2 * Copyright (c) 2009, 2010, 2011 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-provider.h"
26 #include "byte-order.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
36 #include "mac-learning.h"
37 #include "multipath.h"
44 #include "ofp-print.h"
45 #include "ofproto-dpif-sflow.h"
46 #include "poll-loop.h"
48 #include "unaligned.h"
50 #include "vlan-bitmap.h"
53 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
55 COVERAGE_DEFINE(ofproto_dpif_ctlr_action);
56 COVERAGE_DEFINE(ofproto_dpif_expired);
57 COVERAGE_DEFINE(ofproto_dpif_no_packet_in);
58 COVERAGE_DEFINE(ofproto_dpif_xlate);
59 COVERAGE_DEFINE(facet_changed_rule);
60 COVERAGE_DEFINE(facet_invalidated);
61 COVERAGE_DEFINE(facet_revalidate);
62 COVERAGE_DEFINE(facet_unexpected);
64 /* Maximum depth of flow table recursion (due to resubmit actions) in a
65 * flow translation. */
66 #define MAX_RESUBMIT_RECURSION 32
68 /* Number of implemented OpenFlow tables. */
69 enum { N_TABLES = 255 };
70 BUILD_ASSERT_DECL(N_TABLES >= 1 && N_TABLES <= 255);
78 long long int used; /* Time last used; time created if not used. */
82 * - Do include packets and bytes from facets that have been deleted or
83 * whose own statistics have been folded into the rule.
85 * - Do include packets and bytes sent "by hand" that were accounted to
86 * the rule without any facet being involved (this is a rare corner
87 * case in rule_execute()).
89 * - Do not include packet or bytes that can be obtained from any facet's
90 * packet_count or byte_count member or that can be obtained from the
91 * datapath by, e.g., dpif_flow_get() for any subfacet.
93 uint64_t packet_count; /* Number of packets received. */
94 uint64_t byte_count; /* Number of bytes received. */
96 tag_type tag; /* Caches rule_calculate_tag() result. */
98 struct list facets; /* List of "struct facet"s. */
101 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
103 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
106 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
107 const struct flow *, uint8_t table);
109 static void flow_push_stats(const struct rule_dpif *, const struct flow *,
110 uint64_t packets, uint64_t bytes,
113 static uint32_t rule_calculate_tag(const struct flow *,
114 const struct flow_wildcards *,
116 static void rule_invalidate(const struct rule_dpif *);
118 #define MAX_MIRRORS 32
119 typedef uint32_t mirror_mask_t;
120 #define MIRROR_MASK_C(X) UINT32_C(X)
121 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
123 struct ofproto_dpif *ofproto; /* Owning ofproto. */
124 size_t idx; /* In ofproto's "mirrors" array. */
125 void *aux; /* Key supplied by ofproto's client. */
126 char *name; /* Identifier for log messages. */
128 /* Selection criteria. */
129 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
130 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
131 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
133 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
134 struct ofbundle *out; /* Output port or NULL. */
135 int out_vlan; /* Output VLAN or -1. */
136 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
139 int64_t packet_count; /* Number of packets sent. */
140 int64_t byte_count; /* Number of bytes sent. */
143 static void mirror_destroy(struct ofmirror *);
144 static void update_mirror_stats(struct ofproto_dpif *ofproto,
145 mirror_mask_t mirrors,
146 uint64_t packets, uint64_t bytes);
149 struct ofproto_dpif *ofproto; /* Owning ofproto. */
150 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
151 void *aux; /* Key supplied by ofproto's client. */
152 char *name; /* Identifier for log messages. */
155 struct list ports; /* Contains "struct ofport"s. */
156 enum port_vlan_mode vlan_mode; /* VLAN mode */
157 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
158 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
159 * NULL if all VLANs are trunked. */
160 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
161 struct bond *bond; /* Nonnull iff more than one port. */
162 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
165 bool floodable; /* True if no port has OFPPC_NO_FLOOD set. */
167 /* Port mirroring info. */
168 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
169 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
170 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
173 static void bundle_remove(struct ofport *);
174 static void bundle_update(struct ofbundle *);
175 static void bundle_destroy(struct ofbundle *);
176 static void bundle_del_port(struct ofport_dpif *);
177 static void bundle_run(struct ofbundle *);
178 static void bundle_wait(struct ofbundle *);
179 static struct ofport_dpif *lookup_input_bundle(struct ofproto_dpif *,
180 uint16_t in_port, bool warn);
182 static void stp_run(struct ofproto_dpif *ofproto);
183 static void stp_wait(struct ofproto_dpif *ofproto);
185 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
187 struct action_xlate_ctx {
188 /* action_xlate_ctx_init() initializes these members. */
191 struct ofproto_dpif *ofproto;
193 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
194 * this flow when actions change header fields. */
197 /* The packet corresponding to 'flow', or a null pointer if we are
198 * revalidating without a packet to refer to. */
199 const struct ofpbuf *packet;
201 /* Should OFPP_NORMAL MAC learning and NXAST_LEARN actions execute? We
202 * want to execute them if we are actually processing a packet, or if we
203 * are accounting for packets that the datapath has processed, but not if
204 * we are just revalidating. */
207 /* If nonnull, called just before executing a resubmit action.
