2 * Copyright (c) 2009, 2010, 2011 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-provider.h"
26 #include "byte-order.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
36 #include "mac-learning.h"
37 #include "multipath.h"
44 #include "ofp-print.h"
45 #include "ofproto-dpif-sflow.h"
46 #include "poll-loop.h"
48 #include "unaligned.h"
50 #include "vlan-bitmap.h"
53 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
55 COVERAGE_DEFINE(ofproto_dpif_ctlr_action);
56 COVERAGE_DEFINE(ofproto_dpif_expired);
57 COVERAGE_DEFINE(ofproto_dpif_no_packet_in);
58 COVERAGE_DEFINE(ofproto_dpif_xlate);
59 COVERAGE_DEFINE(facet_changed_rule);
60 COVERAGE_DEFINE(facet_invalidated);
61 COVERAGE_DEFINE(facet_revalidate);
62 COVERAGE_DEFINE(facet_unexpected);
64 /* Maximum depth of flow table recursion (due to resubmit actions) in a
65 * flow translation. */
66 #define MAX_RESUBMIT_RECURSION 32
68 /* Number of implemented OpenFlow tables. */
69 enum { N_TABLES = 255 };
70 BUILD_ASSERT_DECL(N_TABLES >= 1 && N_TABLES <= 255);
78 long long int used; /* Time last used; time created if not used. */
82 * - Do include packets and bytes from facets that have been deleted or
83 * whose own statistics have been folded into the rule.
85 * - Do include packets and bytes sent "by hand" that were accounted to
86 * the rule without any facet being involved (this is a rare corner
87 * case in rule_execute()).
89 * - Do not include packet or bytes that can be obtained from any facet's
90 * packet_count or byte_count member or that can be obtained from the
91 * datapath by, e.g., dpif_flow_get() for any subfacet.
93 uint64_t packet_count; /* Number of packets received. */
94 uint64_t byte_count; /* Number of bytes received. */
96 tag_type tag; /* Caches rule_calculate_tag() result. */
98 struct list facets; /* List of "struct facet"s. */
101 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
103 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
106 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
107 const struct flow *, uint8_t table);
109 static void flow_push_stats(const struct rule_dpif *, const struct flow *,
110 uint64_t packets, uint64_t bytes,
113 static uint32_t rule_calculate_tag(const struct flow *,
114 const struct flow_wildcards *,
116 static void rule_invalidate(const struct rule_dpif *);
118 #define MAX_MIRRORS 32
119 typedef uint32_t mirror_mask_t;
120 #define MIRROR_MASK_C(X) UINT32_C(X)
121 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
123 struct ofproto_dpif *ofproto; /* Owning ofproto. */
124 size_t idx; /* In ofproto's "mirrors" array. */
125 void *aux; /* Key supplied by ofproto's client. */
126 char *name; /* Identifier for log messages. */
128 /* Selection criteria. */
129 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
130 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
131 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
133 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
134 struct ofbundle *out; /* Output port or NULL. */
135 int out_vlan; /* Output VLAN or -1. */
136 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
139 int64_t packet_count; /* Number of packets sent. */
140 int64_t byte_count; /* Number of bytes sent. */
143 static void mirror_destroy(struct ofmirror *);
144 static void update_mirror_stats(struct ofproto_dpif *ofproto,
145 mirror_mask_t mirrors,
146 uint64_t packets, uint64_t bytes);
149 struct ofproto_dpif *ofproto; /* Owning ofproto. */
150 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
151 void *aux; /* Key supplied by ofproto's client. */
152 char *name; /* Identifier for log messages. */
155 struct list ports; /* Contains "struct ofport"s. */
156 enum port_vlan_mode vlan_mode; /* VLAN mode */
157 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
158 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
159 * NULL if all VLANs are trunked. */
160 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
161 struct bond *bond; /* Nonnull iff more than one port. */
162 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
165 bool floodable; /* True if no port has OFPPC_NO_FLOOD set. */
167 /* Port mirroring info. */
168 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
169 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
170 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
173 static void bundle_remove(struct ofport *);
174 static void bundle_update(struct ofbundle *);
175 static void bundle_destroy(struct ofbundle *);
176 static void bundle_del_port(struct ofport_dpif *);
177 static void bundle_run(struct ofbundle *);
178 static void bundle_wait(struct ofbundle *);
179 static struct ofbundle *lookup_input_bundle(struct ofproto_dpif *,
180 uint16_t in_port, bool warn);
182 static void stp_run(struct ofproto_dpif *ofproto);
183 static void stp_wait(struct ofproto_dpif *ofproto);
185 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
187 struct action_xlate_ctx {
188 /* action_xlate_ctx_init() initializes these members. */
191 struct ofproto_dpif *ofproto;
193 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
194 * this flow when actions change header fields. */
197 /* The packet corresponding to 'flow', or a null pointer if we are
198 * revalidating without a packet to refer to. */
199 const struct ofpbuf *packet;
201 /* Should OFPP_NORMAL MAC learning and NXAST_LEARN actions execute? We
202 * want to execute them if we are actually processing a packet, or if we
203 * are accounting for packets that the datapath has processed, but not if
204 * we are just revalidating. */
207 /* If nonnull, called just before executing a resubmit action.
209 * This is normally null so the client has to set it manually after
210 * calling action_xlate_ctx_init(). */
211 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *);
213 /* xlate_actions() initializes and uses these members. The client might want
214 * to look at them after it returns. */
216 struct ofpbuf *odp_actions; /* Datapath actions. */
217 tag_type tags; /* Tags associated with actions. */
218 bool may_set_up_flow; /* True ordinarily; false if the actions must
219 * be reassessed for every packet. */
220 bool has_learn; /* Actions include NXAST_LEARN? */
221 bool has_normal; /* Actions output to OFPP_NORMAL? */
222 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
223 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
225 /* xlate_actions() initializes and uses these members, but the client has no
226 * reason to look at them. */
228 int recurse; /* Recursion level, via xlate_table_action. */
229 struct flow base_flow; /* Flow at the last commit. */
230 uint32_t original_priority; /* Priority when packet arrived. */
231 uint8_t table_id; /* OpenFlow table ID where flow was found. */
232 uint32_t sflow_n_outputs; /* Number of output ports. */
233 uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
234 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
235 bool exit; /* No further actions should be processed. */
238 static void action_xlate_ctx_init(struct action_xlate_ctx *,
239 struct ofproto_dpif *, const struct flow *,
240 ovs_be16 initial_tci, const struct ofpbuf *);
241 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
242 const union ofp_action *in, size_t n_in);
244 /* An exact-match instantiation of an OpenFlow flow.
246 * A facet associates a "struct flow", which represents the Open vSwitch
247 * userspace idea of an exact-match flow, with one or more subfacets. Each
248 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
249 * the facet. When the kernel module (or other dpif implementation) and Open
250 * vSwitch userspace agree on the definition of a flow key, there is exactly
251 * one subfacet per facet. If the dpif implementation supports more-specific
252 * flow matching than userspace, however, a facet can have more than one
253 * subfacet, each of which corresponds to some distinction in flow that
254 * userspace simply doesn't understand.
256 * Flow expiration works in terms of subfacets, so a facet must have at least
257 * one subfacet or it will never expire, leaking memory. */
260 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
261 struct list list_node; /* In owning rule's 'facets' list. */
262 struct rule_dpif *rule; /* Owning rule. */
265 struct list subfacets;
266 long long int used; /* Time last used; time created if not used. */
273 * - Do include packets and bytes sent "by hand", e.g. with
276 * - Do include packets and bytes that were obtained from the datapath
277 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
278 * DPIF_FP_ZERO_STATS).
280 * - Do not include packets or bytes that can be obtained from the
281 * datapath for any existing subfacet.
283 uint64_t packet_count; /* Number of packets received. */
284 uint64_t byte_count; /* Number of bytes received. */
286 /* Resubmit statistics. */
287 uint64_t prev_packet_count; /* Number of packets from last stats push. */
288 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
289 long long int prev_used; /* Used time from last stats push. */
292 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
293 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
295 /* Properties of datapath actions.
297 * Every subfacet has its own actions because actions can differ slightly
298 * between splintered and non-splintered subfacets due to the VLAN tag
299 * being initially different (present vs. absent). All of them have these
300 * properties in common so we just store one copy of them here. */
301 bool may_install; /* Reassess actions for every packet? */
302 bool has_learn; /* Actions include NXAST_LEARN? */
303 bool has_normal; /* Actions output to OFPP_NORMAL? */
304 tag_type tags; /* Tags that would require revalidation. */
305 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
308 static struct facet *facet_create(struct rule_dpif *, const struct flow *);
309 static void facet_remove(struct ofproto_dpif *, struct facet *);
310 static void facet_free(struct facet *);
312 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
313 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
314 const struct flow *);
315 static bool facet_revalidate(struct ofproto_dpif *, struct facet *);
317 static bool execute_controller_action(struct ofproto_dpif *,
319 const struct nlattr *odp_actions,
321 struct ofpbuf *packet, bool clone);
323 static void facet_flush_stats(struct ofproto_dpif *, struct facet *);
325 static void facet_update_time(struct ofproto_dpif *, struct facet *,
327 static void facet_reset_counters(struct facet *);
328 static void facet_push_stats(struct facet *);
329 static void facet_account(struct ofproto_dpif *, struct facet *);
331 static bool facet_is_controller_flow(struct facet *);
333 /* A dpif flow and actions associated with a facet.
335 * See also the large comment on struct facet. */
338 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
339 struct list list_node; /* In struct facet's 'facets' list. */
340 struct facet *facet; /* Owning facet. */
344 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
345 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
346 * regenerate the ODP flow key from ->facet->flow. */
347 enum odp_key_fitness key_fitness;
351 long long int used; /* Time last used; time created if not used. */
353 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
354 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
358 * These should be essentially identical for every subfacet in a facet, but
359 * may differ in trivial ways due to VLAN splinters. */
360 size_t actions_len; /* Number of bytes in actions[]. */
361 struct nlattr *actions; /* Datapath actions. */
363 bool installed; /* Installed in datapath? */
365 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
366 * splinters can cause it to differ. This value should be removed when
367 * the VLAN splinters feature is no longer needed. */
368 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
371 static struct subfacet *subfacet_create(struct ofproto_dpif *, struct facet *,
372 enum odp_key_fitness,
373 const struct nlattr *key,
374 size_t key_len, ovs_be16 initial_tci);
375 static struct subfacet *subfacet_find(struct ofproto_dpif *,
376 const struct nlattr *key, size_t key_len);
377 static void subfacet_destroy(struct ofproto_dpif *, struct subfacet *);
378 static void subfacet_destroy__(struct ofproto_dpif *, struct subfacet *);
379 static void subfacet_reset_dp_stats(struct subfacet *,
380 struct dpif_flow_stats *);
381 static void subfacet_update_time(struct ofproto_dpif *, struct subfacet *,
383 static void subfacet_update_stats(struct ofproto_dpif *, struct subfacet *,
384 const struct dpif_flow_stats *);
385 static void subfacet_make_actions(struct ofproto_dpif *, struct subfacet *,
386 const struct ofpbuf *packet);
387 static int subfacet_install(struct ofproto_dpif *, struct subfacet *,
388 const struct nlattr *actions, size_t actions_len,
389 struct dpif_flow_stats *);
390 static void subfacet_uninstall(struct ofproto_dpif *, struct subfacet *);
396 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
397 struct list bundle_node; /* In struct ofbundle's "ports" list. */
398 struct cfm *cfm; /* Connectivity Fault Management, if any. */
399 tag_type tag; /* Tag associated with this port. */
400 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
401 bool may_enable; /* May be enabled in bonds. */
404 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
405 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
406 long long int stp_state_entered;
408 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
410 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
412 * This is deprecated. It is only for compatibility with broken device
413 * drivers in old versions of Linux that do not properly support VLANs when
414 * VLAN devices are not used. When broken device drivers are no longer in
415 * widespread use, we will delete these interfaces. */
416 uint16_t realdev_ofp_port;
420 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
421 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
422 * traffic egressing the 'ofport' with that priority should be marked with. */
423 struct priority_to_dscp {
424 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
425 uint32_t priority; /* Priority of this queue (see struct flow). */
427 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
430 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
432 * This is deprecated. It is only for compatibility with broken device drivers
433 * in old versions of Linux that do not properly support VLANs when VLAN
434 * devices are not used. When broken device drivers are no longer in
435 * widespread use, we will delete these interfaces. */
436 struct vlan_splinter {
437 struct hmap_node realdev_vid_node;
438 struct hmap_node vlandev_node;
439 uint16_t realdev_ofp_port;
440 uint16_t vlandev_ofp_port;
444 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
445 uint32_t realdev, ovs_be16 vlan_tci);
446 static uint16_t vsp_vlandev_to_realdev(const struct ofproto_dpif *,
447 uint16_t vlandev, int *vid);
448 static void vsp_remove(struct ofport_dpif *);
449 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
451 static struct ofport_dpif *
452 ofport_dpif_cast(const struct ofport *ofport)
454 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
455 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
458 static void port_run(struct ofport_dpif *);
459 static void port_wait(struct ofport_dpif *);
460 static int set_cfm(struct ofport *, const struct cfm_settings *);
461 static void ofport_clear_priorities(struct ofport_dpif *);
463 struct dpif_completion {
464 struct list list_node;
465 struct ofoperation *op;
468 /* Extra information about a classifier table.
469 * Currently used just for optimized flow revalidation. */
471 /* If either of these is nonnull, then this table has a form that allows
472 * flows to be tagged to avoid revalidating most flows for the most common
473 * kinds of flow table changes. */
474 struct cls_table *catchall_table; /* Table that wildcards all fields. */
475 struct cls_table *other_table; /* Table with any other wildcard set. */
476 uint32_t basis; /* Keeps each table's tags separate. */
479 struct ofproto_dpif {
480 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
489 struct netflow *netflow;
490 struct dpif_sflow *sflow;
491 struct hmap bundles; /* Contains "struct ofbundle"s. */
492 struct mac_learning *ml;
493 struct ofmirror *mirrors[MAX_MIRRORS];
494 bool has_bonded_bundles;
497 struct timer next_expiration;
501 struct hmap subfacets;
504 struct table_dpif tables[N_TABLES];
505 bool need_revalidate;
506 struct tag_set revalidate_set;
508 /* Support for debugging async flow mods. */
509 struct list completions;
511 bool has_bundle_action; /* True when the first bundle action appears. */
515 long long int stp_last_tick;
517 /* VLAN splinters. */
518 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
519 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
522 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
523 * for debugging the asynchronous flow_mod implementation.) */
526 /* All existing ofproto_dpif instances, indexed by ->up.name. */
527 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
529 static void ofproto_dpif_unixctl_init(void);
531 static struct ofproto_dpif *
532 ofproto_dpif_cast(const struct ofproto *ofproto)
534 assert(ofproto->ofproto_class == &ofproto_dpif_class);
535 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
538 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
540 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
543 /* Packet processing. */
544 static void update_learning_table(struct ofproto_dpif *,
545 const struct flow *, int vlan,
548 #define FLOW_MISS_MAX_BATCH 50
549 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
551 /* Flow expiration. */
552 static int expire(struct ofproto_dpif *);
555 static void send_netflow_active_timeouts(struct ofproto_dpif *);
558 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
560 compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions,
561 const struct flow *, uint32_t odp_port);
562 static void add_mirror_actions(struct action_xlate_ctx *ctx,
563 const struct flow *flow);
564 /* Global variables. */
565 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
567 /* Factory functions. */
570 enumerate_types(struct sset *types)
572 dp_enumerate_types(types);
576 enumerate_names(const char *type, struct sset *names)
578 return dp_enumerate_names(type, names);
582 del(const char *type, const char *name)
587 error = dpif_open(name, type, &dpif);
589 error = dpif_delete(dpif);
595 /* Basic life-cycle. */
597 static struct ofproto *
600 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
605 dealloc(struct ofproto *ofproto_)
607 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
612 construct(struct ofproto *ofproto_, int *n_tablesp)
614 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
615 const char *name = ofproto->up.name;
619 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
621 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
625 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
626 ofproto->n_matches = 0;
628 dpif_flow_flush(ofproto->dpif);
629 dpif_recv_purge(ofproto->dpif);
631 error = dpif_recv_set_mask(ofproto->dpif,
632 ((1u << DPIF_UC_MISS) |
633 (1u << DPIF_UC_ACTION)));
635 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
636 dpif_close(ofproto->dpif);
640 ofproto->netflow = NULL;
641 ofproto->sflow = NULL;
643 hmap_init(&ofproto->bundles);
644 ofproto->ml = mac_learning_create();
645 for (i = 0; i < MAX_MIRRORS; i++) {
646 ofproto->mirrors[i] = NULL;
648 ofproto->has_bonded_bundles = false;
650 timer_set_duration(&ofproto->next_expiration, 1000);
652 hmap_init(&ofproto->facets);
653 hmap_init(&ofproto->subfacets);
655 for (i = 0; i < N_TABLES; i++) {
656 struct table_dpif *table = &ofproto->tables[i];
658 table->catchall_table = NULL;
659 table->other_table = NULL;
660 table->basis = random_uint32();
662 ofproto->need_revalidate = false;
663 tag_set_init(&ofproto->revalidate_set);
665 list_init(&ofproto->completions);
667 ofproto_dpif_unixctl_init();
669 ofproto->has_bundle_action = false;
671 hmap_init(&ofproto->vlandev_map);
672 hmap_init(&ofproto->realdev_vid_map);
674 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
675 hash_string(ofproto->up.name, 0));
677 *n_tablesp = N_TABLES;
682 complete_operations(struct ofproto_dpif *ofproto)
684 struct dpif_completion *c, *next;
686 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
687 ofoperation_complete(c->op, 0);
688 list_remove(&c->list_node);
694 destruct(struct ofproto *ofproto_)
696 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
697 struct rule_dpif *rule, *next_rule;
698 struct classifier *table;
701 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
702 complete_operations(ofproto);
704 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
705 struct cls_cursor cursor;
707 cls_cursor_init(&cursor, table, NULL);
708 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
709 ofproto_rule_destroy(&rule->up);
713 for (i = 0; i < MAX_MIRRORS; i++) {
714 mirror_destroy(ofproto->mirrors[i]);
717 netflow_destroy(ofproto->netflow);
718 dpif_sflow_destroy(ofproto->sflow);
719 hmap_destroy(&ofproto->bundles);
720 mac_learning_destroy(ofproto->ml);
722 hmap_destroy(&ofproto->facets);
723 hmap_destroy(&ofproto->subfacets);
725 hmap_destroy(&ofproto->vlandev_map);
726 hmap_destroy(&ofproto->realdev_vid_map);
728 dpif_close(ofproto->dpif);
732 run_fast(struct ofproto *ofproto_)
734 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
737 /* Handle one or more batches of upcalls, until there's nothing left to do
738 * or until we do a fixed total amount of work.
