass made by update_stats(), because the former function never looks at
3026 * uninstallable subfacets.
3028 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3029 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3030 int buckets[N_BUCKETS] = { 0 };
3031 int total, subtotal, bucket;
3032 struct subfacet *subfacet;
3036 total = hmap_count(&ofproto->subfacets);
3037 if (total <= ofproto->up.flow_eviction_threshold) {
3038 return N_BUCKETS * BUCKET_WIDTH;
3041 /* Build histogram. */
3043 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3044 long long int idle = now - subfacet->used;
3045 int bucket = (idle <= 0 ? 0
3046 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3047 : (unsigned int) idle / BUCKET_WIDTH);
3051 /* Find the first bucket whose flows should be expired. */
3052 subtotal = bucket = 0;
3054 subtotal += buckets[bucket++];
3055 } while (bucket < N_BUCKETS &&
3056 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3058 if (VLOG_IS_DBG_ENABLED()) {
3062 ds_put_cstr(&s, "keep");
3063 for (i = 0; i < N_BUCKETS; i++) {
3065 ds_put_cstr(&s, ", drop");
3068 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3071 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3075 return bucket * BUCKET_WIDTH;
3079 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3081 long long int cutoff = time_msec() - dp_max_idle;
3082 struct subfacet *subfacet, *next_subfacet;
3084 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3085 &ofproto->subfacets) {
3086 if (subfacet->used < cutoff) {
3087 subfacet_destroy(ofproto, subfacet);
3092 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3093 * then delete it entirely. */
3095 rule_expire(struct rule_dpif *rule)
3097 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3098 struct facet *facet, *next_facet;
3102 /* Has 'rule' expired? */
3104 if (rule->up.hard_timeout
3105 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3106 reason = OFPRR_HARD_TIMEOUT;
3107 } else if (rule->up.idle_timeout && list_is_empty(&rule->facets)
3108 && now > rule->used + rule->up.idle_timeout * 1000) {
3109 reason = OFPRR_IDLE_TIMEOUT;
3114 COVERAGE_INC(ofproto_dpif_expired);
3116 /* Update stats. (This is a no-op if the rule expired due to an idle
3117 * timeout, because that only happens when the rule has no facets left.) */
3118 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3119 facet_remove(ofproto, facet);
3122 /* Get rid of the rule. */
3123 ofproto_rule_expire(&rule->up, reason);
3128 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3130 * The caller must already have determined that no facet with an identical
3131 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3132 * the ofproto's classifier table.
3134 * The facet will initially have no subfacets. The caller should create (at
3135 * least) one subfacet with subfacet_create(). */
3136 static struct facet *
3137 facet_create(struct rule_dpif *rule, const struct flow *flow)
3139 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3140 struct facet *facet;
3142 facet = xzalloc(sizeof *facet);
3143 facet->used = time_msec();
3144 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
3145 list_push_back(&rule->facets, &facet->list_node);
3147 facet->flow = *flow;
3148 list_init(&facet->subfacets);
3149 netflow_flow_init(&facet->nf_flow);
3150 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3156 facet_free(struct facet *facet)
3161 /* If the 'actions_len' bytes of actions in 'odp_actions' are just a single
3162 * OVS_ACTION_ATTR_USERSPACE action, executes it internally and returns true.
3163 * Otherwise, returns false without doing anything.
3165 * If 'clone' is true, the caller always retains ownership of 'packet'.
3166 * Otherwise, ownership is transferred to this function if it returns true. */
3168 execute_controller_action(struct ofproto_dpif *ofproto,
3169 const struct flow *flow,
3170 const struct nlattr *odp_actions, size_t actions_len,
3171 struct ofpbuf *packet, bool clone)
3174 && odp_actions->nla_type == OVS_ACTION_ATTR_USERSPACE
3175 && NLA_ALIGN(odp_actions->nla_len) == actions_len) {
3176 /* As an optimization, avoid a round-trip from userspace to kernel to
3177 * userspace. This also avoids possibly filling up kernel packet
3178 * buffers along the way.
3180 * This optimization will not accidentally catch sFlow
3181 * OVS_ACTION_ATTR_USERSPACE actions, since those are encapsulated
3182 * inside OVS_ACTION_ATTR_SAMPLE. */
3183 const struct nlattr *nla;
3185 nla = nl_attr_find_nested(odp_actions, OVS_USERSPACE_ATTR_USERDATA);
3186 send_packet_in_action(ofproto, packet, nl_attr_get_u64(nla), flow,
3194 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3195 * 'packet', which arrived on 'in_port'.
3197 * Takes ownership of 'packet'. */
3199 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3200 const struct nlattr *odp_actions, size_t actions_len,
3201 struct ofpbuf *packet)
3203 struct odputil_keybuf keybuf;
3207 if (execute_controller_action(ofproto, flow, odp_actions, actions_len,
3212 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3213 odp_flow_key_from_flow(&key, flow);
3215 error = dpif_execute(ofproto->dpif, key.data, key.size,
3216 odp_actions, actions_len, packet);
3218 ofpbuf_delete(packet);
3222 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3224 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3225 * rule's statistics, via subfacet_uninstall().
3227 * - Removes 'facet' from its rule and from ofproto->facets.
3230 facet_remove(struct ofproto_dpif *ofproto, struct facet *facet)
3232 struct subfacet *subfacet, *next_subfacet;
3234 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3235 &facet->subfacets) {
3236 subfacet_destroy__(ofproto, subfacet);
3239 facet_flush_stats(ofproto, facet);
3240 hmap_remove(&ofproto->facets, &facet->hmap_node);
3241 list_remove(&facet->list_node);
3246 facet_account(struct ofproto_dpif *ofproto, struct facet *facet)
3249 struct subfacet *subfacet;
3250 const struct nlattr *a;
3254 if (facet->byte_count <= facet->accounted_bytes) {
3257 n_bytes = facet->byte_count - facet->accounted_bytes;
3258 facet->accounted_bytes = facet->byte_count;
3260 /* Feed information from the active flows back into the learning table to
3261 * ensure that table is always in sync with what is actually flowing
3262 * through the datapath. */
3263 if (facet->has_learn || facet->has_normal) {
3264 struct action_xlate_ctx ctx;
3266 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3267 facet->flow.vlan_tci, NULL);
3268 ctx.may_learn = true;
3269 ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions,
3270 facet->rule->up.n_actions));
3273 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3277 /* This loop feeds byte counters to bond_account() for rebalancing to use
3278 * as a basis. We also need to track the actual VLAN on which the packet
3279 * is going to be sent to ensure that it matches the one passed to
3280 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3283 * We use the actions from an arbitrary subfacet because they should all
3284 * be equally valid for our purpose. */
3285 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3286 struct subfacet, list_node);
3287 vlan_tci = facet->flow.vlan_tci;
3288 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3289 subfacet->actions, subfacet->actions_len) {
3290 const struct ovs_action_push_vlan *vlan;
3291 struct ofport_dpif *port;
3293 switch (nl_attr_type(a)) {
3294 case OVS_ACTION_ATTR_OUTPUT:
3295 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3296 if (port && port->bundle && port->bundle->bond) {
3297 bond_account(port->bundle->bond, &facet->flow,
3298 vlan_tci_to_vid(vlan_tci), n_bytes);
3302 case OVS_ACTION_ATTR_POP_VLAN:
3303 vlan_tci = htons(0);
3306 case OVS_ACTION_ATTR_PUSH_VLAN:
3307 vlan = nl_attr_get(a);
3308 vlan_tci = vlan->vlan_tci;
3314 /* Returns true if the only action for 'facet' is to send to the controller.
3315 * (We don't report NetFlow expiration messages for such facets because they
3316 * are just part of the control logic for the network, not real traffic). */
3318 facet_is_controller_flow(struct facet *facet)
3321 && facet->rule->up.n_actions == 1
3322 && action_outputs_to_port(&facet->rule->up.actions[0],
3323 htons(OFPP_CONTROLLER)));
3326 /* Folds all of 'facet''s statistics into its rule. Also updates the
3327 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3328 * 'facet''s statistics in the datapath should have been zeroed and folded into
3329 * its packet and byte counts before this function is called. */
3331 facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet)
3333 struct subfacet *subfacet;
3335 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3336 assert(!subfacet->dp_byte_count);
3337 assert(!subfacet->dp_packet_count);
3340 facet_push_stats(facet);
3341 facet_account(ofproto, facet);
3343 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3344 struct ofexpired expired;
3345 expired.flow = facet->flow;
3346 expired.packet_count = facet->packet_count;
3347 expired.byte_count = facet->byte_count;
3348 expired.used = facet->used;
3349 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3352 facet->rule->packet_count += facet->packet_count;
3353 facet->rule->byte_count += facet->byte_count;
3355 /* Reset counters to prevent double counting if 'facet' ever gets
3357 facet_reset_counters(facet);
3359 netflow_flow_clear(&facet->nf_flow);
3362 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3363 * Returns it if found, otherwise a null pointer.
3365 * The returned facet might need revalidation; use facet_lookup_valid()
3366 * instead if that is important. */
3367 static struct facet *
3368 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
3370 struct facet *facet;
3372 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
3374 if (flow_equal(flow, &facet->flow)) {
3382 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3383 * Returns it if found, otherwise a null pointer.