209 * This is normally null so the client has to set it manually after
210 * calling action_xlate_ctx_init(). */
211 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *);
213 /* xlate_actions() initializes and uses these members. The client might want
214 * to look at them after it returns. */
216 struct ofpbuf *odp_actions; /* Datapath actions. */
217 tag_type tags; /* Tags associated with actions. */
218 bool may_set_up_flow; /* True ordinarily; false if the actions must
219 * be reassessed for every packet. */
220 bool has_learn; /* Actions include NXAST_LEARN? */
221 bool has_normal; /* Actions output to OFPP_NORMAL? */
222 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
223 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
225 /* xlate_actions() initializes and uses these members, but the client has no
226 * reason to look at them. */
228 int recurse; /* Recursion level, via xlate_table_action. */
229 struct flow base_flow; /* Flow at the last commit. */
230 uint32_t original_priority; /* Priority when packet arrived. */
231 uint8_t table_id; /* OpenFlow table ID where flow was found. */
232 uint32_t sflow_n_outputs; /* Number of output ports. */
233 uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
234 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
235 bool exit; /* No further actions should be processed. */
238 static void action_xlate_ctx_init(struct action_xlate_ctx *,
239 struct ofproto_dpif *, const struct flow *,
240 ovs_be16 initial_tci, const struct ofpbuf *);
241 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
242 const union ofp_action *in, size_t n_in);
244 /* An exact-match instantiation of an OpenFlow flow.
246 * A facet associates a "struct flow", which represents the Open vSwitch
247 * userspace idea of an exact-match flow, with one or more subfacets. Each
248 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
249 * the facet. When the kernel module (or other dpif implementation) and Open
250 * vSwitch userspace agree on the definition of a flow key, there is exactly
251 * one subfacet per facet. If the dpif implementation supports more-specific
252 * flow matching than userspace, however, a facet can have more than one
253 * subfacet, each of which corresponds to some distinction in flow that
254 * userspace simply doesn't understand.
256 * Flow expiration works in terms of subfacets, so a facet must have at least
257 * one subfacet or it will never expire, leaking memory. */
260 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
261 struct list list_node; /* In owning rule's 'facets' list. */
262 struct rule_dpif *rule; /* Owning rule. */
265 struct list subfacets;
266 long long int used; /* Time last used; time created if not used. */
273 * - Do include packets and bytes sent "by hand", e.g. with
276 * - Do include packets and bytes that were obtained from the datapath
277 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
278 * DPIF_FP_ZERO_STATS).
280 * - Do not include packets or bytes that can be obtained from the
281 * datapath for any existing subfacet.
283 uint64_t packet_count; /* Number of packets received. */
284 uint64_t byte_count; /* Number of bytes received. */
286 /* Resubmit statistics. */
287 uint64_t prev_packet_count; /* Number of packets from last stats push. */
288 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
289 long long int prev_used; /* Used time from last stats push. */
292 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
293 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
295 /* Properties of datapath actions.
