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 void facet_flush_stats(struct ofproto_dpif *, struct facet *);
329 static void facet_update_time(struct ofproto_dpif *, struct facet *,
331 static void facet_reset_counters(struct facet *);
332 static void facet_push_stats(struct facet *);
333 static void facet_account(struct ofproto_dpif *, struct facet *);
335 static bool facet_is_controller_flow(struct facet *);
337 /* A dpif flow and actions associated with a facet.
339 * See also the large comment on struct facet. */
342 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
343 struct list list_node; /* In struct facet's 'facets' list. */
344 struct facet *facet; /* Owning facet. */
348 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
349 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
350 * regenerate the ODP flow key from ->facet->flow. */
351 enum odp_key_fitness key_fitness;
355 long long int used; /* Time last used; time created if not used. */
357 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
358 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
362 * These should be essentially identical for every subfacet in a facet, but
363 * may differ in trivial ways due to VLAN splinters. */
364 size_t actions_len; /* Number of bytes in actions[]. */
365 struct nlattr *actions; /* Datapath actions. */
367 bool installed; /* Installed in datapath? */
369 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
370 * splinters can cause it to differ. This value should be removed when
371 * the VLAN splinters feature is no longer needed. */
372 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
375 static struct subfacet *subfacet_create(struct ofproto_dpif *, struct facet *,
376 enum odp_key_fitness,
377 const struct nlattr *key,
378 size_t key_len, ovs_be16 initial_tci);
379 static struct subfacet *subfacet_find(struct ofproto_dpif *,
380 const struct nlattr *key, size_t key_len);
381 static void subfacet_destroy(struct ofproto_dpif *, struct subfacet *);
382 static void subfacet_destroy__(struct ofproto_dpif *, struct subfacet *);
383 static void subfacet_reset_dp_stats(struct subfacet *,
384 struct dpif_flow_stats *);
385 static void subfacet_update_time(struct ofproto_dpif *, struct subfacet *,
387 static void subfacet_update_stats(struct ofproto_dpif *, struct subfacet *,
388 const struct dpif_flow_stats *);
389 static void subfacet_make_actions(struct ofproto_dpif *, struct subfacet *,
390 const struct ofpbuf *packet);
391 static int subfacet_install(struct ofproto_dpif *, struct subfacet *,
392 const struct nlattr *actions, size_t actions_len,
393 struct dpif_flow_stats *);
394 static void subfacet_uninstall(struct ofproto_dpif *, struct subfacet *);
400 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
401 struct list bundle_node; /* In struct ofbundle's "ports" list. */
402 struct cfm *cfm; /* Connectivity Fault Management, if any. */
403 tag_type tag; /* Tag associated with this port. */
404 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
405 bool may_enable; /* May be enabled in bonds. */
408 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
409 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
410 long long int stp_state_entered;
412 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
414 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
416 * This is deprecated. It is only for compatibility with broken device
417 * drivers in old versions of Linux that do not properly support VLANs when
418 * VLAN devices are not used. When broken device drivers are no longer in
419 * widespread use, we will delete these interfaces. */
420 uint16_t realdev_ofp_port;
424 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
425 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
426 * traffic egressing the 'ofport' with that priority should be marked with. */
427 struct priority_to_dscp {
428 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
429 uint32_t priority; /* Priority of this queue (see struct flow). */
431 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
434 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
436 * This is deprecated. It is only for compatibility with broken device drivers
437 * in old versions of Linux that do not properly support VLANs when VLAN
438 * devices are not used. When broken device drivers are no longer in
439 * widespread use, we will delete these interfaces. */
440 struct vlan_splinter {
441 struct hmap_node realdev_vid_node;
442 struct hmap_node vlandev_node;
443 uint16_t realdev_ofp_port;
444 uint16_t vlandev_ofp_port;
448 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
449 uint32_t realdev, ovs_be16 vlan_tci);
450 static uint16_t vsp_vlandev_to_realdev(const struct ofproto_dpif *,
451 uint16_t vlandev, int *vid);
452 static void vsp_remove(struct ofport_dpif *);
453 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
455 static struct ofport_dpif *
456 ofport_dpif_cast(const struct ofport *ofport)
458 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
459 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
462 static void port_run(struct ofport_dpif *);
463 static void port_wait(struct ofport_dpif *);
464 static int set_cfm(struct ofport *, const struct cfm_settings *);
465 static void ofport_clear_priorities(struct ofport_dpif *);
467 struct dpif_completion {
468 struct list list_node;
469 struct ofoperation *op;
472 /* Extra information about a classifier table.
473 * Currently used just for optimized flow revalidation. */
475 /* If either of these is nonnull, then this table has a form that allows
476 * flows to be tagged to avoid revalidating most flows for the most common
477 * kinds of flow table changes. */
478 struct cls_table *catchall_table; /* Table that wildcards all fields. */
479 struct cls_table *other_table; /* Table with any other wildcard set. */
480 uint32_t basis; /* Keeps each table's tags separate. */
483 struct ofproto_dpif {
484 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
493 struct netflow *netflow;
494 struct dpif_sflow *sflow;
495 struct hmap bundles; /* Contains "struct ofbundle"s. */
496 struct mac_learning *ml;
497 struct ofmirror *mirrors[MAX_MIRRORS];
498 bool has_bonded_bundles;
501 struct timer next_expiration;
505 struct hmap subfacets;
508 struct table_dpif tables[N_TABLES];
509 bool need_revalidate;
510 struct tag_set revalidate_set;
512 /* Support for debugging async flow mods. */
513 struct list completions;
515 bool has_bundle_action; /* True when the first bundle action appears. */
516 struct netdev_stats stats; /* To account packets generated and consumed in
521 long long int stp_last_tick;
523 /* VLAN splinters. */
524 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
525 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
528 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
529 * for debugging the asynchronous flow_mod implementation.) */
532 /* All existing ofproto_dpif instances, indexed by ->up.name. */
533 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
535 static void ofproto_dpif_unixctl_init(void);
537 static struct ofproto_dpif *
538 ofproto_dpif_cast(const struct ofproto *ofproto)
540 assert(ofproto->ofproto_class == &ofproto_dpif_class);
541 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
544 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
546 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
549 /* Packet processing. */
550 static void update_learning_table(struct ofproto_dpif *,
551 const struct flow *, int vlan,
554 #define FLOW_MISS_MAX_BATCH 50
555 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
557 /* Flow expiration. */
558 static int expire(struct ofproto_dpif *);
561 static void send_netflow_active_timeouts(struct ofproto_dpif *);
564 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
566 compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions,
567 const struct flow *, uint32_t odp_port);
568 static void add_mirror_actions(struct action_xlate_ctx *ctx,
569 const struct flow *flow);
570 /* Global variables. */
571 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
573 /* Factory functions. */
576 enumerate_types(struct sset *types)
578 dp_enumerate_types(types);
582 enumerate_names(const char *type, struct sset *names)
584 return dp_enumerate_names(type, names);
588 del(const char *type, const char *name)
593 error = dpif_open(name, type, &dpif);
595 error = dpif_delete(dpif);
601 /* Basic life-cycle. */
603 static struct ofproto *
606 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
611 dealloc(struct ofproto *ofproto_)
613 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
618 construct(struct ofproto *ofproto_, int *n_tablesp)
620 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
621 const char *name = ofproto->up.name;
625 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
627 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
631 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
632 ofproto->n_matches = 0;
634 dpif_flow_flush(ofproto->dpif);
635 dpif_recv_purge(ofproto->dpif);
637 error = dpif_recv_set_mask(ofproto->dpif,
638 ((1u << DPIF_UC_MISS) |
639 (1u << DPIF_UC_ACTION)));
641 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
642 dpif_close(ofproto->dpif);
646 ofproto->netflow = NULL;
647 ofproto->sflow = NULL;
649 hmap_init(&ofproto->bundles);
650 ofproto->ml = mac_learning_create();
651 for (i = 0; i < MAX_MIRRORS; i++) {
652 ofproto->mirrors[i] = NULL;
654 ofproto->has_bonded_bundles = false;
656 timer_set_duration(&ofproto->next_expiration, 1000);
658 hmap_init(&ofproto->facets);
659 hmap_init(&ofproto->subfacets);
661 for (i = 0; i < N_TABLES; i++) {
662 struct table_dpif *table = &ofproto->tables[i];
664 table->catchall_table = NULL;
665 table->other_table = NULL;
666 table->basis = random_uint32();
668 ofproto->need_revalidate = false;
669 tag_set_init(&ofproto->revalidate_set);
671 list_init(&ofproto->completions);
673 ofproto_dpif_unixctl_init();
675 ofproto->has_bundle_action = false;
677 hmap_init(&ofproto->vlandev_map);
678 hmap_init(&ofproto->realdev_vid_map);
680 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
681 hash_string(ofproto->up.name, 0));
683 *n_tablesp = N_TABLES;
684 memset(&ofproto->stats, 0, sizeof ofproto->stats);
689 complete_operations(struct ofproto_dpif *ofproto)
691 struct dpif_completion *c, *next;
693 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
694 ofoperation_complete(c->op, 0);
695 list_remove(&c->list_node);
701 destruct(struct ofproto *ofproto_)
703 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
704 struct rule_dpif *rule, *next_rule;
705 struct classifier *table;
708 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
709 complete_operations(ofproto);
711 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
712 struct cls_cursor cursor;
714 cls_cursor_init(&cursor, table, NULL);
715 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
716 ofproto_rule_destroy(&rule->up);
720 for (i = 0; i < MAX_MIRRORS; i++) {
721 mirror_destroy(ofproto->mirrors[i]);
724 netflow_destroy(ofproto->netflow);
725 dpif_sflow_destroy(ofproto->sflow);
726 hmap_destroy(&ofproto->bundles);
727 mac_learning_destroy(ofproto->ml);
729 hmap_destroy(&ofproto->facets);
730 hmap_destroy(&ofproto->subfacets);
732 hmap_destroy(&ofproto->vlandev_map);
733 hmap_destroy(&ofproto->realdev_vid_map);
735 dpif_close(ofproto->dpif);
739 run_fast(struct ofproto *ofproto_)
741 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
744 /* Handle one or more batches of upcalls, until there's nothing left to do
745 * or until we do a fixed total amount of work.
747 * We do work in batches because it can be much cheaper to set up a number
748 * of flows and fire off their patches all at once. We do multiple batches
749 * because in some cases handling a packet can cause another packet to be
750 * queued almost immediately as part of the return flow. Both
751 * optimizations can make major improvements on some benchmarks and
752 * presumably for real traffic as well. */
754 while (work < FLOW_MISS_MAX_BATCH) {
755 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
765 run(struct ofproto *ofproto_)
767 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
768 struct ofport_dpif *ofport;
769 struct ofbundle *bundle;
773 complete_operations(ofproto);
775 dpif_run(ofproto->dpif);
777 error = run_fast(ofproto_);
782 if (timer_expired(&ofproto->next_expiration)) {
783 int delay = expire(ofproto);
784 timer_set_duration(&ofproto->next_expiration, delay);
787 if (ofproto->netflow) {
788 if (netflow_run(ofproto->netflow)) {
789 send_netflow_active_timeouts(ofproto);
792 if (ofproto->sflow) {
793 dpif_sflow_run(ofproto->sflow);
796 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
799 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
804 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
806 /* Now revalidate if there's anything to do. */
807 if (ofproto->need_revalidate
808 || !tag_set_is_empty(&ofproto->revalidate_set)) {
809 struct tag_set revalidate_set = ofproto->revalidate_set;
810 bool revalidate_all = ofproto->need_revalidate;
811 struct facet *facet, *next;
813 /* Clear the revalidation flags. */
814 tag_set_init(&ofproto->revalidate_set);
815 ofproto->need_revalidate = false;
817 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
819 || tag_set_intersects(&revalidate_set, facet->tags)) {
820 facet_revalidate(ofproto, facet);
829 wait(struct ofproto *ofproto_)
831 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
832 struct ofport_dpif *ofport;
833 struct ofbundle *bundle;
835 if (!clogged && !list_is_empty(&ofproto->completions)) {
836 poll_immediate_wake();
839 dpif_wait(ofproto->dpif);
840 dpif_recv_wait(ofproto->dpif);
841 if (ofproto->sflow) {
842 dpif_sflow_wait(ofproto->sflow);
844 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
845 poll_immediate_wake();
847 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
850 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
853 if (ofproto->netflow) {
854 netflow_wait(ofproto->netflow);
856 mac_learning_wait(ofproto->ml);
858 if (ofproto->need_revalidate) {
859 /* Shouldn't happen, but if it does just go around again. */
860 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
861 poll_immediate_wake();
863 timer_wait(&ofproto->next_expiration);
868 flush(struct ofproto *ofproto_)
870 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
871 struct facet *facet, *next_facet;
873 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
874 /* Mark the facet as not installed so that facet_remove() doesn't
875 * bother trying to uninstall it. There is no point in uninstalling it
876 * individually since we are about to blow away all the facets with
877 * dpif_flow_flush(). */
878 struct subfacet *subfacet;
880 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
881 subfacet->installed = false;
882 subfacet->dp_packet_count = 0;
883 subfacet->dp_byte_count = 0;
885 facet_remove(ofproto, facet);
887 dpif_flow_flush(ofproto->dpif);
891 get_features(struct ofproto *ofproto_ OVS_UNUSED,
892 bool *arp_match_ip, uint32_t *actions)
894 *arp_match_ip = true;
895 *actions = ((1u << OFPAT_OUTPUT) |
896 (1u << OFPAT_SET_VLAN_VID) |
897 (1u << OFPAT_SET_VLAN_PCP) |
898 (1u << OFPAT_STRIP_VLAN) |
899 (1u << OFPAT_SET_DL_SRC) |
900 (1u << OFPAT_SET_DL_DST) |
901 (1u << OFPAT_SET_NW_SRC) |
902 (1u << OFPAT_SET_NW_DST) |
903 (1u << OFPAT_SET_NW_TOS) |
904 (1u << OFPAT_SET_TP_SRC) |
905 (1u << OFPAT_SET_TP_DST) |
906 (1u << OFPAT_ENQUEUE));
910 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
912 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
913 struct dpif_dp_stats s;
915 strcpy(ots->name, "classifier");
917 dpif_get_dp_stats(ofproto->dpif, &s);
918 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
919 put_32aligned_be64(&ots->matched_count,
920 htonll(s.n_hit + ofproto->n_matches));
923 static struct ofport *
926 struct ofport_dpif *port = xmalloc(sizeof *port);
931 port_dealloc(struct ofport *port_)
933 struct ofport_dpif *port = ofport_dpif_cast(port_);
938 port_construct(struct ofport *port_)
940 struct ofport_dpif *port = ofport_dpif_cast(port_);
941 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
943 ofproto->need_revalidate = true;
944 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
947 port->tag = tag_create_random();
948 port->may_enable = true;
949 port->stp_port = NULL;
950 port->stp_state = STP_DISABLED;
951 hmap_init(&port->priorities);
952 port->realdev_ofp_port = 0;
953 port->vlandev_vid = 0;
955 if (ofproto->sflow) {
956 dpif_sflow_add_port(ofproto->sflow, port_);
963 port_destruct(struct ofport *port_)
965 struct ofport_dpif *port = ofport_dpif_cast(port_);
966 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
968 ofproto->need_revalidate = true;
969 bundle_remove(port_);
970 set_cfm(port_, NULL);
971 if (ofproto->sflow) {
972 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
975 ofport_clear_priorities(port);
976 hmap_destroy(&port->priorities);
980 port_modified(struct ofport *port_)
982 struct ofport_dpif *port = ofport_dpif_cast(port_);
984 if (port->bundle && port->bundle->bond) {
985 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
990 port_reconfigured(struct ofport *port_, ovs_be32 old_config)
992 struct ofport_dpif *port = ofport_dpif_cast(port_);
993 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
994 ovs_be32 changed = old_config ^ port->up.opp.config;
996 if (changed & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
997 OFPPC_NO_FWD | OFPPC_NO_FLOOD)) {
998 ofproto->need_revalidate = true;
1000 if (changed & htonl(OFPPC_NO_FLOOD) && port->bundle) {
1001 bundle_update(port->bundle);
1007 set_sflow(struct ofproto *ofproto_,
1008 const struct ofproto_sflow_options *sflow_options)
1010 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1011 struct dpif_sflow *ds = ofproto->sflow;