3385 * The returned facet is guaranteed to be valid. */
3386 static struct facet *
3387 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
3389 struct facet *facet = facet_find(ofproto, flow);
3391 /* The facet we found might not be valid, since we could be in need of
3392 * revalidation. If it is not valid, don't return it. */
3394 && (ofproto->need_revalidate
3395 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))
3396 && !facet_revalidate(ofproto, facet)) {
3397 COVERAGE_INC(facet_invalidated);
3404 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
3406 * - If the rule found is different from 'facet''s current rule, moves
3407 * 'facet' to the new rule and recompiles its actions.
3409 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3410 * where it is and recompiles its actions anyway.
3412 * - If there is none, destroys 'facet'.
3414 * Returns true if 'facet' still exists, false if it has been destroyed. */
3416 facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet)
3419 struct nlattr *odp_actions;
3422 struct actions *new_actions;
3424 struct action_xlate_ctx ctx;
3425 struct rule_dpif *new_rule;
3426 struct subfacet *subfacet;
3427 bool actions_changed;
3430 COVERAGE_INC(facet_revalidate);
3432 /* Determine the new rule. */
3433 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3435 /* No new rule, so delete the facet. */
3436 facet_remove(ofproto, facet);
3440 /* Calculate new datapath actions.
3442 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3443 * emit a NetFlow expiration and, if so, we need to have the old state
3444 * around to properly compose it. */
3446 /* If the datapath actions changed or the installability changed,
3447 * then we need to talk to the datapath. */
3450 memset(&ctx, 0, sizeof ctx);
3451 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3452 struct ofpbuf *odp_actions;
3453 bool should_install;
3455 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3456 subfacet->initial_tci, NULL);
3457 odp_actions = xlate_actions(&ctx, new_rule->up.actions,
3458 new_rule->up.n_actions);
3459 actions_changed = (subfacet->actions_len != odp_actions->size
3460 || memcmp(subfacet->actions, odp_actions->data,
3461 subfacet->actions_len));
3463 should_install = (ctx.may_set_up_flow
3464 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3465 if (actions_changed || should_install != subfacet->installed) {
3466 if (should_install) {
3467 struct dpif_flow_stats stats;
3469 subfacet_install(ofproto, subfacet,
3470 odp_actions->data, odp_actions->size, &stats);
3471 subfacet_update_stats(ofproto, subfacet, &stats);
3473 subfacet_uninstall(ofproto, subfacet);
3477 new_actions = xcalloc(list_size(&facet->subfacets),
3478 sizeof *new_actions);
3480 new_actions[i].odp_actions = xmemdup(odp_actions->data,
3482 new_actions[i].actions_len = odp_actions->size;
3485 ofpbuf_delete(odp_actions);
3489 facet_flush_stats(ofproto, facet);
3492 /* Update 'facet' now that we've taken care of all the old state. */
3493 facet->tags = ctx.tags;
3494 facet->nf_flow.output_iface = ctx.nf_output_iface;
3495 facet->may_install = ctx.may_set_up_flow;
3496 facet->has_learn = ctx.has_learn;
3497 facet->has_normal = ctx.has_normal;
3498 facet->mirrors = ctx.mirrors;
3501 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3502 if (new_actions[i].odp_actions) {
3503 free(subfacet->actions);
3504 subfacet->actions = new_actions[i].odp_actions;
3505 subfacet->actions_len = new_actions[i].actions_len;
3511 if (facet->rule != new_rule) {
3512 COVERAGE_INC(facet_changed_rule);
3513 list_remove(&facet->list_node);
3514 list_push_back(&new_rule->facets, &facet->list_node);
3515 facet->rule = new_rule;
3516 facet->used = new_rule->up.created;
3517 facet->prev_used = facet->used;
3523 /* Updates 'facet''s used time. Caller is responsible for calling
3524 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3526 facet_update_time(struct ofproto_dpif *ofproto, struct facet *facet,
3529 if (used > facet->used) {
3531 if (used > facet->rule->used) {
3532 facet->rule->used = used;
3534 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3539 facet_reset_counters(struct facet *facet)
3541 facet->packet_count = 0;
3542 facet->byte_count = 0;
3543 facet->prev_packet_count = 0;
3544 facet->prev_byte_count = 0;
3545 facet->accounted_bytes = 0;
3549 facet_push_stats(struct facet *facet)
3551 uint64_t new_packets, new_bytes;
3553 assert(facet->packet_count >= facet->prev_packet_count);
3554 assert(facet->byte_count >= facet->prev_byte_count);
3555 assert(facet->used >= facet->prev_used);
3557 new_packets = facet->packet_count - facet->prev_packet_count;
3558 new_bytes = facet->byte_count - facet->prev_byte_count;
3560 if (new_packets || new_bytes || facet->used > facet->prev_used) {
3561 facet->prev_packet_count = facet->packet_count;
3562 facet->prev_byte_count = facet->byte_count;
3563 facet->prev_used = facet->used;
3565 flow_push_stats(facet->rule, &facet->flow,
3566 new_packets, new_bytes, facet->used);
3568 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
3569 facet->mirrors, new_packets, new_bytes);
3573 struct ofproto_push {
3574 struct action_xlate_ctx ctx;
3581 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
3583 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3586 rule->packet_count += push->packets;
3587 rule->byte_count += push->bytes;
3588 rule->used = MAX(push->used, rule->used);
3592 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3593 * 'rule''s actions and mirrors. */
3595 flow_push_stats(const struct rule_dpif *rule,
3596 const struct flow *flow, uint64_t packets, uint64_t bytes,
3599 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3600 struct ofproto_push push;
3602 push.packets = packets;
3606 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, NULL);
3607 push.ctx.resubmit_hook = push_resubmit;
3608 ofpbuf_delete(xlate_actions(&push.ctx,
3609 rule->up.actions, rule->up.n_actions));
3614 static struct subfacet *
3615 subfacet_find__(struct ofproto_dpif *ofproto,
3616 const struct nlattr *key, size_t key_len, uint32_t key_hash,
3617 const struct flow *flow)
3619 struct subfacet *subfacet;
3621 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
3622 &ofproto->subfacets) {
3624 ? (subfacet->key_len == key_len
3625 && !memcmp(key, subfacet->key, key_len))
3626 : flow_equal(flow, &subfacet->facet->flow)) {
3634 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
3635 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
3636 * there is one, otherwise creates and returns a new subfacet.
3638 * If the returned subfacet is new, then subfacet->actions will be NULL, in
3639 * which case the caller must populate the actions with
3640 * subfacet_make_actions(). */
3641 static struct subfacet *
3642 subfacet_create(struct ofproto_dpif *ofproto, struct facet *facet,
3643 enum odp_key_fitness key_fitness,
3644 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
3646 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3647 struct subfacet *subfacet;
3649 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
3651 if (subfacet->facet == facet) {
3655 /* This shouldn't happen. */
3656 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
3657 subfacet_destroy(ofproto, subfacet);
3660 subfacet = xzalloc(sizeof *subfacet);
3661 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
3662 list_push_back(&facet->subfacets, &subfacet->list_node);
3663 subfacet->facet = facet;
3664 subfacet->used = time_msec();
3665 subfacet->key_fitness = key_fitness;
3666 if (key_fitness != ODP_FIT_PERFECT) {
3667 subfacet->key = xmemdup(key, key_len);
3668 subfacet->key_len = key_len;
3670 subfacet->installed = false;
3671 subfacet->initial_tci = initial_tci;
3676 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
3677 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
3678 static struct subfacet *
3679 subfacet_find(struct ofproto_dpif *ofproto,
3680 const struct nlattr *key, size_t key_len)
3682 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3683 enum odp_key_fitness fitness;
3686 fitness = odp_flow_key_to_flow(key, key_len, &flow);
3687 if (fitness == ODP_FIT_ERROR) {
3691 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
3694 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
3695 * its facet within 'ofproto', and frees it. */
3697 subfacet_destroy__(struct ofproto_dpif *ofproto, struct subfacet *subfacet)
3699 subfacet_uninstall(ofproto, subfacet);
3700 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
3701 list_remove(&subfacet->list_node);
3702 free(subfacet->key);
3703 free(subfacet->actions);
3707 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
3708 * last remaining subfacet in its facet destroys the facet too. */
3710 subfacet_destroy(struct ofproto_dpif *ofproto, struct subfacet *subfacet)
3712 struct facet *facet = subfacet->facet;
3714 subfacet_destroy__(ofproto, subfacet);
3715 if (list_is_empty(&facet->subfacets)) {
3716 facet_remove(ofproto, facet);
3720 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
3721 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
3722 * for use as temporary storage. */
3724 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
3727 if (!subfacet->key) {
3728 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
3729 odp_flow_key_from_flow(key, &subfacet->facet->flow);
3731 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
3735 /* Composes the datapath actions for 'subfacet' based on its rule's actions. */
3737 subfacet_make_actions(struct ofproto_dpif *p, struct subfacet *subfacet,
3738 const struct ofpbuf *packet)
3740 struct facet *facet = subfacet->facet;
3741 const struct rule_dpif *rule = facet->rule;
3742 struct ofpbuf *odp_actions;
3743 struct action_xlate_ctx ctx;
3745 action_xlate_ctx_init(&ctx, p, &facet->flow, subfacet->initial_tci,
3747 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3748 facet->tags = ctx.tags;
3749 facet->may_install = ctx.may_set_up_flow;
3750 facet->has_learn = ctx.has_learn;
3751 facet->has_normal = ctx.has_normal;
3752 facet->nf_flow.output_iface = ctx.nf_output_iface;
3753 facet->mirrors = ctx.mirrors;
3755 if (subfacet->actions_len != odp_actions->size
3756 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
3757 free(subfacet->actions);
3758 subfacet->actions_len = odp_actions->size;
3759 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
3762 ofpbuf_delete(odp_actions);
3765 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
3766 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
3767 * in the datapath will be zeroed and 'stats' will be updated with traffic new
3768 * since 'subfacet' was last updated.