297 * Every subfacet has its own actions because actions can differ slightly
298 * between splintered and non-splintered subfacets due to the VLAN tag
299 * being initially different (present vs. absent). All of them have these
300 * properties in common so we just store one copy of them here. */
301 bool may_install; /* Reassess actions for every packet? */
302 bool has_learn; /* Actions include NXAST_LEARN? */
303 bool has_normal; /* Actions output to OFPP_NORMAL? */
304 tag_type tags; /* Tags that would require revalidation. */
305 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
308 static struct facet *facet_create(struct rule_dpif *, const struct flow *);
309 static void facet_remove(struct ofproto_dpif *, struct facet *);
310 static void facet_free(struct facet *);
312 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
313 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
314 const struct flow *);
315 static bool facet_revalidate(struct ofproto_dpif *, struct facet *);
317 static bool execute_controller_action(struct ofproto_dpif *,
319 const struct nlattr *odp_actions,
321 struct ofpbuf *packet);
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 {
488 struct netflow *netflow;
489 struct dpif_sflow *sflow;
490 struct hmap bundles; /* Contains "struct ofbundle"s. */
491 struct mac_learning *ml;
492 struct ofmirror *mirrors[MAX_MIRRORS];
493 bool has_bonded_bundles;
496 struct timer next_expiration;
500 struct hmap subfacets;
503 struct table_dpif tables[N_TABLES];
504 bool need_revalidate;
505 struct tag_set revalidate_set;
507 /* Support for debugging async flow mods. */
508 struct list completions;
510 bool has_bundle_action; /* True when the first bundle action appears. */
511 struct netdev_stats stats; /* To account packets generated and consumed in
516 long long int stp_last_tick;
518 /* VLAN splinters. */
519 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
520 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
523 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
524 * for debugging the asynchronous flow_mod implementation.) */
527 static void ofproto_dpif_unixctl_init(void);
529 static struct ofproto_dpif *
530 ofproto_dpif_cast(const struct ofproto *ofproto)
532 assert(ofproto->ofproto_class == &ofproto_dpif_class);
533 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
536 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
538 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
541 /* Packet processing. */
542 static void update_learning_table(struct ofproto_dpif *,
543 const struct flow *, int vlan,
546 #define FLOW_MISS_MAX_BATCH 50
547 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
549 /* Flow expiration. */
550 static int expire(struct ofproto_dpif *);
553 static void send_netflow_active_timeouts(struct ofproto_dpif *);
556 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
558 compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions,
559 const struct flow *, uint32_t odp_port);
560 static void add_mirror_actions(struct action_xlate_ctx *ctx,
561 const struct flow *flow);
562 /* Global variables. */
563 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
565 /* Factory functions. */
568 enumerate_types(struct sset *types)
570 dp_enumerate_types(types);
574 enumerate_names(const char *type, struct sset *names)
576 return dp_enumerate_names(type, names);
580 del(const char *type, const char *name)
585 error = dpif_open(name, type, &dpif);
587 error = dpif_delete(dpif);
593 /* Basic life-cycle. */
595 static struct ofproto *
598 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
603 dealloc(struct ofproto *ofproto_)
605 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
610 construct(struct ofproto *ofproto_, int *n_tablesp)
612 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
613 const char *name = ofproto->up.name;
617 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
619 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
623 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
624 ofproto->n_matches = 0;
626 dpif_flow_flush(ofproto->dpif);
627 dpif_recv_purge(ofproto->dpif);
629 error = dpif_recv_set_mask(ofproto->dpif,
630 ((1u << DPIF_UC_MISS) |
631 (1u << DPIF_UC_ACTION)));
633 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
634 dpif_close(ofproto->dpif);
638 ofproto->netflow = NULL;
639 ofproto->sflow = NULL;
641 hmap_init(&ofproto->bundles);
642 ofproto->ml = mac_learning_create();
643 for (i = 0; i < MAX_MIRRORS; i++) {
644 ofproto->mirrors[i] = NULL;
646 ofproto->has_bonded_bundles = false;
648 timer_set_duration(&ofproto->next_expiration, 1000);
650 hmap_init(&ofproto->facets);
651 hmap_init(&ofproto->subfacets);
653 for (i = 0; i < N_TABLES; i++) {
654 struct table_dpif *table = &ofproto->tables[i];
656 table->catchall_table = NULL;
657 table->other_table = NULL;
658 table->basis = random_uint32();
660 ofproto->need_revalidate = false;
661 tag_set_init(&ofproto->revalidate_set);
663 list_init(&ofproto->completions);
665 ofproto_dpif_unixctl_init();
667 ofproto->has_bundle_action = false;
669 hmap_init(&ofproto->vlandev_map);
670 hmap_init(&ofproto->realdev_vid_map);
672 *n_tablesp = N_TABLES;
673 memset(&ofproto->stats, 0, sizeof ofproto->stats);
678 complete_operations(struct ofproto_dpif *ofproto)
680 struct dpif_completion *c, *next;
682 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
683 ofoperation_complete(c->op, 0);
684 list_remove(&c->list_node);
690 destruct(struct ofproto *ofproto_)
692 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
693 struct rule_dpif *rule, *next_rule;
694 struct classifier *table;
697 complete_operations(ofproto);
699 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
700 struct cls_cursor cursor;
702 cls_cursor_init(&cursor, table, NULL);
703 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
704 ofproto_rule_destroy(&rule->up);
708 for (i = 0; i < MAX_MIRRORS; i++) {
709 mirror_destroy(ofproto->mirrors[i]);
712 netflow_destroy(ofproto->netflow);
713 dpif_sflow_destroy(ofproto->sflow);
714 hmap_destroy(&ofproto->bundles);
715 mac_learning_destroy(ofproto->ml);
717 hmap_destroy(&ofproto->facets);
718 hmap_destroy(&ofproto->subfacets);
720 hmap_destroy(&ofproto->vlandev_map);
721 hmap_destroy(&ofproto->realdev_vid_map);
723 dpif_close(ofproto->dpif);
727 run_fast(struct ofproto *ofproto_)
729 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
732 /* Handle one or more batches of upcalls, until there's nothing left to do
733 * or until we do a fixed total amount of work.