740 * We do work in batches because it can be much cheaper to set up a number
741 * of flows and fire off their patches all at once. We do multiple batches
742 * because in some cases handling a packet can cause another packet to be
743 * queued almost immediately as part of the return flow. Both
744 * optimizations can make major improvements on some benchmarks and
745 * presumably for real traffic as well. */
747 while (work < FLOW_MISS_MAX_BATCH) {
748 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
758 run(struct ofproto *ofproto_)
760 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
761 struct ofport_dpif *ofport;
762 struct ofbundle *bundle;
766 complete_operations(ofproto);
768 dpif_run(ofproto->dpif);
770 error = run_fast(ofproto_);
775 if (timer_expired(&ofproto->next_expiration)) {
776 int delay = expire(ofproto);
777 timer_set_duration(&ofproto->next_expiration, delay);
780 if (ofproto->netflow) {
781 if (netflow_run(ofproto->netflow)) {
782 send_netflow_active_timeouts(ofproto);
785 if (ofproto->sflow) {
786 dpif_sflow_run(ofproto->sflow);
789 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
792 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
797 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
799 /* Now revalidate if there's anything to do. */
800 if (ofproto->need_revalidate
801 || !tag_set_is_empty(&ofproto->revalidate_set)) {
802 struct tag_set revalidate_set = ofproto->revalidate_set;
803 bool revalidate_all = ofproto->need_revalidate;
804 struct facet *facet, *next;
806 /* Clear the revalidation flags. */
807 tag_set_init(&ofproto->revalidate_set);
808 ofproto->need_revalidate = false;
810 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
812 || tag_set_intersects(&revalidate_set, facet->tags)) {
813 facet_revalidate(ofproto, facet);
822 wait(struct ofproto *ofproto_)
824 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
825 struct ofport_dpif *ofport;
826 struct ofbundle *bundle;
828 if (!clogged && !list_is_empty(&ofproto->completions)) {
829 poll_immediate_wake();
832 dpif_wait(ofproto->dpif);
833 dpif_recv_wait(ofproto->dpif);
834 if (ofproto->sflow) {
835 dpif_sflow_wait(ofproto->sflow);
837 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
838 poll_immediate_wake();
840 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
843 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
846 if (ofproto->netflow) {
847 netflow_wait(ofproto->netflow);
849 mac_learning_wait(ofproto->ml);
851 if (ofproto->need_revalidate) {
852 /* Shouldn't happen, but if it does just go around again. */
853 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
854 poll_immediate_wake();
856 timer_wait(&ofproto->next_expiration);
861 flush(struct ofproto *ofproto_)
863 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
864 struct facet *facet, *next_facet;
866 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
867 /* Mark the facet as not installed so that facet_remove() doesn't
868 * bother trying to uninstall it. There is no point in uninstalling it
869 * individually since we are about to blow away all the facets with
870 * dpif_flow_flush(). */
871 struct subfacet *subfacet;
873 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
874 subfacet->installed = false;
875 subfacet->dp_packet_count = 0;
876 subfacet->dp_byte_count = 0;
878 facet_remove(ofproto, facet);
880 dpif_flow_flush(ofproto->dpif);
884 get_features(struct ofproto *ofproto_ OVS_UNUSED,
885 bool *arp_match_ip, uint32_t *actions)
887 *arp_match_ip = true;
888 *actions = ((1u << OFPAT_OUTPUT) |
889 (1u << OFPAT_SET_VLAN_VID) |
890 (1u << OFPAT_SET_VLAN_PCP) |
891 (1u << OFPAT_STRIP_VLAN) |
892 (1u << OFPAT_SET_DL_SRC) |
893 (1u << OFPAT_SET_DL_DST) |
894 (1u << OFPAT_SET_NW_SRC) |
895 (1u << OFPAT_SET_NW_DST) |
896 (1u << OFPAT_SET_NW_TOS) |
897 (1u << OFPAT_SET_TP_SRC) |
898 (1u << OFPAT_SET_TP_DST) |
899 (1u << OFPAT_ENQUEUE));
903 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
905 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
906 struct dpif_dp_stats s;
908 strcpy(ots->name, "classifier");
910 dpif_get_dp_stats(ofproto->dpif, &s);
911 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
912 put_32aligned_be64(&ots->matched_count,
913 htonll(s.n_hit + ofproto->n_matches));
916 static struct ofport *
919 struct ofport_dpif *port = xmalloc(sizeof *port);
924 port_dealloc(struct ofport *port_)
926 struct ofport_dpif *port = ofport_dpif_cast(port_);
931 port_construct(struct ofport *port_)
933 struct ofport_dpif *port = ofport_dpif_cast(port_);
934 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
936 ofproto->need_revalidate = true;
937 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
940 port->tag = tag_create_random();
941 port->may_enable = true;
942 port->stp_port = NULL;
943 port->stp_state = STP_DISABLED;
944 hmap_init(&port->priorities);
945 port->realdev_ofp_port = 0;
946 port->vlandev_vid = 0;
948 if (ofproto->sflow) {
949 dpif_sflow_add_port(ofproto->sflow, port->odp_port,
950 netdev_get_name(port->up.netdev));
957 port_destruct(struct ofport *port_)
959 struct ofport_dpif *port = ofport_dpif_cast(port_);
960 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
962 ofproto->need_revalidate = true;
963 bundle_remove(port_);
964 set_cfm(port_, NULL);
965 if (ofproto->sflow) {
966 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
969 ofport_clear_priorities(port);
970 hmap_destroy(&port->priorities);
974 port_modified(struct ofport *port_)
976 struct ofport_dpif *port = ofport_dpif_cast(port_);
978 if (port->bundle && port->bundle->bond) {
979 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
984 port_reconfigured(struct ofport *port_, ovs_be32 old_config)
986 struct ofport_dpif *port = ofport_dpif_cast(port_);
987 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
988 ovs_be32 changed = old_config ^ port->up.opp.config;
990 if (changed & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
991 OFPPC_NO_FWD | OFPPC_NO_FLOOD)) {
992 ofproto->need_revalidate = true;
994 if (changed & htonl(OFPPC_NO_FLOOD) && port->bundle) {
995 bundle_update(port->bundle);
1001 set_sflow(struct ofproto *ofproto_,
1002 const struct ofproto_sflow_options *sflow_options)
1004 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1005 struct dpif_sflow *ds = ofproto->sflow;
1007 if (sflow_options) {
1009 struct ofport_dpif *ofport;
1011 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1012 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1013 dpif_sflow_add_port(ds, ofport->odp_port,
1014 netdev_get_name(ofport->up.netdev));
1016 ofproto->need_revalidate = true;
1018 dpif_sflow_set_options(ds, sflow_options);
1021 dpif_sflow_destroy(ds);
1022 ofproto->need_revalidate = true;
1023 ofproto->sflow = NULL;
1030 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1032 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1039 struct ofproto_dpif *ofproto;
1041 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1042 ofproto->need_revalidate = true;
1043 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1046 if (cfm_configure(ofport->cfm, s)) {
1052 cfm_destroy(ofport->cfm);
1058 get_cfm_fault(const struct ofport *ofport_)
1060 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1062 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1066 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1069 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1072 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1079 /* Spanning Tree. */
1082 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1084 struct ofproto_dpif *ofproto = ofproto_;
1085 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1086 struct ofport_dpif *ofport;
1088 ofport = stp_port_get_aux(sp);
1090 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1091 ofproto->up.name, port_num);
1093 struct eth_header *eth = pkt->l2;
1095 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1096 if (eth_addr_is_zero(eth->eth_src)) {
1097 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1098 "with unknown MAC", ofproto->up.name, port_num);
1100 send_packet(ofport, pkt);
1106 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1108 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1110 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1112 /* Only revalidate flows if the configuration changed. */
1113 if (!s != !ofproto->stp) {
1114 ofproto->need_revalidate = true;
1118 if (!ofproto->stp) {
1119 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1120 send_bpdu_cb, ofproto);
1121 ofproto->stp_last_tick = time_msec();
1124 stp_set_bridge_id(ofproto->stp, s->system_id);
1125 stp_set_bridge_priority(ofproto->stp, s->priority);
1126 stp_set_hello_time(ofproto->stp, s->hello_time);
1127 stp_set_max_age(ofproto->stp, s->max_age);
1128 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1130 stp_destroy(ofproto->stp);
1131 ofproto->stp = NULL;
1138 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1140 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1144 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1145 s->designated_root = stp_get_designated_root(ofproto->stp);
1146 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1155 update_stp_port_state(struct ofport_dpif *ofport)
1157 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1158 enum stp_state state;
1160 /* Figure out new state. */
1161 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1165 if (ofport->stp_state != state) {
1169 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1170 netdev_get_name(ofport->up.netdev),
1171 stp_state_name(ofport->stp_state),
1172 stp_state_name(state));
1173 if (stp_learn_in_state(ofport->stp_state)
1174 != stp_learn_in_state(state)) {
1175 /* xxx Learning action flows should also be flushed. */
1176 mac_learning_flush(ofproto->ml);
1178 fwd_change = stp_forward_in_state(ofport->stp_state)
1179 != stp_forward_in_state(state);
1181 ofproto->need_revalidate = true;
1182 ofport->stp_state = state;
1183 ofport->stp_state_entered = time_msec();
1185 if (fwd_change && ofport->bundle) {
1186 bundle_update(ofport->bundle);
1189 /* Update the STP state bits in the OpenFlow port description. */
1190 of_state = (ofport->up.opp.state & htonl(~OFPPS_STP_MASK))
1191 | htonl(state == STP_LISTENING ? OFPPS_STP_LISTEN
1192 : state == STP_LEARNING ? OFPPS_STP_LEARN
1193 : state == STP_FORWARDING ? OFPPS_STP_FORWARD
1194 : state == STP_BLOCKING ? OFPPS_STP_BLOCK
1196 ofproto_port_set_state(&ofport->up, of_state);
1200 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1201 * caller is responsible for assigning STP port numbers and ensuring
1202 * there are no duplicates. */
1204 set_stp_port(struct ofport *ofport_,
1205 const struct ofproto_port_stp_settings *s)
1207 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1208 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1209 struct stp_port *sp = ofport->stp_port;
1211 if (!s || !s->enable) {
1213 ofport->stp_port = NULL;
1214 stp_port_disable(sp);
1215 update_stp_port_state(ofport);
1218 } else if (sp && stp_port_no(sp) != s->port_num
1219 && ofport == stp_port_get_aux(sp)) {
1220 /* The port-id changed, so disable the old one if it's not
1221 * already in use by another port. */
1222 stp_port_disable(sp);
1225 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1226 stp_port_enable(sp);
1228 stp_port_set_aux(sp, ofport);
1229 stp_port_set_priority(sp, s->priority);
1230 stp_port_set_path_cost(sp, s->path_cost);
1232 update_stp_port_state(ofport);
1238 get_stp_port_status(struct ofport *ofport_,
1239 struct ofproto_port_stp_status *s)
1241 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1242 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1243 struct stp_port *sp = ofport->stp_port;
1245 if (!ofproto->stp || !sp) {
1251 s->port_id = stp_port_get_id(sp);
1252 s->state = stp_port_get_state(sp);
1253 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1254 s->role = stp_port_get_role(sp);
1255 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1261 stp_run(struct ofproto_dpif *ofproto)
1264 long long int now = time_msec();
1265 long long int elapsed = now - ofproto->stp_last_tick;
1266 struct stp_port *sp;
1269 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1270 ofproto->stp_last_tick = now;
1272 while (stp_get_changed_port(ofproto->stp, &sp)) {
1273 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1276 update_stp_port_state(ofport);
1283 stp_wait(struct ofproto_dpif *ofproto)
1286 poll_timer_wait(1000);
1290 /* Returns true if STP should process 'flow'. */
1292 stp_should_process_flow(const struct flow *flow)
1294 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1298 stp_process_packet(const struct ofport_dpif *ofport,
1299 const struct ofpbuf *packet)
1301 struct ofpbuf payload = *packet;
1302 struct eth_header *eth = payload.data;
1303 struct stp_port *sp = ofport->stp_port;
1305 /* Sink packets on ports that have STP disabled when the bridge has
1307 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1311 /* Trim off padding on payload. */
1312 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1313 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1316 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1317 stp_received_bpdu(sp, payload.data, payload.size);
1321 static struct priority_to_dscp *
1322 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1324 struct priority_to_dscp *pdscp;
1327 hash = hash_int(priority, 0);
1328 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1329 if (pdscp->priority == priority) {
1337 ofport_clear_priorities(struct ofport_dpif *ofport)
1339 struct priority_to_dscp *pdscp, *next;
1341 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1342 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1348 set_queues(struct ofport *ofport_,
1349 const struct ofproto_port_queue *qdscp_list,
1352 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1353 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1354 struct hmap new = HMAP_INITIALIZER(&new);
1357 for (i = 0; i < n_qdscp; i++) {
1358 struct priority_to_dscp *pdscp;
1362 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1363 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1368 pdscp = get_priority(ofport, priority);
1370 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1372 pdscp = xmalloc(sizeof *pdscp);
1373 pdscp->priority = priority;
1375 ofproto->need_revalidate = true;
1378 if (pdscp->dscp != dscp) {
1380 ofproto->need_revalidate = true;
1383 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1386 if (!hmap_is_empty(&ofport->priorities)) {
1387 ofport_clear_priorities(ofport);
1388 ofproto->need_revalidate = true;
1391 hmap_swap(&new, &ofport->priorities);
1399 /* Expires all MAC learning entries associated with 'bundle' and forces its
1400 * ofproto to revalidate every flow.