3770 * Returns 0 if successful, otherwise a positive errno value. */
3772 subfacet_install(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3773 const struct nlattr *actions, size_t actions_len,
3774 struct dpif_flow_stats *stats)
3776 struct odputil_keybuf keybuf;
3777 enum dpif_flow_put_flags flags;
3781 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3783 flags |= DPIF_FP_ZERO_STATS;
3786 subfacet_get_key(subfacet, &keybuf, &key);
3787 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
3788 actions, actions_len, stats);
3791 subfacet_reset_dp_stats(subfacet, stats);
3797 /* If 'subfacet' is installed in the datapath, uninstalls it. */
3799 subfacet_uninstall(struct ofproto_dpif *p, struct subfacet *subfacet)
3801 if (subfacet->installed) {
3802 struct odputil_keybuf keybuf;
3803 struct dpif_flow_stats stats;
3807 subfacet_get_key(subfacet, &keybuf, &key);
3808 error = dpif_flow_del(p->dpif, key.data, key.size, &stats);
3809 subfacet_reset_dp_stats(subfacet, &stats);
3811 subfacet_update_stats(p, subfacet, &stats);
3813 subfacet->installed = false;
3815 assert(subfacet->dp_packet_count == 0);
3816 assert(subfacet->dp_byte_count == 0);
3820 /* Resets 'subfacet''s datapath statistics counters. This should be called
3821 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
3822 * non-null, it should contain the statistics returned by dpif when 'subfacet'
3823 * was reset in the datapath. 'stats' will be modified to include only
3824 * statistics new since 'subfacet' was last updated. */
3826 subfacet_reset_dp_stats(struct subfacet *subfacet,
3827 struct dpif_flow_stats *stats)
3830 && subfacet->dp_packet_count <= stats->n_packets
3831 && subfacet->dp_byte_count <= stats->n_bytes) {
3832 stats->n_packets -= subfacet->dp_packet_count;
3833 stats->n_bytes -= subfacet->dp_byte_count;
3836 subfacet->dp_packet_count = 0;
3837 subfacet->dp_byte_count = 0;
3840 /* Updates 'subfacet''s used time. The caller is responsible for calling
3841 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
3843 subfacet_update_time(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3846 if (used > subfacet->used) {
3847 subfacet->used = used;
3848 facet_update_time(ofproto, subfacet->facet, used);
3852 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
3854 * Because of the meaning of a subfacet's counters, it only makes sense to do
3855 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
3856 * represents a packet that was sent by hand or if it represents statistics
3857 * that have been cleared out of the datapath. */
3859 subfacet_update_stats(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3860 const struct dpif_flow_stats *stats)
3862 if (stats->n_packets || stats->used > subfacet->used) {
3863 struct facet *facet = subfacet->facet;
3865 subfacet_update_time(ofproto, subfacet, stats->used);
3866 facet->packet_count += stats->n_packets;
3867 facet->byte_count += stats->n_bytes;
3868 facet_push_stats(facet);
3869 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3875 static struct rule_dpif *
3876 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
3879 struct cls_rule *cls_rule;
3880 struct classifier *cls;
3882 if (table_id >= N_TABLES) {
3886 cls = &ofproto->up.tables[table_id];
3887 if (flow->nw_frag & FLOW_NW_FRAG_ANY
3888 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3889 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
3890 * are unavailable. */
3891 struct flow ofpc_normal_flow = *flow;
3892 ofpc_normal_flow.tp_src = htons(0);
3893 ofpc_normal_flow.tp_dst = htons(0);
3894 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
3896 cls_rule = classifier_lookup(cls, flow);
3898 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
3902 complete_operation(struct rule_dpif *rule)
3904 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3906 rule_invalidate(rule);
3908 struct dpif_completion *c = xmalloc(sizeof *c);
3909 c->op = rule->up.pending;
3910 list_push_back(&ofproto->completions, &c->list_node);
3912 ofoperation_complete(rule->up.pending, 0);
3916 static struct rule *
3919 struct rule_dpif *rule = xmalloc(sizeof *rule);
3924 rule_dealloc(struct rule *rule_)
3926 struct rule_dpif *rule = rule_dpif_cast(rule_);
3931 rule_construct(struct rule *rule_)
3933 struct rule_dpif *rule = rule_dpif_cast(rule_);
3934 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3935 struct rule_dpif *victim;
3939 error = validate_actions(rule->up.actions, rule->up.n_actions,
3940 &rule->up.cr.flow, ofproto->max_ports);
3945 rule->used = rule->up.created;
3946 rule->packet_count = 0;
3947 rule->byte_count = 0;
3949 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
3950 if (victim && !list_is_empty(&victim->facets)) {
3951 struct facet *facet;
3953 rule->facets = victim->facets;
3954 list_moved(&rule->facets);
3955 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3956 /* XXX: We're only clearing our local counters here. It's possible
3957 * that quite a few packets are unaccounted for in the datapath
3958 * statistics. These will be accounted to the new rule instead of
3959 * cleared as required. This could be fixed by clearing out the
3960 * datapath statistics for this facet, but currently it doesn't
3962 facet_reset_counters(facet);
3966 /* Must avoid list_moved() in this case. */
3967 list_init(&rule->facets);
3970 table_id = rule->up.table_id;
3971 rule->tag = (victim ? victim->tag
3973 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
3974 ofproto->tables[table_id].basis));
3976 complete_operation(rule);
3981 rule_destruct(struct rule *rule_)
3983 struct rule_dpif *rule = rule_dpif_cast(rule_);
3984 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3985 struct facet *facet, *next_facet;
3987 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3988 facet_revalidate(ofproto, facet);
3991 complete_operation(rule);
3995 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
3997 struct rule_dpif *rule = rule_dpif_cast(rule_);
3998 struct facet *facet;
4000 /* Start from historical data for 'rule' itself that are no longer tracked
4001 * in facets. This counts, for example, facets that have expired. */
4002 *packets = rule->packet_count;
4003 *bytes = rule->byte_count;
4005 /* Add any statistics that are tracked by facets. This includes
4006 * statistical data recently updated by ofproto_update_stats() as well as
4007 * stats for packets that were executed "by hand" via dpif_execute(). */
4008 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4009 *packets += facet->packet_count;
4010 *bytes += facet->byte_count;
4015 rule_execute(struct rule *rule_, const struct flow *flow,
4016 struct ofpbuf *packet)
4018 struct rule_dpif *rule = rule_dpif_cast(rule_);
4019 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4020 struct action_xlate_ctx ctx;
4021 struct ofpbuf *odp_actions;
4024 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, packet);
4025 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
4026 size = packet->size;
4027 if (execute_odp_actions(ofproto, flow, odp_actions->data,
4028 odp_actions->size, packet)) {
4029 rule->used = time_msec();
4030 rule->packet_count++;
4031 rule->byte_count += size;
4032 flow_push_stats(rule, flow, 1, size, rule->used);
4034 ofpbuf_delete(odp_actions);
4040 rule_modify_actions(struct rule *rule_)
4042 struct rule_dpif *rule = rule_dpif_cast(rule_);
4043 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4046 error = validate_actions(rule->up.actions, rule->up.n_actions,
4047 &rule->up.cr.flow, ofproto->max_ports);
4049 ofoperation_complete(rule->up.pending, error);
4053 complete_operation(rule);
4056 /* Sends 'packet' out 'ofport'.
4057 * May modify 'packet'.