735 * We do work in batches because it can be much cheaper to set up a number
736 * of flows and fire off their patches all at once. We do multiple batches
737 * because in some cases handling a packet can cause another packet to be
738 * queued almost immediately as part of the return flow. Both
739 * optimizations can make major improvements on some benchmarks and
740 * presumably for real traffic as well. */
742 while (work < FLOW_MISS_MAX_BATCH) {
743 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
753 run(struct ofproto *ofproto_)
755 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
756 struct ofport_dpif *ofport;
757 struct ofbundle *bundle;
761 complete_operations(ofproto);
763 dpif_run(ofproto->dpif);
765 error = run_fast(ofproto_);
770 if (timer_expired(&ofproto->next_expiration)) {
771 int delay = expire(ofproto);
772 timer_set_duration(&ofproto->next_expiration, delay);
775 if (ofproto->netflow) {
776 if (netflow_run(ofproto->netflow)) {
777 send_netflow_active_timeouts(ofproto);
780 if (ofproto->sflow) {
781 dpif_sflow_run(ofproto->sflow);
784 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
787 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
792 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
794 /* Now revalidate if there's anything to do. */
795 if (ofproto->need_revalidate
796 || !tag_set_is_empty(&ofproto->revalidate_set)) {
797 struct tag_set revalidate_set = ofproto->revalidate_set;
798 bool revalidate_all = ofproto->need_revalidate;
799 struct facet *facet, *next;
801 /* Clear the revalidation flags. */
802 tag_set_init(&ofproto->revalidate_set);
803 ofproto->need_revalidate = false;
805 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
807 || tag_set_intersects(&revalidate_set, facet->tags)) {
808 facet_revalidate(ofproto, facet);
817 wait(struct ofproto *ofproto_)
819 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
820 struct ofport_dpif *ofport;
821 struct ofbundle *bundle;
823 if (!clogged && !list_is_empty(&ofproto->completions)) {
824 poll_immediate_wake();
827 dpif_wait(ofproto->dpif);
828 dpif_recv_wait(ofproto->dpif);
829 if (ofproto->sflow) {
830 dpif_sflow_wait(ofproto->sflow);
832 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
833 poll_immediate_wake();
835 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
838 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
841 if (ofproto->netflow) {
842 netflow_wait(ofproto->netflow);
844 mac_learning_wait(ofproto->ml);
846 if (ofproto->need_revalidate) {
847 /* Shouldn't happen, but if it does just go around again. */
848 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
849 poll_immediate_wake();
851 timer_wait(&ofproto->next_expiration);
856 flush(struct ofproto *ofproto_)
858 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
859 struct facet *facet, *next_facet;
861 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
862 /* Mark the facet as not installed so that facet_remove() doesn't
863 * bother trying to uninstall it. There is no point in uninstalling it
864 * individually since we are about to blow away all the facets with
865 * dpif_flow_flush(). */
866 struct subfacet *subfacet;
868 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
869 subfacet->installed = false;
870 subfacet->dp_packet_count = 0;
871 subfacet->dp_byte_count = 0;
873 facet_remove(ofproto, facet);
875 dpif_flow_flush(ofproto->dpif);
879 get_features(struct ofproto *ofproto_ OVS_UNUSED,
880 bool *arp_match_ip, uint32_t *actions)
882 *arp_match_ip = true;
883 *actions = ((1u << OFPAT_OUTPUT) |
884 (1u << OFPAT_SET_VLAN_VID) |
885 (1u << OFPAT_SET_VLAN_PCP) |
886 (1u << OFPAT_STRIP_VLAN) |
887 (1u << OFPAT_SET_DL_SRC) |
888 (1u << OFPAT_SET_DL_DST) |
889 (1u << OFPAT_SET_NW_SRC) |
890 (1u << OFPAT_SET_NW_DST) |
891 (1u << OFPAT_SET_NW_TOS) |
892 (1u << OFPAT_SET_TP_SRC) |
893 (1u << OFPAT_SET_TP_DST) |
894 (1u << OFPAT_ENQUEUE));
898 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
900 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
901 struct dpif_dp_stats s;
903 strcpy(ots->name, "classifier");
905 dpif_get_dp_stats(ofproto->dpif, &s);
906 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
907 put_32aligned_be64(&ots->matched_count,
908 htonll(s.n_hit + ofproto->n_matches));
911 static struct ofport *
914 struct ofport_dpif *port = xmalloc(sizeof *port);
919 port_dealloc(struct ofport *port_)