1402 * Normally MAC learning entries are removed only from the ofproto associated
1403 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1404 * are removed from every ofproto. When patch ports and SLB bonds are in use
1405 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1406 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1407 * with the host from which it migrated. */
1409 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1411 struct ofproto_dpif *ofproto = bundle->ofproto;
1412 struct mac_learning *ml = ofproto->ml;
1413 struct mac_entry *mac, *next_mac;
1415 ofproto->need_revalidate = true;
1416 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1417 if (mac->port.p == bundle) {
1419 struct ofproto_dpif *o;
1421 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1423 struct mac_entry *e;
1425 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1428 tag_set_add(&o->revalidate_set, e->tag);
1429 mac_learning_expire(o->ml, e);
1435 mac_learning_expire(ml, mac);
1440 static struct ofbundle *
1441 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1443 struct ofbundle *bundle;
1445 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1446 &ofproto->bundles) {
1447 if (bundle->aux == aux) {
1454 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1455 * ones that are found to 'bundles'. */
1457 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1458 void **auxes, size_t n_auxes,
1459 struct hmapx *bundles)
1463 hmapx_init(bundles);
1464 for (i = 0; i < n_auxes; i++) {
1465 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1467 hmapx_add(bundles, bundle);
1473 bundle_update(struct ofbundle *bundle)
1475 struct ofport_dpif *port;
1477 bundle->floodable = true;
1478 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1479 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
1480 bundle->floodable = false;
1487 bundle_del_port(struct ofport_dpif *port)
1489 struct ofbundle *bundle = port->bundle;
1491 bundle->ofproto->need_revalidate = true;
1493 list_remove(&port->bundle_node);
1494 port->bundle = NULL;
1497 lacp_slave_unregister(bundle->lacp, port);
1500 bond_slave_unregister(bundle->bond, port);
1503 bundle_update(bundle);
1507 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1508 struct lacp_slave_settings *lacp,
1509 uint32_t bond_stable_id)
1511 struct ofport_dpif *port;
1513 port = get_ofp_port(bundle->ofproto, ofp_port);
1518 if (port->bundle != bundle) {
1519 bundle->ofproto->need_revalidate = true;
1521 bundle_del_port(port);
1524 port->bundle = bundle;
1525 list_push_back(&bundle->ports, &port->bundle_node);
1526 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
1527 bundle->floodable = false;
1531 port->bundle->ofproto->need_revalidate = true;
1532 lacp_slave_register(bundle->lacp, port, lacp);
1535 port->bond_stable_id = bond_stable_id;
1541 bundle_destroy(struct ofbundle *bundle)
1543 struct ofproto_dpif *ofproto;
1544 struct ofport_dpif *port, *next_port;
1551 ofproto = bundle->ofproto;
1552 for (i = 0; i < MAX_MIRRORS; i++) {
1553 struct ofmirror *m = ofproto->mirrors[i];
1555 if (m->out == bundle) {
1557 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1558 || hmapx_find_and_delete(&m->dsts, bundle)) {
1559 ofproto->need_revalidate = true;
1564 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1565 bundle_del_port(port);
1568 bundle_flush_macs(bundle, true);
1569 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1571 free(bundle->trunks);
1572 lacp_destroy(bundle->lacp);
1573 bond_destroy(bundle->bond);
1578 bundle_set(struct ofproto *ofproto_, void *aux,
1579 const struct ofproto_bundle_settings *s)
1581 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1582 bool need_flush = false;
1583 struct ofport_dpif *port;
1584 struct ofbundle *bundle;
1585 unsigned long *trunks;
1591 bundle_destroy(bundle_lookup(ofproto, aux));
1595 assert(s->n_slaves == 1 || s->bond != NULL);
1596 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1598 bundle = bundle_lookup(ofproto, aux);
1600 bundle = xmalloc(sizeof *bundle);
1602 bundle->ofproto = ofproto;
1603 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1604 hash_pointer(aux, 0));
1606 bundle->name = NULL;
1608 list_init(&bundle->ports);
1609 bundle->vlan_mode = PORT_VLAN_TRUNK;
1611 bundle->trunks = NULL;
1612 bundle->use_priority_tags = s->use_priority_tags;
1613 bundle->lacp = NULL;
1614 bundle->bond = NULL;
1616 bundle->floodable = true;
1618 bundle->src_mirrors = 0;
1619 bundle->dst_mirrors = 0;
1620 bundle->mirror_out = 0;
1623 if (!bundle->name || strcmp(s->name, bundle->name)) {
1625 bundle->name = xstrdup(s->name);
1630 if (!bundle->lacp) {
1631 ofproto->need_revalidate = true;
1632 bundle->lacp = lacp_create();
1634 lacp_configure(bundle->lacp, s->lacp);
1636 lacp_destroy(bundle->lacp);
1637 bundle->lacp = NULL;
1640 /* Update set of ports. */
1642 for (i = 0; i < s->n_slaves; i++) {
1643 if (!bundle_add_port(bundle, s->slaves[i],
1644 s->lacp ? &s->lacp_slaves[i] : NULL,
1645 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1649 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1650 struct ofport_dpif *next_port;
1652 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1653 for (i = 0; i < s->n_slaves; i++) {
1654 if (s->slaves[i] == port->up.ofp_port) {
1659 bundle_del_port(port);
1663 assert(list_size(&bundle->ports) <= s->n_slaves);
1665 if (list_is_empty(&bundle->ports)) {
1666 bundle_destroy(bundle);
1670 /* Set VLAN tagging mode */
1671 if (s->vlan_mode != bundle->vlan_mode
1672 || s->use_priority_tags != bundle->use_priority_tags) {
1673 bundle->vlan_mode = s->vlan_mode;
1674 bundle->use_priority_tags = s->use_priority_tags;
1679 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1680 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1682 if (vlan != bundle->vlan) {
1683 bundle->vlan = vlan;
1687 /* Get trunked VLANs. */
1688 switch (s->vlan_mode) {
1689 case PORT_VLAN_ACCESS:
1693 case PORT_VLAN_TRUNK:
1694 trunks = (unsigned long *) s->trunks;
1697 case PORT_VLAN_NATIVE_UNTAGGED:
1698 case PORT_VLAN_NATIVE_TAGGED:
1699 if (vlan != 0 && (!s->trunks
1700 || !bitmap_is_set(s->trunks, vlan)
1701 || bitmap_is_set(s->trunks, 0))) {
1702 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1704 trunks = bitmap_clone(s->trunks, 4096);
1706 trunks = bitmap_allocate1(4096);
1708 bitmap_set1(trunks, vlan);
1709 bitmap_set0(trunks, 0);
1711 trunks = (unsigned long *) s->trunks;
1718 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1719 free(bundle->trunks);
1720 if (trunks == s->trunks) {
1721 bundle->trunks = vlan_bitmap_clone(trunks);
1723 bundle->trunks = trunks;
1728 if (trunks != s->trunks) {
1733 if (!list_is_short(&bundle->ports)) {
1734 bundle->ofproto->has_bonded_bundles = true;
1736 if (bond_reconfigure(bundle->bond, s->bond)) {
1737 ofproto->need_revalidate = true;
1740 bundle->bond = bond_create(s->bond);
1741 ofproto->need_revalidate = true;
1744 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1745 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1749 bond_destroy(bundle->bond);
1750 bundle->bond = NULL;
1753 /* If we changed something that would affect MAC learning, un-learn
1754 * everything on this port and force flow revalidation. */
1756 bundle_flush_macs(bundle, false);
1763 bundle_remove(struct ofport *port_)
1765 struct ofport_dpif *port = ofport_dpif_cast(port_);
1766 struct ofbundle *bundle = port->bundle;
1769 bundle_del_port(port);
1770 if (list_is_empty(&bundle->ports)) {
1771 bundle_destroy(bundle);
1772 } else if (list_is_short(&bundle->ports)) {
1773 bond_destroy(bundle->bond);
1774 bundle->bond = NULL;
1780 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
1782 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1783 struct ofport_dpif *port = port_;
1784 uint8_t ea[ETH_ADDR_LEN];
1787 error = netdev_get_etheraddr(port->up.netdev, ea);
1789 struct ofpbuf packet;
1792 ofpbuf_init(&packet, 0);
1793 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1795 memcpy(packet_pdu, pdu, pdu_size);
1797 send_packet(port, &packet);
1798 ofpbuf_uninit(&packet);
1800 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1801 "%s (%s)", port->bundle->name,
1802 netdev_get_name(port->up.netdev), strerror(error));
1807 bundle_send_learning_packets(struct ofbundle *bundle)
1809 struct ofproto_dpif *ofproto = bundle->ofproto;
1810 int error, n_packets, n_errors;
1811 struct mac_entry *e;
1813 error = n_packets = n_errors = 0;
1814 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1815 if (e->port.p != bundle) {
1816 struct ofpbuf *learning_packet;
1817 struct ofport_dpif *port;
1820 learning_packet = bond_compose_learning_packet(bundle->bond, e->mac,
1823 ret = send_packet(port, learning_packet);
1824 ofpbuf_delete(learning_packet);
1834 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1835 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1836 "packets, last error was: %s",
1837 bundle->name, n_errors, n_packets, strerror(error));
1839 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1840 bundle->name, n_packets);
1845 bundle_run(struct ofbundle *bundle)
1848 lacp_run(bundle->lacp, send_pdu_cb);
1851 struct ofport_dpif *port;
1853 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1854 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1857 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1858 lacp_negotiated(bundle->lacp));
1859 if (bond_should_send_learning_packets(bundle->bond)) {
1860 bundle_send_learning_packets(bundle);
1866 bundle_wait(struct ofbundle *bundle)
1869 lacp_wait(bundle->lacp);
1872 bond_wait(bundle->bond);
1879 mirror_scan(struct ofproto_dpif *ofproto)
1883 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1884 if (!ofproto->mirrors[idx]) {
1891 static struct ofmirror *
1892 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1896 for (i = 0; i < MAX_MIRRORS; i++) {
1897 struct ofmirror *mirror = ofproto->mirrors[i];
1898 if (mirror && mirror->aux == aux) {
1906 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
1908 mirror_update_dups(struct ofproto_dpif *ofproto)
1912 for (i = 0; i < MAX_MIRRORS; i++) {
1913 struct ofmirror *m = ofproto->mirrors[i];
1916 m->dup_mirrors = MIRROR_MASK_C(1) << i;
1920 for (i = 0; i < MAX_MIRRORS; i++) {
1921 struct ofmirror *m1 = ofproto->mirrors[i];
1928 for (j = i + 1; j < MAX_MIRRORS; j++) {
1929 struct ofmirror *m2 = ofproto->mirrors[j];
1931 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
1932 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
1933 m2->dup_mirrors |= m1->dup_mirrors;
1940 mirror_set(struct ofproto *ofproto_, void *aux,
1941 const struct ofproto_mirror_settings *s)
1943 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1944 mirror_mask_t mirror_bit;
1945 struct ofbundle *bundle;
1946 struct ofmirror *mirror;
1947 struct ofbundle *out;
1948 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
1949 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
1952 mirror = mirror_lookup(ofproto, aux);
1954 mirror_destroy(mirror);
1960 idx = mirror_scan(ofproto);
1962 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
1964 ofproto->up.name, MAX_MIRRORS, s->name);
1968 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
1969 mirror->ofproto = ofproto;
1972 mirror->out_vlan = -1;
1973 mirror->name = NULL;
1976 if (!mirror->name || strcmp(s->name, mirror->name)) {
1978 mirror->name = xstrdup(s->name);
1981 /* Get the new configuration. */
1982 if (s->out_bundle) {
1983 out = bundle_lookup(ofproto, s->out_bundle);
1985 mirror_destroy(mirror);
1991 out_vlan = s->out_vlan;
1993 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
1994 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
1996 /* If the configuration has not changed, do nothing. */
1997 if (hmapx_equals(&srcs, &mirror->srcs)
1998 && hmapx_equals(&dsts, &mirror->dsts)
1999 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2000 && mirror->out == out
2001 && mirror->out_vlan == out_vlan)
2003 hmapx_destroy(&srcs);
2004 hmapx_destroy(&dsts);
2008 hmapx_swap(&srcs, &mirror->srcs);
2009 hmapx_destroy(&srcs);
2011 hmapx_swap(&dsts, &mirror->dsts);
2012 hmapx_destroy(&dsts);
2014 free(mirror->vlans);
2015 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2018 mirror->out_vlan = out_vlan;
2020 /* Update bundles. */
2021 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2022 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2023 if (hmapx_contains(&mirror->srcs, bundle)) {
2024 bundle->src_mirrors |= mirror_bit;
2026 bundle->src_mirrors &= ~mirror_bit;
2029 if (hmapx_contains(&mirror->dsts, bundle)) {
2030 bundle->dst_mirrors |= mirror_bit;
2032 bundle->dst_mirrors &= ~mirror_bit;
2035 if (mirror->out == bundle) {
2036 bundle->mirror_out |= mirror_bit;
2038 bundle->mirror_out &= ~mirror_bit;
2042 ofproto->need_revalidate = true;
2043 mac_learning_flush(ofproto->ml);
2044 mirror_update_dups(ofproto);
2050 mirror_destroy(struct ofmirror *mirror)
2052 struct ofproto_dpif *ofproto;
2053 mirror_mask_t mirror_bit;
2054 struct ofbundle *bundle;
2060 ofproto = mirror->ofproto;
2061 ofproto->need_revalidate = true;
2062 mac_learning_flush(ofproto->ml);
2064 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2065 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2066 bundle->src_mirrors &= ~mirror_bit;
2067 bundle->dst_mirrors &= ~mirror_bit;
2068 bundle->mirror_out &= ~mirror_bit;
2071 hmapx_destroy(&mirror->srcs);
2072 hmapx_destroy(&mirror->dsts);
2073 free(mirror->vlans);
2075 ofproto->mirrors[mirror->idx] = NULL;
2079 mirror_update_dups(ofproto);
2083 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2084 uint64_t *packets, uint64_t *bytes)
2086 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2087 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2090 *packets = *bytes = UINT64_MAX;
2094 *packets = mirror->packet_count;
2095 *bytes = mirror->byte_count;
2101 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2103 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2104 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2105 ofproto->need_revalidate = true;
2106 mac_learning_flush(ofproto->ml);
2112 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2114 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2115 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2116 return bundle && bundle->mirror_out != 0;
2120 forward_bpdu_changed(struct ofproto *ofproto_)
2122 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2123 /* Revalidate cached flows whenever forward_bpdu option changes. */
2124 ofproto->need_revalidate = true;
2129 static struct ofport_dpif *
2130 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
2132 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2133 return ofport ? ofport_dpif_cast(ofport) : NULL;
2136 static struct ofport_dpif *
2137 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
2139 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2143 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2144 struct dpif_port *dpif_port)
2146 ofproto_port->name = dpif_port->name;
2147 ofproto_port->type = dpif_port->type;
2148 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2152 port_run(struct ofport_dpif *ofport)
2154 bool enable = netdev_get_carrier(ofport->up.netdev);
2157 cfm_run(ofport->cfm);
2159 if (cfm_should_send_ccm(ofport->cfm)) {
2160 struct ofpbuf packet;
2162 ofpbuf_init(&packet, 0);
2163 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr);
2164 send_packet(ofport, &packet);
2165 ofpbuf_uninit(&packet);
2168 enable = enable && !cfm_get_fault(ofport->cfm)
2169 && cfm_get_opup(ofport->cfm);
2172 if (ofport->bundle) {
2173 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2176 if (ofport->may_enable != enable) {
2177 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2179 if (ofproto->has_bundle_action) {
2180 ofproto->need_revalidate = true;
2184 ofport->may_enable = enable;
2188 port_wait(struct ofport_dpif *ofport)
2191 cfm_wait(ofport->cfm);
2196 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2197 struct ofproto_port *ofproto_port)
2199 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2200 struct dpif_port dpif_port;
2203 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2205 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2211 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2213 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2217 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2219 *ofp_portp = odp_port_to_ofp_port(odp_port);
2225 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2227 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2230 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2232 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2234 /* The caller is going to close ofport->up.netdev. If this is a
2235 * bonded port, then the bond is using that netdev, so remove it
2236 * from the bond. The client will need to reconfigure everything
2237 * after deleting ports, so then the slave will get re-added. */
2238 bundle_remove(&ofport->up);
2244 struct port_dump_state {
2245 struct dpif_port_dump dump;
2250 port_dump_start(const struct ofproto *ofproto_, void **statep)
2252 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2253 struct port_dump_state *state;
2255 *statep = state = xmalloc(sizeof *state);
2256 dpif_port_dump_start(&state->dump, ofproto->dpif);
2257 state->done = false;
2262 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2263 struct ofproto_port *port)
2265 struct port_dump_state *state = state_;
2266 struct dpif_port dpif_port;
2268 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2269 ofproto_port_from_dpif_port(port, &dpif_port);
2272 int error = dpif_port_dump_done(&state->dump);
2274 return error ? error : EOF;
2279 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2281 struct port_dump_state *state = state_;
2284 dpif_port_dump_done(&state->dump);
2291 port_poll(const struct ofproto *ofproto_, char **devnamep)
2293 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2294 return dpif_port_poll(ofproto->dpif, devnamep);
2298 port_poll_wait(const struct ofproto *ofproto_)
2300 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2301 dpif_port_poll_wait(ofproto->dpif);
2305 port_is_lacp_current(const struct ofport *ofport_)
2307 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2308 return (ofport->bundle && ofport->bundle->lacp
2309 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2313 /* Upcall handling. */
2315 /* Flow miss batching.
2317 * Some dpifs implement operations faster when you hand them off in a batch.
2318 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2319 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2320 * more packets, plus possibly installing the flow in the dpif.
2322 * So far we only batch the operations that affect flow setup time the most.
2323 * It's possible to batch more than that, but the benefit might be minimal. */
2325 struct hmap_node hmap_node;
2327 enum odp_key_fitness key_fitness;
2328 const struct nlattr *key;
2330 ovs_be16 initial_tci;
2331 struct list packets;
2334 struct flow_miss_op {
2335 union dpif_op dpif_op;
2336 struct subfacet *subfacet;
2339 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2340 * OpenFlow controller as necessary according to their individual
2343 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
2344 * ownership is transferred to this function. */
2346 send_packet_in_miss(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
2347 const struct flow *flow, bool clone)
2349 struct ofputil_packet_in pin;
2351 pin.packet = packet;
2352 pin.in_port = flow->in_port;
2353 pin.reason = OFPR_NO_MATCH;
2354 pin.buffer_id = 0; /* not yet known */
2355 pin.send_len = 0; /* not used for flow table misses */
2356 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
2357 clone ? NULL : packet);
2360 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_ACTION to each
2361 * OpenFlow controller as necessary according to their individual
2364 * 'send_len' should be the number of bytes of 'packet' to send to the
2365 * controller, as specified in the action that caused the packet to be sent.