4058 * Returns 0 if successful, otherwise a positive errno value. */
4060 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4062 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4063 struct ofpbuf key, odp_actions;
4064 struct odputil_keybuf keybuf;
4069 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
4070 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4072 if (odp_port != ofport->odp_port) {
4073 eth_pop_vlan(packet);
4074 flow.vlan_tci = htons(0);
4077 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4078 odp_flow_key_from_flow(&key, &flow);
4080 ofpbuf_init(&odp_actions, 32);
4081 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4083 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4084 error = dpif_execute(ofproto->dpif,
4086 odp_actions.data, odp_actions.size,
4088 ofpbuf_uninit(&odp_actions);
4091 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4092 ofproto->up.name, odp_port, strerror(error));
4094 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
4098 /* OpenFlow to datapath action translation. */
4100 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
4101 struct action_xlate_ctx *ctx);
4102 static void xlate_normal(struct action_xlate_ctx *);
4105 put_userspace_action(const struct ofproto_dpif *ofproto,
4106 struct ofpbuf *odp_actions,
4107 const struct flow *flow,
4108 const struct user_action_cookie *cookie)
4112 pid = dpif_port_get_pid(ofproto->dpif,
4113 ofp_port_to_odp_port(flow->in_port));
4115 return odp_put_userspace_action(pid, cookie, odp_actions);
4118 /* Compose SAMPLE action for sFlow. */
4120 compose_sflow_action(const struct ofproto_dpif *ofproto,
4121 struct ofpbuf *odp_actions,
4122 const struct flow *flow,
4125 uint32_t port_ifindex;
4126 uint32_t probability;
4127 struct user_action_cookie cookie;
4128 size_t sample_offset, actions_offset;
4129 int cookie_offset, n_output;
4131 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4135 if (odp_port == OVSP_NONE) {
4139 port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4143 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4145 /* Number of packets out of UINT_MAX to sample. */
4146 probability = dpif_sflow_get_probability(ofproto->sflow);
4147 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4149 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4151 cookie.type = USER_ACTION_COOKIE_SFLOW;
4152 cookie.data = port_ifindex;
4153 cookie.n_output = n_output;
4154 cookie.vlan_tci = 0;
4155 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4157 nl_msg_end_nested(odp_actions, actions_offset);
4158 nl_msg_end_nested(odp_actions, sample_offset);
4159 return cookie_offset;
4162 /* SAMPLE action must be first action in any given list of actions.
4163 * At this point we do not have all information required to build it. So try to
4164 * build sample action as complete as possible. */
4166 add_sflow_action(struct action_xlate_ctx *ctx)
4168 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4170 &ctx->flow, OVSP_NONE);
4171 ctx->sflow_odp_port = 0;
4172 ctx->sflow_n_outputs = 0;
4175 /* Fix SAMPLE action according to data collected while composing ODP actions.
4176 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4177 * USERSPACE action's user-cookie which is required for sflow. */
4179 fix_sflow_action(struct action_xlate_ctx *ctx)
4181 const struct flow *base = &ctx->base_flow;
4182 struct user_action_cookie *cookie;
4184 if (!ctx->user_cookie_offset) {
4188 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4190 assert(cookie != NULL);
4191 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4193 if (ctx->sflow_n_outputs) {
4194 cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow,
4195 ctx->sflow_odp_port);
4197 if (ctx->sflow_n_outputs >= 255) {
4198 cookie->n_output = 255;
4200 cookie->n_output = ctx->sflow_n_outputs;
4202 cookie->vlan_tci = base->vlan_tci;
4206 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4209 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4210 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4211 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4212 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4216 struct priority_to_dscp *pdscp;
4218 if (ofport->up.opp.config & htonl(OFPPC_NO_FWD)
4219 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4223 pdscp = get_priority(ofport, ctx->flow.skb_priority);
4225 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4226 ctx->flow.nw_tos |= pdscp->dscp;
4229 /* We may not have an ofport record for this port, but it doesn't hurt
4230 * to allow forwarding to it anyhow. Maybe such a port will appear
4231 * later and we're pre-populating the flow table. */
4234 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4235 ctx->flow.vlan_tci);
4236 if (out_port != odp_port) {
4237 ctx->flow.vlan_tci = htons(0);
4239 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4240 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4242 ctx->sflow_odp_port = odp_port;
4243 ctx->sflow_n_outputs++;
4244 ctx->nf_output_iface = ofp_port;
4245 ctx->flow.vlan_tci = flow_vlan_tci;
4246 ctx->flow.nw_tos = flow_nw_tos;
4250 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4252 compose_output_action__(ctx, ofp_port, true);
4256 xlate_table_action(struct action_xlate_ctx *ctx,
4257 uint16_t in_port, uint8_t table_id)
4259 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4260 struct ofproto_dpif *ofproto = ctx->ofproto;
4261 struct rule_dpif *rule;
4262 uint16_t old_in_port;
4263 uint8_t old_table_id;
4265 old_table_id = ctx->table_id;
4266 ctx->table_id = table_id;
4268 /* Look up a flow with 'in_port' as the input port. */
4269 old_in_port = ctx->flow.in_port;
4270 ctx->flow.in_port = in_port;
4271 rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
4274 if (table_id > 0 && table_id < N_TABLES) {
4275 struct table_dpif *table = &ofproto->tables[table_id];
4276 if (table->other_table) {
4279 : rule_calculate_tag(&ctx->flow,
4280 &table->other_table->wc,
4285 /* Restore the original input port. Otherwise OFPP_NORMAL and
4286 * OFPP_IN_PORT will have surprising behavior. */
4287 ctx->flow.in_port = old_in_port;
4289 if (ctx->resubmit_hook) {
4290 ctx->resubmit_hook(ctx, rule);
4295 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
4299 ctx->table_id = old_table_id;
4301 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4303 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
4304 MAX_RESUBMIT_RECURSION);
4309 xlate_resubmit_table(struct action_xlate_ctx *ctx,
4310 const struct nx_action_resubmit *nar)
4315 in_port = (nar->in_port == htons(OFPP_IN_PORT)
4317 : ntohs(nar->in_port));
4318 table_id = nar->table == 255 ? ctx->table_id : nar->table;
4320 xlate_table_action(ctx, in_port, table_id);
4324 flood_packets(struct action_xlate_ctx *ctx, bool all)
4326 struct ofport_dpif *ofport;
4328 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
4329 uint16_t ofp_port = ofport->up.ofp_port;
4331 if (ofp_port == ctx->flow.in_port) {
4336 compose_output_action__(ctx, ofp_port, false);
4337 } else if (!(ofport->up.opp.config & htonl(OFPPC_NO_FLOOD))) {
4338 compose_output_action(ctx, ofp_port);
4342 ctx->nf_output_iface = NF_OUT_FLOOD;
4346 compose_controller_action(struct action_xlate_ctx *ctx, int len)
4348 struct user_action_cookie cookie;
4350 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4351 cookie.type = USER_ACTION_COOKIE_CONTROLLER;
4353 cookie.n_output = 0;
4354 cookie.vlan_tci = 0;
4355 put_userspace_action(ctx->ofproto, ctx->odp_actions, &ctx->flow, &cookie);
4359 xlate_output_action__(struct action_xlate_ctx *ctx,
4360 uint16_t port, uint16_t max_len)
4362 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
4364 ctx->nf_output_iface = NF_OUT_DROP;
4368 compose_output_action(ctx, ctx->flow.in_port);
4371 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
4377 flood_packets(ctx, false);
4380 flood_packets(ctx, true);
4382 case OFPP_CONTROLLER:
4383 compose_controller_action(ctx, max_len);
4386 compose_output_action(ctx, OFPP_LOCAL);
4391 if (port != ctx->flow.in_port) {
4392 compose_output_action(ctx, port);
4397 if (prev_nf_output_iface == NF_OUT_FLOOD) {
4398 ctx->nf_output_iface = NF_OUT_FLOOD;
4399 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
4400 ctx->nf_output_iface = prev_nf_output_iface;
4401 } else if (prev_nf_output_iface != NF_OUT_DROP &&
4402 ctx->nf_output_iface != NF_OUT_FLOOD) {
4403 ctx->nf_output_iface = NF_OUT_MULTI;
4408 xlate_output_reg_action(struct action_xlate_ctx *ctx,
4409 const struct nx_action_output_reg *naor)
4413 ofp_port = nxm_read_field_bits(naor->src, naor->ofs_nbits, &ctx->flow);
4415 if (ofp_port <= UINT16_MAX) {
4416 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
4421 xlate_output_action(struct action_xlate_ctx *ctx,
4422 const struct ofp_action_output *oao)
4424 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
4428 xlate_enqueue_action(struct action_xlate_ctx *ctx,
4429 const struct ofp_action_enqueue *oae)
4432 uint32_t flow_priority, priority;
4435 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
4438 /* Fall back to ordinary output action. */
4439 xlate_output_action__(ctx, ntohs(oae->port), 0);
4443 /* Figure out datapath output port. */
4444 ofp_port = ntohs(oae->port);
4445 if (ofp_port == OFPP_IN_PORT) {
4446 ofp_port = ctx->flow.in_port;
4447 } else if (ofp_port == ctx->flow.in_port) {
4451 /* Add datapath actions. */
4452 flow_priority = ctx->flow.skb_priority;
4453 ctx->flow.skb_priority = priority;
4454 compose_output_action(ctx, ofp_port);
4455 ctx->flow.skb_priority = flow_priority;