921 struct ofport_dpif *port = ofport_dpif_cast(port_);
926 port_construct(struct ofport *port_)
928 struct ofport_dpif *port = ofport_dpif_cast(port_);
929 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
931 ofproto->need_revalidate = true;
932 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
935 port->tag = tag_create_random();
936 port->may_enable = true;
937 port->stp_port = NULL;
938 port->stp_state = STP_DISABLED;
939 hmap_init(&port->priorities);
940 port->realdev_ofp_port = 0;
941 port->vlandev_vid = 0;
943 if (ofproto->sflow) {
944 dpif_sflow_add_port(ofproto->sflow, port_);
951 port_destruct(struct ofport *port_)
953 struct ofport_dpif *port = ofport_dpif_cast(port_);
954 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
956 ofproto->need_revalidate = true;
957 bundle_remove(port_);
958 set_cfm(port_, NULL);
959 if (ofproto->sflow) {
960 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
963 ofport_clear_priorities(port);
964 hmap_destroy(&port->priorities);
968 port_modified(struct ofport *port_)
970 struct ofport_dpif *port = ofport_dpif_cast(port_);
972 if (port->bundle && port->bundle->bond) {
973 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
978 port_reconfigured(struct ofport *port_, ovs_be32 old_config)
980 struct ofport_dpif *port = ofport_dpif_cast(port_);
981 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
982 ovs_be32 changed = old_config ^ port->up.opp.config;
984 if (changed & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
985 OFPPC_NO_FWD | OFPPC_NO_FLOOD)) {
986 ofproto->need_revalidate = true;
988 if (changed & htonl(OFPPC_NO_FLOOD) && port->bundle) {
989 bundle_update(port->bundle);
995 set_sflow(struct ofproto *ofproto_,
996 const struct ofproto_sflow_options *sflow_options)
998 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
999 struct dpif_sflow *ds = ofproto->sflow;
1001 if (sflow_options) {
1003 struct ofport_dpif *ofport;
1005 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1006 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1007 dpif_sflow_add_port(ds, &ofport->up);
1009 ofproto->need_revalidate = true;
1011 dpif_sflow_set_options(ds, sflow_options);
1014 dpif_sflow_destroy(ds);
1015 ofproto->need_revalidate = true;
1016 ofproto->sflow = NULL;
1023 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1025 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1032 struct ofproto_dpif *ofproto;
1034 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1035 ofproto->need_revalidate = true;
1036 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1039 if (cfm_configure(ofport->cfm, s)) {
1045 cfm_destroy(ofport->cfm);
1051 get_cfm_fault(const struct ofport *ofport_)
1053 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1055 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1059 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1062 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1065 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1072 /* Spanning Tree. */
1075 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1077 struct ofproto_dpif *ofproto = ofproto_;
1078 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1079 struct ofport_dpif *ofport;
1081 ofport = stp_port_get_aux(sp);
1083 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1084 ofproto->up.name, port_num);
1086 struct eth_header *eth = pkt->l2;
1088 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1089 if (eth_addr_is_zero(eth->eth_src)) {
1090 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1091 "with unknown MAC", ofproto->up.name, port_num);
1093 send_packet(ofport, pkt);
1099 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1101 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1103 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1105 /* Only revalidate flows if the configuration changed. */
1106 if (!s != !ofproto->stp) {
1107 ofproto->need_revalidate = true;
1111 if (!ofproto->stp) {
1112 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1113 send_bpdu_cb, ofproto);
1114 ofproto->stp_last_tick = time_msec();
1117 stp_set_bridge_id(ofproto->stp, s->system_id);
1118 stp_set_bridge_priority(ofproto->stp, s->priority);
1119 stp_set_hello_time(ofproto->stp, s->hello_time);
1120 stp_set_max_age(ofproto->stp, s->max_age);
1121 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1123 stp_destroy(ofproto->stp);
1124 ofproto->stp = NULL;