2367 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
2368 * Otherwise, ownership is transferred to this function. */
2370 send_packet_in_action(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
2371 uint64_t userdata, const struct flow *flow, bool clone)
2373 struct ofputil_packet_in pin;
2374 struct user_action_cookie cookie;
2376 memcpy(&cookie, &userdata, sizeof(cookie));
2378 pin.packet = packet;
2379 pin.in_port = flow->in_port;
2380 pin.reason = OFPR_ACTION;
2381 pin.buffer_id = 0; /* not yet known */
2382 pin.send_len = cookie.data;
2383 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
2384 clone ? NULL : packet);
2388 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2389 const struct ofpbuf *packet)
2391 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2397 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2399 cfm_process_heartbeat(ofport->cfm, packet);
2402 } else if (ofport->bundle && ofport->bundle->lacp
2403 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2405 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2408 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2410 stp_process_packet(ofport, packet);
2417 static struct flow_miss *
2418 flow_miss_create(struct hmap *todo, const struct flow *flow,
2419 enum odp_key_fitness key_fitness,
2420 const struct nlattr *key, size_t key_len,
2421 ovs_be16 initial_tci)
2423 uint32_t hash = flow_hash(flow, 0);
2424 struct flow_miss *miss;
2426 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2427 if (flow_equal(&miss->flow, flow)) {
2432 miss = xmalloc(sizeof *miss);
2433 hmap_insert(todo, &miss->hmap_node, hash);
2435 miss->key_fitness = key_fitness;
2437 miss->key_len = key_len;
2438 miss->initial_tci = initial_tci;
2439 list_init(&miss->packets);
2444 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2445 struct flow_miss_op *ops, size_t *n_ops)
2447 const struct flow *flow = &miss->flow;
2448 struct ofpbuf *packet, *next_packet;
2449 struct subfacet *subfacet;
2450 struct facet *facet;
2452 facet = facet_lookup_valid(ofproto, flow);
2454 struct rule_dpif *rule;
2456 rule = rule_dpif_lookup(ofproto, flow, 0);
2458 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
2459 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
2461 if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) {
2462 COVERAGE_INC(ofproto_dpif_no_packet_in);
2463 /* XXX install 'drop' flow entry */
2467 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
2471 LIST_FOR_EACH_SAFE (packet, next_packet, list_node,
2473 list_remove(&packet->list_node);
2474 send_packet_in_miss(ofproto, packet, flow, false);
2480 facet = facet_create(rule, flow);
2483 subfacet = subfacet_create(ofproto, facet,
2484 miss->key_fitness, miss->key, miss->key_len,
2487 LIST_FOR_EACH_SAFE (packet, next_packet, list_node, &miss->packets) {
2488 struct dpif_flow_stats stats;
2490 list_remove(&packet->list_node);
2491 ofproto->n_matches++;
2493 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2495 * Extra-special case for fail-open mode.
2497 * We are in fail-open mode and the packet matched the fail-open
2498 * rule, but we are connected to a controller too. We should send
2499 * the packet up to the controller in the hope that it will try to
2500 * set up a flow and thereby allow us to exit fail-open.
2502 * See the top-level comment in fail-open.c for more information.
2504 send_packet_in_miss(ofproto, packet, flow, true);
2507 if (!facet->may_install || !subfacet->actions) {
2508 subfacet_make_actions(ofproto, subfacet, packet);
2511 /* Credit statistics to subfacet for this packet. We must do this now
2512 * because execute_controller_action() below may destroy 'packet'. */
2513 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2514 subfacet_update_stats(ofproto, subfacet, &stats);
2516 if (!execute_controller_action(ofproto, &facet->flow,
2518 subfacet->actions_len, packet, true)) {
2519 struct flow_miss_op *op = &ops[(*n_ops)++];
2520 struct dpif_execute *execute = &op->dpif_op.execute;
2522 if (flow->vlan_tci != subfacet->initial_tci) {
2523 /* This packet was received on a VLAN splinter port. We added
2524 * a VLAN to the packet to make the packet resemble the flow,
2525 * but the actions were composed assuming that the packet
2526 * contained no VLAN. So, we must remove the VLAN header from
2527 * the packet before trying to execute the actions. */
2528 eth_pop_vlan(packet);
2531 op->subfacet = subfacet;
2532 execute->type = DPIF_OP_EXECUTE;
2533 execute->key = miss->key;
2534 execute->key_len = miss->key_len;
2536 = (facet->may_install
2538 : xmemdup(subfacet->actions, subfacet->actions_len));
2539 execute->actions_len = subfacet->actions_len;
2540 execute->packet = packet;
2544 if (facet->may_install && subfacet->key_fitness != ODP_FIT_TOO_LITTLE) {
2545 struct flow_miss_op *op = &ops[(*n_ops)++];
2546 struct dpif_flow_put *put = &op->dpif_op.flow_put;
2548 op->subfacet = subfacet;
2549 put->type = DPIF_OP_FLOW_PUT;
2550 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2551 put->key = miss->key;
2552 put->key_len = miss->key_len;
2553 put->actions = subfacet->actions;
2554 put->actions_len = subfacet->actions_len;
2559 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
2560 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
2561 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
2562 * what a flow key should contain.
2564 * This function also includes some logic to help make VLAN splinters
2565 * transparent to the rest of the upcall processing logic. In particular, if
2566 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
2567 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
2568 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
2570 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
2571 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
2572 * (This differs from the value returned in flow->vlan_tci only for packets
2573 * received on VLAN splinters.)
2575 static enum odp_key_fitness
2576 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
2577 const struct nlattr *key, size_t key_len,
2578 struct flow *flow, ovs_be16 *initial_tci,
2579 struct ofpbuf *packet)
2581 enum odp_key_fitness fitness;
2585 fitness = odp_flow_key_to_flow(key, key_len, flow);
2586 if (fitness == ODP_FIT_ERROR) {
2589 *initial_tci = flow->vlan_tci;
2591 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
2593 /* Cause the flow to be processed as if it came in on the real device
2594 * with the VLAN device's VLAN ID. */
2595 flow->in_port = realdev;
2596 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
2598 /* Make the packet resemble the flow, so that it gets sent to an
2599 * OpenFlow controller properly, so that it looks correct for
2600 * sFlow, and so that flow_extract() will get the correct vlan_tci
2601 * if it is called on 'packet'.
2603 * The allocated space inside 'packet' probably also contains
2604 * 'key', that is, both 'packet' and 'key' are probably part of a
2605 * struct dpif_upcall (see the large comment on that structure
2606 * definition), so pushing data on 'packet' is in general not a
2607 * good idea since it could overwrite 'key' or free it as a side
2608 * effect. However, it's OK in this special case because we know
2609 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
2610 * will just overwrite the 4-byte "struct nlattr", which is fine
2611 * since we don't need that header anymore. */
2612 eth_push_vlan(packet, flow->vlan_tci);
2615 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
2616 if (fitness == ODP_FIT_PERFECT) {
2617 fitness = ODP_FIT_TOO_MUCH;
2625 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
2628 struct dpif_upcall *upcall;
2629 struct flow_miss *miss, *next_miss;
2630 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
2631 union dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
2640 /* Construct the to-do list.
2642 * This just amounts to extracting the flow from each packet and sticking
2643 * the packets that have the same flow in the same "flow_miss" structure so
2644 * that we can process them together. */
2646 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
2647 enum odp_key_fitness fitness;
2648 struct flow_miss *miss;
2649 ovs_be16 initial_tci;
2652 /* Obtain metadata and check userspace/kernel agreement on flow match,
2653 * then set 'flow''s header pointers. */
2654 fitness = ofproto_dpif_extract_flow_key(ofproto,
2655 upcall->key, upcall->key_len,
2656 &flow, &initial_tci,
2658 if (fitness == ODP_FIT_ERROR) {
2659 ofpbuf_delete(upcall->packet);
2662 flow_extract(upcall->packet, flow.priority, flow.tun_id,
2663 flow.in_port, &flow);
2665 /* Handle 802.1ag, LACP, and STP specially. */
2666 if (process_special(ofproto, &flow, upcall->packet)) {
2667 ofpbuf_delete(upcall->packet);
2668 ofproto->n_matches++;
2672 /* Add other packets to a to-do list. */
2673 miss = flow_miss_create(&todo, &flow, fitness,
2674 upcall->key, upcall->key_len, initial_tci);
2675 list_push_back(&miss->packets, &upcall->packet->list_node);
2678 /* Process each element in the to-do list, constructing the set of
2679 * operations to batch. */
2681 HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) {
2682 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
2683 ofpbuf_list_delete(&miss->packets);
2684 hmap_remove(&todo, &miss->hmap_node);
2687 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
2688 hmap_destroy(&todo);
2690 /* Execute batch. */
2691 for (i = 0; i < n_ops; i++) {
2692 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
2694 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
2696 /* Free memory and update facets. */
2697 for (i = 0; i < n_ops; i++) {
2698 struct flow_miss_op *op = &flow_miss_ops[i];
2699 struct dpif_execute *execute;
2700 struct dpif_flow_put *put;
2702 switch (op->dpif_op.type) {
2703 case DPIF_OP_EXECUTE:
2704 execute = &op->dpif_op.execute;
2705 if (op->subfacet->actions != execute->actions) {
2706 free((struct nlattr *) execute->actions);
2708 ofpbuf_delete((struct ofpbuf *) execute->packet);
2711 case DPIF_OP_FLOW_PUT:
2712 put = &op->dpif_op.flow_put;
2714 op->subfacet->installed = true;
2722 handle_userspace_upcall(struct ofproto_dpif *ofproto,
2723 struct dpif_upcall *upcall)
2725 struct user_action_cookie cookie;
2726 enum odp_key_fitness fitness;
2727 ovs_be16 initial_tci;
2730 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
2732 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
2733 upcall->key_len, &flow,
2734 &initial_tci, upcall->packet);
2735 if (fitness == ODP_FIT_ERROR) {
2736 ofpbuf_delete(upcall->packet);
2740 if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
2741 if (ofproto->sflow) {
2742 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
2745 ofpbuf_delete(upcall->packet);
2746 } else if (cookie.type == USER_ACTION_COOKIE_CONTROLLER) {
2747 COVERAGE_INC(ofproto_dpif_ctlr_action);
2748 send_packet_in_action(ofproto, upcall->packet, upcall->userdata,
2751 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
2752 ofpbuf_delete(upcall->packet);
2757 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
2759 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
2763 assert (max_batch <= FLOW_MISS_MAX_BATCH);
2766 for (i = 0; i < max_batch; i++) {
2767 struct dpif_upcall *upcall = &misses[n_misses];
2770 error = dpif_recv(ofproto->dpif, upcall);
2775 switch (upcall->type) {
2776 case DPIF_UC_ACTION:
2777 handle_userspace_upcall(ofproto, upcall);
2781 /* Handle it later. */
2785 case DPIF_N_UC_TYPES:
2787 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32,
2793 handle_miss_upcalls(ofproto, misses, n_misses);
2798 /* Flow expiration. */
2800 static int subfacet_max_idle(const struct ofproto_dpif *);
2801 static void update_stats(struct ofproto_dpif *);
2802 static void rule_expire(struct rule_dpif *);
2803 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
2805 /* This function is called periodically by run(). Its job is to collect
2806 * updates for the flows that have been installed into the datapath, most
2807 * importantly when they last were used, and then use that information to
2808 * expire flows that have not been used recently.
2810 * Returns the number of milliseconds after which it should be called again. */
2812 expire(struct ofproto_dpif *ofproto)
2814 struct rule_dpif *rule, *next_rule;
2815 struct classifier *table;
2818 /* Update stats for each flow in the datapath. */
2819 update_stats(ofproto);
2821 /* Expire subfacets that have been idle too long. */
2822 dp_max_idle = subfacet_max_idle(ofproto);
2823 expire_subfacets(ofproto, dp_max_idle);
2825 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
2826 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
2827 struct cls_cursor cursor;
2829 cls_cursor_init(&cursor, table, NULL);
2830 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
2835 /* All outstanding data in existing flows has been accounted, so it's a
2836 * good time to do bond rebalancing. */
2837 if (ofproto->has_bonded_bundles) {
2838 struct ofbundle *bundle;
2840 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2842 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
2847 return MIN(dp_max_idle, 1000);
2850 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
2852 * This function also pushes statistics updates to rules which each facet
2853 * resubmits into. Generally these statistics will be accurate. However, if a
2854 * facet changes the rule it resubmits into at some time in between
2855 * update_stats() runs, it is possible that statistics accrued to the
2856 * old rule will be incorrectly attributed to the new rule. This could be
2857 * avoided by calling update_stats() whenever rules are created or
2858 * deleted. However, the performance impact of making so many calls to the
2859 * datapath do not justify the benefit of having perfectly accurate statistics.
2862 update_stats(struct ofproto_dpif *p)
2864 const struct dpif_flow_stats *stats;
2865 struct dpif_flow_dump dump;
2866 const struct nlattr *key;
2869 dpif_flow_dump_start(&dump, p->dpif);
2870 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
2871 struct subfacet *subfacet;
2873 subfacet = subfacet_find(p, key, key_len);
2874 if (subfacet && subfacet->installed) {
2875 struct facet *facet = subfacet->facet;
2877 if (stats->n_packets >= subfacet->dp_packet_count) {
2878 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
2879 facet->packet_count += extra;
2881 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
2884 if (stats->n_bytes >= subfacet->dp_byte_count) {
2885 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
2887 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
2890 subfacet->dp_packet_count = stats->n_packets;
2891 subfacet->dp_byte_count = stats->n_bytes;
2893 subfacet_update_time(p, subfacet, stats->used);
2894 facet_account(p, facet);
2895 facet_push_stats(facet);
2897 if (!VLOG_DROP_WARN(&rl)) {
2901 odp_flow_key_format(key, key_len, &s);
2902 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
2906 COVERAGE_INC(facet_unexpected);
2907 /* There's a flow in the datapath that we know nothing about, or a
2908 * flow that shouldn't be installed but was anyway. Delete it. */
2909 dpif_flow_del(p->dpif, key, key_len, NULL);
2912 dpif_flow_dump_done(&dump);
2915 /* Calculates and returns the number of milliseconds of idle time after which
2916 * subfacets should expire from the datapath. When a subfacet expires, we fold
2917 * its statistics into its facet, and when a facet's last subfacet expires, we
2918 * fold its statistic into its rule. */
2920 subfacet_max_idle(const struct ofproto_dpif *ofproto)
2923 * Idle time histogram.
2925 * Most of the time a switch has a relatively small number of subfacets.
2926 * When this is the case we might as well keep statistics for all of them
2927 * in userspace and to cache them in the kernel datapath for performance as
2930 * As the number of subfacets increases, the memory required to maintain
2931 * statistics about them in userspace and in the kernel becomes
2932 * significant. However, with a large number of subfacets it is likely
2933 * that only a few of them are "heavy hitters" that consume a large amount
2934 * of bandwidth. At this point, only heavy hitters are worth caching in
2935 * the kernel and maintaining in userspaces; other subfacets we can
2938 * The technique used to compute the idle time is to build a histogram with
2939 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
2940 * that is installed in the kernel gets dropped in the appropriate bucket.
2941 * After the histogram has been built, we compute the cutoff so that only
2942 * the most-recently-used 1% of subfacets (but at least
2943 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
2944 * the most-recently-used bucket of subfacets is kept, so actually an
2945 * arbitrary number of subfacets can be kept in any given expiration run
2946 * (though the next run will delete most of those unless they receive
2949 * This requires a second pass through the subfacets, in addition to the
2950 * pass made by update_stats(), because the former function never looks at
2951 * uninstallable subfacets.
2953 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
2954 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
2955 int buckets[N_BUCKETS] = { 0 };
2956 int total, subtotal, bucket;
2957 struct subfacet *subfacet;
2961 total = hmap_count(&ofproto->subfacets);
2962 if (total <= ofproto->up.flow_eviction_threshold) {
2963 return N_BUCKETS * BUCKET_WIDTH;
2966 /* Build histogram. */
2968 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
2969 long long int idle = now - subfacet->used;
2970 int bucket = (idle <= 0 ? 0
2971 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
2972 : (unsigned int) idle / BUCKET_WIDTH);
2976 /* Find the first bucket whose flows should be expired. */
2977 subtotal = bucket = 0;
2979 subtotal += buckets[bucket++];
2980 } while (bucket < N_BUCKETS &&
2981 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
2983 if (VLOG_IS_DBG_ENABLED()) {
2987 ds_put_cstr(&s, "keep");
2988 for (i = 0; i < N_BUCKETS; i++) {
2990 ds_put_cstr(&s, ", drop");
2993 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
2996 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3000 return bucket * BUCKET_WIDTH;
3004 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3006 long long int cutoff = time_msec() - dp_max_idle;
3007 struct subfacet *subfacet, *next_subfacet;
3009 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3010 &ofproto->subfacets) {
3011 if (subfacet->used < cutoff) {
3012 subfacet_destroy(ofproto, subfacet);
3017 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3018 * then delete it entirely. */
3020 rule_expire(struct rule_dpif *rule)
3022 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3023 struct facet *facet, *next_facet;
3027 /* Has 'rule' expired? */
3029 if (rule->up.hard_timeout
3030 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3031 reason = OFPRR_HARD_TIMEOUT;
3032 } else if (rule->up.idle_timeout && list_is_empty(&rule->facets)
3033 && now > rule->used + rule->up.idle_timeout * 1000) {
3034 reason = OFPRR_IDLE_TIMEOUT;
3039 COVERAGE_INC(ofproto_dpif_expired);
3041 /* Update stats. (This is a no-op if the rule expired due to an idle
3042 * timeout, because that only happens when the rule has no facets left.) */
3043 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3044 facet_remove(ofproto, facet);
3047 /* Get rid of the rule. */
3048 ofproto_rule_expire(&rule->up, reason);
3053 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3055 * The caller must already have determined that no facet with an identical
3056 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3057 * the ofproto's classifier table.
3059 * The facet will initially have no subfacets. The caller should create (at
3060 * least) one subfacet with subfacet_create(). */
3061 static struct facet *
3062 facet_create(struct rule_dpif *rule, const struct flow *flow)
3064 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3065 struct facet *facet;
3067 facet = xzalloc(sizeof *facet);
3068 facet->used = time_msec();
3069 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
3070 list_push_back(&rule->facets, &facet->list_node);
3072 facet->flow = *flow;
3073 list_init(&facet->subfacets);
3074 netflow_flow_init(&facet->nf_flow);
3075 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3081 facet_free(struct facet *facet)
3086 /* If the 'actions_len' bytes of actions in 'odp_actions' are just a single
3087 * OVS_ACTION_ATTR_USERSPACE action, executes it internally and returns true.
3088 * Otherwise, returns false without doing anything.