4457 /* Update NetFlow output port. */
4458 if (ctx->nf_output_iface == NF_OUT_DROP) {
4459 ctx->nf_output_iface = ofp_port;
4460 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
4461 ctx->nf_output_iface = NF_OUT_MULTI;
4466 xlate_set_queue_action(struct action_xlate_ctx *ctx,
4467 const struct nx_action_set_queue *nasq)
4472 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
4475 /* Couldn't translate queue to a priority, so ignore. A warning
4476 * has already been logged. */
4480 ctx->flow.skb_priority = priority;
4483 struct xlate_reg_state {
4489 xlate_autopath(struct action_xlate_ctx *ctx,
4490 const struct nx_action_autopath *naa)
4492 uint16_t ofp_port = ntohl(naa->id);
4493 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
4495 if (!port || !port->bundle) {
4496 ofp_port = OFPP_NONE;
4497 } else if (port->bundle->bond) {
4498 /* Autopath does not support VLAN hashing. */
4499 struct ofport_dpif *slave = bond_choose_output_slave(
4500 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
4502 ofp_port = slave->up.ofp_port;
4505 autopath_execute(naa, &ctx->flow, ofp_port);
4509 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
4511 struct ofproto_dpif *ofproto = ofproto_;
4512 struct ofport_dpif *port;
4522 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
4525 port = get_ofp_port(ofproto, ofp_port);
4526 return port ? port->may_enable : false;
4531 xlate_learn_action(struct action_xlate_ctx *ctx,
4532 const struct nx_action_learn *learn)
4534 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
4535 struct ofputil_flow_mod fm;
4538 learn_execute(learn, &ctx->flow, &fm);
4540 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
4541 if (error && !VLOG_DROP_WARN(&rl)) {
4542 char *msg = ofputil_error_to_string(error);
4543 VLOG_WARN("learning action failed to modify flow table (%s)", msg);
4551 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
4553 if (port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
4554 ? htonl(OFPPC_NO_RECV_STP)
4555 : htonl(OFPPC_NO_RECV))) {
4559 /* Only drop packets here if both forwarding and learning are
4560 * disabled. If just learning is enabled, we need to have
4561 * OFPP_NORMAL and the learning action have a look at the packet
4562 * before we can drop it. */
4563 if (!stp_forward_in_state(port->stp_state)
4564 && !stp_learn_in_state(port->stp_state)) {
4572 do_xlate_actions(const union ofp_action *in, size_t n_in,
4573 struct action_xlate_ctx *ctx)
4575 const struct ofport_dpif *port;
4576 const union ofp_action *ia;
4579 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
4580 if (port && !may_receive(port, ctx)) {
4581 /* Drop this flow. */
4585 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
4586 const struct ofp_action_dl_addr *oada;
4587 const struct nx_action_resubmit *nar;
4588 const struct nx_action_set_tunnel *nast;
4589 const struct nx_action_set_queue *nasq;
4590 const struct nx_action_multipath *nam;
4591 const struct nx_action_autopath *naa;
4592 const struct nx_action_bundle *nab;
4593 const struct nx_action_output_reg *naor;
4594 enum ofputil_action_code code;
4601 code = ofputil_decode_action_unsafe(ia);
4603 case OFPUTIL_OFPAT_OUTPUT:
4604 xlate_output_action(ctx, &ia->output);
4607 case OFPUTIL_OFPAT_SET_VLAN_VID:
4608 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
4609 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
4612 case OFPUTIL_OFPAT_SET_VLAN_PCP:
4613 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
4614 ctx->flow.vlan_tci |= htons(
4615 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
4618 case OFPUTIL_OFPAT_STRIP_VLAN:
4619 ctx->flow.vlan_tci = htons(0);
4622 case OFPUTIL_OFPAT_SET_DL_SRC:
4623 oada = ((struct ofp_action_dl_addr *) ia);
4624 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
4627 case OFPUTIL_OFPAT_SET_DL_DST:
4628 oada = ((struct ofp_action_dl_addr *) ia);
4629 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
4632 case OFPUTIL_OFPAT_SET_NW_SRC:
4633 ctx->flow.nw_src = ia->nw_addr.nw_addr;
4636 case OFPUTIL_OFPAT_SET_NW_DST:
4637 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
4640 case OFPUTIL_OFPAT_SET_NW_TOS:
4641 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4642 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
4645 case OFPUTIL_OFPAT_SET_TP_SRC:
4646 ctx->flow.tp_src = ia->tp_port.tp_port;
4649 case OFPUTIL_OFPAT_SET_TP_DST:
4650 ctx->flow.tp_dst = ia->tp_port.tp_port;
4653 case OFPUTIL_OFPAT_ENQUEUE:
4654 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
4657 case OFPUTIL_NXAST_RESUBMIT:
4658 nar = (const struct nx_action_resubmit *) ia;
4659 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
4662 case OFPUTIL_NXAST_RESUBMIT_TABLE:
4663 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
4666 case OFPUTIL_NXAST_SET_TUNNEL:
4667 nast = (const struct nx_action_set_tunnel *) ia;
4668 tun_id = htonll(ntohl(nast->tun_id));
4669 ctx->flow.tun_id = tun_id;
4672 case OFPUTIL_NXAST_SET_QUEUE:
4673 nasq = (const struct nx_action_set_queue *) ia;
4674 xlate_set_queue_action(ctx, nasq);
4677 case OFPUTIL_NXAST_POP_QUEUE:
4678 ctx->flow.skb_priority = ctx->orig_skb_priority;
4681 case OFPUTIL_NXAST_REG_MOVE:
4682 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
4686 case OFPUTIL_NXAST_REG_LOAD:
4687 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
4691 case OFPUTIL_NXAST_NOTE:
4692 /* Nothing to do. */
4695 case OFPUTIL_NXAST_SET_TUNNEL64:
4696 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
4697 ctx->flow.tun_id = tun_id;
4700 case OFPUTIL_NXAST_MULTIPATH:
4701 nam = (const struct nx_action_multipath *) ia;
4702 multipath_execute(nam, &ctx->flow);
4705 case OFPUTIL_NXAST_AUTOPATH:
4706 naa = (const struct nx_action_autopath *) ia;
4707 xlate_autopath(ctx, naa);
4710 case OFPUTIL_NXAST_BUNDLE:
4711 ctx->ofproto->has_bundle_action = true;
4712 nab = (const struct nx_action_bundle *) ia;
4713 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
4718 case OFPUTIL_NXAST_BUNDLE_LOAD:
4719 ctx->ofproto->has_bundle_action = true;
4720 nab = (const struct nx_action_bundle *) ia;
4721 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
4725 case OFPUTIL_NXAST_OUTPUT_REG:
4726 naor = (const struct nx_action_output_reg *) ia;
4727 xlate_output_reg_action(ctx, naor);
4730 case OFPUTIL_NXAST_LEARN:
4731 ctx->has_learn = true;
4732 if (ctx->may_learn) {
4733 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
4737 case OFPUTIL_NXAST_EXIT:
4743 /* We've let OFPP_NORMAL and the learning action look at the packet,
4744 * so drop it now if forwarding is disabled. */
4745 if (port && !stp_forward_in_state(port->stp_state)) {
4746 ofpbuf_clear(ctx->odp_actions);
4747 add_sflow_action(ctx);
4752 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
4753 struct ofproto_dpif *ofproto, const struct flow *flow,
4754 ovs_be16 initial_tci, const struct ofpbuf *packet)
4756 ctx->ofproto = ofproto;
4758 ctx->base_flow = ctx->flow;
4759 ctx->base_flow.tun_id = 0;
4760 ctx->base_flow.vlan_tci = initial_tci;
4761 ctx->packet = packet;
4762 ctx->may_learn = packet != NULL;
4763 ctx->resubmit_hook = NULL;
4766 static struct ofpbuf *
4767 xlate_actions(struct action_xlate_ctx *ctx,
4768 const union ofp_action *in, size_t n_in)
4770 struct flow orig_flow = ctx->flow;
4772 COVERAGE_INC(ofproto_dpif_xlate);
4774 ctx->odp_actions = ofpbuf_new(512);
4775 ofpbuf_reserve(ctx->odp_actions, NL_A_U32_SIZE);
4777 ctx->may_set_up_flow = true;
4778 ctx->has_learn = false;
4779 ctx->has_normal = false;
4780 ctx->nf_output_iface = NF_OUT_DROP;
4783 ctx->orig_skb_priority = ctx->flow.skb_priority;
4787 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
4788 switch (ctx->ofproto->up.frag_handling) {
4789 case OFPC_FRAG_NORMAL:
4790 /* We must pretend that transport ports are unavailable. */
4791 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
4792 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
4795 case OFPC_FRAG_DROP:
4796 return ctx->odp_actions;
4798 case OFPC_FRAG_REASM:
4801 case OFPC_FRAG_NX_MATCH:
4802 /* Nothing to do. */
4807 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
4808 ctx->may_set_up_flow = false;
4809 return ctx->odp_actions;
4811 add_sflow_action(ctx);
4812 do_xlate_actions(in, n_in, ctx);
4814 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
4815 ctx->odp_actions->data,
4816 ctx->odp_actions->size)) {
4817 ctx->may_set_up_flow = false;
4819 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
4821 compose_output_action(ctx, OFPP_LOCAL);
4824 add_mirror_actions(ctx, &orig_flow);
4825 fix_sflow_action(ctx);
4828 return ctx->odp_actions;
4831 /* OFPP_NORMAL implementation. */
4833 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
4835 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
4836 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
4837 * the bundle on which the packet was received, returns the VLAN to which the
4840 * Both 'vid' and the return value are in the range 0...4095. */
4842 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
4844 switch (in_bundle->vlan_mode) {
4845 case PORT_VLAN_ACCESS:
4846 return in_bundle->vlan;
4849 case PORT_VLAN_TRUNK:
4852 case PORT_VLAN_NATIVE_UNTAGGED:
4853 case PORT_VLAN_NATIVE_TAGGED:
4854 return vid ? vid : in_bundle->vlan;
4861 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
4862 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
4865 * 'vid' should be the VID obtained from the 802.1Q header that was received as
4866 * part of a packet (specify 0 if there was no 802.1Q header), in the range
4869 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