1013 if (sflow_options) {
1015 struct ofport_dpif *ofport;
1017 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1018 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1019 dpif_sflow_add_port(ds, &ofport->up);
1021 ofproto->need_revalidate = true;
1023 dpif_sflow_set_options(ds, sflow_options);
1026 dpif_sflow_destroy(ds);
1027 ofproto->need_revalidate = true;
1028 ofproto->sflow = NULL;
1035 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1037 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1044 struct ofproto_dpif *ofproto;
1046 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1047 ofproto->need_revalidate = true;
1048 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1051 if (cfm_configure(ofport->cfm, s)) {
1057 cfm_destroy(ofport->cfm);
1063 get_cfm_fault(const struct ofport *ofport_)
1065 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1067 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1071 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1074 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1077 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1084 /* Spanning Tree. */
1087 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1089 struct ofproto_dpif *ofproto = ofproto_;
1090 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1091 struct ofport_dpif *ofport;
1093 ofport = stp_port_get_aux(sp);
1095 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1096 ofproto->up.name, port_num);
1098 struct eth_header *eth = pkt->l2;
1100 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1101 if (eth_addr_is_zero(eth->eth_src)) {
1102 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1103 "with unknown MAC", ofproto->up.name, port_num);
1105 send_packet(ofport, pkt);
1111 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1113 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1115 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1117 /* Only revalidate flows if the configuration changed. */
1118 if (!s != !ofproto->stp) {
1119 ofproto->need_revalidate = true;
1123 if (!ofproto->stp) {
1124 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1125 send_bpdu_cb, ofproto);
1126 ofproto->stp_last_tick = time_msec();
1129 stp_set_bridge_id(ofproto->stp, s->system_id);
1130 stp_set_bridge_priority(ofproto->stp, s->priority);
1131 stp_set_hello_time(ofproto->stp, s->hello_time);
1132 stp_set_max_age(ofproto->stp, s->max_age);
1133 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1135 stp_destroy(ofproto->stp);
1136 ofproto->stp = NULL;
1143 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1145 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1149 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1150 s->designated_root = stp_get_designated_root(ofproto->stp);
1151 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1160 update_stp_port_state(struct ofport_dpif *ofport)
1162 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1163 enum stp_state state;
1165 /* Figure out new state. */
1166 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1170 if (ofport->stp_state != state) {
1174 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1175 netdev_get_name(ofport->up.netdev),
1176 stp_state_name(ofport->stp_state),
1177 stp_state_name(state));
1178 if (stp_learn_in_state(ofport->stp_state)
1179 != stp_learn_in_state(state)) {
1180 /* xxx Learning action flows should also be flushed. */
1181 mac_learning_flush(ofproto->ml);
1183 fwd_change = stp_forward_in_state(ofport->stp_state)
1184 != stp_forward_in_state(state);
1186 ofproto->need_revalidate = true;
1187 ofport->stp_state = state;
1188 ofport->stp_state_entered = time_msec();
1190 if (fwd_change && ofport->bundle) {
1191 bundle_update(ofport->bundle);
1194 /* Update the STP state bits in the OpenFlow port description. */
1195 of_state = (ofport->up.opp.state & htonl(~OFPPS_STP_MASK))
1196 | htonl(state == STP_LISTENING ? OFPPS_STP_LISTEN
1197 : state == STP_LEARNING ? OFPPS_STP_LEARN
1198 : state == STP_FORWARDING ? OFPPS_STP_FORWARD
1199 : state == STP_BLOCKING ? OFPPS_STP_BLOCK
1201 ofproto_port_set_state(&ofport->up, of_state);
1205 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1206 * caller is responsible for assigning STP port numbers and ensuring
1207 * there are no duplicates. */
1209 set_stp_port(struct ofport *ofport_,
1210 const struct ofproto_port_stp_settings *s)
1212 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1213 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1214 struct stp_port *sp = ofport->stp_port;
1216 if (!s || !s->enable) {
1218 ofport->stp_port = NULL;
1219 stp_port_disable(sp);
1220 update_stp_port_state(ofport);
1223 } else if (sp && stp_port_no(sp) != s->port_num
1224 && ofport == stp_port_get_aux(sp)) {
1225 /* The port-id changed, so disable the old one if it's not
1226 * already in use by another port. */
1227 stp_port_disable(sp);
1230 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1231 stp_port_enable(sp);
1233 stp_port_set_aux(sp, ofport);
1234 stp_port_set_priority(sp, s->priority);
1235 stp_port_set_path_cost(sp, s->path_cost);
1237 update_stp_port_state(ofport);
1243 get_stp_port_status(struct ofport *ofport_,
1244 struct ofproto_port_stp_status *s)
1246 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1247 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1248 struct stp_port *sp = ofport->stp_port;
1250 if (!ofproto->stp || !sp) {
1256 s->port_id = stp_port_get_id(sp);
1257 s->state = stp_port_get_state(sp);
1258 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1259 s->role = stp_port_get_role(sp);
1260 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1266 stp_run(struct ofproto_dpif *ofproto)
1269 long long int now = time_msec();
1270 long long int elapsed = now - ofproto->stp_last_tick;
1271 struct stp_port *sp;
1274 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1275 ofproto->stp_last_tick = now;
1277 while (stp_get_changed_port(ofproto->stp, &sp)) {
1278 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1281 update_stp_port_state(ofport);
1288 stp_wait(struct ofproto_dpif *ofproto)
1291 poll_timer_wait(1000);
1295 /* Returns true if STP should process 'flow'. */
1297 stp_should_process_flow(const struct flow *flow)
1299 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1303 stp_process_packet(const struct ofport_dpif *ofport,
1304 const struct ofpbuf *packet)
1306 struct ofpbuf payload = *packet;
1307 struct eth_header *eth = payload.data;
1308 struct stp_port *sp = ofport->stp_port;
1310 /* Sink packets on ports that have STP disabled when the bridge has
1312 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1316 /* Trim off padding on payload. */
1317 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1318 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1321 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1322 stp_received_bpdu(sp, payload.data, payload.size);
1326 static struct priority_to_dscp *
1327 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1329 struct priority_to_dscp *pdscp;
1332 hash = hash_int(priority, 0);
1333 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1334 if (pdscp->priority == priority) {
1342 ofport_clear_priorities(struct ofport_dpif *ofport)
1344 struct priority_to_dscp *pdscp, *next;
1346 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1347 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1353 set_queues(struct ofport *ofport_,
1354 const struct ofproto_port_queue *qdscp_list,
1357 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1358 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1359 struct hmap new = HMAP_INITIALIZER(&new);
1362 for (i = 0; i < n_qdscp; i++) {
1363 struct priority_to_dscp *pdscp;
1367 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1368 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1373 pdscp = get_priority(ofport, priority);
1375 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1377 pdscp = xmalloc(sizeof *pdscp);
1378 pdscp->priority = priority;
1380 ofproto->need_revalidate = true;
1383 if (pdscp->dscp != dscp) {
1385 ofproto->need_revalidate = true;
1388 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1391 if (!hmap_is_empty(&ofport->priorities)) {
1392 ofport_clear_priorities(ofport);
1393 ofproto->need_revalidate = true;
1396 hmap_swap(&new, &ofport->priorities);
1404 /* Expires all MAC learning entries associated with 'bundle' and forces its
1405 * ofproto to revalidate every flow.
1407 * Normally MAC learning entries are removed only from the ofproto associated
1408 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1409 * are removed from every ofproto. When patch ports and SLB bonds are in use
1410 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1411 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1412 * with the host from which it migrated. */
1414 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1416 struct ofproto_dpif *ofproto = bundle->ofproto;
1417 struct mac_learning *ml = ofproto->ml;
1418 struct mac_entry *mac, *next_mac;
1420 ofproto->need_revalidate = true;
1421 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1422 if (mac->port.p == bundle) {
1424 struct ofproto_dpif *o;
1426 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1428 struct mac_entry *e;
1430 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1433 tag_set_add(&o->revalidate_set, e->tag);
1434 mac_learning_expire(o->ml, e);
1440 mac_learning_expire(ml, mac);
1445 static struct ofbundle *
1446 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1448 struct ofbundle *bundle;
1450 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1451 &ofproto->bundles) {
1452 if (bundle->aux == aux) {
1459 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1460 * ones that are found to 'bundles'. */
1462 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1463 void **auxes, size_t n_auxes,
1464 struct hmapx *bundles)
1468 hmapx_init(bundles);
1469 for (i = 0; i < n_auxes; i++) {
1470 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1472 hmapx_add(bundles, bundle);
1478 bundle_update(struct ofbundle *bundle)
1480 struct ofport_dpif *port;
1482 bundle->floodable = true;
1483 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1484 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
1485 bundle->floodable = false;
1492 bundle_del_port(struct ofport_dpif *port)
1494 struct ofbundle *bundle = port->bundle;
1496 bundle->ofproto->need_revalidate = true;
1498 list_remove(&port->bundle_node);
1499 port->bundle = NULL;
1502 lacp_slave_unregister(bundle->lacp, port);
1505 bond_slave_unregister(bundle->bond, port);
1508 bundle_update(bundle);
1512 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1513 struct lacp_slave_settings *lacp,
1514 uint32_t bond_stable_id)
1516 struct ofport_dpif *port;
1518 port = get_ofp_port(bundle->ofproto, ofp_port);
1523 if (port->bundle != bundle) {
1524 bundle->ofproto->need_revalidate = true;
1526 bundle_del_port(port);
1529 port->bundle = bundle;
1530 list_push_back(&bundle->ports, &port->bundle_node);
1531 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
1532 bundle->floodable = false;
1536 port->bundle->ofproto->need_revalidate = true;
1537 lacp_slave_register(bundle->lacp, port, lacp);
1540 port->bond_stable_id = bond_stable_id;
1546 bundle_destroy(struct ofbundle *bundle)
1548 struct ofproto_dpif *ofproto;
1549 struct ofport_dpif *port, *next_port;
1556 ofproto = bundle->ofproto;
1557 for (i = 0; i < MAX_MIRRORS; i++) {
1558 struct ofmirror *m = ofproto->mirrors[i];
1560 if (m->out == bundle) {
1562 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1563 || hmapx_find_and_delete(&m->dsts, bundle)) {
1564 ofproto->need_revalidate = true;
1569 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1570 bundle_del_port(port);
1573 bundle_flush_macs(bundle, true);
1574 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1576 free(bundle->trunks);
1577 lacp_destroy(bundle->lacp);
1578 bond_destroy(bundle->bond);
1583 bundle_set(struct ofproto *ofproto_, void *aux,
1584 const struct ofproto_bundle_settings *s)
1586 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1587 bool need_flush = false;
1588 struct ofport_dpif *port;
1589 struct ofbundle *bundle;
1590 unsigned long *trunks;
1596 bundle_destroy(bundle_lookup(ofproto, aux));
1600 assert(s->n_slaves == 1 || s->bond != NULL);
1601 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1603 bundle = bundle_lookup(ofproto, aux);
1605 bundle = xmalloc(sizeof *bundle);
1607 bundle->ofproto = ofproto;
1608 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1609 hash_pointer(aux, 0));
1611 bundle->name = NULL;
1613 list_init(&bundle->ports);
1614 bundle->vlan_mode = PORT_VLAN_TRUNK;
1616 bundle->trunks = NULL;
1617 bundle->use_priority_tags = s->use_priority_tags;
1618 bundle->lacp = NULL;
1619 bundle->bond = NULL;
1621 bundle->floodable = true;
1623 bundle->src_mirrors = 0;
1624 bundle->dst_mirrors = 0;
1625 bundle->mirror_out = 0;
1628 if (!bundle->name || strcmp(s->name, bundle->name)) {
1630 bundle->name = xstrdup(s->name);
1635 if (!bundle->lacp) {
1636 ofproto->need_revalidate = true;
1637 bundle->lacp = lacp_create();
1639 lacp_configure(bundle->lacp, s->lacp);
1641 lacp_destroy(bundle->lacp);
1642 bundle->lacp = NULL;
1645 /* Update set of ports. */
1647 for (i = 0; i < s->n_slaves; i++) {
1648 if (!bundle_add_port(bundle, s->slaves[i],
1649 s->lacp ? &s->lacp_slaves[i] : NULL,
1650 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1654 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1655 struct ofport_dpif *next_port;
1657 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1658 for (i = 0; i < s->n_slaves; i++) {
1659 if (s->slaves[i] == port->up.ofp_port) {
1664 bundle_del_port(port);
1668 assert(list_size(&bundle->ports) <= s->n_slaves);
1670 if (list_is_empty(&bundle->ports)) {
1671 bundle_destroy(bundle);
1675 /* Set VLAN tagging mode */
1676 if (s->vlan_mode != bundle->vlan_mode
1677 || s->use_priority_tags != bundle->use_priority_tags) {
1678 bundle->vlan_mode = s->vlan_mode;
1679 bundle->use_priority_tags = s->use_priority_tags;
1684 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1685 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1687 if (vlan != bundle->vlan) {
1688 bundle->vlan = vlan;
1692 /* Get trunked VLANs. */
1693 switch (s->vlan_mode) {
1694 case PORT_VLAN_ACCESS:
1698 case PORT_VLAN_TRUNK:
1699 trunks = (unsigned long *) s->trunks;
1702 case PORT_VLAN_NATIVE_UNTAGGED:
1703 case PORT_VLAN_NATIVE_TAGGED:
1704 if (vlan != 0 && (!s->trunks
1705 || !bitmap_is_set(s->trunks, vlan)
1706 || bitmap_is_set(s->trunks, 0))) {
1707 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1709 trunks = bitmap_clone(s->trunks, 4096);
1711 trunks = bitmap_allocate1(4096);
1713 bitmap_set1(trunks, vlan);
1714 bitmap_set0(trunks, 0);
1716 trunks = (unsigned long *) s->trunks;
1723 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1724 free(bundle->trunks);
1725 if (trunks == s->trunks) {
1726 bundle->trunks = vlan_bitmap_clone(trunks);
1728 bundle->trunks = trunks;
1733 if (trunks != s->trunks) {
1738 if (!list_is_short(&bundle->ports)) {
1739 bundle->ofproto->has_bonded_bundles = true;
1741 if (bond_reconfigure(bundle->bond, s->bond)) {
1742 ofproto->need_revalidate = true;
1745 bundle->bond = bond_create(s->bond);
1746 ofproto->need_revalidate = true;
1749 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1750 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1754 bond_destroy(bundle->bond);
1755 bundle->bond = NULL;
1758 /* If we changed something that would affect MAC learning, un-learn
1759 * everything on this port and force flow revalidation. */
1761 bundle_flush_macs(bundle, false);
1768 bundle_remove(struct ofport *port_)
1770 struct ofport_dpif *port = ofport_dpif_cast(port_);
1771 struct ofbundle *bundle = port->bundle;
1774 bundle_del_port(port);
1775 if (list_is_empty(&bundle->ports)) {
1776 bundle_destroy(bundle);
1777 } else if (list_is_short(&bundle->ports)) {
1778 bond_destroy(bundle->bond);
1779 bundle->bond = NULL;
1785 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
1787 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1788 struct ofport_dpif *port = port_;
1789 uint8_t ea[ETH_ADDR_LEN];
1792 error = netdev_get_etheraddr(port->up.netdev, ea);
1794 struct ofpbuf packet;
1797 ofpbuf_init(&packet, 0);
1798 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1800 memcpy(packet_pdu, pdu, pdu_size);
1802 send_packet(port, &packet);
1803 ofpbuf_uninit(&packet);
1805 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1806 "%s (%s)", port->bundle->name,
1807 netdev_get_name(port->up.netdev), strerror(error));
1812 bundle_send_learning_packets(struct ofbundle *bundle)
1814 struct ofproto_dpif *ofproto = bundle->ofproto;
1815 int error, n_packets, n_errors;
1816 struct mac_entry *e;
1818 error = n_packets = n_errors = 0;
1819 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1820 if (e->port.p != bundle) {
1821 struct ofpbuf *learning_packet;
1822 struct ofport_dpif *port;
1826 /* The assignment to "port" is unnecessary but makes "grep"ing for
1827 * struct ofport_dpif more effective. */
1828 learning_packet = bond_compose_learning_packet(bundle->bond,
1832 ret = send_packet(port, learning_packet);
1833 ofpbuf_delete(learning_packet);
1843 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1844 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1845 "packets, last error was: %s",
1846 bundle->name, n_errors, n_packets, strerror(error));
1848 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1849 bundle->name, n_packets);
1854 bundle_run(struct ofbundle *bundle)
1857 lacp_run(bundle->lacp, send_pdu_cb);
1860 struct ofport_dpif *port;
1862 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1863 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1866 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1867 lacp_negotiated(bundle->lacp));