3090 * If 'clone' is true, the caller always retains ownership of 'packet'.
3091 * Otherwise, ownership is transferred to this function if it returns true. */
3093 execute_controller_action(struct ofproto_dpif *ofproto,
3094 const struct flow *flow,
3095 const struct nlattr *odp_actions, size_t actions_len,
3096 struct ofpbuf *packet, bool clone)
3099 && odp_actions->nla_type == OVS_ACTION_ATTR_USERSPACE
3100 && NLA_ALIGN(odp_actions->nla_len) == actions_len) {
3101 /* As an optimization, avoid a round-trip from userspace to kernel to
3102 * userspace. This also avoids possibly filling up kernel packet
3103 * buffers along the way.
3105 * This optimization will not accidentally catch sFlow
3106 * OVS_ACTION_ATTR_USERSPACE actions, since those are encapsulated
3107 * inside OVS_ACTION_ATTR_SAMPLE. */
3108 const struct nlattr *nla;
3110 nla = nl_attr_find_nested(odp_actions, OVS_USERSPACE_ATTR_USERDATA);
3111 send_packet_in_action(ofproto, packet, nl_attr_get_u64(nla), flow,
3119 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3120 * 'packet', which arrived on 'in_port'.
3122 * Takes ownership of 'packet'. */
3124 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3125 const struct nlattr *odp_actions, size_t actions_len,
3126 struct ofpbuf *packet)
3128 struct odputil_keybuf keybuf;
3132 if (execute_controller_action(ofproto, flow, odp_actions, actions_len,
3137 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3138 odp_flow_key_from_flow(&key, flow);
3140 error = dpif_execute(ofproto->dpif, key.data, key.size,
3141 odp_actions, actions_len, packet);
3143 ofpbuf_delete(packet);
3147 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3149 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3150 * rule's statistics, via subfacet_uninstall().
3152 * - Removes 'facet' from its rule and from ofproto->facets.
3155 facet_remove(struct ofproto_dpif *ofproto, struct facet *facet)
3157 struct subfacet *subfacet, *next_subfacet;
3159 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3160 &facet->subfacets) {
3161 subfacet_destroy__(ofproto, subfacet);
3164 facet_flush_stats(ofproto, facet);
3165 hmap_remove(&ofproto->facets, &facet->hmap_node);
3166 list_remove(&facet->list_node);
3171 facet_account(struct ofproto_dpif *ofproto, struct facet *facet)
3174 struct subfacet *subfacet;
3175 const struct nlattr *a;
3179 if (facet->byte_count <= facet->accounted_bytes) {
3182 n_bytes = facet->byte_count - facet->accounted_bytes;
3183 facet->accounted_bytes = facet->byte_count;
3185 /* Feed information from the active flows back into the learning table to
3186 * ensure that table is always in sync with what is actually flowing
3187 * through the datapath. */
3188 if (facet->has_learn || facet->has_normal) {
3189 struct action_xlate_ctx ctx;
3191 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3192 facet->flow.vlan_tci, NULL);
3193 ctx.may_learn = true;
3194 ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions,
3195 facet->rule->up.n_actions));
3198 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3202 /* This loop feeds byte counters to bond_account() for rebalancing to use
3203 * as a basis. We also need to track the actual VLAN on which the packet
3204 * is going to be sent to ensure that it matches the one passed to
3205 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3208 * We use the actions from an arbitrary subfacet because they should all
3209 * be equally valid for our purpose. */
3210 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3211 struct subfacet, list_node);
3212 vlan_tci = facet->flow.vlan_tci;
3213 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3214 subfacet->actions, subfacet->actions_len) {
3215 const struct ovs_action_push_vlan *vlan;
3216 struct ofport_dpif *port;
3218 switch (nl_attr_type(a)) {
3219 case OVS_ACTION_ATTR_OUTPUT:
3220 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3221 if (port && port->bundle && port->bundle->bond) {
3222 bond_account(port->bundle->bond, &facet->flow,
3223 vlan_tci_to_vid(vlan_tci), n_bytes);
3227 case OVS_ACTION_ATTR_POP_VLAN:
3228 vlan_tci = htons(0);
3231 case OVS_ACTION_ATTR_PUSH_VLAN:
3232 vlan = nl_attr_get(a);
3233 vlan_tci = vlan->vlan_tci;
3239 /* Returns true if the only action for 'facet' is to send to the controller.
3240 * (We don't report NetFlow expiration messages for such facets because they
3241 * are just part of the control logic for the network, not real traffic). */
3243 facet_is_controller_flow(struct facet *facet)
3246 && facet->rule->up.n_actions == 1
3247 && action_outputs_to_port(&facet->rule->up.actions[0],
3248 htons(OFPP_CONTROLLER)));
3251 /* Folds all of 'facet''s statistics into its rule. Also updates the
3252 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3253 * 'facet''s statistics in the datapath should have been zeroed and folded into
3254 * its packet and byte counts before this function is called. */
3256 facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet)
3258 struct subfacet *subfacet;
3260 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3261 assert(!subfacet->dp_byte_count);
3262 assert(!subfacet->dp_packet_count);
3265 facet_push_stats(facet);
3266 facet_account(ofproto, facet);
3268 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3269 struct ofexpired expired;
3270 expired.flow = facet->flow;
3271 expired.packet_count = facet->packet_count;
3272 expired.byte_count = facet->byte_count;
3273 expired.used = facet->used;
3274 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3277 facet->rule->packet_count += facet->packet_count;
3278 facet->rule->byte_count += facet->byte_count;
3280 /* Reset counters to prevent double counting if 'facet' ever gets
3282 facet_reset_counters(facet);
3284 netflow_flow_clear(&facet->nf_flow);
3287 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3288 * Returns it if found, otherwise a null pointer.
3290 * The returned facet might need revalidation; use facet_lookup_valid()
3291 * instead if that is important. */
3292 static struct facet *
3293 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
3295 struct facet *facet;
3297 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
3299 if (flow_equal(flow, &facet->flow)) {
3307 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3308 * Returns it if found, otherwise a null pointer.
3310 * The returned facet is guaranteed to be valid. */
3311 static struct facet *
3312 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
3314 struct facet *facet = facet_find(ofproto, flow);
3316 /* The facet we found might not be valid, since we could be in need of
3317 * revalidation. If it is not valid, don't return it. */
3319 && (ofproto->need_revalidate
3320 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))
3321 && !facet_revalidate(ofproto, facet)) {
3322 COVERAGE_INC(facet_invalidated);
3329 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
3331 * - If the rule found is different from 'facet''s current rule, moves
3332 * 'facet' to the new rule and recompiles its actions.
3334 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3335 * where it is and recompiles its actions anyway.
3337 * - If there is none, destroys 'facet'.
3339 * Returns true if 'facet' still exists, false if it has been destroyed. */
3341 facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet)
3344 struct nlattr *odp_actions;
3347 struct actions *new_actions;
3349 struct action_xlate_ctx ctx;
3350 struct rule_dpif *new_rule;
3351 struct subfacet *subfacet;
3352 bool actions_changed;
3355 COVERAGE_INC(facet_revalidate);
3357 /* Determine the new rule. */
3358 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3360 /* No new rule, so delete the facet. */
3361 facet_remove(ofproto, facet);
3365 /* Calculate new datapath actions.
3367 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3368 * emit a NetFlow expiration and, if so, we need to have the old state
3369 * around to properly compose it. */
3371 /* If the datapath actions changed or the installability changed,
3372 * then we need to talk to the datapath. */
3375 memset(&ctx, 0, sizeof ctx);
3376 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3377 struct ofpbuf *odp_actions;
3378 bool should_install;
3380 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3381 subfacet->initial_tci, NULL);
3382 odp_actions = xlate_actions(&ctx, new_rule->up.actions,
3383 new_rule->up.n_actions);
3384 actions_changed = (subfacet->actions_len != odp_actions->size
3385 || memcmp(subfacet->actions, odp_actions->data,
3386 subfacet->actions_len));
3388 should_install = (ctx.may_set_up_flow
3389 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3390 if (actions_changed || should_install != subfacet->installed) {
3391 if (should_install) {
3392 struct dpif_flow_stats stats;
3394 subfacet_install(ofproto, subfacet,
3395 odp_actions->data, odp_actions->size, &stats);
3396 subfacet_update_stats(ofproto, subfacet, &stats);
3398 subfacet_uninstall(ofproto, subfacet);
3402 new_actions = xcalloc(list_size(&facet->subfacets),
3403 sizeof *new_actions);
3405 new_actions[i].odp_actions = xmemdup(odp_actions->data,
3407 new_actions[i].actions_len = odp_actions->size;
3410 ofpbuf_delete(odp_actions);
3414 facet_flush_stats(ofproto, facet);
3417 /* Update 'facet' now that we've taken care of all the old state. */
3418 facet->tags = ctx.tags;
3419 facet->nf_flow.output_iface = ctx.nf_output_iface;
3420 facet->may_install = ctx.may_set_up_flow;
3421 facet->has_learn = ctx.has_learn;
3422 facet->has_normal = ctx.has_normal;
3423 facet->mirrors = ctx.mirrors;
3426 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3427 if (new_actions[i].odp_actions) {
3428 free(subfacet->actions);
3429 subfacet->actions = new_actions[i].odp_actions;
3430 subfacet->actions_len = new_actions[i].actions_len;
3436 if (facet->rule != new_rule) {
3437 COVERAGE_INC(facet_changed_rule);
3438 list_remove(&facet->list_node);
3439 list_push_back(&new_rule->facets, &facet->list_node);
3440 facet->rule = new_rule;
3441 facet->used = new_rule->up.created;
3442 facet->prev_used = facet->used;
3448 /* Updates 'facet''s used time. Caller is responsible for calling
3449 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3451 facet_update_time(struct ofproto_dpif *ofproto, struct facet *facet,
3454 if (used > facet->used) {
3456 if (used > facet->rule->used) {
3457 facet->rule->used = used;
3459 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3464 facet_reset_counters(struct facet *facet)
3466 facet->packet_count = 0;
3467 facet->byte_count = 0;
3468 facet->prev_packet_count = 0;
3469 facet->prev_byte_count = 0;
3470 facet->accounted_bytes = 0;
3474 facet_push_stats(struct facet *facet)
3476 uint64_t new_packets, new_bytes;
3478 assert(facet->packet_count >= facet->prev_packet_count);
3479 assert(facet->byte_count >= facet->prev_byte_count);
3480 assert(facet->used >= facet->prev_used);
3482 new_packets = facet->packet_count - facet->prev_packet_count;
3483 new_bytes = facet->byte_count - facet->prev_byte_count;
3485 if (new_packets || new_bytes || facet->used > facet->prev_used) {
3486 facet->prev_packet_count = facet->packet_count;
3487 facet->prev_byte_count = facet->byte_count;
3488 facet->prev_used = facet->used;
3490 flow_push_stats(facet->rule, &facet->flow,
3491 new_packets, new_bytes, facet->used);
3493 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
3494 facet->mirrors, new_packets, new_bytes);
3498 struct ofproto_push {
3499 struct action_xlate_ctx ctx;
3506 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
3508 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3511 rule->packet_count += push->packets;
3512 rule->byte_count += push->bytes;
3513 rule->used = MAX(push->used, rule->used);
3517 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3518 * 'rule''s actions and mirrors. */
3520 flow_push_stats(const struct rule_dpif *rule,
3521 const struct flow *flow, uint64_t packets, uint64_t bytes,
3524 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3525 struct ofproto_push push;
3527 push.packets = packets;
3531 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, NULL);
3532 push.ctx.resubmit_hook = push_resubmit;
3533 ofpbuf_delete(xlate_actions(&push.ctx,
3534 rule->up.actions, rule->up.n_actions));
3539 static struct subfacet *
3540 subfacet_find__(struct ofproto_dpif *ofproto,
3541 const struct nlattr *key, size_t key_len, uint32_t key_hash,
3542 const struct flow *flow)
3544 struct subfacet *subfacet;
3546 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
3547 &ofproto->subfacets) {
3549 ? (subfacet->key_len == key_len
3550 && !memcmp(key, subfacet->key, key_len))
3551 : flow_equal(flow, &subfacet->facet->flow)) {
3559 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
3560 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
3561 * there is one, otherwise creates and returns a new subfacet.
3563 * If the returned subfacet is new, then subfacet->actions will be NULL, in
3564 * which case the caller must populate the actions with
3565 * subfacet_make_actions(). */
3566 static struct subfacet *
3567 subfacet_create(struct ofproto_dpif *ofproto, struct facet *facet,
3568 enum odp_key_fitness key_fitness,
3569 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
3571 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3572 struct subfacet *subfacet;
3574 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
3576 if (subfacet->facet == facet) {
3580 /* This shouldn't happen. */
3581 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
3582 subfacet_destroy(ofproto, subfacet);
3585 subfacet = xzalloc(sizeof *subfacet);
3586 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
3587 list_push_back(&facet->subfacets, &subfacet->list_node);
3588 subfacet->facet = facet;
3589 subfacet->used = time_msec();
3590 subfacet->key_fitness = key_fitness;
3591 if (key_fitness != ODP_FIT_PERFECT) {
3592 subfacet->key = xmemdup(key, key_len);
3593 subfacet->key_len = key_len;
3595 subfacet->installed = false;
3596 subfacet->initial_tci = initial_tci;
3601 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
3602 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
3603 static struct subfacet *
3604 subfacet_find(struct ofproto_dpif *ofproto,
3605 const struct nlattr *key, size_t key_len)
3607 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3608 enum odp_key_fitness fitness;
3611 fitness = odp_flow_key_to_flow(key, key_len, &flow);
3612 if (fitness == ODP_FIT_ERROR) {
3616 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
3619 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
3620 * its facet within 'ofproto', and frees it. */
3622 subfacet_destroy__(struct ofproto_dpif *ofproto, struct subfacet *subfacet)
3624 subfacet_uninstall(ofproto, subfacet);
3625 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
3626 list_remove(&subfacet->list_node);
3627 free(subfacet->key);
3628 free(subfacet->actions);
3632 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
3633 * last remaining subfacet in its facet destroys the facet too. */
3635 subfacet_destroy(struct ofproto_dpif *ofproto, struct subfacet *subfacet)
3637 struct facet *facet = subfacet->facet;
3639 subfacet_destroy__(ofproto, subfacet);
3640 if (list_is_empty(&facet->subfacets)) {
3641 facet_remove(ofproto, facet);
3645 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
3646 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
3647 * for use as temporary storage. */
3649 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
3652 if (!subfacet->key) {
3653 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
3654 odp_flow_key_from_flow(key, &subfacet->facet->flow);
3656 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
3660 /* Composes the datapath actions for 'subfacet' based on its rule's actions. */
3662 subfacet_make_actions(struct ofproto_dpif *p, struct subfacet *subfacet,
3663 const struct ofpbuf *packet)
3665 struct facet *facet = subfacet->facet;
3666 const struct rule_dpif *rule = facet->rule;
3667 struct ofpbuf *odp_actions;
3668 struct action_xlate_ctx ctx;
3670 action_xlate_ctx_init(&ctx, p, &facet->flow, subfacet->initial_tci,
3672 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3673 facet->tags = ctx.tags;
3674 facet->may_install = ctx.may_set_up_flow;
3675 facet->has_learn = ctx.has_learn;
3676 facet->has_normal = ctx.has_normal;
3677 facet->nf_flow.output_iface = ctx.nf_output_iface;
3678 facet->mirrors = ctx.mirrors;
3680 if (subfacet->actions_len != odp_actions->size
3681 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
3682 free(subfacet->actions);
3683 subfacet->actions_len = odp_actions->size;
3684 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
3687 ofpbuf_delete(odp_actions);
3690 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
3691 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
3692 * in the datapath will be zeroed and 'stats' will be updated with traffic new
3693 * since 'subfacet' was last updated.