4871 /* Allow any VID on the OFPP_NONE port. */
4872 if (in_bundle == &ofpp_none_bundle) {
4876 switch (in_bundle->vlan_mode) {
4877 case PORT_VLAN_ACCESS:
4880 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4881 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
4882 "packet received on port %s configured as VLAN "
4883 "%"PRIu16" access port",
4884 in_bundle->ofproto->up.name, vid,
4885 in_bundle->name, in_bundle->vlan);
4891 case PORT_VLAN_NATIVE_UNTAGGED:
4892 case PORT_VLAN_NATIVE_TAGGED:
4894 /* Port must always carry its native VLAN. */
4898 case PORT_VLAN_TRUNK:
4899 if (!ofbundle_includes_vlan(in_bundle, vid)) {
4901 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4902 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
4903 "received on port %s not configured for trunking "
4905 in_bundle->ofproto->up.name, vid,
4906 in_bundle->name, vid);
4918 /* Given 'vlan', the VLAN that a packet belongs to, and
4919 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
4920 * that should be included in the 802.1Q header. (If the return value is 0,
4921 * then the 802.1Q header should only be included in the packet if there is a
4924 * Both 'vlan' and the return value are in the range 0...4095. */
4926 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
4928 switch (out_bundle->vlan_mode) {
4929 case PORT_VLAN_ACCESS:
4932 case PORT_VLAN_TRUNK:
4933 case PORT_VLAN_NATIVE_TAGGED:
4936 case PORT_VLAN_NATIVE_UNTAGGED:
4937 return vlan == out_bundle->vlan ? 0 : vlan;
4945 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
4948 struct ofport_dpif *port;
4950 ovs_be16 tci, old_tci;
4952 vid = output_vlan_to_vid(out_bundle, vlan);
4953 if (!out_bundle->bond) {
4954 port = ofbundle_get_a_port(out_bundle);
4956 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
4959 /* No slaves enabled, so drop packet. */
4964 old_tci = ctx->flow.vlan_tci;
4966 if (tci || out_bundle->use_priority_tags) {
4967 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
4969 tci |= htons(VLAN_CFI);
4972 ctx->flow.vlan_tci = tci;
4974 compose_output_action(ctx, port->up.ofp_port);
4975 ctx->flow.vlan_tci = old_tci;
4979 mirror_mask_ffs(mirror_mask_t mask)
4981 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
4986 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
4988 return (bundle->vlan_mode != PORT_VLAN_ACCESS
4989 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
4993 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
4995 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
4998 /* Returns an arbitrary interface within 'bundle'. */
4999 static struct ofport_dpif *
5000 ofbundle_get_a_port(const struct ofbundle *bundle)
5002 return CONTAINER_OF(list_front(&bundle->ports),
5003 struct ofport_dpif, bundle_node);
5007 vlan_is_mirrored(const struct ofmirror *m, int vlan)
5009 return !m->vlans || bitmap_is_set(m->vlans, vlan);
5012 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
5013 * to a VLAN. In general most packets may be mirrored but we want to drop
5014 * protocols that may confuse switches. */
5016 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
5018 /* If you change this function's behavior, please update corresponding
5019 * documentation in vswitch.xml at the same time. */
5020 if (dst[0] != 0x01) {
5021 /* All the currently banned MACs happen to start with 01 currently, so
5022 * this is a quick way to eliminate most of the good ones. */
5024 if (eth_addr_is_reserved(dst)) {
5025 /* Drop STP, IEEE pause frames, and other reserved protocols
5026 * (01-80-c2-00-00-0x). */
5030 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
5032 if ((dst[3] & 0xfe) == 0xcc &&
5033 (dst[4] & 0xfe) == 0xcc &&
5034 (dst[5] & 0xfe) == 0xcc) {
5035 /* Drop the following protocols plus others following the same
5038 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
5039 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
5040 STP Uplink Fast (01-00-0c-cd-cd-cd) */
5044 if (!(dst[3] | dst[4] | dst[5])) {
5045 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
5054 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5056 struct ofproto_dpif *ofproto = ctx->ofproto;
5057 mirror_mask_t mirrors;
5058 struct ofbundle *in_bundle;
5061 const struct nlattr *a;
5064 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5065 ctx->packet != NULL);
5069 mirrors = in_bundle->src_mirrors;
5071 /* Drop frames on bundles reserved for mirroring. */
5072 if (in_bundle->mirror_out) {
5073 if (ctx->packet != NULL) {
5074 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5075 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5076 "%s, which is reserved exclusively for mirroring",
5077 ctx->ofproto->up.name, in_bundle->name);
5083 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5084 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5087 vlan = input_vid_to_vlan(in_bundle, vid);
5089 /* Look at the output ports to check for destination selections. */
5091 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5092 ctx->odp_actions->size) {
5093 enum ovs_action_attr type = nl_attr_type(a);
5094 struct ofport_dpif *ofport;
5096 if (type != OVS_ACTION_ATTR_OUTPUT) {
5100 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5101 if (ofport && ofport->bundle) {
5102 mirrors |= ofport->bundle->dst_mirrors;
5110 /* Restore the original packet before adding the mirror actions. */
5111 ctx->flow = *orig_flow;
5116 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5118 if (!vlan_is_mirrored(m, vlan)) {
5119 mirrors &= mirrors - 1;
5123 mirrors &= ~m->dup_mirrors;
5124 ctx->mirrors |= m->dup_mirrors;
5126 output_normal(ctx, m->out, vlan);
5127 } else if (eth_dst_may_rspan(orig_flow->dl_dst)
5128 && vlan != m->out_vlan) {
5129 struct ofbundle *bundle;
5131 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5132 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5133 && !bundle->mirror_out) {
5134 output_normal(ctx, bundle, m->out_vlan);
5142 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5143 uint64_t packets, uint64_t bytes)
5149 for (; mirrors; mirrors &= mirrors - 1) {
5152 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5155 /* In normal circumstances 'm' will not be NULL. However,
5156 * if mirrors are reconfigured, we can temporarily get out
5157 * of sync in facet_revalidate(). We could "correct" the
5158 * mirror list before reaching here, but doing that would
5159 * not properly account the traffic stats we've currently
5160 * accumulated for previous mirror configuration. */
5164 m->packet_count += packets;
5165 m->byte_count += bytes;
5169 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
5170 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
5171 * indicate this; newer upstream kernels use gratuitous ARP requests. */
5173 is_gratuitous_arp(const struct flow *flow)
5175 return (flow->dl_type == htons(ETH_TYPE_ARP)
5176 && eth_addr_is_broadcast(flow->dl_dst)
5177 && (flow->nw_proto == ARP_OP_REPLY
5178 || (flow->nw_proto == ARP_OP_REQUEST
5179 && flow->nw_src == flow->nw_dst)));
5183 update_learning_table(struct ofproto_dpif *ofproto,
5184 const struct flow *flow, int vlan,
5185 struct ofbundle *in_bundle)
5187 struct mac_entry *mac;
5189 /* Don't learn the OFPP_NONE port. */
5190 if (in_bundle == &ofpp_none_bundle) {
5194 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
5198 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
5199 if (is_gratuitous_arp(flow)) {
5200 /* We don't want to learn from gratuitous ARP packets that are
5201 * reflected back over bond slaves so we lock the learning table. */
5202 if (!in_bundle->bond) {
5203 mac_entry_set_grat_arp_lock(mac);
5204 } else if (mac_entry_is_grat_arp_locked(mac)) {
5209 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
5210 /* The log messages here could actually be useful in debugging,
5211 * so keep the rate limit relatively high. */
5212 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5213 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
5214 "on port %s in VLAN %d",
5215 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
5216 in_bundle->name, vlan);
5218 mac->port.p = in_bundle;
5219 tag_set_add(&ofproto->revalidate_set,
5220 mac_learning_changed(ofproto->ml, mac));
5224 static struct ofbundle *
5225 lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn)
5227 struct ofport_dpif *ofport;
5229 /* Special-case OFPP_NONE, which a controller may use as the ingress
5230 * port for traffic that it is sourcing. */
5231 if (in_port == OFPP_NONE) {
5232 return &ofpp_none_bundle;
5235 /* Find the port and bundle for the received packet. */
5236 ofport = get_ofp_port(ofproto, in_port);
5237 if (ofport && ofport->bundle) {
5238 return ofport->bundle;
5241 /* Odd. A few possible reasons here:
5243 * - We deleted a port but there are still a few packets queued up
5246 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
5247 * we don't know about.
5249 * - The ofproto client didn't configure the port as part of a bundle.
5252 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5254 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
5255 "port %"PRIu16, ofproto->up.name, in_port);
5260 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
5261 * dropped. Returns true if they may be forwarded, false if they should be
5264 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
5265 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
5267 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
5268 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
5269 * checked by input_vid_is_valid().