1868 if (bond_should_send_learning_packets(bundle->bond)) {
1869 bundle_send_learning_packets(bundle);
1875 bundle_wait(struct ofbundle *bundle)
1878 lacp_wait(bundle->lacp);
1881 bond_wait(bundle->bond);
1888 mirror_scan(struct ofproto_dpif *ofproto)
1892 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1893 if (!ofproto->mirrors[idx]) {
1900 static struct ofmirror *
1901 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1905 for (i = 0; i < MAX_MIRRORS; i++) {
1906 struct ofmirror *mirror = ofproto->mirrors[i];
1907 if (mirror && mirror->aux == aux) {
1915 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
1917 mirror_update_dups(struct ofproto_dpif *ofproto)
1921 for (i = 0; i < MAX_MIRRORS; i++) {
1922 struct ofmirror *m = ofproto->mirrors[i];
1925 m->dup_mirrors = MIRROR_MASK_C(1) << i;
1929 for (i = 0; i < MAX_MIRRORS; i++) {
1930 struct ofmirror *m1 = ofproto->mirrors[i];
1937 for (j = i + 1; j < MAX_MIRRORS; j++) {
1938 struct ofmirror *m2 = ofproto->mirrors[j];
1940 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
1941 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
1942 m2->dup_mirrors |= m1->dup_mirrors;
1949 mirror_set(struct ofproto *ofproto_, void *aux,
1950 const struct ofproto_mirror_settings *s)
1952 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1953 mirror_mask_t mirror_bit;
1954 struct ofbundle *bundle;
1955 struct ofmirror *mirror;
1956 struct ofbundle *out;
1957 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
1958 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
1961 mirror = mirror_lookup(ofproto, aux);
1963 mirror_destroy(mirror);
1969 idx = mirror_scan(ofproto);
1971 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
1973 ofproto->up.name, MAX_MIRRORS, s->name);
1977 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
1978 mirror->ofproto = ofproto;
1981 mirror->out_vlan = -1;
1982 mirror->name = NULL;
1985 if (!mirror->name || strcmp(s->name, mirror->name)) {
1987 mirror->name = xstrdup(s->name);
1990 /* Get the new configuration. */
1991 if (s->out_bundle) {
1992 out = bundle_lookup(ofproto, s->out_bundle);
1994 mirror_destroy(mirror);
2000 out_vlan = s->out_vlan;
2002 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2003 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2005 /* If the configuration has not changed, do nothing. */
2006 if (hmapx_equals(&srcs, &mirror->srcs)
2007 && hmapx_equals(&dsts, &mirror->dsts)
2008 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2009 && mirror->out == out
2010 && mirror->out_vlan == out_vlan)
2012 hmapx_destroy(&srcs);
2013 hmapx_destroy(&dsts);
2017 hmapx_swap(&srcs, &mirror->srcs);
2018 hmapx_destroy(&srcs);
2020 hmapx_swap(&dsts, &mirror->dsts);
2021 hmapx_destroy(&dsts);
2023 free(mirror->vlans);
2024 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2027 mirror->out_vlan = out_vlan;
2029 /* Update bundles. */
2030 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2031 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2032 if (hmapx_contains(&mirror->srcs, bundle)) {
2033 bundle->src_mirrors |= mirror_bit;
2035 bundle->src_mirrors &= ~mirror_bit;
2038 if (hmapx_contains(&mirror->dsts, bundle)) {
2039 bundle->dst_mirrors |= mirror_bit;
2041 bundle->dst_mirrors &= ~mirror_bit;
2044 if (mirror->out == bundle) {
2045 bundle->mirror_out |= mirror_bit;
2047 bundle->mirror_out &= ~mirror_bit;
2051 ofproto->need_revalidate = true;
2052 mac_learning_flush(ofproto->ml);
2053 mirror_update_dups(ofproto);
2059 mirror_destroy(struct ofmirror *mirror)
2061 struct ofproto_dpif *ofproto;
2062 mirror_mask_t mirror_bit;
2063 struct ofbundle *bundle;
2069 ofproto = mirror->ofproto;
2070 ofproto->need_revalidate = true;
2071 mac_learning_flush(ofproto->ml);
2073 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2074 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2075 bundle->src_mirrors &= ~mirror_bit;
2076 bundle->dst_mirrors &= ~mirror_bit;
2077 bundle->mirror_out &= ~mirror_bit;
2080 hmapx_destroy(&mirror->srcs);
2081 hmapx_destroy(&mirror->dsts);
2082 free(mirror->vlans);
2084 ofproto->mirrors[mirror->idx] = NULL;
2088 mirror_update_dups(ofproto);
2092 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2093 uint64_t *packets, uint64_t *bytes)
2095 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2096 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2099 *packets = *bytes = UINT64_MAX;
2103 *packets = mirror->packet_count;
2104 *bytes = mirror->byte_count;
2110 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2112 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2113 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2114 ofproto->need_revalidate = true;
2115 mac_learning_flush(ofproto->ml);
2121 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2123 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2124 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2125 return bundle && bundle->mirror_out != 0;
2129 forward_bpdu_changed(struct ofproto *ofproto_)
2131 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2132 /* Revalidate cached flows whenever forward_bpdu option changes. */
2133 ofproto->need_revalidate = true;
2138 static struct ofport_dpif *
2139 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
2141 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2142 return ofport ? ofport_dpif_cast(ofport) : NULL;
2145 static struct ofport_dpif *
2146 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
2148 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2152 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2153 struct dpif_port *dpif_port)
2155 ofproto_port->name = dpif_port->name;
2156 ofproto_port->type = dpif_port->type;
2157 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2161 port_run(struct ofport_dpif *ofport)
2163 bool enable = netdev_get_carrier(ofport->up.netdev);
2166 cfm_run(ofport->cfm);
2168 if (cfm_should_send_ccm(ofport->cfm)) {
2169 struct ofpbuf packet;
2171 ofpbuf_init(&packet, 0);
2172 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr);
2173 send_packet(ofport, &packet);
2174 ofpbuf_uninit(&packet);
2177 enable = enable && !cfm_get_fault(ofport->cfm)
2178 && cfm_get_opup(ofport->cfm);
2181 if (ofport->bundle) {
2182 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2185 if (ofport->may_enable != enable) {
2186 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2188 if (ofproto->has_bundle_action) {
2189 ofproto->need_revalidate = true;
2193 ofport->may_enable = enable;
2197 port_wait(struct ofport_dpif *ofport)
2200 cfm_wait(ofport->cfm);
2205 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2206 struct ofproto_port *ofproto_port)
2208 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2209 struct dpif_port dpif_port;
2212 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2214 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2220 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2222 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2226 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2228 *ofp_portp = odp_port_to_ofp_port(odp_port);
2234 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2236 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2239 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2241 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2243 /* The caller is going to close ofport->up.netdev. If this is a
2244 * bonded port, then the bond is using that netdev, so remove it
2245 * from the bond. The client will need to reconfigure everything
2246 * after deleting ports, so then the slave will get re-added. */
2247 bundle_remove(&ofport->up);
2254 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2256 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2259 error = netdev_get_stats(ofport->up.netdev, stats);
2261 if (!error && ofport->odp_port == OVSP_LOCAL) {
2262 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2264 /* ofproto->stats.tx_packets represents packets that we created
2265 * internally and sent to some port (e.g. packets sent with
2266 * send_packet()). Account for them as if they had come from
2267 * OFPP_LOCAL and got forwarded. */
2269 if (stats->rx_packets != UINT64_MAX) {
2270 stats->rx_packets += ofproto->stats.tx_packets;
2273 if (stats->rx_bytes != UINT64_MAX) {
2274 stats->rx_bytes += ofproto->stats.tx_bytes;
2277 /* ofproto->stats.rx_packets represents packets that were received on
2278 * some port and we processed internally and dropped (e.g. STP).
2279 * Account fro them as if they had been forwarded to OFPP_LOCAL. */
2281 if (stats->tx_packets != UINT64_MAX) {
2282 stats->tx_packets += ofproto->stats.rx_packets;
2285 if (stats->tx_bytes != UINT64_MAX) {
2286 stats->tx_bytes += ofproto->stats.rx_bytes;
2293 /* Account packets for LOCAL port. */
2295 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2296 size_t tx_size, size_t rx_size)
2298 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2301 ofproto->stats.rx_packets++;
2302 ofproto->stats.rx_bytes += rx_size;
2305 ofproto->stats.tx_packets++;
2306 ofproto->stats.tx_bytes += tx_size;
2310 struct port_dump_state {
2311 struct dpif_port_dump dump;
2316 port_dump_start(const struct ofproto *ofproto_, void **statep)
2318 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2319 struct port_dump_state *state;
2321 *statep = state = xmalloc(sizeof *state);
2322 dpif_port_dump_start(&state->dump, ofproto->dpif);
2323 state->done = false;
2328 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2329 struct ofproto_port *port)
2331 struct port_dump_state *state = state_;
2332 struct dpif_port dpif_port;
2334 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2335 ofproto_port_from_dpif_port(port, &dpif_port);
2338 int error = dpif_port_dump_done(&state->dump);
2340 return error ? error : EOF;
2345 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2347 struct port_dump_state *state = state_;
2350 dpif_port_dump_done(&state->dump);
2357 port_poll(const struct ofproto *ofproto_, char **devnamep)
2359 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2360 return dpif_port_poll(ofproto->dpif, devnamep);
2364 port_poll_wait(const struct ofproto *ofproto_)
2366 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2367 dpif_port_poll_wait(ofproto->dpif);
2371 port_is_lacp_current(const struct ofport *ofport_)
2373 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2374 return (ofport->bundle && ofport->bundle->lacp
2375 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2379 /* Upcall handling. */
2381 /* Flow miss batching.
2383 * Some dpifs implement operations faster when you hand them off in a batch.
2384 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2385 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2386 * more packets, plus possibly installing the flow in the dpif.
2388 * So far we only batch the operations that affect flow setup time the most.
2389 * It's possible to batch more than that, but the benefit might be minimal. */
2391 struct hmap_node hmap_node;
2393 enum odp_key_fitness key_fitness;
2394 const struct nlattr *key;
2396 ovs_be16 initial_tci;
2397 struct list packets;
2400 struct flow_miss_op {
2401 union dpif_op dpif_op;
2402 struct subfacet *subfacet;
2405 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2406 * OpenFlow controller as necessary according to their individual
2407 * configurations. */
2409 send_packet_in_miss(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
2410 const struct flow *flow)
2412 struct ofputil_packet_in pin;
2414 pin.packet = packet->data;
2415 pin.packet_len = packet->size;
2416 pin.total_len = packet->size;
2417 pin.reason = OFPR_NO_MATCH;
2418 pin.buffer_id = 0; /* not yet known */
2419 pin.send_len = 0; /* not used for flow table misses */
2421 flow_get_metadata(flow, &pin.fmd);
2423 /* Registers aren't meaningful on a miss. */
2424 memset(pin.fmd.reg_masks, 0, sizeof pin.fmd.reg_masks);
2426 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow);
2430 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2431 const struct ofpbuf *packet)
2433 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2439 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2441 cfm_process_heartbeat(ofport->cfm, packet);
2444 } else if (ofport->bundle && ofport->bundle->lacp
2445 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2447 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2450 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2452 stp_process_packet(ofport, packet);
2459 static struct flow_miss *
2460 flow_miss_create(struct hmap *todo, const struct flow *flow,
2461 enum odp_key_fitness key_fitness,
2462 const struct nlattr *key, size_t key_len,
2463 ovs_be16 initial_tci)
2465 uint32_t hash = flow_hash(flow, 0);
2466 struct flow_miss *miss;
2468 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2469 if (flow_equal(&miss->flow, flow)) {
2474 miss = xmalloc(sizeof *miss);
2475 hmap_insert(todo, &miss->hmap_node, hash);
2477 miss->key_fitness = key_fitness;
2479 miss->key_len = key_len;
2480 miss->initial_tci = initial_tci;
2481 list_init(&miss->packets);
2486 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2487 struct flow_miss_op *ops, size_t *n_ops)
2489 const struct flow *flow = &miss->flow;
2490 struct ofpbuf *packet, *next_packet;
2491 struct subfacet *subfacet;
2492 struct facet *facet;
2494 facet = facet_lookup_valid(ofproto, flow);
2496 struct rule_dpif *rule;
2498 rule = rule_dpif_lookup(ofproto, flow, 0);
2500 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
2501 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
2503 if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) {
2504 COVERAGE_INC(ofproto_dpif_no_packet_in);
2505 /* XXX install 'drop' flow entry */
2509 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
2513 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2514 send_packet_in_miss(ofproto, packet, flow);
2520 facet = facet_create(rule, flow);
2523 subfacet = subfacet_create(ofproto, facet,
2524 miss->key_fitness, miss->key, miss->key_len,
2527 LIST_FOR_EACH_SAFE (packet, next_packet, list_node, &miss->packets) {
2528 struct dpif_flow_stats stats;
2529 struct flow_miss_op *op;
2530 struct dpif_execute *execute;
2532 list_remove(&packet->list_node);
2533 ofproto->n_matches++;
2535 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2537 * Extra-special case for fail-open mode.
2539 * We are in fail-open mode and the packet matched the fail-open
2540 * rule, but we are connected to a controller too. We should send
2541 * the packet up to the controller in the hope that it will try to
2542 * set up a flow and thereby allow us to exit fail-open.
2544 * See the top-level comment in fail-open.c for more information.
2546 send_packet_in_miss(ofproto, packet, flow);
2549 if (!facet->may_install || !subfacet->actions) {
2550 subfacet_make_actions(ofproto, subfacet, packet);
2553 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2554 subfacet_update_stats(ofproto, subfacet, &stats);
2556 if (flow->vlan_tci != subfacet->initial_tci) {
2557 /* This packet was received on a VLAN splinter port. We added
2558 * a VLAN to the packet to make the packet resemble the flow,
2559 * but the actions were composed assuming that the packet
2560 * contained no VLAN. So, we must remove the VLAN header from
2561 * the packet before trying to execute the actions. */
2562 eth_pop_vlan(packet);
2565 op = &ops[(*n_ops)++];
2566 execute = &op->dpif_op.execute;
2567 op->subfacet = subfacet;
2568 execute->type = DPIF_OP_EXECUTE;
2569 execute->key = miss->key;
2570 execute->key_len = miss->key_len;
2571 execute->actions = (facet->may_install
2573 : xmemdup(subfacet->actions,
2574 subfacet->actions_len));
2575 execute->actions_len = subfacet->actions_len;
2576 execute->packet = packet;
2579 if (facet->may_install && subfacet->key_fitness != ODP_FIT_TOO_LITTLE) {
2580 struct flow_miss_op *op = &ops[(*n_ops)++];
2581 struct dpif_flow_put *put = &op->dpif_op.flow_put;
2583 op->subfacet = subfacet;
2584 put->type = DPIF_OP_FLOW_PUT;
2585 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2586 put->key = miss->key;
2587 put->key_len = miss->key_len;
2588 put->actions = subfacet->actions;
2589 put->actions_len = subfacet->actions_len;
2594 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
2595 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
2596 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
2597 * what a flow key should contain.
2599 * This function also includes some logic to help make VLAN splinters
2600 * transparent to the rest of the upcall processing logic. In particular, if
2601 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
2602 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
2603 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
2605 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
2606 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
2607 * (This differs from the value returned in flow->vlan_tci only for packets
2608 * received on VLAN splinters.)
2610 static enum odp_key_fitness
2611 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
2612 const struct nlattr *key, size_t key_len,
2613 struct flow *flow, ovs_be16 *initial_tci,
2614 struct ofpbuf *packet)
2616 enum odp_key_fitness fitness;
2620 fitness = odp_flow_key_to_flow(key, key_len, flow);
2621 if (fitness == ODP_FIT_ERROR) {
2624 *initial_tci = flow->vlan_tci;
2626 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
2628 /* Cause the flow to be processed as if it came in on the real device
2629 * with the VLAN device's VLAN ID. */
2630 flow->in_port = realdev;
2631 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
2633 /* Make the packet resemble the flow, so that it gets sent to an
2634 * OpenFlow controller properly, so that it looks correct for
2635 * sFlow, and so that flow_extract() will get the correct vlan_tci
2636 * if it is called on 'packet'.