3695 * Returns 0 if successful, otherwise a positive errno value. */
3697 subfacet_install(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3698 const struct nlattr *actions, size_t actions_len,
3699 struct dpif_flow_stats *stats)
3701 struct odputil_keybuf keybuf;
3702 enum dpif_flow_put_flags flags;
3706 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3708 flags |= DPIF_FP_ZERO_STATS;
3711 subfacet_get_key(subfacet, &keybuf, &key);
3712 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
3713 actions, actions_len, stats);
3716 subfacet_reset_dp_stats(subfacet, stats);
3722 /* If 'subfacet' is installed in the datapath, uninstalls it. */
3724 subfacet_uninstall(struct ofproto_dpif *p, struct subfacet *subfacet)
3726 if (subfacet->installed) {
3727 struct odputil_keybuf keybuf;
3728 struct dpif_flow_stats stats;
3732 subfacet_get_key(subfacet, &keybuf, &key);
3733 error = dpif_flow_del(p->dpif, key.data, key.size, &stats);
3734 subfacet_reset_dp_stats(subfacet, &stats);
3736 subfacet_update_stats(p, subfacet, &stats);
3738 subfacet->installed = false;
3740 assert(subfacet->dp_packet_count == 0);
3741 assert(subfacet->dp_byte_count == 0);
3745 /* Resets 'subfacet''s datapath statistics counters. This should be called
3746 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
3747 * non-null, it should contain the statistics returned by dpif when 'subfacet'
3748 * was reset in the datapath. 'stats' will be modified to include only
3749 * statistics new since 'subfacet' was last updated. */
3751 subfacet_reset_dp_stats(struct subfacet *subfacet,
3752 struct dpif_flow_stats *stats)
3755 && subfacet->dp_packet_count <= stats->n_packets
3756 && subfacet->dp_byte_count <= stats->n_bytes) {
3757 stats->n_packets -= subfacet->dp_packet_count;
3758 stats->n_bytes -= subfacet->dp_byte_count;
3761 subfacet->dp_packet_count = 0;
3762 subfacet->dp_byte_count = 0;
3765 /* Updates 'subfacet''s used time. The caller is responsible for calling
3766 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
3768 subfacet_update_time(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3771 if (used > subfacet->used) {
3772 subfacet->used = used;
3773 facet_update_time(ofproto, subfacet->facet, used);
3777 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
3779 * Because of the meaning of a subfacet's counters, it only makes sense to do
3780 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
3781 * represents a packet that was sent by hand or if it represents statistics
3782 * that have been cleared out of the datapath. */
3784 subfacet_update_stats(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3785 const struct dpif_flow_stats *stats)
3787 if (stats->n_packets || stats->used > subfacet->used) {
3788 struct facet *facet = subfacet->facet;
3790 subfacet_update_time(ofproto, subfacet, stats->used);
3791 facet->packet_count += stats->n_packets;
3792 facet->byte_count += stats->n_bytes;
3793 facet_push_stats(facet);
3794 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3800 static struct rule_dpif *
3801 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
3804 struct cls_rule *cls_rule;
3805 struct classifier *cls;
3807 if (table_id >= N_TABLES) {
3811 cls = &ofproto->up.tables[table_id];
3812 if (flow->nw_frag & FLOW_NW_FRAG_ANY
3813 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3814 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
3815 * are unavailable. */
3816 struct flow ofpc_normal_flow = *flow;
3817 ofpc_normal_flow.tp_src = htons(0);
3818 ofpc_normal_flow.tp_dst = htons(0);
3819 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
3821 cls_rule = classifier_lookup(cls, flow);
3823 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
3827 complete_operation(struct rule_dpif *rule)
3829 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3831 rule_invalidate(rule);
3833 struct dpif_completion *c = xmalloc(sizeof *c);
3834 c->op = rule->up.pending;
3835 list_push_back(&ofproto->completions, &c->list_node);
3837 ofoperation_complete(rule->up.pending, 0);
3841 static struct rule *
3844 struct rule_dpif *rule = xmalloc(sizeof *rule);
3849 rule_dealloc(struct rule *rule_)
3851 struct rule_dpif *rule = rule_dpif_cast(rule_);
3856 rule_construct(struct rule *rule_)
3858 struct rule_dpif *rule = rule_dpif_cast(rule_);
3859 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3860 struct rule_dpif *victim;
3864 error = validate_actions(rule->up.actions, rule->up.n_actions,
3865 &rule->up.cr.flow, ofproto->max_ports);
3870 rule->used = rule->up.created;
3871 rule->packet_count = 0;
3872 rule->byte_count = 0;
3874 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
3875 if (victim && !list_is_empty(&victim->facets)) {
3876 struct facet *facet;
3878 rule->facets = victim->facets;
3879 list_moved(&rule->facets);
3880 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3881 /* XXX: We're only clearing our local counters here. It's possible
3882 * that quite a few packets are unaccounted for in the datapath
3883 * statistics. These will be accounted to the new rule instead of
3884 * cleared as required. This could be fixed by clearing out the
3885 * datapath statistics for this facet, but currently it doesn't
3887 facet_reset_counters(facet);
3891 /* Must avoid list_moved() in this case. */
3892 list_init(&rule->facets);
3895 table_id = rule->up.table_id;
3896 rule->tag = (victim ? victim->tag
3898 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
3899 ofproto->tables[table_id].basis));
3901 complete_operation(rule);
3906 rule_destruct(struct rule *rule_)
3908 struct rule_dpif *rule = rule_dpif_cast(rule_);
3909 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3910 struct facet *facet, *next_facet;
3912 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3913 facet_revalidate(ofproto, facet);
3916 complete_operation(rule);
3920 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
3922 struct rule_dpif *rule = rule_dpif_cast(rule_);
3923 struct facet *facet;
3925 /* Start from historical data for 'rule' itself that are no longer tracked
3926 * in facets. This counts, for example, facets that have expired. */
3927 *packets = rule->packet_count;
3928 *bytes = rule->byte_count;
3930 /* Add any statistics that are tracked by facets. This includes
3931 * statistical data recently updated by ofproto_update_stats() as well as
3932 * stats for packets that were executed "by hand" via dpif_execute(). */
3933 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3934 *packets += facet->packet_count;
3935 *bytes += facet->byte_count;
3940 rule_execute(struct rule *rule_, const struct flow *flow,
3941 struct ofpbuf *packet)
3943 struct rule_dpif *rule = rule_dpif_cast(rule_);
3944 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3945 struct action_xlate_ctx ctx;
3946 struct ofpbuf *odp_actions;
3949 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, packet);
3950 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3951 size = packet->size;
3952 if (execute_odp_actions(ofproto, flow, odp_actions->data,
3953 odp_actions->size, packet)) {
3954 rule->used = time_msec();
3955 rule->packet_count++;
3956 rule->byte_count += size;
3957 flow_push_stats(rule, flow, 1, size, rule->used);
3959 ofpbuf_delete(odp_actions);
3965 rule_modify_actions(struct rule *rule_)
3967 struct rule_dpif *rule = rule_dpif_cast(rule_);
3968 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3971 error = validate_actions(rule->up.actions, rule->up.n_actions,
3972 &rule->up.cr.flow, ofproto->max_ports);
3974 ofoperation_complete(rule->up.pending, error);
3978 complete_operation(rule);
3981 /* Sends 'packet' out 'ofport'.
3982 * May modify 'packet'.
3983 * Returns 0 if successful, otherwise a positive errno value. */
3985 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
3987 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3988 struct ofpbuf key, odp_actions;
3989 struct odputil_keybuf keybuf;
3994 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
3995 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
3997 if (odp_port != ofport->odp_port) {
3998 eth_pop_vlan(packet);
3999 flow.vlan_tci = htons(0);
4002 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4003 odp_flow_key_from_flow(&key, &flow);
4005 ofpbuf_init(&odp_actions, 32);
4006 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4008 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4009 error = dpif_execute(ofproto->dpif,
4011 odp_actions.data, odp_actions.size,
4013 ofpbuf_uninit(&odp_actions);
4016 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4017 ofproto->up.name, odp_port, strerror(error));
4022 /* OpenFlow to datapath action translation. */
4024 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
4025 struct action_xlate_ctx *ctx);
4026 static void xlate_normal(struct action_xlate_ctx *);
4029 put_userspace_action(const struct ofproto_dpif *ofproto,
4030 struct ofpbuf *odp_actions,
4031 const struct flow *flow,
4032 const struct user_action_cookie *cookie)
4036 pid = dpif_port_get_pid(ofproto->dpif,
4037 ofp_port_to_odp_port(flow->in_port));
4039 return odp_put_userspace_action(pid, cookie, odp_actions);
4042 /* Compose SAMPLE action for sFlow. */
4044 compose_sflow_action(const struct ofproto_dpif *ofproto,
4045 struct ofpbuf *odp_actions,
4046 const struct flow *flow,
4049 uint32_t port_ifindex;
4050 uint32_t probability;
4051 struct user_action_cookie cookie;
4052 size_t sample_offset, actions_offset;
4053 int cookie_offset, n_output;
4055 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4059 if (odp_port == OVSP_NONE) {
4063 port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4067 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4069 /* Number of packets out of UINT_MAX to sample. */
4070 probability = dpif_sflow_get_probability(ofproto->sflow);
4071 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4073 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4075 cookie.type = USER_ACTION_COOKIE_SFLOW;
4076 cookie.data = port_ifindex;
4077 cookie.n_output = n_output;
4078 cookie.vlan_tci = 0;
4079 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4081 nl_msg_end_nested(odp_actions, actions_offset);
4082 nl_msg_end_nested(odp_actions, sample_offset);
4083 return cookie_offset;
4086 /* SAMPLE action must be first action in any given list of actions.
4087 * At this point we do not have all information required to build it. So try to
4088 * build sample action as complete as possible. */
4090 add_sflow_action(struct action_xlate_ctx *ctx)
4092 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4094 &ctx->flow, OVSP_NONE);
4095 ctx->sflow_odp_port = 0;
4096 ctx->sflow_n_outputs = 0;
4099 /* Fix SAMPLE action according to data collected while composing ODP actions.
4100 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4101 * USERSPACE action's user-cookie which is required for sflow. */
4103 fix_sflow_action(struct action_xlate_ctx *ctx)
4105 const struct flow *base = &ctx->base_flow;
4106 struct user_action_cookie *cookie;
4108 if (!ctx->user_cookie_offset) {
4112 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4114 assert(cookie != NULL);
4115 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4117 if (ctx->sflow_n_outputs) {
4118 cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow,
4119 ctx->sflow_odp_port);
4121 if (ctx->sflow_n_outputs >= 255) {
4122 cookie->n_output = 255;
4124 cookie->n_output = ctx->sflow_n_outputs;
4126 cookie->vlan_tci = base->vlan_tci;
4130 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4133 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4134 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4135 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4136 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4140 struct priority_to_dscp *pdscp;
4142 if (ofport->up.opp.config & htonl(OFPPC_NO_FWD)
4143 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4147 pdscp = get_priority(ofport, ctx->flow.priority);
4149 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4150 ctx->flow.nw_tos |= pdscp->dscp;
4153 /* We may not have an ofport record for this port, but it doesn't hurt
4154 * to allow forwarding to it anyhow. Maybe such a port will appear
4155 * later and we're pre-populating the flow table. */
4158 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4159 ctx->flow.vlan_tci);
4160 if (out_port != odp_port) {
4161 ctx->flow.vlan_tci = htons(0);
4163 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4164 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4166 ctx->sflow_odp_port = odp_port;
4167 ctx->sflow_n_outputs++;
4168 ctx->nf_output_iface = ofp_port;
4169 ctx->flow.vlan_tci = flow_vlan_tci;
4170 ctx->flow.nw_tos = flow_nw_tos;
4174 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4176 compose_output_action__(ctx, ofp_port, true);
4180 xlate_table_action(struct action_xlate_ctx *ctx,
4181 uint16_t in_port, uint8_t table_id)
4183 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4184 struct ofproto_dpif *ofproto = ctx->ofproto;
4185 struct rule_dpif *rule;
4186 uint16_t old_in_port;
4187 uint8_t old_table_id;
4189 old_table_id = ctx->table_id;
4190 ctx->table_id = table_id;
4192 /* Look up a flow with 'in_port' as the input port. */
4193 old_in_port = ctx->flow.in_port;
4194 ctx->flow.in_port = in_port;
4195 rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
4198 if (table_id > 0 && table_id < N_TABLES) {
4199 struct table_dpif *table = &ofproto->tables[table_id];
4200 if (table->other_table) {
4203 : rule_calculate_tag(&ctx->flow,
4204 &table->other_table->wc,
4209 /* Restore the original input port. Otherwise OFPP_NORMAL and
4210 * OFPP_IN_PORT will have surprising behavior. */
4211 ctx->flow.in_port = old_in_port;
4213 if (ctx->resubmit_hook) {
4214 ctx->resubmit_hook(ctx, rule);
4219 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
4223 ctx->table_id = old_table_id;
4225 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4227 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
4228 MAX_RESUBMIT_RECURSION);
4233 xlate_resubmit_table(struct action_xlate_ctx *ctx,
4234 const struct nx_action_resubmit *nar)
4239 in_port = (nar->in_port == htons(OFPP_IN_PORT)
4241 : ntohs(nar->in_port));
4242 table_id = nar->table == 255 ? ctx->table_id : nar->table;
4244 xlate_table_action(ctx, in_port, table_id);
4248 flood_packets(struct action_xlate_ctx *ctx, bool all)
4250 struct ofport_dpif *ofport;
4252 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
4253 uint16_t ofp_port = ofport->up.ofp_port;
4255 if (ofp_port == ctx->flow.in_port) {
4260 compose_output_action__(ctx, ofp_port, false);
4261 } else if (!(ofport->up.opp.config & htonl(OFPPC_NO_FLOOD))) {
4262 compose_output_action(ctx, ofp_port);
4266 ctx->nf_output_iface = NF_OUT_FLOOD;
4270 compose_controller_action(struct action_xlate_ctx *ctx, int len)
4272 struct user_action_cookie cookie;
4274 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4275 cookie.type = USER_ACTION_COOKIE_CONTROLLER;
4277 cookie.n_output = 0;
4278 cookie.vlan_tci = 0;
4279 put_userspace_action(ctx->ofproto, ctx->odp_actions, &ctx->flow, &cookie);
4283 xlate_output_action__(struct action_xlate_ctx *ctx,
4284 uint16_t port, uint16_t max_len)
4286 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
4288 ctx->nf_output_iface = NF_OUT_DROP;
4292 compose_output_action(ctx, ctx->flow.in_port);
4295 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
4301 flood_packets(ctx, false);
4304 flood_packets(ctx, true);
4306 case OFPP_CONTROLLER:
4307 compose_controller_action(ctx, max_len);
4310 compose_output_action(ctx, OFPP_LOCAL);
4315 if (port != ctx->flow.in_port) {
4316 compose_output_action(ctx, port);
4321 if (prev_nf_output_iface == NF_OUT_FLOOD) {
4322 ctx->nf_output_iface = NF_OUT_FLOOD;
4323 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
4324 ctx->nf_output_iface = prev_nf_output_iface;
4325 } else if (prev_nf_output_iface != NF_OUT_DROP &&
4326 ctx->nf_output_iface != NF_OUT_FLOOD) {
4327 ctx->nf_output_iface = NF_OUT_MULTI;
4332 xlate_output_reg_action(struct action_xlate_ctx *ctx,
4333 const struct nx_action_output_reg *naor)
4337 ofp_port = nxm_read_field_bits(naor->src, naor->ofs_nbits, &ctx->flow);
4339 if (ofp_port <= UINT16_MAX) {
4340 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
4345 xlate_output_action(struct action_xlate_ctx *ctx,
4346 const struct ofp_action_output *oao)
4348 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
4352 xlate_enqueue_action(struct action_xlate_ctx *ctx,
4353 const struct ofp_action_enqueue *oae)
4356 uint32_t flow_priority, priority;
4359 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
4362 /* Fall back to ordinary output action. */
4363 xlate_output_action__(ctx, ntohs(oae->port), 0);
4367 /* Figure out datapath output port. */
4368 ofp_port = ntohs(oae->port);
4369 if (ofp_port == OFPP_IN_PORT) {
4370 ofp_port = ctx->flow.in_port;
4371 } else if (ofp_port == ctx->flow.in_port) {
4375 /* Add datapath actions. */
4376 flow_priority = ctx->flow.priority;
4377 ctx->flow.priority = priority;
4378 compose_output_action(ctx, ofp_port);
4379 ctx->flow.priority = flow_priority;
4381 /* Update NetFlow output port. */
4382 if (ctx->nf_output_iface == NF_OUT_DROP) {
4383 ctx->nf_output_iface = ofp_port;
4384 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
4385 ctx->nf_output_iface = NF_OUT_MULTI;
4390 xlate_set_queue_action(struct action_xlate_ctx *ctx,
4391 const struct nx_action_set_queue *nasq)
4396 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
4399 /* Couldn't translate queue to a priority, so ignore. A warning
4400 * has already been logged. */
4404 ctx->flow.priority = priority;
4407 struct xlate_reg_state {
4413 xlate_autopath(struct action_xlate_ctx *ctx,
4414 const struct nx_action_autopath *naa)
4416 uint16_t ofp_port = ntohl(naa->id);
4417 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
4419 if (!port || !port->bundle) {
4420 ofp_port = OFPP_NONE;
4421 } else if (port->bundle->bond) {
4422 /* Autopath does not support VLAN hashing. */
4423 struct ofport_dpif *slave = bond_choose_output_slave(
4424 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
4426 ofp_port = slave->up.ofp_port;
4429 autopath_execute(naa, &ctx->flow, ofp_port);
4433 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
4435 struct ofproto_dpif *ofproto = ofproto_;
4436 struct ofport_dpif *port;
4446 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
4449 port = get_ofp_port(ofproto, ofp_port);
4450 return port ? port->may_enable : false;
4455 xlate_learn_action(struct action_xlate_ctx *ctx,
4456 const struct nx_action_learn *learn)
4458 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
4459 struct ofputil_flow_mod fm;
4462 learn_execute(learn, &ctx->flow, &fm);
4464 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
4465 if (error && !VLOG_DROP_WARN(&rl)) {
4466 char *msg = ofputil_error_to_string(error);
4467 VLOG_WARN("learning action failed to modify flow table (%s)", msg);
4475 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
4477 if (port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
4478 ? htonl(OFPPC_NO_RECV_STP)
4479 : htonl(OFPPC_NO_RECV))) {
4483 /* Only drop packets here if both forwarding and learning are
4484 * disabled. If just learning is enabled, we need to have
4485 * OFPP_NORMAL and the learning action have a look at the packet
4486 * before we can drop it. */
4487 if (!stp_forward_in_state(port->stp_state)
4488 && !stp_learn_in_state(port->stp_state)) {
4496 do_xlate_actions(const union ofp_action *in, size_t n_in,