1131 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1133 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1137 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1138 s->designated_root = stp_get_designated_root(ofproto->stp);
1139 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1148 update_stp_port_state(struct ofport_dpif *ofport)
1150 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1151 enum stp_state state;
1153 /* Figure out new state. */
1154 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1158 if (ofport->stp_state != state) {
1162 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1163 netdev_get_name(ofport->up.netdev),
1164 stp_state_name(ofport->stp_state),
1165 stp_state_name(state));
1166 if (stp_learn_in_state(ofport->stp_state)
1167 != stp_learn_in_state(state)) {
1168 /* xxx Learning action flows should also be flushed. */
1169 mac_learning_flush(ofproto->ml);
1171 fwd_change = stp_forward_in_state(ofport->stp_state)
1172 != stp_forward_in_state(state);
1174 ofproto->need_revalidate = true;
1175 ofport->stp_state = state;
1176 ofport->stp_state_entered = time_msec();
1178 if (fwd_change && ofport->bundle) {
1179 bundle_update(ofport->bundle);
1182 /* Update the STP state bits in the OpenFlow port description. */
1183 of_state = (ofport->up.opp.state & htonl(~OFPPS_STP_MASK))
1184 | htonl(state == STP_LISTENING ? OFPPS_STP_LISTEN
1185 : state == STP_LEARNING ? OFPPS_STP_LEARN
1186 : state == STP_FORWARDING ? OFPPS_STP_FORWARD
1187 : state == STP_BLOCKING ? OFPPS_STP_BLOCK
1189 ofproto_port_set_state(&ofport->up, of_state);
1193 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1194 * caller is responsible for assigning STP port numbers and ensuring
1195 * there are no duplicates. */
1197 set_stp_port(struct ofport *ofport_,
1198 const struct ofproto_port_stp_settings *s)
1200 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1201 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1202 struct stp_port *sp = ofport->stp_port;
1204 if (!s || !s->enable) {
1206 ofport->stp_port = NULL;
1207 stp_port_disable(sp);
1208 update_stp_port_state(ofport);
1211 } else if (sp && stp_port_no(sp) != s->port_num
1212 && ofport == stp_port_get_aux(sp)) {
1213 /* The port-id changed, so disable the old one if it's not
1214 * already in use by another port. */
1215 stp_port_disable(sp);
1218 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1219 stp_port_enable(sp);
1221 stp_port_set_aux(sp, ofport);
1222 stp_port_set_priority(sp, s->priority);
1223 stp_port_set_path_cost(sp, s->path_cost);
1225 update_stp_port_state(ofport);
1231 get_stp_port_status(struct ofport *ofport_,
1232 struct ofproto_port_stp_status *s)
1234 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1235 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1236 struct stp_port *sp = ofport->stp_port;
1238 if (!ofproto->stp || !sp) {
1244 s->port_id = stp_port_get_id(sp);
1245 s->state = stp_port_get_state(sp);
1246 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1247 s->role = stp_port_get_role(sp);
1248 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1254 stp_run(struct ofproto_dpif *ofproto)
1257 long long int now = time_msec();
1258 long long int elapsed = now - ofproto->stp_last_tick;
1259 struct stp_port *sp;
1262 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1263 ofproto->stp_last_tick = now;
1265 while (stp_get_changed_port(ofproto->stp, &sp)) {
1266 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1269 update_stp_port_state(ofport);
1276 stp_wait(struct ofproto_dpif *ofproto)
1279 poll_timer_wait(1000);
1283 /* Returns true if STP should process 'flow'. */
1285 stp_should_process_flow(const struct flow *flow)
1287 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1291 stp_process_packet(const struct ofport_dpif *ofport,
1292 const struct ofpbuf *packet)
1294 struct ofpbuf payload = *packet;
1295 struct eth_header *eth = payload.data;
1296 struct stp_port *sp = ofport->stp_port;
1298 /* Sink packets on ports that have STP disabled when the bridge has
1300 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1304 /* Trim off padding on payload. */
1305 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1306 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1309 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1310 stp_received_bpdu(sp, payload.data, payload.size);
1314 static struct priority_to_dscp *