2638 * The allocated space inside 'packet' probably also contains
2639 * 'key', that is, both 'packet' and 'key' are probably part of a
2640 * struct dpif_upcall (see the large comment on that structure
2641 * definition), so pushing data on 'packet' is in general not a
2642 * good idea since it could overwrite 'key' or free it as a side
2643 * effect. However, it's OK in this special case because we know
2644 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
2645 * will just overwrite the 4-byte "struct nlattr", which is fine
2646 * since we don't need that header anymore. */
2647 eth_push_vlan(packet, flow->vlan_tci);
2650 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
2651 if (fitness == ODP_FIT_PERFECT) {
2652 fitness = ODP_FIT_TOO_MUCH;
2660 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
2663 struct dpif_upcall *upcall;
2664 struct flow_miss *miss, *next_miss;
2665 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
2666 union dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
2675 /* Construct the to-do list.
2677 * This just amounts to extracting the flow from each packet and sticking
2678 * the packets that have the same flow in the same "flow_miss" structure so
2679 * that we can process them together. */
2681 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
2682 enum odp_key_fitness fitness;
2683 struct flow_miss *miss;
2684 ovs_be16 initial_tci;
2687 /* Obtain metadata and check userspace/kernel agreement on flow match,
2688 * then set 'flow''s header pointers. */
2689 fitness = ofproto_dpif_extract_flow_key(ofproto,
2690 upcall->key, upcall->key_len,
2691 &flow, &initial_tci,
2693 if (fitness == ODP_FIT_ERROR) {
2694 ofpbuf_delete(upcall->packet);
2697 flow_extract(upcall->packet, flow.skb_priority, flow.tun_id,
2698 flow.in_port, &flow);
2700 /* Handle 802.1ag, LACP, and STP specially. */
2701 if (process_special(ofproto, &flow, upcall->packet)) {
2702 ofproto_update_local_port_stats(&ofproto->up,
2703 0, upcall->packet->size);
2704 ofpbuf_delete(upcall->packet);
2705 ofproto->n_matches++;
2709 /* Add other packets to a to-do list. */
2710 miss = flow_miss_create(&todo, &flow, fitness,
2711 upcall->key, upcall->key_len, initial_tci);
2712 list_push_back(&miss->packets, &upcall->packet->list_node);
2715 /* Process each element in the to-do list, constructing the set of
2716 * operations to batch. */
2718 HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) {
2719 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
2720 ofpbuf_list_delete(&miss->packets);
2721 hmap_remove(&todo, &miss->hmap_node);
2724 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
2725 hmap_destroy(&todo);
2727 /* Execute batch. */
2728 for (i = 0; i < n_ops; i++) {
2729 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
2731 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
2733 /* Free memory and update facets. */
2734 for (i = 0; i < n_ops; i++) {
2735 struct flow_miss_op *op = &flow_miss_ops[i];
2736 struct dpif_execute *execute;
2737 struct dpif_flow_put *put;
2739 switch (op->dpif_op.type) {
2740 case DPIF_OP_EXECUTE:
2741 execute = &op->dpif_op.execute;
2742 if (op->subfacet->actions != execute->actions) {
2743 free((struct nlattr *) execute->actions);
2745 ofpbuf_delete((struct ofpbuf *) execute->packet);
2748 case DPIF_OP_FLOW_PUT:
2749 put = &op->dpif_op.flow_put;
2751 op->subfacet->installed = true;
2759 handle_userspace_upcall(struct ofproto_dpif *ofproto,
2760 struct dpif_upcall *upcall)
2762 struct user_action_cookie cookie;
2763 enum odp_key_fitness fitness;
2764 ovs_be16 initial_tci;
2767 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
2769 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
2770 upcall->key_len, &flow,
2771 &initial_tci, upcall->packet);
2772 if (fitness == ODP_FIT_ERROR) {
2773 ofpbuf_delete(upcall->packet);
2777 if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
2778 if (ofproto->sflow) {
2779 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
2783 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
2785 ofpbuf_delete(upcall->packet);
2789 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
2791 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
2795 assert (max_batch <= FLOW_MISS_MAX_BATCH);
2798 for (i = 0; i < max_batch; i++) {
2799 struct dpif_upcall *upcall = &misses[n_misses];
2802 error = dpif_recv(ofproto->dpif, upcall);
2807 switch (upcall->type) {
2808 case DPIF_UC_ACTION:
2809 handle_userspace_upcall(ofproto, upcall);
2813 /* Handle it later. */
2817 case DPIF_N_UC_TYPES:
2819 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32,
2825 handle_miss_upcalls(ofproto, misses, n_misses);
2830 /* Flow expiration. */
2832 static int subfacet_max_idle(const struct ofproto_dpif *);
2833 static void update_stats(struct ofproto_dpif *);
2834 static void rule_expire(struct rule_dpif *);
2835 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
2837 /* This function is called periodically by run(). Its job is to collect
2838 * updates for the flows that have been installed into the datapath, most
2839 * importantly when they last were used, and then use that information to
2840 * expire flows that have not been used recently.
2842 * Returns the number of milliseconds after which it should be called again. */
2844 expire(struct ofproto_dpif *ofproto)
2846 struct rule_dpif *rule, *next_rule;
2847 struct classifier *table;
2850 /* Update stats for each flow in the datapath. */
2851 update_stats(ofproto);
2853 /* Expire subfacets that have been idle too long. */
2854 dp_max_idle = subfacet_max_idle(ofproto);
2855 expire_subfacets(ofproto, dp_max_idle);
2857 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
2858 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
2859 struct cls_cursor cursor;
2861 cls_cursor_init(&cursor, table, NULL);
2862 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
2867 /* All outstanding data in existing flows has been accounted, so it's a
2868 * good time to do bond rebalancing. */
2869 if (ofproto->has_bonded_bundles) {
2870 struct ofbundle *bundle;
2872 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2874 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
2879 return MIN(dp_max_idle, 1000);
2882 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
2884 * This function also pushes statistics updates to rules which each facet
2885 * resubmits into. Generally these statistics will be accurate. However, if a
2886 * facet changes the rule it resubmits into at some time in between
2887 * update_stats() runs, it is possible that statistics accrued to the
2888 * old rule will be incorrectly attributed to the new rule. This could be
2889 * avoided by calling update_stats() whenever rules are created or
2890 * deleted. However, the performance impact of making so many calls to the
2891 * datapath do not justify the benefit of having perfectly accurate statistics.
2894 update_stats(struct ofproto_dpif *p)
2896 const struct dpif_flow_stats *stats;
2897 struct dpif_flow_dump dump;
2898 const struct nlattr *key;
2901 dpif_flow_dump_start(&dump, p->dpif);
2902 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
2903 struct subfacet *subfacet;
2905 subfacet = subfacet_find(p, key, key_len);
2906 if (subfacet && subfacet->installed) {
2907 struct facet *facet = subfacet->facet;
2909 if (stats->n_packets >= subfacet->dp_packet_count) {
2910 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
2911 facet->packet_count += extra;
2913 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
2916 if (stats->n_bytes >= subfacet->dp_byte_count) {
2917 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
2919 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
2922 subfacet->dp_packet_count = stats->n_packets;
2923 subfacet->dp_byte_count = stats->n_bytes;
2925 subfacet_update_time(p, subfacet, stats->used);
2926 facet_account(p, facet);
2927 facet_push_stats(facet);
2929 if (!VLOG_DROP_WARN(&rl)) {
2933 odp_flow_key_format(key, key_len, &s);
2934 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
2938 COVERAGE_INC(facet_unexpected);
2939 /* There's a flow in the datapath that we know nothing about, or a
2940 * flow that shouldn't be installed but was anyway. Delete it. */
2941 dpif_flow_del(p->dpif, key, key_len, NULL);
2944 dpif_flow_dump_done(&dump);
2947 /* Calculates and returns the number of milliseconds of idle time after which
2948 * subfacets should expire from the datapath. When a subfacet expires, we fold
2949 * its statistics into its facet, and when a facet's last subfacet expires, we
2950 * fold its statistic into its rule. */
2952 subfacet_max_idle(const struct ofproto_dpif *ofproto)
2955 * Idle time histogram.
2957 * Most of the time a switch has a relatively small number of subfacets.
2958 * When this is the case we might as well keep statistics for all of them
2959 * in userspace and to cache them in the kernel datapath for performance as
2962 * As the number of subfacets increases, the memory required to maintain
2963 * statistics about them in userspace and in the kernel becomes
2964 * significant. However, with a large number of subfacets it is likely
2965 * that only a few of them are "heavy hitters" that consume a large amount
2966 * of bandwidth. At this point, only heavy hitters are worth caching in
2967 * the kernel and maintaining in userspaces; other subfacets we can
2970 * The technique used to compute the idle time is to build a histogram with
2971 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
2972 * that is installed in the kernel gets dropped in the appropriate bucket.
2973 * After the histogram has been built, we compute the cutoff so that only
2974 * the most-recently-used 1% of subfacets (but at least
2975 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
2976 * the most-recently-used bucket of subfacets is kept, so actually an
2977 * arbitrary number of subfacets can be kept in any given expiration run
2978 * (though the next run will delete most of those unless they receive
2981 * This requires a second pass through the subfacets, in addition to the
2982 * pass made by update_stats(), because the former function never looks at
2983 * uninstallable subfacets.
2985 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
2986 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
2987 int buckets[N_BUCKETS] = { 0 };
2988 int total, subtotal, bucket;
2989 struct subfacet *subfacet;
2993 total = hmap_count(&ofproto->subfacets);
2994 if (total <= ofproto->up.flow_eviction_threshold) {
2995 return N_BUCKETS * BUCKET_WIDTH;
2998 /* Build histogram. */
3000 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3001 long long int idle = now - subfacet->used;
3002 int bucket = (idle <= 0 ? 0
3003 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3004 : (unsigned int) idle / BUCKET_WIDTH);
3008 /* Find the first bucket whose flows should be expired. */
3009 subtotal = bucket = 0;
3011 subtotal += buckets[bucket++];
3012 } while (bucket < N_BUCKETS &&
3013 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3015 if (VLOG_IS_DBG_ENABLED()) {
3019 ds_put_cstr(&s, "keep");
3020 for (i = 0; i < N_BUCKETS; i++) {
3022 ds_put_cstr(&s, ", drop");
3025 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3028 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3032 return bucket * BUCKET_WIDTH;
3036 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3038 long long int cutoff = time_msec() - dp_max_idle;
3039 struct subfacet *subfacet, *next_subfacet;
3041 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3042 &ofproto->subfacets) {
3043 if (subfacet->used < cutoff) {
3044 subfacet_destroy(ofproto, subfacet);
3049 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3050 * then delete it entirely. */
3052 rule_expire(struct rule_dpif *rule)
3054 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3055 struct facet *facet, *next_facet;
3059 /* Has 'rule' expired? */
3061 if (rule->up.hard_timeout
3062 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3063 reason = OFPRR_HARD_TIMEOUT;
3064 } else if (rule->up.idle_timeout && list_is_empty(&rule->facets)
3065 && now > rule->used + rule->up.idle_timeout * 1000) {
3066 reason = OFPRR_IDLE_TIMEOUT;
3071 COVERAGE_INC(ofproto_dpif_expired);
3073 /* Update stats. (This is a no-op if the rule expired due to an idle
3074 * timeout, because that only happens when the rule has no facets left.) */
3075 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3076 facet_remove(ofproto, facet);
3079 /* Get rid of the rule. */
3080 ofproto_rule_expire(&rule->up, reason);
3085 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3087 * The caller must already have determined that no facet with an identical
3088 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3089 * the ofproto's classifier table.
3091 * The facet will initially have no subfacets. The caller should create (at
3092 * least) one subfacet with subfacet_create(). */
3093 static struct facet *
3094 facet_create(struct rule_dpif *rule, const struct flow *flow)
3096 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3097 struct facet *facet;
3099 facet = xzalloc(sizeof *facet);
3100 facet->used = time_msec();
3101 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
3102 list_push_back(&rule->facets, &facet->list_node);
3104 facet->flow = *flow;
3105 list_init(&facet->subfacets);
3106 netflow_flow_init(&facet->nf_flow);
3107 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3113 facet_free(struct facet *facet)
3118 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3119 * 'packet', which arrived on 'in_port'.
3121 * Takes ownership of 'packet'. */
3123 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3124 const struct nlattr *odp_actions, size_t actions_len,
3125 struct ofpbuf *packet)
3127 struct odputil_keybuf keybuf;
3131 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3132 odp_flow_key_from_flow(&key, flow);
3134 error = dpif_execute(ofproto->dpif, key.data, key.size,
3135 odp_actions, actions_len, packet);
3137 ofpbuf_delete(packet);
3141 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3143 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3144 * rule's statistics, via subfacet_uninstall().
3146 * - Removes 'facet' from its rule and from ofproto->facets.
3149 facet_remove(struct ofproto_dpif *ofproto, struct facet *facet)
3151 struct subfacet *subfacet, *next_subfacet;
3153 assert(!list_is_empty(&facet->subfacets));
3155 /* First uninstall all of the subfacets to get final statistics. */
3156 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3157 subfacet_uninstall(ofproto, subfacet);
3160 /* Flush the final stats to the rule.
3162 * This might require us to have at least one subfacet around so that we
3163 * can use its actions for accounting in facet_account(), which is why we
3164 * have uninstalled but not yet destroyed the subfacets. */
3165 facet_flush_stats(ofproto, facet);
3167 /* Now we're really all done so destroy everything. */
3168 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3169 &facet->subfacets) {
3170 subfacet_destroy__(ofproto, subfacet);
3172 hmap_remove(&ofproto->facets, &facet->hmap_node);
3173 list_remove(&facet->list_node);
3178 facet_account(struct ofproto_dpif *ofproto, struct facet *facet)
3181 struct subfacet *subfacet;
3182 const struct nlattr *a;
3186 if (facet->byte_count <= facet->accounted_bytes) {
3189 n_bytes = facet->byte_count - facet->accounted_bytes;
3190 facet->accounted_bytes = facet->byte_count;
3192 /* Feed information from the active flows back into the learning table to
3193 * ensure that table is always in sync with what is actually flowing
3194 * through the datapath. */
3195 if (facet->has_learn || facet->has_normal) {
3196 struct action_xlate_ctx ctx;
3198 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3199 facet->flow.vlan_tci, NULL);
3200 ctx.may_learn = true;
3201 ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions,
3202 facet->rule->up.n_actions));
3205 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3209 /* This loop feeds byte counters to bond_account() for rebalancing to use
3210 * as a basis. We also need to track the actual VLAN on which the packet
3211 * is going to be sent to ensure that it matches the one passed to
3212 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3215 * We use the actions from an arbitrary subfacet because they should all
3216 * be equally valid for our purpose. */
3217 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3218 struct subfacet, list_node);
3219 vlan_tci = facet->flow.vlan_tci;
3220 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3221 subfacet->actions, subfacet->actions_len) {
3222 const struct ovs_action_push_vlan *vlan;
3223 struct ofport_dpif *port;
3225 switch (nl_attr_type(a)) {
3226 case OVS_ACTION_ATTR_OUTPUT:
3227 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3228 if (port && port->bundle && port->bundle->bond) {
3229 bond_account(port->bundle->bond, &facet->flow,
3230 vlan_tci_to_vid(vlan_tci), n_bytes);
3234 case OVS_ACTION_ATTR_POP_VLAN:
3235 vlan_tci = htons(0);
3238 case OVS_ACTION_ATTR_PUSH_VLAN:
3239 vlan = nl_attr_get(a);
3240 vlan_tci = vlan->vlan_tci;
3246 /* Returns true if the only action for 'facet' is to send to the controller.