5271 * May also add tags to '*tags', although the current implementation only does
5272 * so in one special case.
5275 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
5276 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
5278 struct ofbundle *in_bundle = in_port->bundle;
5280 /* Drop frames for reserved multicast addresses
5281 * only if forward_bpdu option is absent. */
5282 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
5286 if (in_bundle->bond) {
5287 struct mac_entry *mac;
5289 switch (bond_check_admissibility(in_bundle->bond, in_port,
5290 flow->dl_dst, tags)) {
5297 case BV_DROP_IF_MOVED:
5298 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
5299 if (mac && mac->port.p != in_bundle &&
5300 (!is_gratuitous_arp(flow)
5301 || mac_entry_is_grat_arp_locked(mac))) {
5312 xlate_normal(struct action_xlate_ctx *ctx)
5314 struct ofport_dpif *in_port;
5315 struct ofbundle *in_bundle;
5316 struct mac_entry *mac;
5320 ctx->has_normal = true;
5322 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
5323 ctx->packet != NULL);
5328 /* We know 'in_port' exists unless it is "ofpp_none_bundle",
5329 * since lookup_input_bundle() succeeded. */
5330 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5332 /* Drop malformed frames. */
5333 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
5334 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
5335 if (ctx->packet != NULL) {
5336 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5337 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
5338 "VLAN tag received on port %s",
5339 ctx->ofproto->up.name, in_bundle->name);
5344 /* Drop frames on bundles reserved for mirroring. */
5345 if (in_bundle->mirror_out) {
5346 if (ctx->packet != NULL) {
5347 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5348 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5349 "%s, which is reserved exclusively for mirroring",
5350 ctx->ofproto->up.name, in_bundle->name);
5356 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
5357 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5360 vlan = input_vid_to_vlan(in_bundle, vid);
5362 /* Check other admissibility requirements. */
5364 !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
5368 /* Learn source MAC. */
5369 if (ctx->may_learn) {
5370 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
5373 /* Determine output bundle. */
5374 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
5377 if (mac->port.p != in_bundle) {
5378 output_normal(ctx, mac->port.p, vlan);
5381 struct ofbundle *bundle;
5383 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
5384 if (bundle != in_bundle
5385 && ofbundle_includes_vlan(bundle, vlan)
5386 && bundle->floodable
5387 && !bundle->mirror_out) {
5388 output_normal(ctx, bundle, vlan);
5391 ctx->nf_output_iface = NF_OUT_FLOOD;
5395 /* Optimized flow revalidation.
5397 * It's a difficult problem, in general, to tell which facets need to have
5398 * their actions recalculated whenever the OpenFlow flow table changes. We
5399 * don't try to solve that general problem: for most kinds of OpenFlow flow
5400 * table changes, we recalculate the actions for every facet. This is
5401 * relatively expensive, but it's good enough if the OpenFlow flow table
5402 * doesn't change very often.
5404 * However, we can expect one particular kind of OpenFlow flow table change to
5405 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5406 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5407 * table, we add a special case that applies to flow tables in which every rule
5408 * has the same form (that is, the same wildcards), except that the table is
5409 * also allowed to have a single "catch-all" flow that matches all packets. We
5410 * optimize this case by tagging all of the facets that resubmit into the table
5411 * and invalidating the same tag whenever a flow changes in that table. The
5412 * end result is that we revalidate just the facets that need it (and sometimes
5413 * a few more, but not all of the facets or even all of the facets that
5414 * resubmit to the table modified by MAC learning). */
5416 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
5417 * into an OpenFlow table with the given 'basis'. */
5419 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
5422 if (flow_wildcards_is_catchall(wc)) {
5425 struct flow tag_flow = *flow;
5426 flow_zero_wildcards(&tag_flow, wc);
5427 return tag_create_deterministic(flow_hash(&tag_flow, secret));
5431 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5432 * taggability of that table.
5434 * This function must be called after *each* change to a flow table. If you
5435 * skip calling it on some changes then the pointer comparisons at the end can
5436 * be invalid if you get unlucky. For example, if a flow removal causes a
5437 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5438 * different wildcards to be created with the same address, then this function
5439 * will incorrectly skip revalidation. */
5441 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5443 struct table_dpif *table = &ofproto->tables[table_id];
5444 const struct classifier *cls = &ofproto->up.tables[table_id];
5445 struct cls_table *catchall, *other;
5446 struct cls_table *t;
5448 catchall = other = NULL;
5450 switch (hmap_count(&cls->tables)) {
5452 /* We could tag this OpenFlow table but it would make the logic a
5453 * little harder and it's a corner case that doesn't seem worth it
5459 HMAP_FOR_EACH (t, hmap_node, &cls->tables) {
5460 if (cls_table_is_catchall(t)) {
5462 } else if (!other) {
5465 /* Indicate that we can't tag this by setting both tables to
5466 * NULL. (We know that 'catchall' is already NULL.) */
5473 /* Can't tag this table. */
5477 if (table->catchall_table != catchall || table->other_table != other) {
5478 table->catchall_table = catchall;
5479 table->other_table = other;
5480 ofproto->need_revalidate = true;
5484 /* Given 'rule' that has changed in some way (either it is a rule being
5485 * inserted, a rule being deleted, or a rule whose actions are being
5486 * modified), marks facets for revalidation to ensure that packets will be
5487 * forwarded correctly according to the new state of the flow table.
5489 * This function must be called after *each* change to a flow table. See
5490 * the comment on table_update_taggable() for more information. */
5492 rule_invalidate(const struct rule_dpif *rule)
5494 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5496 table_update_taggable(ofproto, rule->up.table_id);
5498 if (!ofproto->need_revalidate) {
5499 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5501 if (table->other_table && rule->tag) {
5502 tag_set_add(&ofproto->revalidate_set, rule->tag);
5504 ofproto->need_revalidate = true;
5510 set_frag_handling(struct ofproto *ofproto_,
5511 enum ofp_config_flags frag_handling)
5513 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5515 if (frag_handling != OFPC_FRAG_REASM) {
5516 ofproto->need_revalidate = true;
5524 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5525 const struct flow *flow,
5526 const union ofp_action *ofp_actions, size_t n_ofp_actions)
5528 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5531 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
5532 return ofp_mkerr_nicira(OFPET_BAD_REQUEST, NXBRC_BAD_IN_PORT);
5535 error = validate_actions(ofp_actions, n_ofp_actions, flow,
5536 ofproto->max_ports);
5538 struct odputil_keybuf keybuf;
5539 struct action_xlate_ctx ctx;
5540 struct ofpbuf *odp_actions;
5543 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5544 odp_flow_key_from_flow(&key, flow);
5546 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, packet);
5547 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
5548 dpif_execute(ofproto->dpif, key.data, key.size,
5549 odp_actions->data, odp_actions->size, packet);
5550 ofpbuf_delete(odp_actions);
5558 set_netflow(struct ofproto *ofproto_,
5559 const struct netflow_options *netflow_options)
5561 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5563 if (netflow_options) {
5564 if (!ofproto->netflow) {
5565 ofproto->netflow = netflow_create();
5567 return netflow_set_options(ofproto->netflow, netflow_options);
5569 netflow_destroy(ofproto->netflow);
5570 ofproto->netflow = NULL;
5576 get_netflow_ids(const struct ofproto *ofproto_,
5577 uint8_t *engine_type, uint8_t *engine_id)
5579 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5581 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
5585 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5587 if (!facet_is_controller_flow(facet) &&
5588 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5589 struct subfacet *subfacet;
5590 struct ofexpired expired;
5592 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5593 if (subfacet->installed) {
5594 struct dpif_flow_stats stats;
5596 subfacet_install(ofproto, subfacet, subfacet->actions,
5597 subfacet->actions_len, &stats);
5598 subfacet_update_stats(ofproto, subfacet, &stats);
5602 expired.flow = facet->flow;
5603 expired.packet_count = facet->packet_count;
5604 expired.byte_count = facet->byte_count;
5605 expired.used = facet->used;
5606 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5611 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5613 struct facet *facet;
5615 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5616 send_active_timeout(ofproto, facet);
5620 static struct ofproto_dpif *
5621 ofproto_dpif_lookup(const char *name)
5623 struct ofproto_dpif *ofproto;
5625 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5626 hash_string(name, 0), &all_ofproto_dpifs) {
5627 if (!strcmp(ofproto->up.name, name)) {
5635 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc OVS_UNUSED,
5636 const char *argv[], void *aux OVS_UNUSED)
5638 const struct ofproto_dpif *ofproto;
5640 ofproto = ofproto_dpif_lookup(argv[1]);
5642 unixctl_command_reply(conn, 501, "no such bridge");
5645 mac_learning_flush(ofproto->ml);
5647 unixctl_command_reply(conn, 200, "table successfully flushed");