4497 struct action_xlate_ctx *ctx)
4499 const struct ofport_dpif *port;
4500 const union ofp_action *ia;
4503 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
4504 if (port && !may_receive(port, ctx)) {
4505 /* Drop this flow. */
4509 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
4510 const struct ofp_action_dl_addr *oada;
4511 const struct nx_action_resubmit *nar;
4512 const struct nx_action_set_tunnel *nast;
4513 const struct nx_action_set_queue *nasq;
4514 const struct nx_action_multipath *nam;
4515 const struct nx_action_autopath *naa;
4516 const struct nx_action_bundle *nab;
4517 const struct nx_action_output_reg *naor;
4518 enum ofputil_action_code code;
4525 code = ofputil_decode_action_unsafe(ia);
4527 case OFPUTIL_OFPAT_OUTPUT:
4528 xlate_output_action(ctx, &ia->output);
4531 case OFPUTIL_OFPAT_SET_VLAN_VID:
4532 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
4533 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
4536 case OFPUTIL_OFPAT_SET_VLAN_PCP:
4537 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
4538 ctx->flow.vlan_tci |= htons(
4539 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
4542 case OFPUTIL_OFPAT_STRIP_VLAN:
4543 ctx->flow.vlan_tci = htons(0);
4546 case OFPUTIL_OFPAT_SET_DL_SRC:
4547 oada = ((struct ofp_action_dl_addr *) ia);
4548 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
4551 case OFPUTIL_OFPAT_SET_DL_DST:
4552 oada = ((struct ofp_action_dl_addr *) ia);
4553 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
4556 case OFPUTIL_OFPAT_SET_NW_SRC:
4557 ctx->flow.nw_src = ia->nw_addr.nw_addr;
4560 case OFPUTIL_OFPAT_SET_NW_DST:
4561 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
4564 case OFPUTIL_OFPAT_SET_NW_TOS:
4565 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4566 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
4569 case OFPUTIL_OFPAT_SET_TP_SRC:
4570 ctx->flow.tp_src = ia->tp_port.tp_port;
4573 case OFPUTIL_OFPAT_SET_TP_DST:
4574 ctx->flow.tp_dst = ia->tp_port.tp_port;
4577 case OFPUTIL_OFPAT_ENQUEUE:
4578 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
4581 case OFPUTIL_NXAST_RESUBMIT:
4582 nar = (const struct nx_action_resubmit *) ia;
4583 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
4586 case OFPUTIL_NXAST_RESUBMIT_TABLE:
4587 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
4590 case OFPUTIL_NXAST_SET_TUNNEL:
4591 nast = (const struct nx_action_set_tunnel *) ia;
4592 tun_id = htonll(ntohl(nast->tun_id));
4593 ctx->flow.tun_id = tun_id;
4596 case OFPUTIL_NXAST_SET_QUEUE:
4597 nasq = (const struct nx_action_set_queue *) ia;
4598 xlate_set_queue_action(ctx, nasq);
4601 case OFPUTIL_NXAST_POP_QUEUE:
4602 ctx->flow.priority = ctx->original_priority;
4605 case OFPUTIL_NXAST_REG_MOVE:
4606 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
4610 case OFPUTIL_NXAST_REG_LOAD:
4611 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
4615 case OFPUTIL_NXAST_NOTE:
4616 /* Nothing to do. */
4619 case OFPUTIL_NXAST_SET_TUNNEL64:
4620 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
4621 ctx->flow.tun_id = tun_id;
4624 case OFPUTIL_NXAST_MULTIPATH:
4625 nam = (const struct nx_action_multipath *) ia;
4626 multipath_execute(nam, &ctx->flow);
4629 case OFPUTIL_NXAST_AUTOPATH:
4630 naa = (const struct nx_action_autopath *) ia;
4631 xlate_autopath(ctx, naa);
4634 case OFPUTIL_NXAST_BUNDLE:
4635 ctx->ofproto->has_bundle_action = true;
4636 nab = (const struct nx_action_bundle *) ia;
4637 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
4642 case OFPUTIL_NXAST_BUNDLE_LOAD:
4643 ctx->ofproto->has_bundle_action = true;
4644 nab = (const struct nx_action_bundle *) ia;
4645 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
4649 case OFPUTIL_NXAST_OUTPUT_REG:
4650 naor = (const struct nx_action_output_reg *) ia;
4651 xlate_output_reg_action(ctx, naor);
4654 case OFPUTIL_NXAST_LEARN:
4655 ctx->has_learn = true;
4656 if (ctx->may_learn) {
4657 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
4661 case OFPUTIL_NXAST_EXIT:
4667 /* We've let OFPP_NORMAL and the learning action look at the packet,
4668 * so drop it now if forwarding is disabled. */
4669 if (port && !stp_forward_in_state(port->stp_state)) {
4670 ofpbuf_clear(ctx->odp_actions);
4671 add_sflow_action(ctx);
4676 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
4677 struct ofproto_dpif *ofproto, const struct flow *flow,
4678 ovs_be16 initial_tci, const struct ofpbuf *packet)
4680 ctx->ofproto = ofproto;
4682 ctx->base_flow = ctx->flow;
4683 ctx->base_flow.tun_id = 0;
4684 ctx->base_flow.vlan_tci = initial_tci;
4685 ctx->packet = packet;
4686 ctx->may_learn = packet != NULL;
4687 ctx->resubmit_hook = NULL;
4690 static struct ofpbuf *
4691 xlate_actions(struct action_xlate_ctx *ctx,
4692 const union ofp_action *in, size_t n_in)
4694 struct flow orig_flow = ctx->flow;
4696 COVERAGE_INC(ofproto_dpif_xlate);
4698 ctx->odp_actions = ofpbuf_new(512);
4699 ofpbuf_reserve(ctx->odp_actions, NL_A_U32_SIZE);
4701 ctx->may_set_up_flow = true;
4702 ctx->has_learn = false;
4703 ctx->has_normal = false;
4704 ctx->nf_output_iface = NF_OUT_DROP;
4707 ctx->original_priority = ctx->flow.priority;
4711 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
4712 switch (ctx->ofproto->up.frag_handling) {
4713 case OFPC_FRAG_NORMAL:
4714 /* We must pretend that transport ports are unavailable. */
4715 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
4716 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
4719 case OFPC_FRAG_DROP:
4720 return ctx->odp_actions;
4722 case OFPC_FRAG_REASM:
4725 case OFPC_FRAG_NX_MATCH:
4726 /* Nothing to do. */
4731 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
4732 ctx->may_set_up_flow = false;
4733 return ctx->odp_actions;
4735 add_sflow_action(ctx);
4736 do_xlate_actions(in, n_in, ctx);
4738 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
4739 ctx->odp_actions->data,
4740 ctx->odp_actions->size)) {
4741 ctx->may_set_up_flow = false;
4743 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
4745 compose_output_action(ctx, OFPP_LOCAL);
4748 add_mirror_actions(ctx, &orig_flow);
4749 fix_sflow_action(ctx);
4752 return ctx->odp_actions;
4755 /* OFPP_NORMAL implementation. */
4757 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
4759 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
4760 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
4761 * the bundle on which the packet was received, returns the VLAN to which the
4764 * Both 'vid' and the return value are in the range 0...4095. */
4766 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
4768 switch (in_bundle->vlan_mode) {
4769 case PORT_VLAN_ACCESS:
4770 return in_bundle->vlan;
4773 case PORT_VLAN_TRUNK:
4776 case PORT_VLAN_NATIVE_UNTAGGED:
4777 case PORT_VLAN_NATIVE_TAGGED:
4778 return vid ? vid : in_bundle->vlan;
4785 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
4786 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
4789 * 'vid' should be the VID obtained from the 802.1Q header that was received as
4790 * part of a packet (specify 0 if there was no 802.1Q header), in the range
4793 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
4795 switch (in_bundle->vlan_mode) {
4796 case PORT_VLAN_ACCESS:
4799 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4800 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
4801 "packet received on port %s configured as VLAN "
4802 "%"PRIu16" access port",
4803 in_bundle->ofproto->up.name, vid,
4804 in_bundle->name, in_bundle->vlan);
4810 case PORT_VLAN_NATIVE_UNTAGGED:
4811 case PORT_VLAN_NATIVE_TAGGED:
4813 /* Port must always carry its native VLAN. */
4817 case PORT_VLAN_TRUNK:
4818 if (!ofbundle_includes_vlan(in_bundle, vid)) {
4820 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4821 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
4822 "received on port %s not configured for trunking "
4824 in_bundle->ofproto->up.name, vid,
4825 in_bundle->name, vid);
4837 /* Given 'vlan', the VLAN that a packet belongs to, and
4838 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
4839 * that should be included in the 802.1Q header. (If the return value is 0,
4840 * then the 802.1Q header should only be included in the packet if there is a
4843 * Both 'vlan' and the return value are in the range 0...4095. */
4845 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
4847 switch (out_bundle->vlan_mode) {
4848 case PORT_VLAN_ACCESS:
4851 case PORT_VLAN_TRUNK:
4852 case PORT_VLAN_NATIVE_TAGGED:
4855 case PORT_VLAN_NATIVE_UNTAGGED:
4856 return vlan == out_bundle->vlan ? 0 : vlan;
4864 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
4867 struct ofport_dpif *port;
4869 ovs_be16 tci, old_tci;
4871 vid = output_vlan_to_vid(out_bundle, vlan);
4872 if (!out_bundle->bond) {
4873 port = ofbundle_get_a_port(out_bundle);
4875 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
4878 /* No slaves enabled, so drop packet. */
4883 old_tci = ctx->flow.vlan_tci;
4885 if (tci || out_bundle->use_priority_tags) {
4886 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
4888 tci |= htons(VLAN_CFI);
4891 ctx->flow.vlan_tci = tci;
4893 compose_output_action(ctx, port->up.ofp_port);
4894 ctx->flow.vlan_tci = old_tci;
4898 mirror_mask_ffs(mirror_mask_t mask)
4900 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
4905 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
4907 return (bundle->vlan_mode != PORT_VLAN_ACCESS
4908 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
4912 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
4914 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
4917 /* Returns an arbitrary interface within 'bundle'. */
4918 static struct ofport_dpif *
4919 ofbundle_get_a_port(const struct ofbundle *bundle)
4921 return CONTAINER_OF(list_front(&bundle->ports),
4922 struct ofport_dpif, bundle_node);
4926 vlan_is_mirrored(const struct ofmirror *m, int vlan)
4928 return !m->vlans || bitmap_is_set(m->vlans, vlan);
4931 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
4932 * to a VLAN. In general most packets may be mirrored but we want to drop
4933 * protocols that may confuse switches. */
4935 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
4937 /* If you change this function's behavior, please update corresponding
4938 * documentation in vswitch.xml at the same time. */
4939 if (dst[0] != 0x01) {
4940 /* All the currently banned MACs happen to start with 01 currently, so
4941 * this is a quick way to eliminate most of the good ones. */
4943 if (eth_addr_is_reserved(dst)) {
4944 /* Drop STP, IEEE pause frames, and other reserved protocols
4945 * (01-80-c2-00-00-0x). */
4949 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
4951 if ((dst[3] & 0xfe) == 0xcc &&
4952 (dst[4] & 0xfe) == 0xcc &&
4953 (dst[5] & 0xfe) == 0xcc) {
4954 /* Drop the following protocols plus others following the same
4957 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
4958 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
4959 STP Uplink Fast (01-00-0c-cd-cd-cd) */
4963 if (!(dst[3] | dst[4] | dst[5])) {
4964 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
4973 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
4975 struct ofproto_dpif *ofproto = ctx->ofproto;
4976 mirror_mask_t mirrors;
4977 struct ofbundle *in_bundle;
4980 const struct nlattr *a;
4983 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
4984 ctx->packet != NULL);
4988 mirrors = in_bundle->src_mirrors;
4990 /* Drop frames on bundles reserved for mirroring. */
4991 if (in_bundle->mirror_out) {
4992 if (ctx->packet != NULL) {
4993 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4994 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
4995 "%s, which is reserved exclusively for mirroring",
4996 ctx->ofproto->up.name, in_bundle->name);
5002 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5003 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5006 vlan = input_vid_to_vlan(in_bundle, vid);
5008 /* Look at the output ports to check for destination selections. */
5010 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5011 ctx->odp_actions->size) {
5012 enum ovs_action_attr type = nl_attr_type(a);
5013 struct ofport_dpif *ofport;
5015 if (type != OVS_ACTION_ATTR_OUTPUT) {
5019 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5020 if (ofport && ofport->bundle) {
5021 mirrors |= ofport->bundle->dst_mirrors;
5029 /* Restore the original packet before adding the mirror actions. */
5030 ctx->flow = *orig_flow;
5035 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5037 if (!vlan_is_mirrored(m, vlan)) {
5038 mirrors &= mirrors - 1;
5042 mirrors &= ~m->dup_mirrors;
5043 ctx->mirrors |= m->dup_mirrors;
5045 output_normal(ctx, m->out, vlan);
5046 } else if (eth_dst_may_rspan(orig_flow->dl_dst)
5047 && vlan != m->out_vlan) {
5048 struct ofbundle *bundle;
5050 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5051 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5052 && !bundle->mirror_out) {
5053 output_normal(ctx, bundle, m->out_vlan);
5061 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5062 uint64_t packets, uint64_t bytes)
5068 for (; mirrors; mirrors &= mirrors - 1) {
5071 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5074 /* In normal circumstances 'm' will not be NULL. However,
5075 * if mirrors are reconfigured, we can temporarily get out
5076 * of sync in facet_revalidate(). We could "correct" the
5077 * mirror list before reaching here, but doing that would
5078 * not properly account the traffic stats we've currently
5079 * accumulated for previous mirror configuration. */
5083 m->packet_count += packets;
5084 m->byte_count += bytes;
5088 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
5089 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
5090 * indicate this; newer upstream kernels use gratuitous ARP requests. */
5092 is_gratuitous_arp(const struct flow *flow)
5094 return (flow->dl_type == htons(ETH_TYPE_ARP)
5095 && eth_addr_is_broadcast(flow->dl_dst)
5096 && (flow->nw_proto == ARP_OP_REPLY
5097 || (flow->nw_proto == ARP_OP_REQUEST
5098 && flow->nw_src == flow->nw_dst)));
5102 update_learning_table(struct ofproto_dpif *ofproto,
5103 const struct flow *flow, int vlan,
5104 struct ofbundle *in_bundle)
5106 struct mac_entry *mac;
5108 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
5112 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
5113 if (is_gratuitous_arp(flow)) {
5114 /* We don't want to learn from gratuitous ARP packets that are
5115 * reflected back over bond slaves so we lock the learning table. */
5116 if (!in_bundle->bond) {
5117 mac_entry_set_grat_arp_lock(mac);
5118 } else if (mac_entry_is_grat_arp_locked(mac)) {
5123 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
5124 /* The log messages here could actually be useful in debugging,
5125 * so keep the rate limit relatively high. */
5126 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5127 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
5128 "on port %s in VLAN %d",
5129 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
5130 in_bundle->name, vlan);
5132 mac->port.p = in_bundle;
5133 tag_set_add(&ofproto->revalidate_set,
5134 mac_learning_changed(ofproto->ml, mac));
5138 static struct ofbundle *
5139 lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn)
5141 struct ofport_dpif *ofport;
5143 /* Find the port and bundle for the received packet. */
5144 ofport = get_ofp_port(ofproto, in_port);
5145 if (ofport && ofport->bundle) {
5146 return ofport->bundle;
5149 /* Odd. A few possible reasons here:
5151 * - We deleted a port but there are still a few packets queued up
5154 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
5155 * we don't know about.
5157 * - The ofproto client didn't configure the port as part of a bundle.
5160 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5162 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
5163 "port %"PRIu16, ofproto->up.name, in_port);
5168 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
5169 * dropped. Returns true if they may be forwarded, false if they should be
5172 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
5173 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
5175 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
5176 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
5177 * checked by input_vid_is_valid().
5179 * May also add tags to '*tags', although the current implementation only does
5180 * so in one special case.
5183 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
5184 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
5186 struct ofbundle *in_bundle = in_port->bundle;
5188 /* Drop frames for reserved multicast addresses
5189 * only if forward_bpdu option is absent. */
5190 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
5194 if (in_bundle->bond) {
5195 struct mac_entry *mac;
5197 switch (bond_check_admissibility(in_bundle->bond, in_port,
5198 flow->dl_dst, tags)) {
5205 case BV_DROP_IF_MOVED:
5206 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
5207 if (mac && mac->port.p != in_bundle &&
5208 (!is_gratuitous_arp(flow)
5209 || mac_entry_is_grat_arp_locked(mac))) {
5220 xlate_normal(struct action_xlate_ctx *ctx)
5222 struct ofport_dpif *in_port;
5223 struct ofbundle *in_bundle;
5224 struct mac_entry *mac;
5228 ctx->has_normal = true;
5230 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
5231 ctx->packet != NULL);
5236 /* We know 'in_port' exists, since lookup_input_bundle() succeeded. */
5237 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5239 /* Drop malformed frames. */
5240 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
5241 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
5242 if (ctx->packet != NULL) {
5243 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5244 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
5245 "VLAN tag received on port %s",
5246 ctx->ofproto->up.name, in_bundle->name);
5251 /* Drop frames on bundles reserved for mirroring. */
5252 if (in_bundle->mirror_out) {
5253 if (ctx->packet != NULL) {
5254 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5255 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5256 "%s, which is reserved exclusively for mirroring",
5257 ctx->ofproto->up.name, in_bundle->name);
5263 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
5264 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5267 vlan = input_vid_to_vlan(in_bundle, vid);
5269 /* Check other admissibility requirements. */
5270 if (!is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
5274 /* Learn source MAC. */
5275 if (ctx->may_learn) {
5276 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
5279 /* Determine output bundle. */
5280 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
5283 if (mac->port.p != in_bundle) {
5284 output_normal(ctx, mac->port.p, vlan);
5286 } else if (!ctx->packet && !eth_addr_is_multicast(ctx->flow.dl_dst)) {
5287 /* If we are revalidating but don't have a learning entry then eject
5288 * the flow. Installing a flow that floods packets opens up a window
5289 * of time where we could learn from a packet reflected on a bond and
5290 * blackhole packets before the learning table is updated to reflect
5291 * the correct port. */
5292 ctx->may_set_up_flow = false;
5295 struct ofbundle *bundle;
5297 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
5298 if (bundle != in_bundle
5299 && ofbundle_includes_vlan(bundle, vlan)
5300 && bundle->floodable
5301 && !bundle->mirror_out) {
5302 output_normal(ctx, bundle, vlan);
5305 ctx->nf_output_iface = NF_OUT_FLOOD;
5309 /* Optimized flow revalidation.