3247 * (We don't report NetFlow expiration messages for such facets because they
3248 * are just part of the control logic for the network, not real traffic). */
3250 facet_is_controller_flow(struct facet *facet)
3253 && facet->rule->up.n_actions == 1
3254 && action_outputs_to_port(&facet->rule->up.actions[0],
3255 htons(OFPP_CONTROLLER)));
3258 /* Folds all of 'facet''s statistics into its rule. Also updates the
3259 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3260 * 'facet''s statistics in the datapath should have been zeroed and folded into
3261 * its packet and byte counts before this function is called. */
3263 facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet)
3265 struct subfacet *subfacet;
3267 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3268 assert(!subfacet->dp_byte_count);
3269 assert(!subfacet->dp_packet_count);
3272 facet_push_stats(facet);
3273 facet_account(ofproto, facet);
3275 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3276 struct ofexpired expired;
3277 expired.flow = facet->flow;
3278 expired.packet_count = facet->packet_count;
3279 expired.byte_count = facet->byte_count;
3280 expired.used = facet->used;
3281 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3284 facet->rule->packet_count += facet->packet_count;
3285 facet->rule->byte_count += facet->byte_count;
3287 /* Reset counters to prevent double counting if 'facet' ever gets
3289 facet_reset_counters(facet);
3291 netflow_flow_clear(&facet->nf_flow);
3294 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3295 * Returns it if found, otherwise a null pointer.
3297 * The returned facet might need revalidation; use facet_lookup_valid()
3298 * instead if that is important. */
3299 static struct facet *
3300 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
3302 struct facet *facet;
3304 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
3306 if (flow_equal(flow, &facet->flow)) {
3314 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3315 * Returns it if found, otherwise a null pointer.
3317 * The returned facet is guaranteed to be valid. */
3318 static struct facet *
3319 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
3321 struct facet *facet = facet_find(ofproto, flow);
3323 /* The facet we found might not be valid, since we could be in need of
3324 * revalidation. If it is not valid, don't return it. */
3326 && (ofproto->need_revalidate
3327 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))
3328 && !facet_revalidate(ofproto, facet)) {
3329 COVERAGE_INC(facet_invalidated);
3336 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
3338 * - If the rule found is different from 'facet''s current rule, moves
3339 * 'facet' to the new rule and recompiles its actions.
3341 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3342 * where it is and recompiles its actions anyway.
3344 * - If there is none, destroys 'facet'.
3346 * Returns true if 'facet' still exists, false if it has been destroyed. */
3348 facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet)
3351 struct nlattr *odp_actions;
3354 struct actions *new_actions;
3356 struct action_xlate_ctx ctx;
3357 struct rule_dpif *new_rule;
3358 struct subfacet *subfacet;
3359 bool actions_changed;
3362 COVERAGE_INC(facet_revalidate);
3364 /* Determine the new rule. */
3365 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3367 /* No new rule, so delete the facet. */
3368 facet_remove(ofproto, facet);
3372 /* Calculate new datapath actions.
3374 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3375 * emit a NetFlow expiration and, if so, we need to have the old state
3376 * around to properly compose it. */
3378 /* If the datapath actions changed or the installability changed,
3379 * then we need to talk to the datapath. */
3382 memset(&ctx, 0, sizeof ctx);
3383 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3384 struct ofpbuf *odp_actions;
3385 bool should_install;
3387 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3388 subfacet->initial_tci, NULL);
3389 odp_actions = xlate_actions(&ctx, new_rule->up.actions,
3390 new_rule->up.n_actions);
3391 actions_changed = (subfacet->actions_len != odp_actions->size
3392 || memcmp(subfacet->actions, odp_actions->data,
3393 subfacet->actions_len));
3395 should_install = (ctx.may_set_up_flow
3396 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3397 if (actions_changed || should_install != subfacet->installed) {
3398 if (should_install) {
3399 struct dpif_flow_stats stats;
3401 subfacet_install(ofproto, subfacet,
3402 odp_actions->data, odp_actions->size, &stats);
3403 subfacet_update_stats(ofproto, subfacet, &stats);
3405 subfacet_uninstall(ofproto, subfacet);
3409 new_actions = xcalloc(list_size(&facet->subfacets),
3410 sizeof *new_actions);
3412 new_actions[i].odp_actions = xmemdup(odp_actions->data,
3414 new_actions[i].actions_len = odp_actions->size;
3417 ofpbuf_delete(odp_actions);
3421 facet_flush_stats(ofproto, facet);
3424 /* Update 'facet' now that we've taken care of all the old state. */
3425 facet->tags = ctx.tags;
3426 facet->nf_flow.output_iface = ctx.nf_output_iface;
3427 facet->may_install = ctx.may_set_up_flow;
3428 facet->has_learn = ctx.has_learn;
3429 facet->has_normal = ctx.has_normal;
3430 facet->mirrors = ctx.mirrors;
3433 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3434 if (new_actions[i].odp_actions) {
3435 free(subfacet->actions);
3436 subfacet->actions = new_actions[i].odp_actions;
3437 subfacet->actions_len = new_actions[i].actions_len;
3443 if (facet->rule != new_rule) {
3444 COVERAGE_INC(facet_changed_rule);
3445 list_remove(&facet->list_node);
3446 list_push_back(&new_rule->facets, &facet->list_node);
3447 facet->rule = new_rule;
3448 facet->used = new_rule->up.created;
3449 facet->prev_used = facet->used;
3455 /* Updates 'facet''s used time. Caller is responsible for calling
3456 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3458 facet_update_time(struct ofproto_dpif *ofproto, struct facet *facet,
3461 if (used > facet->used) {
3463 if (used > facet->rule->used) {
3464 facet->rule->used = used;
3466 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3471 facet_reset_counters(struct facet *facet)
3473 facet->packet_count = 0;
3474 facet->byte_count = 0;
3475 facet->prev_packet_count = 0;
3476 facet->prev_byte_count = 0;
3477 facet->accounted_bytes = 0;
3481 facet_push_stats(struct facet *facet)
3483 uint64_t new_packets, new_bytes;
3485 assert(facet->packet_count >= facet->prev_packet_count);
3486 assert(facet->byte_count >= facet->prev_byte_count);
3487 assert(facet->used >= facet->prev_used);
3489 new_packets = facet->packet_count - facet->prev_packet_count;
3490 new_bytes = facet->byte_count - facet->prev_byte_count;
3492 if (new_packets || new_bytes || facet->used > facet->prev_used) {
3493 facet->prev_packet_count = facet->packet_count;
3494 facet->prev_byte_count = facet->byte_count;
3495 facet->prev_used = facet->used;
3497 flow_push_stats(facet->rule, &facet->flow,
3498 new_packets, new_bytes, facet->used);
3500 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
3501 facet->mirrors, new_packets, new_bytes);
3505 struct ofproto_push {
3506 struct action_xlate_ctx ctx;
3513 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
3515 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3518 rule->packet_count += push->packets;
3519 rule->byte_count += push->bytes;
3520 rule->used = MAX(push->used, rule->used);
3524 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3525 * 'rule''s actions and mirrors. */
3527 flow_push_stats(const struct rule_dpif *rule,
3528 const struct flow *flow, uint64_t packets, uint64_t bytes,
3531 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3532 struct ofproto_push push;
3534 push.packets = packets;
3538 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, NULL);
3539 push.ctx.resubmit_hook = push_resubmit;
3540 ofpbuf_delete(xlate_actions(&push.ctx,
3541 rule->up.actions, rule->up.n_actions));
3546 static struct subfacet *
3547 subfacet_find__(struct ofproto_dpif *ofproto,
3548 const struct nlattr *key, size_t key_len, uint32_t key_hash,
3549 const struct flow *flow)
3551 struct subfacet *subfacet;
3553 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
3554 &ofproto->subfacets) {
3556 ? (subfacet->key_len == key_len
3557 && !memcmp(key, subfacet->key, key_len))
3558 : flow_equal(flow, &subfacet->facet->flow)) {
3566 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
3567 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
3568 * there is one, otherwise creates and returns a new subfacet.
3570 * If the returned subfacet is new, then subfacet->actions will be NULL, in
3571 * which case the caller must populate the actions with
3572 * subfacet_make_actions(). */
3573 static struct subfacet *
3574 subfacet_create(struct ofproto_dpif *ofproto, struct facet *facet,
3575 enum odp_key_fitness key_fitness,
3576 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
3578 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3579 struct subfacet *subfacet;
3581 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
3583 if (subfacet->facet == facet) {
3587 /* This shouldn't happen. */
3588 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
3589 subfacet_destroy(ofproto, subfacet);
3592 subfacet = xzalloc(sizeof *subfacet);
3593 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
3594 list_push_back(&facet->subfacets, &subfacet->list_node);
3595 subfacet->facet = facet;
3596 subfacet->used = time_msec();
3597 subfacet->key_fitness = key_fitness;
3598 if (key_fitness != ODP_FIT_PERFECT) {
3599 subfacet->key = xmemdup(key, key_len);
3600 subfacet->key_len = key_len;
3602 subfacet->installed = false;
3603 subfacet->initial_tci = initial_tci;
3608 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
3609 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
3610 static struct subfacet *
3611 subfacet_find(struct ofproto_dpif *ofproto,
3612 const struct nlattr *key, size_t key_len)
3614 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3615 enum odp_key_fitness fitness;
3618 fitness = odp_flow_key_to_flow(key, key_len, &flow);
3619 if (fitness == ODP_FIT_ERROR) {
3623 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
3626 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
3627 * its facet within 'ofproto', and frees it. */
3629 subfacet_destroy__(struct ofproto_dpif *ofproto, struct subfacet *subfacet)
3631 subfacet_uninstall(ofproto, subfacet);
3632 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
3633 list_remove(&subfacet->list_node);
3634 free(subfacet->key);
3635 free(subfacet->actions);
3639 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
3640 * last remaining subfacet in its facet destroys the facet too. */
3642 subfacet_destroy(struct ofproto_dpif *ofproto, struct subfacet *subfacet)
3644 struct facet *facet = subfacet->facet;
3646 if (list_is_singleton(&facet->subfacets)) {
3647 /* facet_remove() needs at least one subfacet (it will remove it). */
3648 facet_remove(ofproto, facet);
3650 subfacet_destroy__(ofproto, subfacet);
3654 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
3655 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
3656 * for use as temporary storage. */
3658 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
3661 if (!subfacet->key) {
3662 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
3663 odp_flow_key_from_flow(key, &subfacet->facet->flow);
3665 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
3669 /* Composes the datapath actions for 'subfacet' based on its rule's actions. */
3671 subfacet_make_actions(struct ofproto_dpif *p, struct subfacet *subfacet,
3672 const struct ofpbuf *packet)
3674 struct facet *facet = subfacet->facet;
3675 const struct rule_dpif *rule = facet->rule;
3676 struct ofpbuf *odp_actions;
3677 struct action_xlate_ctx ctx;
3679 action_xlate_ctx_init(&ctx, p, &facet->flow, subfacet->initial_tci,
3681 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3682 facet->tags = ctx.tags;
3683 facet->may_install = ctx.may_set_up_flow;
3684 facet->has_learn = ctx.has_learn;
3685 facet->has_normal = ctx.has_normal;
3686 facet->nf_flow.output_iface = ctx.nf_output_iface;
3687 facet->mirrors = ctx.mirrors;
3689 if (subfacet->actions_len != odp_actions->size
3690 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
3691 free(subfacet->actions);
3692 subfacet->actions_len = odp_actions->size;
3693 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
3696 ofpbuf_delete(odp_actions);
3699 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
3700 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
3701 * in the datapath will be zeroed and 'stats' will be updated with traffic new
3702 * since 'subfacet' was last updated.
3704 * Returns 0 if successful, otherwise a positive errno value. */
3706 subfacet_install(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3707 const struct nlattr *actions, size_t actions_len,
3708 struct dpif_flow_stats *stats)
3710 struct odputil_keybuf keybuf;
3711 enum dpif_flow_put_flags flags;
3715 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3717 flags |= DPIF_FP_ZERO_STATS;
3720 subfacet_get_key(subfacet, &keybuf, &key);
3721 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
3722 actions, actions_len, stats);
3725 subfacet_reset_dp_stats(subfacet, stats);
3731 /* If 'subfacet' is installed in the datapath, uninstalls it. */
3733 subfacet_uninstall(struct ofproto_dpif *p, struct subfacet *subfacet)
3735 if (subfacet->installed) {
3736 struct odputil_keybuf keybuf;
3737 struct dpif_flow_stats stats;
3741 subfacet_get_key(subfacet, &keybuf, &key);
3742 error = dpif_flow_del(p->dpif, key.data, key.size, &stats);
3743 subfacet_reset_dp_stats(subfacet, &stats);
3745 subfacet_update_stats(p, subfacet, &stats);
3747 subfacet->installed = false;
3749 assert(subfacet->dp_packet_count == 0);
3750 assert(subfacet->dp_byte_count == 0);
3754 /* Resets 'subfacet''s datapath statistics counters. This should be called
3755 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
3756 * non-null, it should contain the statistics returned by dpif when 'subfacet'
3757 * was reset in the datapath. 'stats' will be modified to include only
3758 * statistics new since 'subfacet' was last updated. */
3760 subfacet_reset_dp_stats(struct subfacet *subfacet,
3761 struct dpif_flow_stats *stats)
3764 && subfacet->dp_packet_count <= stats->n_packets
3765 && subfacet->dp_byte_count <= stats->n_bytes) {
3766 stats->n_packets -= subfacet->dp_packet_count;
3767 stats->n_bytes -= subfacet->dp_byte_count;
3770 subfacet->dp_packet_count = 0;
3771 subfacet->dp_byte_count = 0;
3774 /* Updates 'subfacet''s used time. The caller is responsible for calling
3775 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
3777 subfacet_update_time(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3780 if (used > subfacet->used) {
3781 subfacet->used = used;
3782 facet_update_time(ofproto, subfacet->facet, used);
3786 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
3788 * Because of the meaning of a subfacet's counters, it only makes sense to do
3789 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
3790 * represents a packet that was sent by hand or if it represents statistics
3791 * that have been cleared out of the datapath. */
3793 subfacet_update_stats(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3794 const struct dpif_flow_stats *stats)
3796 if (stats->n_packets || stats->used > subfacet->used) {
3797 struct facet *facet = subfacet->facet;
3799 subfacet_update_time(ofproto, subfacet, stats->used);
3800 facet->packet_count += stats->n_packets;
3801 facet->byte_count += stats->n_bytes;
3802 facet_push_stats(facet);
3803 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3809 static struct rule_dpif *
3810 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
3813 struct cls_rule *cls_rule;
3814 struct classifier *cls;
3816 if (table_id >= N_TABLES) {
3820 cls = &ofproto->up.tables[table_id];
3821 if (flow->nw_frag & FLOW_NW_FRAG_ANY
3822 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3823 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
3824 * are unavailable. */
3825 struct flow ofpc_normal_flow = *flow;
3826 ofpc_normal_flow.tp_src = htons(0);
3827 ofpc_normal_flow.tp_dst = htons(0);
3828 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
3830 cls_rule = classifier_lookup(cls, flow);
3832 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
3836 complete_operation(struct rule_dpif *rule)
3838 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3840 rule_invalidate(rule);
3842 struct dpif_completion *c = xmalloc(sizeof *c);
3843 c->op = rule->up.pending;
3844 list_push_back(&ofproto->completions, &c->list_node);
3846 ofoperation_complete(rule->up.pending, 0);
3850 static struct rule *
3853 struct rule_dpif *rule = xmalloc(sizeof *rule);
3858 rule_dealloc(struct rule *rule_)
3860 struct rule_dpif *rule = rule_dpif_cast(rule_);
3865 rule_construct(struct rule *rule_)
3867 struct rule_dpif *rule = rule_dpif_cast(rule_);
3868 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3869 struct rule_dpif *victim;
3873 error = validate_actions(rule->up.actions, rule->up.n_actions,
3874 &rule->up.cr.flow, ofproto->max_ports);
3879 rule->used = rule->up.created;
3880 rule->packet_count = 0;
3881 rule->byte_count = 0;
3883 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
3884 if (victim && !list_is_empty(&victim->facets)) {
3885 struct facet *facet;
3887 rule->facets = victim->facets;
3888 list_moved(&rule->facets);
3889 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3890 /* XXX: We're only clearing our local counters here. It's possible
3891 * that quite a few packets are unaccounted for in the datapath
3892 * statistics. These will be accounted to the new rule instead of
3893 * cleared as required. This could be fixed by clearing out the
3894 * datapath statistics for this facet, but currently it doesn't
3896 facet_reset_counters(facet);
3900 /* Must avoid list_moved() in this case. */
3901 list_init(&rule->facets);
3904 table_id = rule->up.table_id;
3905 rule->tag = (victim ? victim->tag
3907 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
3908 ofproto->tables[table_id].basis));
3910 complete_operation(rule);
3915 rule_destruct(struct rule *rule_)
3917 struct rule_dpif *rule = rule_dpif_cast(rule_);
3918 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3919 struct facet *facet, *next_facet;
3921 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3922 facet_revalidate(ofproto, facet);
3925 complete_operation(rule);
3929 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
3931 struct rule_dpif *rule = rule_dpif_cast(rule_);
3932 struct facet *facet;
3934 /* Start from historical data for 'rule' itself that are no longer tracked
3935 * in facets. This counts, for example, facets that have expired. */
3936 *packets = rule->packet_count;
3937 *bytes = rule->byte_count;
3939 /* Add any statistics that are tracked by facets. This includes
3940 * statistical data recently updated by ofproto_update_stats() as well as
3941 * stats for packets that were executed "by hand" via dpif_execute(). */
3942 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3943 *packets += facet->packet_count;
3944 *bytes += facet->byte_count;
3949 rule_execute(struct rule *rule_, const struct flow *flow,
3950 struct ofpbuf *packet)
3952 struct rule_dpif *rule = rule_dpif_cast(rule_);
3953 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3954 struct action_xlate_ctx ctx;
3955 struct ofpbuf *odp_actions;
3958 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, packet);
3959 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3960 size = packet->size;
3961 if (execute_odp_actions(ofproto, flow, odp_actions->data,
3962 odp_actions->size, packet)) {
3963 rule->used = time_msec();
3964 rule->packet_count++;
3965 rule->byte_count += size;
3966 flow_push_stats(rule, flow, 1, size, rule->used);
3968 ofpbuf_delete(odp_actions);
3974 rule_modify_actions(struct rule *rule_)
3976 struct rule_dpif *rule = rule_dpif_cast(rule_);
3977 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3980 error = validate_actions(rule->up.actions, rule->up.n_actions,
3981 &rule->up.cr.flow, ofproto->max_ports);
3983 ofoperation_complete(rule->up.pending, error);
3987 complete_operation(rule);
3990 /* Sends 'packet' out 'ofport'.