5651 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5652 const char *argv[], void *aux OVS_UNUSED)
5654 struct ds ds = DS_EMPTY_INITIALIZER;
5655 const struct ofproto_dpif *ofproto;
5656 const struct mac_entry *e;
5658 ofproto = ofproto_dpif_lookup(argv[1]);
5660 unixctl_command_reply(conn, 501, "no such bridge");
5664 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5665 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5666 struct ofbundle *bundle = e->port.p;
5667 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
5668 ofbundle_get_a_port(bundle)->odp_port,
5669 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
5671 unixctl_command_reply(conn, 200, ds_cstr(&ds));
5675 struct ofproto_trace {
5676 struct action_xlate_ctx ctx;
5682 trace_format_rule(struct ds *result, uint8_t table_id, int level,
5683 const struct rule_dpif *rule)
5685 ds_put_char_multiple(result, '\t', level);
5687 ds_put_cstr(result, "No match\n");
5691 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5692 table_id, ntohll(rule->up.flow_cookie));
5693 cls_rule_format(&rule->up.cr, result);
5694 ds_put_char(result, '\n');
5696 ds_put_char_multiple(result, '\t', level);
5697 ds_put_cstr(result, "OpenFlow ");
5698 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
5699 ds_put_char(result, '\n');
5703 trace_format_flow(struct ds *result, int level, const char *title,
5704 struct ofproto_trace *trace)
5706 ds_put_char_multiple(result, '\t', level);
5707 ds_put_format(result, "%s: ", title);
5708 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
5709 ds_put_cstr(result, "unchanged");
5711 flow_format(result, &trace->ctx.flow);
5712 trace->flow = trace->ctx.flow;
5714 ds_put_char(result, '\n');
5718 trace_format_regs(struct ds *result, int level, const char *title,
5719 struct ofproto_trace *trace)
5723 ds_put_char_multiple(result, '\t', level);
5724 ds_put_format(result, "%s:", title);
5725 for (i = 0; i < FLOW_N_REGS; i++) {
5726 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5728 ds_put_char(result, '\n');
5732 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
5734 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
5735 struct ds *result = trace->result;
5737 ds_put_char(result, '\n');
5738 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
5739 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
5740 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
5744 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5745 void *aux OVS_UNUSED)
5747 const char *dpname = argv[1];
5748 struct ofproto_dpif *ofproto;
5749 struct ofpbuf odp_key;
5750 struct ofpbuf *packet;
5751 struct rule_dpif *rule;
5752 ovs_be16 initial_tci;
5758 ofpbuf_init(&odp_key, 0);
5761 ofproto = ofproto_dpif_lookup(dpname);
5763 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
5767 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
5768 /* ofproto/trace dpname flow [-generate] */
5769 const char *flow_s = argv[2];
5770 const char *generate_s = argv[3];
5773 /* Convert string to datapath key. */
5774 ofpbuf_init(&odp_key, 0);
5775 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
5777 unixctl_command_reply(conn, 501, "Bad flow syntax");
5781 /* Convert odp_key to flow. */
5782 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
5783 odp_key.size, &flow,
5784 &initial_tci, NULL);
5785 if (error == ODP_FIT_ERROR) {
5786 unixctl_command_reply(conn, 501, "Invalid flow");
5790 /* Generate a packet, if requested. */
5792 packet = ofpbuf_new(0);
5793 flow_compose(packet, &flow);
5795 } else if (argc == 6) {
5796 /* ofproto/trace dpname priority tun_id in_port packet */
5797 const char *priority_s = argv[2];
5798 const char *tun_id_s = argv[3];
5799 const char *in_port_s = argv[4];
5800 const char *packet_s = argv[5];
5801 uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s));
5802 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
5803 uint32_t priority = atoi(priority_s);
5806 msg = eth_from_hex(packet_s, &packet);
5808 unixctl_command_reply(conn, 501, msg);
5812 ds_put_cstr(&result, "Packet: ");
5813 s = ofp_packet_to_string(packet->data, packet->size, packet->size);
5814 ds_put_cstr(&result, s);
5817 flow_extract(packet, priority, tun_id, in_port, &flow);
5818 initial_tci = flow.vlan_tci;
5820 unixctl_command_reply(conn, 501, "Bad command syntax");
5824 ds_put_cstr(&result, "Flow: ");
5825 flow_format(&result, &flow);
5826 ds_put_char(&result, '\n');
5828 rule = rule_dpif_lookup(ofproto, &flow, 0);
5829 trace_format_rule(&result, 0, 0, rule);
5831 struct ofproto_trace trace;
5832 struct ofpbuf *odp_actions;
5834 trace.result = &result;
5836 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, initial_tci, packet);
5837 trace.ctx.resubmit_hook = trace_resubmit;
5838 odp_actions = xlate_actions(&trace.ctx,
5839 rule->up.actions, rule->up.n_actions);
5841 ds_put_char(&result, '\n');
5842 trace_format_flow(&result, 0, "Final flow", &trace);
5843 ds_put_cstr(&result, "Datapath actions: ");
5844 format_odp_actions(&result, odp_actions->data, odp_actions->size);
5845 ofpbuf_delete(odp_actions);
5847 if (!trace.ctx.may_set_up_flow) {
5849 ds_put_cstr(&result, "\nThis flow is not cachable.");
5851 ds_put_cstr(&result, "\nThe datapath actions are incomplete--"
5852 "for complete actions, please supply a packet.");
5857 unixctl_command_reply(conn, 200, ds_cstr(&result));
5860 ds_destroy(&result);
5861 ofpbuf_delete(packet);
5862 ofpbuf_uninit(&odp_key);
5866 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5867 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5870 unixctl_command_reply(conn, 200, NULL);
5874 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5875 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5878 unixctl_command_reply(conn, 200, NULL);
5882 ofproto_dpif_unixctl_init(void)
5884 static bool registered;
5890 unixctl_command_register(
5892 "bridge {tun_id in_port packet | odp_flow [-generate]}",
5893 2, 4, ofproto_unixctl_trace, NULL);
5894 unixctl_command_register("fdb/flush", "bridge", 1, 1,
5895 ofproto_unixctl_fdb_flush, NULL);
5896 unixctl_command_register("fdb/show", "bridge", 1, 1,
5897 ofproto_unixctl_fdb_show, NULL);
5898 unixctl_command_register("ofproto/clog", "", 0, 0,
5899 ofproto_dpif_clog, NULL);
5900 unixctl_command_register("ofproto/unclog", "", 0, 0,
5901 ofproto_dpif_unclog, NULL);
5904 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5906 * This is deprecated. It is only for compatibility with broken device drivers
5907 * in old versions of Linux that do not properly support VLANs when VLAN
5908 * devices are not used. When broken device drivers are no longer in
5909 * widespread use, we will delete these interfaces. */
5912 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
5914 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
5915 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
5917 if (realdev_ofp_port == ofport->realdev_ofp_port
5918 && vid == ofport->vlandev_vid) {
5922 ofproto->need_revalidate = true;
5924 if (ofport->realdev_ofp_port) {
5927 if (realdev_ofp_port && ofport->bundle) {
5928 /* vlandevs are enslaved to their realdevs, so they are not allowed to
5929 * themselves be part of a bundle. */
5930 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
5933 ofport->realdev_ofp_port = realdev_ofp_port;
5934 ofport->vlandev_vid = vid;
5936 if (realdev_ofp_port) {
5937 vsp_add(ofport, realdev_ofp_port, vid);
5944 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
5946 return hash_2words(realdev_ofp_port, vid);
5950 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
5951 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
5953 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
5954 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
5955 int vid = vlan_tci_to_vid(vlan_tci);
5956 const struct vlan_splinter *vsp;
5958 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
5959 hash_realdev_vid(realdev_ofp_port, vid),
5960 &ofproto->realdev_vid_map) {
5961 if (vsp->realdev_ofp_port == realdev_ofp_port
5962 && vsp->vid == vid) {
5963 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
5967 return realdev_odp_port;
5970 static struct vlan_splinter *
5971 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
5973 struct vlan_splinter *vsp;
5975 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
5976 &ofproto->vlandev_map) {
5977 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
5986 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
5987 uint16_t vlandev_ofp_port, int *vid)
5989 if (!hmap_is_empty(&ofproto->vlandev_map)) {
5990 const struct vlan_splinter *vsp;
5992 vsp = vlandev_find(ofproto, vlandev_ofp_port);
5997 return vsp->realdev_ofp_port;
6004 vsp_remove(struct ofport_dpif *port)
6006 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6007 struct vlan_splinter *vsp;
6009 vsp = vlandev_find(ofproto, port->up.ofp_port);
6011 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6012 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6015 port->realdev_ofp_port = 0;
6017 VLOG_ERR("missing vlan device record");
6022 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
6024 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6026 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6027 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6028 == realdev_ofp_port)) {
6029 struct vlan_splinter *vsp;
6031 vsp = xmalloc(sizeof *vsp);
6032 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6033 hash_int(port->up.ofp_port, 0));
6034 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6035 hash_realdev_vid(realdev_ofp_port, vid));
6036 vsp->realdev_ofp_port = realdev_ofp_port;
6037 vsp->vlandev_ofp_port = port->up.ofp_port;
6040 port->realdev_ofp_port = realdev_ofp_port;
6042 VLOG_ERR("duplicate vlan device record");
6046 const struct ofproto_class ofproto_dpif_class = {
6075 port_is_lacp_current,
6076 NULL, /* rule_choose_table */
6083 rule_modify_actions,
6091 get_cfm_remote_mpids,
6095 get_stp_port_status,
6102 is_mirror_output_bundle,
6103 forward_bpdu_changed,