5311 * It's a difficult problem, in general, to tell which facets need to have
5312 * their actions recalculated whenever the OpenFlow flow table changes. We
5313 * don't try to solve that general problem: for most kinds of OpenFlow flow
5314 * table changes, we recalculate the actions for every facet. This is
5315 * relatively expensive, but it's good enough if the OpenFlow flow table
5316 * doesn't change very often.
5318 * However, we can expect one particular kind of OpenFlow flow table change to
5319 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5320 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5321 * table, we add a special case that applies to flow tables in which every rule
5322 * has the same form (that is, the same wildcards), except that the table is
5323 * also allowed to have a single "catch-all" flow that matches all packets. We
5324 * optimize this case by tagging all of the facets that resubmit into the table
5325 * and invalidating the same tag whenever a flow changes in that table. The
5326 * end result is that we revalidate just the facets that need it (and sometimes
5327 * a few more, but not all of the facets or even all of the facets that
5328 * resubmit to the table modified by MAC learning). */
5330 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
5331 * into an OpenFlow table with the given 'basis'. */
5333 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
5336 if (flow_wildcards_is_catchall(wc)) {
5339 struct flow tag_flow = *flow;
5340 flow_zero_wildcards(&tag_flow, wc);
5341 return tag_create_deterministic(flow_hash(&tag_flow, secret));
5345 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5346 * taggability of that table.
5348 * This function must be called after *each* change to a flow table. If you
5349 * skip calling it on some changes then the pointer comparisons at the end can
5350 * be invalid if you get unlucky. For example, if a flow removal causes a
5351 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5352 * different wildcards to be created with the same address, then this function
5353 * will incorrectly skip revalidation. */
5355 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5357 struct table_dpif *table = &ofproto->tables[table_id];
5358 const struct classifier *cls = &ofproto->up.tables[table_id];
5359 struct cls_table *catchall, *other;
5360 struct cls_table *t;
5362 catchall = other = NULL;
5364 switch (hmap_count(&cls->tables)) {
5366 /* We could tag this OpenFlow table but it would make the logic a
5367 * little harder and it's a corner case that doesn't seem worth it
5373 HMAP_FOR_EACH (t, hmap_node, &cls->tables) {
5374 if (cls_table_is_catchall(t)) {
5376 } else if (!other) {
5379 /* Indicate that we can't tag this by setting both tables to
5380 * NULL. (We know that 'catchall' is already NULL.) */
5387 /* Can't tag this table. */
5391 if (table->catchall_table != catchall || table->other_table != other) {
5392 table->catchall_table = catchall;
5393 table->other_table = other;
5394 ofproto->need_revalidate = true;
5398 /* Given 'rule' that has changed in some way (either it is a rule being
5399 * inserted, a rule being deleted, or a rule whose actions are being
5400 * modified), marks facets for revalidation to ensure that packets will be
5401 * forwarded correctly according to the new state of the flow table.
5403 * This function must be called after *each* change to a flow table. See
5404 * the comment on table_update_taggable() for more information. */
5406 rule_invalidate(const struct rule_dpif *rule)
5408 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5410 table_update_taggable(ofproto, rule->up.table_id);
5412 if (!ofproto->need_revalidate) {
5413 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5415 if (table->other_table && rule->tag) {
5416 tag_set_add(&ofproto->revalidate_set, rule->tag);
5418 ofproto->need_revalidate = true;
5424 set_frag_handling(struct ofproto *ofproto_,
5425 enum ofp_config_flags frag_handling)
5427 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5429 if (frag_handling != OFPC_FRAG_REASM) {
5430 ofproto->need_revalidate = true;
5438 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5439 const struct flow *flow,
5440 const union ofp_action *ofp_actions, size_t n_ofp_actions)
5442 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5445 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
5446 return ofp_mkerr_nicira(OFPET_BAD_REQUEST, NXBRC_BAD_IN_PORT);
5449 error = validate_actions(ofp_actions, n_ofp_actions, flow,
5450 ofproto->max_ports);
5452 struct odputil_keybuf keybuf;
5453 struct action_xlate_ctx ctx;
5454 struct ofpbuf *odp_actions;
5457 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5458 odp_flow_key_from_flow(&key, flow);
5460 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, packet);
5461 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
5462 dpif_execute(ofproto->dpif, key.data, key.size,
5463 odp_actions->data, odp_actions->size, packet);
5464 ofpbuf_delete(odp_actions);
5472 set_netflow(struct ofproto *ofproto_,
5473 const struct netflow_options *netflow_options)
5475 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5477 if (netflow_options) {
5478 if (!ofproto->netflow) {
5479 ofproto->netflow = netflow_create();
5481 return netflow_set_options(ofproto->netflow, netflow_options);
5483 netflow_destroy(ofproto->netflow);
5484 ofproto->netflow = NULL;
5490 get_netflow_ids(const struct ofproto *ofproto_,
5491 uint8_t *engine_type, uint8_t *engine_id)
5493 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5495 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
5499 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5501 if (!facet_is_controller_flow(facet) &&
5502 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5503 struct subfacet *subfacet;
5504 struct ofexpired expired;
5506 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5507 if (subfacet->installed) {
5508 struct dpif_flow_stats stats;
5510 subfacet_install(ofproto, subfacet, subfacet->actions,
5511 subfacet->actions_len, &stats);
5512 subfacet_update_stats(ofproto, subfacet, &stats);
5516 expired.flow = facet->flow;
5517 expired.packet_count = facet->packet_count;
5518 expired.byte_count = facet->byte_count;
5519 expired.used = facet->used;
5520 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5525 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5527 struct facet *facet;
5529 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5530 send_active_timeout(ofproto, facet);
5534 static struct ofproto_dpif *
5535 ofproto_dpif_lookup(const char *name)
5537 struct ofproto_dpif *ofproto;
5539 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5540 hash_string(name, 0), &all_ofproto_dpifs) {
5541 if (!strcmp(ofproto->up.name, name)) {
5549 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn,
5550 const char *args, void *aux OVS_UNUSED)
5552 const struct ofproto_dpif *ofproto;
5554 ofproto = ofproto_dpif_lookup(args);
5556 unixctl_command_reply(conn, 501, "no such bridge");
5559 mac_learning_flush(ofproto->ml);
5561 unixctl_command_reply(conn, 200, "table successfully flushed");
5565 ofproto_unixctl_fdb_show(struct unixctl_conn *conn,
5566 const char *args, void *aux OVS_UNUSED)
5568 struct ds ds = DS_EMPTY_INITIALIZER;
5569 const struct ofproto_dpif *ofproto;
5570 const struct mac_entry *e;
5572 ofproto = ofproto_dpif_lookup(args);
5574 unixctl_command_reply(conn, 501, "no such bridge");
5578 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5579 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5580 struct ofbundle *bundle = e->port.p;
5581 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
5582 ofbundle_get_a_port(bundle)->odp_port,
5583 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
5585 unixctl_command_reply(conn, 200, ds_cstr(&ds));
5589 struct ofproto_trace {
5590 struct action_xlate_ctx ctx;
5596 trace_format_rule(struct ds *result, uint8_t table_id, int level,
5597 const struct rule_dpif *rule)
5599 ds_put_char_multiple(result, '\t', level);
5601 ds_put_cstr(result, "No match\n");
5605 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5606 table_id, ntohll(rule->up.flow_cookie));
5607 cls_rule_format(&rule->up.cr, result);
5608 ds_put_char(result, '\n');
5610 ds_put_char_multiple(result, '\t', level);
5611 ds_put_cstr(result, "OpenFlow ");
5612 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
5613 ds_put_char(result, '\n');
5617 trace_format_flow(struct ds *result, int level, const char *title,
5618 struct ofproto_trace *trace)
5620 ds_put_char_multiple(result, '\t', level);
5621 ds_put_format(result, "%s: ", title);
5622 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
5623 ds_put_cstr(result, "unchanged");
5625 flow_format(result, &trace->ctx.flow);
5626 trace->flow = trace->ctx.flow;
5628 ds_put_char(result, '\n');
5632 trace_format_regs(struct ds *result, int level, const char *title,
5633 struct ofproto_trace *trace)
5637 ds_put_char_multiple(result, '\t', level);
5638 ds_put_format(result, "%s:", title);
5639 for (i = 0; i < FLOW_N_REGS; i++) {
5640 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5642 ds_put_char(result, '\n');
5646 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
5648 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
5649 struct ds *result = trace->result;
5651 ds_put_char(result, '\n');
5652 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
5653 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
5654 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
5658 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
5659 void *aux OVS_UNUSED)
5661 char *dpname, *arg1, *arg2, *arg3, *arg4;
5662 char *args = xstrdup(args_);
5663 char *save_ptr = NULL;
5664 struct ofproto_dpif *ofproto;
5665 struct ofpbuf odp_key;
5666 struct ofpbuf *packet;
5667 struct rule_dpif *rule;
5668 ovs_be16 initial_tci;
5674 ofpbuf_init(&odp_key, 0);
5677 dpname = strtok_r(args, " ", &save_ptr);
5679 unixctl_command_reply(conn, 501, "Bad command syntax");
5683 ofproto = ofproto_dpif_lookup(dpname);
5685 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
5689 arg1 = strtok_r(NULL, " ", &save_ptr);
5690 arg2 = strtok_r(NULL, " ", &save_ptr);
5691 arg3 = strtok_r(NULL, " ", &save_ptr);
5692 arg4 = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
5693 if (dpname && arg1 && (!arg2 || !strcmp(arg2, "-generate")) && !arg3) {
5694 /* ofproto/trace dpname flow [-generate] */
5697 /* Convert string to datapath key. */
5698 ofpbuf_init(&odp_key, 0);
5699 error = odp_flow_key_from_string(arg1, NULL, &odp_key);
5701 unixctl_command_reply(conn, 501, "Bad flow syntax");
5705 /* Convert odp_key to flow. */
5706 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
5707 odp_key.size, &flow,
5708 &initial_tci, NULL);
5709 if (error == ODP_FIT_ERROR) {
5710 unixctl_command_reply(conn, 501, "Invalid flow");
5714 /* Generate a packet, if requested. */
5716 packet = ofpbuf_new(0);
5717 flow_compose(packet, &flow);
5719 } else if (dpname && arg1 && arg2 && arg3 && arg4) {
5720 /* ofproto/trace dpname priority tun_id in_port packet */
5725 priority = atoi(arg1);
5726 tun_id = htonll(strtoull(arg2, NULL, 0));
5727 in_port = ofp_port_to_odp_port(atoi(arg3));
5729 packet = ofpbuf_new(strlen(args) / 2);
5730 arg4 = ofpbuf_put_hex(packet, arg4, NULL);
5731 arg4 += strspn(arg4, " ");
5732 if (*arg4 != '\0') {
5733 unixctl_command_reply(conn, 501, "Trailing garbage in command");
5736 if (packet->size < ETH_HEADER_LEN) {
5737 unixctl_command_reply(conn, 501,
5738 "Packet data too short for Ethernet");
5742 ds_put_cstr(&result, "Packet: ");
5743 s = ofp_packet_to_string(packet->data, packet->size, packet->size);
5744 ds_put_cstr(&result, s);
5747 flow_extract(packet, priority, tun_id, in_port, &flow);
5748 initial_tci = flow.vlan_tci;
5750 unixctl_command_reply(conn, 501, "Bad command syntax");
5754 ds_put_cstr(&result, "Flow: ");
5755 flow_format(&result, &flow);
5756 ds_put_char(&result, '\n');
5758 rule = rule_dpif_lookup(ofproto, &flow, 0);
5759 trace_format_rule(&result, 0, 0, rule);
5761 struct ofproto_trace trace;
5762 struct ofpbuf *odp_actions;
5764 trace.result = &result;
5766 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, initial_tci, packet);
5767 trace.ctx.resubmit_hook = trace_resubmit;
5768 odp_actions = xlate_actions(&trace.ctx,
5769 rule->up.actions, rule->up.n_actions);
5771 ds_put_char(&result, '\n');
5772 trace_format_flow(&result, 0, "Final flow", &trace);
5773 ds_put_cstr(&result, "Datapath actions: ");
5774 format_odp_actions(&result, odp_actions->data, odp_actions->size);
5775 ofpbuf_delete(odp_actions);
5777 if (!trace.ctx.may_set_up_flow) {
5779 ds_put_cstr(&result, "\nThis flow is not cachable.");
5781 ds_put_cstr(&result, "\nThe datapath actions are incomplete--"
5782 "for complete actions, please supply a packet.");
5787 unixctl_command_reply(conn, 200, ds_cstr(&result));
5790 ds_destroy(&result);
5791 ofpbuf_delete(packet);
5792 ofpbuf_uninit(&odp_key);
5797 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED,
5798 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
5801 unixctl_command_reply(conn, 200, NULL);
5805 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED,
5806 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
5809 unixctl_command_reply(conn, 200, NULL);
5813 ofproto_dpif_unixctl_init(void)
5815 static bool registered;
5821 unixctl_command_register("ofproto/trace",
5822 "bridge {tun_id in_port packet | odp_flow [-generate]}",
5823 ofproto_unixctl_trace, NULL);
5824 unixctl_command_register("fdb/flush", "bridge", ofproto_unixctl_fdb_flush,
5826 unixctl_command_register("fdb/show", "bridge", ofproto_unixctl_fdb_show,
5828 unixctl_command_register("ofproto/clog", "", ofproto_dpif_clog, NULL);
5829 unixctl_command_register("ofproto/unclog", "", ofproto_dpif_unclog, NULL);
5832 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5834 * This is deprecated. It is only for compatibility with broken device drivers
5835 * in old versions of Linux that do not properly support VLANs when VLAN
5836 * devices are not used. When broken device drivers are no longer in
5837 * widespread use, we will delete these interfaces. */
5840 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
5842 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
5843 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
5845 if (realdev_ofp_port == ofport->realdev_ofp_port
5846 && vid == ofport->vlandev_vid) {
5850 ofproto->need_revalidate = true;
5852 if (ofport->realdev_ofp_port) {
5855 if (realdev_ofp_port && ofport->bundle) {
5856 /* vlandevs are enslaved to their realdevs, so they are not allowed to
5857 * themselves be part of a bundle. */
5858 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
5861 ofport->realdev_ofp_port = realdev_ofp_port;
5862 ofport->vlandev_vid = vid;
5864 if (realdev_ofp_port) {
5865 vsp_add(ofport, realdev_ofp_port, vid);
5872 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
5874 return hash_2words(realdev_ofp_port, vid);
5878 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
5879 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
5881 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
5882 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
5883 int vid = vlan_tci_to_vid(vlan_tci);
5884 const struct vlan_splinter *vsp;
5886 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
5887 hash_realdev_vid(realdev_ofp_port, vid),
5888 &ofproto->realdev_vid_map) {
5889 if (vsp->realdev_ofp_port == realdev_ofp_port
5890 && vsp->vid == vid) {
5891 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
5895 return realdev_odp_port;
5898 static struct vlan_splinter *
5899 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
5901 struct vlan_splinter *vsp;
5903 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
5904 &ofproto->vlandev_map) {
5905 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
5914 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
5915 uint16_t vlandev_ofp_port, int *vid)
5917 if (!hmap_is_empty(&ofproto->vlandev_map)) {
5918 const struct vlan_splinter *vsp;
5920 vsp = vlandev_find(ofproto, vlandev_ofp_port);
5925 return vsp->realdev_ofp_port;
5932 vsp_remove(struct ofport_dpif *port)
5934 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
5935 struct vlan_splinter *vsp;
5937 vsp = vlandev_find(ofproto, port->up.ofp_port);
5939 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
5940 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
5943 port->realdev_ofp_port = 0;
5945 VLOG_ERR("missing vlan device record");
5950 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
5952 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
5954 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
5955 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
5956 == realdev_ofp_port)) {
5957 struct vlan_splinter *vsp;
5959 vsp = xmalloc(sizeof *vsp);
5960 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
5961 hash_int(port->up.ofp_port, 0));
5962 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
5963 hash_realdev_vid(realdev_ofp_port, vid));
5964 vsp->realdev_ofp_port = realdev_ofp_port;
5965 vsp->vlandev_ofp_port = port->up.ofp_port;
5968 port->realdev_ofp_port = realdev_ofp_port;
5970 VLOG_ERR("duplicate vlan device record");
5974 const struct ofproto_class ofproto_dpif_class = {
6002 port_is_lacp_current,
6003 NULL, /* rule_choose_table */
6010 rule_modify_actions,
6018 get_cfm_remote_mpids,
6022 get_stp_port_status,
6029 is_mirror_output_bundle,
6030 forward_bpdu_changed,