3991 * May modify 'packet'.
3992 * Returns 0 if successful, otherwise a positive errno value. */
3994 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
3996 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3997 struct ofpbuf key, odp_actions;
3998 struct odputil_keybuf keybuf;
4003 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
4004 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4006 if (odp_port != ofport->odp_port) {
4007 eth_pop_vlan(packet);
4008 flow.vlan_tci = htons(0);
4011 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4012 odp_flow_key_from_flow(&key, &flow);
4014 ofpbuf_init(&odp_actions, 32);
4015 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4017 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4018 error = dpif_execute(ofproto->dpif,
4020 odp_actions.data, odp_actions.size,
4022 ofpbuf_uninit(&odp_actions);
4025 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4026 ofproto->up.name, odp_port, strerror(error));
4028 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
4032 /* OpenFlow to datapath action translation. */
4034 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
4035 struct action_xlate_ctx *ctx);
4036 static void xlate_normal(struct action_xlate_ctx *);
4039 put_userspace_action(const struct ofproto_dpif *ofproto,
4040 struct ofpbuf *odp_actions,
4041 const struct flow *flow,
4042 const struct user_action_cookie *cookie)
4046 pid = dpif_port_get_pid(ofproto->dpif,
4047 ofp_port_to_odp_port(flow->in_port));
4049 return odp_put_userspace_action(pid, cookie, odp_actions);
4052 /* Compose SAMPLE action for sFlow. */
4054 compose_sflow_action(const struct ofproto_dpif *ofproto,
4055 struct ofpbuf *odp_actions,
4056 const struct flow *flow,
4059 uint32_t port_ifindex;
4060 uint32_t probability;
4061 struct user_action_cookie cookie;
4062 size_t sample_offset, actions_offset;
4063 int cookie_offset, n_output;
4065 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4069 if (odp_port == OVSP_NONE) {
4073 port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4077 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4079 /* Number of packets out of UINT_MAX to sample. */
4080 probability = dpif_sflow_get_probability(ofproto->sflow);
4081 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4083 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4085 cookie.type = USER_ACTION_COOKIE_SFLOW;
4086 cookie.data = port_ifindex;
4087 cookie.n_output = n_output;
4088 cookie.vlan_tci = 0;
4089 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4091 nl_msg_end_nested(odp_actions, actions_offset);
4092 nl_msg_end_nested(odp_actions, sample_offset);
4093 return cookie_offset;
4096 /* SAMPLE action must be first action in any given list of actions.
4097 * At this point we do not have all information required to build it. So try to
4098 * build sample action as complete as possible. */
4100 add_sflow_action(struct action_xlate_ctx *ctx)
4102 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4104 &ctx->flow, OVSP_NONE);
4105 ctx->sflow_odp_port = 0;
4106 ctx->sflow_n_outputs = 0;
4109 /* Fix SAMPLE action according to data collected while composing ODP actions.
4110 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4111 * USERSPACE action's user-cookie which is required for sflow. */
4113 fix_sflow_action(struct action_xlate_ctx *ctx)
4115 const struct flow *base = &ctx->base_flow;
4116 struct user_action_cookie *cookie;
4118 if (!ctx->user_cookie_offset) {
4122 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4124 assert(cookie != NULL);
4125 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4127 if (ctx->sflow_n_outputs) {
4128 cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow,
4129 ctx->sflow_odp_port);
4131 if (ctx->sflow_n_outputs >= 255) {
4132 cookie->n_output = 255;
4134 cookie->n_output = ctx->sflow_n_outputs;
4136 cookie->vlan_tci = base->vlan_tci;
4140 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4143 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4144 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4145 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4146 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4150 struct priority_to_dscp *pdscp;
4152 if (ofport->up.opp.config & htonl(OFPPC_NO_FWD)
4153 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4157 pdscp = get_priority(ofport, ctx->flow.skb_priority);
4159 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4160 ctx->flow.nw_tos |= pdscp->dscp;
4163 /* We may not have an ofport record for this port, but it doesn't hurt
4164 * to allow forwarding to it anyhow. Maybe such a port will appear
4165 * later and we're pre-populating the flow table. */
4168 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4169 ctx->flow.vlan_tci);
4170 if (out_port != odp_port) {
4171 ctx->flow.vlan_tci = htons(0);
4173 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4174 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4176 ctx->sflow_odp_port = odp_port;
4177 ctx->sflow_n_outputs++;
4178 ctx->nf_output_iface = ofp_port;
4179 ctx->flow.vlan_tci = flow_vlan_tci;
4180 ctx->flow.nw_tos = flow_nw_tos;
4184 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4186 compose_output_action__(ctx, ofp_port, true);
4190 xlate_table_action(struct action_xlate_ctx *ctx,
4191 uint16_t in_port, uint8_t table_id)
4193 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4194 struct ofproto_dpif *ofproto = ctx->ofproto;
4195 struct rule_dpif *rule;
4196 uint16_t old_in_port;
4197 uint8_t old_table_id;
4199 old_table_id = ctx->table_id;
4200 ctx->table_id = table_id;
4202 /* Look up a flow with 'in_port' as the input port. */
4203 old_in_port = ctx->flow.in_port;
4204 ctx->flow.in_port = in_port;
4205 rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
4208 if (table_id > 0 && table_id < N_TABLES) {
4209 struct table_dpif *table = &ofproto->tables[table_id];
4210 if (table->other_table) {
4213 : rule_calculate_tag(&ctx->flow,
4214 &table->other_table->wc,
4219 /* Restore the original input port. Otherwise OFPP_NORMAL and
4220 * OFPP_IN_PORT will have surprising behavior. */
4221 ctx->flow.in_port = old_in_port;
4223 if (ctx->resubmit_hook) {
4224 ctx->resubmit_hook(ctx, rule);
4229 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
4233 ctx->table_id = old_table_id;
4235 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4237 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
4238 MAX_RESUBMIT_RECURSION);
4243 xlate_resubmit_table(struct action_xlate_ctx *ctx,
4244 const struct nx_action_resubmit *nar)
4249 in_port = (nar->in_port == htons(OFPP_IN_PORT)
4251 : ntohs(nar->in_port));
4252 table_id = nar->table == 255 ? ctx->table_id : nar->table;
4254 xlate_table_action(ctx, in_port, table_id);
4258 flood_packets(struct action_xlate_ctx *ctx, bool all)
4260 struct ofport_dpif *ofport;
4262 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
4263 uint16_t ofp_port = ofport->up.ofp_port;
4265 if (ofp_port == ctx->flow.in_port) {
4270 compose_output_action__(ctx, ofp_port, false);
4271 } else if (!(ofport->up.opp.config & htonl(OFPPC_NO_FLOOD))) {
4272 compose_output_action(ctx, ofp_port);
4276 ctx->nf_output_iface = NF_OUT_FLOOD;
4280 execute_controller_action(struct action_xlate_ctx *ctx, int len)
4282 struct ofputil_packet_in pin;
4283 struct ofpbuf *packet;
4285 ctx->may_set_up_flow = false;
4290 packet = ofpbuf_clone(ctx->packet);
4292 if (packet->l2 && packet->l3) {
4293 struct eth_header *eh;
4295 eth_pop_vlan(packet);
4297 assert(eh->eth_type == ctx->flow.dl_type);
4298 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
4299 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
4301 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
4302 eth_push_vlan(packet, ctx->flow.vlan_tci);
4306 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4307 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
4308 ctx->flow.nw_tos, ctx->flow.nw_ttl);
4312 if (ctx->flow.nw_proto == IPPROTO_TCP) {
4313 packet_set_tcp_port(packet, ctx->flow.tp_src,
4315 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
4316 packet_set_udp_port(packet, ctx->flow.tp_src,
4323 pin.packet = packet->data;
4324 pin.packet_len = packet->size;
4325 pin.reason = OFPR_ACTION;
4328 pin.total_len = packet->size;
4329 flow_get_metadata(&ctx->flow, &pin.fmd);
4331 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin, &ctx->flow);
4332 ofpbuf_delete(packet);
4336 xlate_output_action__(struct action_xlate_ctx *ctx,
4337 uint16_t port, uint16_t max_len)
4339 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
4341 ctx->nf_output_iface = NF_OUT_DROP;
4345 compose_output_action(ctx, ctx->flow.in_port);
4348 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
4354 flood_packets(ctx, false);
4357 flood_packets(ctx, true);
4359 case OFPP_CONTROLLER:
4360 execute_controller_action(ctx, max_len);
4363 compose_output_action(ctx, OFPP_LOCAL);
4368 if (port != ctx->flow.in_port) {
4369 compose_output_action(ctx, port);
4374 if (prev_nf_output_iface == NF_OUT_FLOOD) {
4375 ctx->nf_output_iface = NF_OUT_FLOOD;
4376 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
4377 ctx->nf_output_iface = prev_nf_output_iface;
4378 } else if (prev_nf_output_iface != NF_OUT_DROP &&
4379 ctx->nf_output_iface != NF_OUT_FLOOD) {
4380 ctx->nf_output_iface = NF_OUT_MULTI;
4385 xlate_output_reg_action(struct action_xlate_ctx *ctx,
4386 const struct nx_action_output_reg *naor)
4390 ofp_port = nxm_read_field_bits(naor->src, naor->ofs_nbits, &ctx->flow);
4392 if (ofp_port <= UINT16_MAX) {
4393 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
4398 xlate_output_action(struct action_xlate_ctx *ctx,
4399 const struct ofp_action_output *oao)
4401 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
4405 xlate_enqueue_action(struct action_xlate_ctx *ctx,
4406 const struct ofp_action_enqueue *oae)
4409 uint32_t flow_priority, priority;
4412 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
4415 /* Fall back to ordinary output action. */
4416 xlate_output_action__(ctx, ntohs(oae->port), 0);
4420 /* Figure out datapath output port. */
4421 ofp_port = ntohs(oae->port);
4422 if (ofp_port == OFPP_IN_PORT) {
4423 ofp_port = ctx->flow.in_port;
4424 } else if (ofp_port == ctx->flow.in_port) {
4428 /* Add datapath actions. */
4429 flow_priority = ctx->flow.skb_priority;
4430 ctx->flow.skb_priority = priority;
4431 compose_output_action(ctx, ofp_port);
4432 ctx->flow.skb_priority = flow_priority;
4434 /* Update NetFlow output port. */
4435 if (ctx->nf_output_iface == NF_OUT_DROP) {
4436 ctx->nf_output_iface = ofp_port;
4437 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
4438 ctx->nf_output_iface = NF_OUT_MULTI;
4443 xlate_set_queue_action(struct action_xlate_ctx *ctx,
4444 const struct nx_action_set_queue *nasq)
4449 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
4452 /* Couldn't translate queue to a priority, so ignore. A warning
4453 * has already been logged. */
4457 ctx->flow.skb_priority = priority;
4460 struct xlate_reg_state {
4466 xlate_autopath(struct action_xlate_ctx *ctx,
4467 const struct nx_action_autopath *naa)
4469 uint16_t ofp_port = ntohl(naa->id);
4470 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
4472 if (!port || !port->bundle) {
4473 ofp_port = OFPP_NONE;
4474 } else if (port->bundle->bond) {
4475 /* Autopath does not support VLAN hashing. */
4476 struct ofport_dpif *slave = bond_choose_output_slave(
4477 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
4479 ofp_port = slave->up.ofp_port;
4482 autopath_execute(naa, &ctx->flow, ofp_port);
4486 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
4488 struct ofproto_dpif *ofproto = ofproto_;
4489 struct ofport_dpif *port;
4499 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
4502 port = get_ofp_port(ofproto, ofp_port);
4503 return port ? port->may_enable : false;
4508 xlate_learn_action(struct action_xlate_ctx *ctx,
4509 const struct nx_action_learn *learn)
4511 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
4512 struct ofputil_flow_mod fm;
4515 learn_execute(learn, &ctx->flow, &fm);
4517 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
4518 if (error && !VLOG_DROP_WARN(&rl)) {
4519 char *msg = ofputil_error_to_string(error);
4520 VLOG_WARN("learning action failed to modify flow table (%s)", msg);
4528 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
4530 if (port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
4531 ? htonl(OFPPC_NO_RECV_STP)
4532 : htonl(OFPPC_NO_RECV))) {
4536 /* Only drop packets here if both forwarding and learning are
4537 * disabled. If just learning is enabled, we need to have
4538 * OFPP_NORMAL and the learning action have a look at the packet
4539 * before we can drop it. */
4540 if (!stp_forward_in_state(port->stp_state)
4541 && !stp_learn_in_state(port->stp_state)) {
4549 do_xlate_actions(const union ofp_action *in, size_t n_in,
4550 struct action_xlate_ctx *ctx)
4552 const struct ofport_dpif *port;
4553 const union ofp_action *ia;
4556 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
4557 if (port && !may_receive(port, ctx)) {
4558 /* Drop this flow. */
4562 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
4563 const struct ofp_action_dl_addr *oada;
4564 const struct nx_action_resubmit *nar;
4565 const struct nx_action_set_tunnel *nast;
4566 const struct nx_action_set_queue *nasq;
4567 const struct nx_action_multipath *nam;
4568 const struct nx_action_autopath *naa;
4569 const struct nx_action_bundle *nab;
4570 const struct nx_action_output_reg *naor;
4571 enum ofputil_action_code code;
4578 code = ofputil_decode_action_unsafe(ia);
4580 case OFPUTIL_OFPAT_OUTPUT:
4581 xlate_output_action(ctx, &ia->output);
4584 case OFPUTIL_OFPAT_SET_VLAN_VID:
4585 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
4586 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
4589 case OFPUTIL_OFPAT_SET_VLAN_PCP:
4590 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
4591 ctx->flow.vlan_tci |= htons(
4592 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
4595 case OFPUTIL_OFPAT_STRIP_VLAN:
4596 ctx->flow.vlan_tci = htons(0);
4599 case OFPUTIL_OFPAT_SET_DL_SRC:
4600 oada = ((struct ofp_action_dl_addr *) ia);
4601 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
4604 case OFPUTIL_OFPAT_SET_DL_DST:
4605 oada = ((struct ofp_action_dl_addr *) ia);
4606 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
4609 case OFPUTIL_OFPAT_SET_NW_SRC:
4610 ctx->flow.nw_src = ia->nw_addr.nw_addr;
4613 case OFPUTIL_OFPAT_SET_NW_DST:
4614 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
4617 case OFPUTIL_OFPAT_SET_NW_TOS:
4618 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4619 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
4622 case OFPUTIL_OFPAT_SET_TP_SRC:
4623 ctx->flow.tp_src = ia->tp_port.tp_port;
4626 case OFPUTIL_OFPAT_SET_TP_DST:
4627 ctx->flow.tp_dst = ia->tp_port.tp_port;
4630 case OFPUTIL_OFPAT_ENQUEUE:
4631 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
4634 case OFPUTIL_NXAST_RESUBMIT:
4635 nar = (const struct nx_action_resubmit *) ia;
4636 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
4639 case OFPUTIL_NXAST_RESUBMIT_TABLE:
4640 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
4643 case OFPUTIL_NXAST_SET_TUNNEL:
4644 nast = (const struct nx_action_set_tunnel *) ia;
4645 tun_id = htonll(ntohl(nast->tun_id));
4646 ctx->flow.tun_id = tun_id;
4649 case OFPUTIL_NXAST_SET_QUEUE:
4650 nasq = (const struct nx_action_set_queue *) ia;
4651 xlate_set_queue_action(ctx, nasq);
4654 case OFPUTIL_NXAST_POP_QUEUE:
4655 ctx->flow.skb_priority = ctx->orig_skb_priority;
4658 case OFPUTIL_NXAST_REG_MOVE:
4659 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
4663 case OFPUTIL_NXAST_REG_LOAD:
4664 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
4668 case OFPUTIL_NXAST_NOTE:
4669 /* Nothing to do. */
4672 case OFPUTIL_NXAST_SET_TUNNEL64:
4673 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
4674 ctx->flow.tun_id = tun_id;
4677 case OFPUTIL_NXAST_MULTIPATH:
4678 nam = (const struct nx_action_multipath *) ia;
4679 multipath_execute(nam, &ctx->flow);
4682 case OFPUTIL_NXAST_AUTOPATH:
4683 naa = (const struct nx_action_autopath *) ia;
4684 xlate_autopath(ctx, naa);