1315 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1317 struct priority_to_dscp *pdscp;
1320 hash = hash_int(priority, 0);
1321 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1322 if (pdscp->priority == priority) {
1330 ofport_clear_priorities(struct ofport_dpif *ofport)
1332 struct priority_to_dscp *pdscp, *next;
1334 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1335 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1341 set_queues(struct ofport *ofport_,
1342 const struct ofproto_port_queue *qdscp_list,
1345 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1346 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1347 struct hmap new = HMAP_INITIALIZER(&new);
1350 for (i = 0; i < n_qdscp; i++) {
1351 struct priority_to_dscp *pdscp;
1355 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1356 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1361 pdscp = get_priority(ofport, priority);
1363 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1365 pdscp = xmalloc(sizeof *pdscp);
1366 pdscp->priority = priority;
1368 ofproto->need_revalidate = true;
1371 if (pdscp->dscp != dscp) {
1373 ofproto->need_revalidate = true;
1376 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1379 if (!hmap_is_empty(&ofport->priorities)) {
1380 ofport_clear_priorities(ofport);
1381 ofproto->need_revalidate = true;
1384 hmap_swap(&new, &ofport->priorities);
1392 /* Expires all MAC learning entries associated with 'port' and forces ofproto
1393 * to revalidate every flow. */
1395 bundle_flush_macs(struct ofbundle *bundle)
1397 struct ofproto_dpif *ofproto = bundle->ofproto;
1398 struct mac_learning *ml = ofproto->ml;
1399 struct mac_entry *mac, *next_mac;
1401 ofproto->need_revalidate = true;
1402 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1403 if (mac->port.p == bundle) {
1404 mac_learning_expire(ml, mac);
1409 static struct ofbundle *
1410 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1412 struct ofbundle *bundle;
1414 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1415 &ofproto->bundles) {
1416 if (bundle->aux == aux) {
1423 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1424 * ones that are found to 'bundles'. */
1426 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1427 void **auxes, size_t n_auxes,
1428 struct hmapx *bundles)
1432 hmapx_init(bundles);
1433 for (i = 0; i < n_auxes; i++) {
1434 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1436 hmapx_add(bundles, bundle);
1442 bundle_update(struct ofbundle *bundle)
1444 struct ofport_dpif *port;
1446 bundle->floodable = true;
1447 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1448 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
1449 bundle->floodable = false;
1456 bundle_del_port(struct ofport_dpif *port)
1458 struct ofbundle *bundle = port->bundle;
1460 bundle->ofproto->need_revalidate = true;
1462 list_remove(&port->bundle_node);
1463 port->bundle = NULL;
1466 lacp_slave_unregister(bundle->lacp, port);
1469 bond_slave_unregister(bundle->bond, port);
1472 bundle_update(bundle);
1476 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1477 struct lacp_slave_settings *lacp,
1478 uint32_t bond_stable_id)
1480 struct ofport_dpif *port;
1482 port = get_ofp_port(bundle->ofproto, ofp_port);
1487 if (port->bundle != bundle) {
1488 bundle->ofproto->need_revalidate = true;
1490 bundle_del_port(port);
1493 port->bundle = bundle;
1494 list_push_back(&bundle->ports, &port->bundle_node);
1495 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
1496 bundle->floodable = false;
1500 port->bundle->ofproto->need_revalidate = true;
1501 lacp_slave_register(bundle->lacp, port, lacp);
1504 port->bond_stable_id = bond_stable_id;
1510 bundle_destroy(struct ofbundle *bundle)
1512 struct ofproto_dpif *ofproto;
1513 struct ofport_dpif *port, *next_port;
1520 ofproto = bundle->ofproto;
1521 for (i = 0; i < MAX_MIRRORS; i++) {
1522 struct ofmirror *m = ofproto->mirrors[i];
1524 if (m->out == bundle) {
1526 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1527 || hmapx_find_and_delete(&m->dsts, bundle)) {
1528 ofproto->need_revalidate = true;
1533 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1534 bundle_del_port(port);
1537 bundle_flush_macs(bundle);
1538 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1540 free(bundle->trunks);
1541 lacp_destroy(bundle->lacp);
1542 bond_destroy(bundle->bond);
1547 bundle_set(struct ofproto *ofproto_, void *aux,