4687 case OFPUTIL_NXAST_BUNDLE:
4688 ctx->ofproto->has_bundle_action = true;
4689 nab = (const struct nx_action_bundle *) ia;
4690 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
4695 case OFPUTIL_NXAST_BUNDLE_LOAD:
4696 ctx->ofproto->has_bundle_action = true;
4697 nab = (const struct nx_action_bundle *) ia;
4698 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
4702 case OFPUTIL_NXAST_OUTPUT_REG:
4703 naor = (const struct nx_action_output_reg *) ia;
4704 xlate_output_reg_action(ctx, naor);
4707 case OFPUTIL_NXAST_LEARN:
4708 ctx->has_learn = true;
4709 if (ctx->may_learn) {
4710 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
4714 case OFPUTIL_NXAST_EXIT:
4720 /* We've let OFPP_NORMAL and the learning action look at the packet,
4721 * so drop it now if forwarding is disabled. */
4722 if (port && !stp_forward_in_state(port->stp_state)) {
4723 ofpbuf_clear(ctx->odp_actions);
4724 add_sflow_action(ctx);
4729 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
4730 struct ofproto_dpif *ofproto, const struct flow *flow,
4731 ovs_be16 initial_tci, const struct ofpbuf *packet)
4733 ctx->ofproto = ofproto;
4735 ctx->base_flow = ctx->flow;
4736 ctx->base_flow.tun_id = 0;
4737 ctx->base_flow.vlan_tci = initial_tci;
4738 ctx->packet = packet;
4739 ctx->may_learn = packet != NULL;
4740 ctx->resubmit_hook = NULL;
4743 static struct ofpbuf *
4744 xlate_actions(struct action_xlate_ctx *ctx,
4745 const union ofp_action *in, size_t n_in)
4747 struct flow orig_flow = ctx->flow;
4749 COVERAGE_INC(ofproto_dpif_xlate);
4751 ctx->odp_actions = ofpbuf_new(512);
4752 ofpbuf_reserve(ctx->odp_actions, NL_A_U32_SIZE);
4754 ctx->may_set_up_flow = true;
4755 ctx->has_learn = false;
4756 ctx->has_normal = false;
4757 ctx->nf_output_iface = NF_OUT_DROP;
4760 ctx->orig_skb_priority = ctx->flow.skb_priority;
4764 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
4765 switch (ctx->ofproto->up.frag_handling) {
4766 case OFPC_FRAG_NORMAL:
4767 /* We must pretend that transport ports are unavailable. */
4768 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
4769 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
4772 case OFPC_FRAG_DROP:
4773 return ctx->odp_actions;
4775 case OFPC_FRAG_REASM:
4778 case OFPC_FRAG_NX_MATCH:
4779 /* Nothing to do. */
4784 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
4785 ctx->may_set_up_flow = false;
4786 return ctx->odp_actions;
4788 add_sflow_action(ctx);
4789 do_xlate_actions(in, n_in, ctx);
4791 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
4792 ctx->odp_actions->data,
4793 ctx->odp_actions->size)) {
4794 ctx->may_set_up_flow = false;
4796 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
4798 compose_output_action(ctx, OFPP_LOCAL);
4801 add_mirror_actions(ctx, &orig_flow);
4802 fix_sflow_action(ctx);
4805 return ctx->odp_actions;
4808 /* OFPP_NORMAL implementation. */
4810 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
4812 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
4813 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
4814 * the bundle on which the packet was received, returns the VLAN to which the
4817 * Both 'vid' and the return value are in the range 0...4095. */
4819 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
4821 switch (in_bundle->vlan_mode) {
4822 case PORT_VLAN_ACCESS:
4823 return in_bundle->vlan;
4826 case PORT_VLAN_TRUNK:
4829 case PORT_VLAN_NATIVE_UNTAGGED:
4830 case PORT_VLAN_NATIVE_TAGGED:
4831 return vid ? vid : in_bundle->vlan;
4838 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
4839 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
4842 * 'vid' should be the VID obtained from the 802.1Q header that was received as
4843 * part of a packet (specify 0 if there was no 802.1Q header), in the range
4846 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
4848 /* Allow any VID on the OFPP_NONE port. */
4849 if (in_bundle == &ofpp_none_bundle) {
4853 switch (in_bundle->vlan_mode) {
4854 case PORT_VLAN_ACCESS:
4857 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4858 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
4859 "packet received on port %s configured as VLAN "
4860 "%"PRIu16" access port",
4861 in_bundle->ofproto->up.name, vid,
4862 in_bundle->name, in_bundle->vlan);
4868 case PORT_VLAN_NATIVE_UNTAGGED:
4869 case PORT_VLAN_NATIVE_TAGGED:
4871 /* Port must always carry its native VLAN. */
4875 case PORT_VLAN_TRUNK:
4876 if (!ofbundle_includes_vlan(in_bundle, vid)) {
4878 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4879 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
4880 "received on port %s not configured for trunking "
4882 in_bundle->ofproto->up.name, vid,
4883 in_bundle->name, vid);
4895 /* Given 'vlan', the VLAN that a packet belongs to, and
4896 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
4897 * that should be included in the 802.1Q header. (If the return value is 0,
4898 * then the 802.1Q header should only be included in the packet if there is a
4901 * Both 'vlan' and the return value are in the range 0...4095. */
4903 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
4905 switch (out_bundle->vlan_mode) {
4906 case PORT_VLAN_ACCESS:
4909 case PORT_VLAN_TRUNK:
4910 case PORT_VLAN_NATIVE_TAGGED:
4913 case PORT_VLAN_NATIVE_UNTAGGED:
4914 return vlan == out_bundle->vlan ? 0 : vlan;
4922 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
4925 struct ofport_dpif *port;
4927 ovs_be16 tci, old_tci;
4929 vid = output_vlan_to_vid(out_bundle, vlan);
4930 if (!out_bundle->bond) {
4931 port = ofbundle_get_a_port(out_bundle);
4933 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
4936 /* No slaves enabled, so drop packet. */
4941 old_tci = ctx->flow.vlan_tci;
4943 if (tci || out_bundle->use_priority_tags) {
4944 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
4946 tci |= htons(VLAN_CFI);
4949 ctx->flow.vlan_tci = tci;
4951 compose_output_action(ctx, port->up.ofp_port);
4952 ctx->flow.vlan_tci = old_tci;
4956 mirror_mask_ffs(mirror_mask_t mask)
4958 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
4963 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
4965 return (bundle->vlan_mode != PORT_VLAN_ACCESS
4966 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
4970 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
4972 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
4975 /* Returns an arbitrary interface within 'bundle'. */
4976 static struct ofport_dpif *
4977 ofbundle_get_a_port(const struct ofbundle *bundle)
4979 return CONTAINER_OF(list_front(&bundle->ports),
4980 struct ofport_dpif, bundle_node);
4984 vlan_is_mirrored(const struct ofmirror *m, int vlan)
4986 return !m->vlans || bitmap_is_set(m->vlans, vlan);
4989 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
4990 * to a VLAN. In general most packets may be mirrored but we want to drop
4991 * protocols that may confuse switches. */
4993 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
4995 /* If you change this function's behavior, please update corresponding
4996 * documentation in vswitch.xml at the same time. */
4997 if (dst[0] != 0x01) {
4998 /* All the currently banned MACs happen to start with 01 currently, so
4999 * this is a quick way to eliminate most of the good ones. */
5001 if (eth_addr_is_reserved(dst)) {
5002 /* Drop STP, IEEE pause frames, and other reserved protocols
5003 * (01-80-c2-00-00-0x). */
5007 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
5009 if ((dst[3] & 0xfe) == 0xcc &&
5010 (dst[4] & 0xfe) == 0xcc &&
5011 (dst[5] & 0xfe) == 0xcc) {
5012 /* Drop the following protocols plus others following the same
5015 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
5016 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
5017 STP Uplink Fast (01-00-0c-cd-cd-cd) */
5021 if (!(dst[3] | dst[4] | dst[5])) {
5022 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
5031 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5033 struct ofproto_dpif *ofproto = ctx->ofproto;
5034 mirror_mask_t mirrors;
5035 struct ofbundle *in_bundle;
5038 const struct nlattr *a;
5041 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5042 ctx->packet != NULL);
5046 mirrors = in_bundle->src_mirrors;
5048 /* Drop frames on bundles reserved for mirroring. */
5049 if (in_bundle->mirror_out) {
5050 if (ctx->packet != NULL) {
5051 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5052 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5053 "%s, which is reserved exclusively for mirroring",
5054 ctx->ofproto->up.name, in_bundle->name);
5060 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5061 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5064 vlan = input_vid_to_vlan(in_bundle, vid);
5066 /* Look at the output ports to check for destination selections. */
5068 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5069 ctx->odp_actions->size) {
5070 enum ovs_action_attr type = nl_attr_type(a);
5071 struct ofport_dpif *ofport;
5073 if (type != OVS_ACTION_ATTR_OUTPUT) {
5077 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5078 if (ofport && ofport->bundle) {
5079 mirrors |= ofport->bundle->dst_mirrors;
5087 /* Restore the original packet before adding the mirror actions. */
5088 ctx->flow = *orig_flow;
5093 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5095 if (!vlan_is_mirrored(m, vlan)) {
5096 mirrors &= mirrors - 1;
5100 mirrors &= ~m->dup_mirrors;
5101 ctx->mirrors |= m->dup_mirrors;
5103 output_normal(ctx, m->out, vlan);
5104 } else if (eth_dst_may_rspan(orig_flow->dl_dst)
5105 && vlan != m->out_vlan) {
5106 struct ofbundle *bundle;
5108 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5109 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5110 && !bundle->mirror_out) {
5111 output_normal(ctx, bundle, m->out_vlan);
5119 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5120 uint64_t packets, uint64_t bytes)
5126 for (; mirrors; mirrors &= mirrors - 1) {
5129 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5132 /* In normal circumstances 'm' will not be NULL. However,
5133 * if mirrors are reconfigured, we can temporarily get out
5134 * of sync in facet_revalidate(). We could "correct" the
5135 * mirror list before reaching here, but doing that would
5136 * not properly account the traffic stats we've currently
5137 * accumulated for previous mirror configuration. */
5141 m->packet_count += packets;
5142 m->byte_count += bytes;
5146 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
5147 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
5148 * indicate this; newer upstream kernels use gratuitous ARP requests. */
5150 is_gratuitous_arp(const struct flow *flow)
5152 return (flow->dl_type == htons(ETH_TYPE_ARP)
5153 && eth_addr_is_broadcast(flow->dl_dst)
5154 && (flow->nw_proto == ARP_OP_REPLY
5155 || (flow->nw_proto == ARP_OP_REQUEST
5156 && flow->nw_src == flow->nw_dst)));
5160 update_learning_table(struct ofproto_dpif *ofproto,
5161 const struct flow *flow, int vlan,
5162 struct ofbundle *in_bundle)
5164 struct mac_entry *mac;
5166 /* Don't learn the OFPP_NONE port. */
5167 if (in_bundle == &ofpp_none_bundle) {
5171 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
5175 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
5176 if (is_gratuitous_arp(flow)) {
5177 /* We don't want to learn from gratuitous ARP packets that are
5178 * reflected back over bond slaves so we lock the learning table. */
5179 if (!in_bundle->bond) {
5180 mac_entry_set_grat_arp_lock(mac);
5181 } else if (mac_entry_is_grat_arp_locked(mac)) {
5186 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
5187 /* The log messages here could actually be useful in debugging,
5188 * so keep the rate limit relatively high. */
5189 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5190 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
5191 "on port %s in VLAN %d",
5192 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
5193 in_bundle->name, vlan);
5195 mac->port.p = in_bundle;
5196 tag_set_add(&ofproto->revalidate_set,
5197 mac_learning_changed(ofproto->ml, mac));
5201 static struct ofbundle *
5202 lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn)
5204 struct ofport_dpif *ofport;
5206 /* Special-case OFPP_NONE, which a controller may use as the ingress
5207 * port for traffic that it is sourcing. */
5208 if (in_port == OFPP_NONE) {
5209 return &ofpp_none_bundle;
5212 /* Find the port and bundle for the received packet. */
5213 ofport = get_ofp_port(ofproto, in_port);
5214 if (ofport && ofport->bundle) {
5215 return ofport->bundle;
5218 /* Odd. A few possible reasons here:
5220 * - We deleted a port but there are still a few packets queued up
5223 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
5224 * we don't know about.
5226 * - The ofproto client didn't configure the port as part of a bundle.
5229 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5231 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
5232 "port %"PRIu16, ofproto->up.name, in_port);
5237 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
5238 * dropped. Returns true if they may be forwarded, false if they should be
5241 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
5242 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
5244 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
5245 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
5246 * checked by input_vid_is_valid().
5248 * May also add tags to '*tags', although the current implementation only does
5249 * so in one special case.
5252 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
5253 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
5255 struct ofbundle *in_bundle = in_port->bundle;
5257 /* Drop frames for reserved multicast addresses
5258 * only if forward_bpdu option is absent. */
5259 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
5263 if (in_bundle->bond) {
5264 struct mac_entry *mac;
5266 switch (bond_check_admissibility(in_bundle->bond, in_port,
5267 flow->dl_dst, tags)) {
5274 case BV_DROP_IF_MOVED:
5275 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
5276 if (mac && mac->port.p != in_bundle &&
5277 (!is_gratuitous_arp(flow)
5278 || mac_entry_is_grat_arp_locked(mac))) {
5289 xlate_normal(struct action_xlate_ctx *ctx)
5291 struct ofport_dpif *in_port;
5292 struct ofbundle *in_bundle;
5293 struct mac_entry *mac;
5297 ctx->has_normal = true;
5299 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
5300 ctx->packet != NULL);
5305 /* We know 'in_port' exists unless it is "ofpp_none_bundle",
5306 * since lookup_input_bundle() succeeded. */
5307 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5309 /* Drop malformed frames. */
5310 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
5311 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
5312 if (ctx->packet != NULL) {
5313 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5314 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
5315 "VLAN tag received on port %s",
5316 ctx->ofproto->up.name, in_bundle->name);
5321 /* Drop frames on bundles reserved for mirroring. */
5322 if (in_bundle->mirror_out) {
5323 if (ctx->packet != NULL) {
5324 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5325 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5326 "%s, which is reserved exclusively for mirroring",
5327 ctx->ofproto->up.name, in_bundle->name);
5333 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
5334 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5337 vlan = input_vid_to_vlan(in_bundle, vid);
5339 /* Check other admissibility requirements. */
5341 !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
5345 /* Learn source MAC. */
5346 if (ctx->may_learn) {
5347 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
5350 /* Determine output bundle. */
5351 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
5354 if (mac->port.p != in_bundle) {
5355 output_normal(ctx, mac->port.p, vlan);
5358 struct ofbundle *bundle;
5360 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
5361 if (bundle != in_bundle
5362 && ofbundle_includes_vlan(bundle, vlan)
5363 && bundle->floodable
5364 && !bundle->mirror_out) {
5365 output_normal(ctx, bundle, vlan);
5368 ctx->nf_output_iface = NF_OUT_FLOOD;
5372 /* Optimized flow revalidation.
5374 * It's a difficult problem, in general, to tell which facets need to have
5375 * their actions recalculated whenever the OpenFlow flow table changes. We
5376 * don't try to solve that general problem: for most kinds of OpenFlow flow
5377 * table changes, we recalculate the actions for every facet. This is
5378 * relatively expensive, but it's good enough if the OpenFlow flow table
5379 * doesn't change very often.
5381 * However, we can expect one particular kind of OpenFlow flow table change to
5382 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5383 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5384 * table, we add a special case that applies to flow tables in which every rule
5385 * has the same form (that is, the same wildcards), except that the table is
5386 * also allowed to have a single "catch-all" flow that matches all packets. We
5387 * optimize this case by tagging all of the facets that resubmit into the table
5388 * and invalidating the same tag whenever a flow changes in that table. The
5389 * end result is that we revalidate just the facets that need it (and sometimes
5390 * a few more, but not all of the facets or even all of the facets that
5391 * resubmit to the table modified by MAC learning). */
5393 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
5394 * into an OpenFlow table with the given 'basis'. */
5396 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
5399 if (flow_wildcards_is_catchall(wc)) {
5402 struct flow tag_flow = *flow;
5403 flow_zero_wildcards(&tag_flow, wc);
5404 return tag_create_deterministic(flow_hash(&tag_flow, secret));
5408 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5409 * taggability of that table.
5411 * This function must be called after *each* change to a flow table. If you
5412 * skip calling it on some changes then the pointer comparisons at the end can
5413 * be invalid if you get unlucky. For example, if a flow removal causes a
5414 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5415 * different wildcards to be created with the same address, then this function
5416 * will incorrectly skip revalidation. */
5418 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5420 struct table_dpif *table = &ofproto->tables[table_id];
5421 const struct classifier *cls = &ofproto->up.tables[table_id];
5422 struct cls_table *catchall, *other;
5423 struct cls_table *t;
5425 catchall = other = NULL;
5427 switch (hmap_count(&cls->tables)) {
5429 /* We could tag this OpenFlow table but it would make the logic a
5430 * little harder and it's a corner case that doesn't seem worth it
5436 HMAP_FOR_EACH (t, hmap_node, &cls->tables) {
5437 if (cls_table_is_catchall(t)) {
5439 } else if (!other) {
5442 /* Indicate that we can't tag this by setting both tables to
5443 * NULL. (We know that 'catchall' is already NULL.) */
5450 /* Can't tag this table. */
5454 if (table->catchall_table != catchall || table->other_table != other) {
5455 table->catchall_table = catchall;
5456 table->other_table = other;
5457 ofproto->need_revalidate = true;
5461 /* Given 'rule' that has changed in some way (either it is a rule being
5462 * inserted, a rule being deleted, or a rule whose actions are being
5463 * modified), marks facets for revalidation to ensure that packets will be
5464 * forwarded correctly according to the new state of the flow table.
5466 * This function must be called after *each* change to a flow table. See
5467 * the comment on table_update_taggable() for more information. */
5469 rule_invalidate(const struct rule_dpif *rule)
5471 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5473 table_update_taggable(ofproto, rule->up.table_id);
5475 if (!ofproto->need_revalidate) {
5476 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5478 if (table->other_table && rule->tag) {
5479 tag_set_add(&ofproto->revalidate_set, rule->tag);
5481 ofproto->need_revalidate = true;
5487 set_frag_handling(struct ofproto *ofproto_,
5488 enum ofp_config_flags frag_handling)
5490 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5492 if (frag_handling != OFPC_FRAG_REASM) {
5493 ofproto->need_revalidate = true;
5501 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5502 const struct flow *flow,
5503 const union ofp_action *ofp_actions, size_t n_ofp_actions)
5505 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5508 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
5509 return ofp_mkerr_nicira(OFPET_BAD_REQUEST, NXBRC_BAD_IN_PORT);
5512 error = validate_actions(ofp_actions, n_ofp_actions, flow,
5513 ofproto->max_ports);
5515 struct odputil_keybuf keybuf;
5516 struct action_xlate_ctx ctx;
5517 struct ofpbuf *odp_actions;
5520 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5521 odp_flow_key_from_flow(&key, flow);
5523 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, packet);
5524 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
5525 dpif_execute(ofproto->dpif, key.data, key.size,
5526 odp_actions->data, odp_actions->size, packet);
5527 ofpbuf_delete(odp_actions);
5535 set_netflow(struct ofproto *ofproto_,
5536 const struct netflow_options *netflow_options)
5538 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5540 if (netflow_options) {
5541 if (!ofproto->netflow) {
5542 ofproto->netflow = netflow_create();
5544 return netflow_set_options(ofproto->netflow, netflow_options);
5546 netflow_destroy(ofproto->netflow);
5547 ofproto->netflow = NULL;
5553 get_netflow_ids(const struct ofproto *ofproto_,
5554 uint8_t *engine_type, uint8_t *engine_id)
5556 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5558 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
5562 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5564 if (!facet_is_controller_flow(facet) &&
5565 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5566 struct subfacet *subfacet;
5567 struct ofexpired expired;
5569 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5570 if (subfacet->installed) {
5571 struct dpif_flow_stats stats;
5573 subfacet_install(ofproto, subfacet, subfacet->actions,
5574 subfacet->actions_len, &stats);
5575 subfacet_update_stats(ofproto, subfacet, &stats);
5579 expired.flow = facet->flow;
5580 expired.packet_count = facet->packet_count;
5581 expired.byte_count = facet->byte_count;
5582 expired.used = facet->used;
5583 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5588 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5590 struct facet *facet;
5592 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5593 send_active_timeout(ofproto, facet);
5597 static struct ofproto_dpif *
5598 ofproto_dpif_lookup(const char *name)
5600 struct ofproto_dpif *ofproto;
5602 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5603 hash_string(name, 0), &all_ofproto_dpifs) {
5604 if (!strcmp(ofproto->up.name, name)) {
5612 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc OVS_UNUSED,
5613 const char *argv[], void *aux OVS_UNUSED)
5615 const struct ofproto_dpif *ofproto;
5617 ofproto = ofproto_dpif_lookup(argv[1]);
5619 unixctl_command_reply(conn, 501, "no such bridge");
5622 mac_learning_flush(ofproto->ml);
5624 unixctl_command_reply(conn, 200, "table successfully flushed");
5628 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5629 const char *argv[], void *aux OVS_UNUSED)
5631 struct ds ds = DS_EMPTY_INITIALIZER;
5632 const struct ofproto_dpif *ofproto;
5633 const struct mac_entry *e;
5635 ofproto = ofproto_dpif_lookup(argv[1]);
5637 unixctl_command_reply(conn, 501, "no such bridge");
5641 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5642 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5643 struct ofbundle *bundle = e->port.p;
5644 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
5645 ofbundle_get_a_port(bundle)->odp_port,
5646 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
5648 unixctl_command_reply(conn, 200, ds_cstr(&ds));
5652 struct ofproto_trace {
5653 struct action_xlate_ctx ctx;
5659 trace_format_rule(struct ds *result, uint8_t table_id, int level,
5660 const struct rule_dpif *rule)
5662 ds_put_char_multiple(result, '\t', level);
5664 ds_put_cstr(result, "No match\n");
5668 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5669 table_id, ntohll(rule->up.flow_cookie));
5670 cls_rule_format(&rule->up.cr, result);
5671 ds_put_char(result, '\n');
5673 ds_put_char_multiple(result, '\t', level);
5674 ds_put_cstr(result, "OpenFlow ");
5675 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
5676 ds_put_char(result, '\n');
5680 trace_format_flow(struct ds *result, int level, const char *title,
5681 struct ofproto_trace *trace)
5683 ds_put_char_multiple(result, '\t', level);
5684 ds_put_format(result, "%s: ", title);
5685 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
5686 ds_put_cstr(result, "unchanged");
5688 flow_format(result, &trace->ctx.flow);
5689 trace->flow = trace->ctx.flow;
5691 ds_put_char(result, '\n');
5695 trace_format_regs(struct ds *result, int level, const char *title,
5696 struct ofproto_trace *trace)
5700 ds_put_char_multiple(result, '\t', level);
5701 ds_put_format(result, "%s:", title);
5702 for (i = 0; i < FLOW_N_REGS; i++) {
5703 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5705 ds_put_char(result, '\n');
5709 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
5711 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
5712 struct ds *result = trace->result;
5714 ds_put_char(result, '\n');
5715 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
5716 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
5717 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
5721 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5722 void *aux OVS_UNUSED)
5724 const char *dpname = argv[1];
5725 struct ofproto_dpif *ofproto;
5726 struct ofpbuf odp_key;
5727 struct ofpbuf *packet;
5728 struct rule_dpif *rule;
5729 ovs_be16 initial_tci;
5735 ofpbuf_init(&odp_key, 0);
5738 ofproto = ofproto_dpif_lookup(dpname);
5740 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
5744 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
5745 /* ofproto/trace dpname flow [-generate] */
5746 const char *flow_s = argv[2];
5747 const char *generate_s = argv[3];
5750 /* Convert string to datapath key. */
5751 ofpbuf_init(&odp_key, 0);
5752 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
5754 unixctl_command_reply(conn, 501, "Bad flow syntax");
5758 /* Convert odp_key to flow. */
5759 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
5760 odp_key.size, &flow,
5761 &initial_tci, NULL);
5762 if (error == ODP_FIT_ERROR) {
5763 unixctl_command_reply(conn, 501, "Invalid flow");
5767 /* Generate a packet, if requested. */
5769 packet = ofpbuf_new(0);
5770 flow_compose(packet, &flow);
5772 } else if (argc == 6) {
5773 /* ofproto/trace dpname priority tun_id in_port packet */
5774 const char *priority_s = argv[2];
5775 const char *tun_id_s = argv[3];
5776 const char *in_port_s = argv[4];
5777 const char *packet_s = argv[5];
5778 uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s));
5779 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
5780 uint32_t priority = atoi(priority_s);
5783 msg = eth_from_hex(packet_s, &packet);
5785 unixctl_command_reply(conn, 501, msg);
5789 ds_put_cstr(&result, "Packet: ");
5790 s = ofp_packet_to_string(packet->data, packet->size);
5791 ds_put_cstr(&result, s);
5794 flow_extract(packet, priority, tun_id, in_port, &flow);
5795 initial_tci = flow.vlan_tci;
5797 unixctl_command_reply(conn, 501, "Bad command syntax");
5801 ds_put_cstr(&result, "Flow: ");
5802 flow_format(&result, &flow);
5803 ds_put_char(&result, '\n');
5805 rule = rule_dpif_lookup(ofproto, &flow, 0);
5806 trace_format_rule(&result, 0, 0, rule);
5808 struct ofproto_trace trace;
5809 struct ofpbuf *odp_actions;
5811 trace.result = &result;
5813 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, initial_tci, packet);
5814 trace.ctx.resubmit_hook = trace_resubmit;
5815 odp_actions = xlate_actions(&trace.ctx,
5816 rule->up.actions, rule->up.n_actions);
5818 ds_put_char(&result, '\n');
5819 trace_format_flow(&result, 0, "Final flow", &trace);
5820 ds_put_cstr(&result, "Datapath actions: ");
5821 format_odp_actions(&result, odp_actions->data, odp_actions->size);
5822 ofpbuf_delete(odp_actions);
5824 if (!trace.ctx.may_set_up_flow) {
5826 ds_put_cstr(&result, "\nThis flow is not cachable.");
5828 ds_put_cstr(&result, "\nThe datapath actions are incomplete--"
5829 "for complete actions, please supply a packet.");
5834 unixctl_command_reply(conn, 200, ds_cstr(&result));
5837 ds_destroy(&result);
5838 ofpbuf_delete(packet);
5839 ofpbuf_uninit(&odp_key);
5843 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5844 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5847 unixctl_command_reply(conn, 200, NULL);
5851 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5852 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5855 unixctl_command_reply(conn, 200, NULL);
5859 ofproto_dpif_unixctl_init(void)
5861 static bool registered;
5867 unixctl_command_register(
5869 "bridge {tun_id in_port packet | odp_flow [-generate]}",
5870 2, 4, ofproto_unixctl_trace, NULL);
5871 unixctl_command_register("fdb/flush", "bridge", 1, 1,
5872 ofproto_unixctl_fdb_flush, NULL);
5873 unixctl_command_register("fdb/show", "bridge", 1, 1,
5874 ofproto_unixctl_fdb_show, NULL);
5875 unixctl_command_register("ofproto/clog", "", 0, 0,
5876 ofproto_dpif_clog, NULL);
5877 unixctl_command_register("ofproto/unclog", "", 0, 0,
5878 ofproto_dpif_unclog, NULL);
5881 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5883 * This is deprecated. It is only for compatibility with broken device drivers
5884 * in old versions of Linux that do not properly support VLANs when VLAN
5885 * devices are not used. When broken device drivers are no longer in
5886 * widespread use, we will delete these interfaces. */
5889 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
5891 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
5892 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
5894 if (realdev_ofp_port == ofport->realdev_ofp_port
5895 && vid == ofport->vlandev_vid) {
5899 ofproto->need_revalidate = true;
5901 if (ofport->realdev_ofp_port) {
5904 if (realdev_ofp_port && ofport->bundle) {
5905 /* vlandevs are enslaved to their realdevs, so they are not allowed to
5906 * themselves be part of a bundle. */
5907 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
5910 ofport->realdev_ofp_port = realdev_ofp_port;
5911 ofport->vlandev_vid = vid;
5913 if (realdev_ofp_port) {
5914 vsp_add(ofport, realdev_ofp_port, vid);
5921 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
5923 return hash_2words(realdev_ofp_port, vid);
5927 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
5928 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
5930 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
5931 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
5932 int vid = vlan_tci_to_vid(vlan_tci);
5933 const struct vlan_splinter *vsp;
5935 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
5936 hash_realdev_vid(realdev_ofp_port, vid),
5937 &ofproto->realdev_vid_map) {
5938 if (vsp->realdev_ofp_port == realdev_ofp_port
5939 && vsp->vid == vid) {
5940 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
5944 return realdev_odp_port;
5947 static struct vlan_splinter *
5948 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
5950 struct vlan_splinter *vsp;
5952 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
5953 &ofproto->vlandev_map) {
5954 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
5963 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
5964 uint16_t vlandev_ofp_port, int *vid)
5966 if (!hmap_is_empty(&ofproto->vlandev_map)) {
5967 const struct vlan_splinter *vsp;
5969 vsp = vlandev_find(ofproto, vlandev_ofp_port);
5974 return vsp->realdev_ofp_port;
5981 vsp_remove(struct ofport_dpif *port)
5983 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
5984 struct vlan_splinter *vsp;
5986 vsp = vlandev_find(ofproto, port->up.ofp_port);
5988 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
5989 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
5992 port->realdev_ofp_port = 0;
5994 VLOG_ERR("missing vlan device record");
5999 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
6001 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6003 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6004 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6005 == realdev_ofp_port)) {
6006 struct vlan_splinter *vsp;
6008 vsp = xmalloc(sizeof *vsp);
6009 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6010 hash_int(port->up.ofp_port, 0));
6011 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6012 hash_realdev_vid(realdev_ofp_port, vid));
6013 vsp->realdev_ofp_port = realdev_ofp_port;
6014 vsp->vlandev_ofp_port = port->up.ofp_port;
6017 port->realdev_ofp_port = realdev_ofp_port;
6019 VLOG_ERR("duplicate vlan device record");
6023 const struct ofproto_class ofproto_dpif_class = {
6052 port_is_lacp_current,
6053 NULL, /* rule_choose_table */
6060 rule_modify_actions,
6068 get_cfm_remote_mpids,
6072 get_stp_port_status,
6079 is_mirror_output_bundle,
6080 forward_bpdu_changed,