1548 const struct ofproto_bundle_settings *s)
1550 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1551 bool need_flush = false;
1552 struct ofport_dpif *port;
1553 struct ofbundle *bundle;
1554 unsigned long *trunks;
1560 bundle_destroy(bundle_lookup(ofproto, aux));
1564 assert(s->n_slaves == 1 || s->bond != NULL);
1565 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1567 bundle = bundle_lookup(ofproto, aux);
1569 bundle = xmalloc(sizeof *bundle);
1571 bundle->ofproto = ofproto;
1572 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1573 hash_pointer(aux, 0));
1575 bundle->name = NULL;
1577 list_init(&bundle->ports);
1578 bundle->vlan_mode = PORT_VLAN_TRUNK;
1580 bundle->trunks = NULL;
1581 bundle->use_priority_tags = s->use_priority_tags;
1582 bundle->lacp = NULL;
1583 bundle->bond = NULL;
1585 bundle->floodable = true;
1587 bundle->src_mirrors = 0;
1588 bundle->dst_mirrors = 0;
1589 bundle->mirror_out = 0;
1592 if (!bundle->name || strcmp(s->name, bundle->name)) {
1594 bundle->name = xstrdup(s->name);
1599 if (!bundle->lacp) {
1600 ofproto->need_revalidate = true;
1601 bundle->lacp = lacp_create();
1603 lacp_configure(bundle->lacp, s->lacp);
1605 lacp_destroy(bundle->lacp);
1606 bundle->lacp = NULL;
1609 /* Update set of ports. */
1611 for (i = 0; i < s->n_slaves; i++) {
1612 if (!bundle_add_port(bundle, s->slaves[i],
1613 s->lacp ? &s->lacp_slaves[i] : NULL,
1614 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1618 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1619 struct ofport_dpif *next_port;
1621 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1622 for (i = 0; i < s->n_slaves; i++) {
1623 if (s->slaves[i] == port->up.ofp_port) {
1628 bundle_del_port(port);
1632 assert(list_size(&bundle->ports) <= s->n_slaves);
1634 if (list_is_empty(&bundle->ports)) {
1635 bundle_destroy(bundle);
1639 /* Set VLAN tagging mode */
1640 if (s->vlan_mode != bundle->vlan_mode
1641 || s->use_priority_tags != bundle->use_priority_tags) {
1642 bundle->vlan_mode = s->vlan_mode;
1643 bundle->use_priority_tags = s->use_priority_tags;
1648 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1649 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1651 if (vlan != bundle->vlan) {
1652 bundle->vlan = vlan;
1656 /* Get trunked VLANs. */
1657 switch (s->vlan_mode) {
1658 case PORT_VLAN_ACCESS:
1662 case PORT_VLAN_TRUNK:
1663 trunks = (unsigned long *) s->trunks;
1666 case PORT_VLAN_NATIVE_UNTAGGED:
1667 case PORT_VLAN_NATIVE_TAGGED:
1668 if (vlan != 0 && (!s->trunks
1669 || !bitmap_is_set(s->trunks, vlan)
1670 || bitmap_is_set(s->trunks, 0))) {
1671 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1673 trunks = bitmap_clone(s->trunks, 4096);
1675 trunks = bitmap_allocate1(4096);
1677 bitmap_set1(trunks, vlan);
1678 bitmap_set0(trunks, 0);
1680 trunks = (unsigned long *) s->trunks;
1687 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1688 free(bundle->trunks);
1689 if (trunks == s->trunks) {
1690 bundle->trunks = vlan_bitmap_clone(trunks);
1692 bundle->trunks = trunks;
1697 if (trunks != s->trunks) {
1702 if (!list_is_short(&bundle->ports)) {
1703 bundle->ofproto->has_bonded_bundles = true;
1705 if (bond_reconfigure(bundle->bond, s->bond)) {
1706 ofproto->need_revalidate = true;
1709 bundle->bond = bond_create(s->bond);
1710 ofproto->need_revalidate = true;
1713 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1714 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1718 bond_destroy(bundle->bond);
1719 bundle->bond = NULL;
1722 /* If we changed something that would affect MAC learning, un-learn
1723 * everything on this port and force flow revalidation. */
1725 bundle_flush_macs(bundle);
1732 bundle_remove(struct ofport *port_)
1734 struct ofport_dpif *port = ofport_dpif_cast(port_);
1735 struct ofbundle *bundle = port->bundle;
1738 bundle_del_port(port);
1739 if (list_is_empty(&bundle->ports)) {
1740 bundle_destroy(bundle);
1741 } else if (list_is_short(&bundle->ports)) {
1742 bond_destroy(bundle->bond);
1743 bundle->bond = NULL;
1749 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
1751 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1752 struct ofport_dpif *port = port_;
1753 uint8_t ea[ETH_ADDR_LEN];
1756 error = netdev_get_etheraddr(port->up.netdev, ea);
1758 struct ofpbuf packet;
1761 ofpbuf_init(&packet, 0);
1762 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1764 memcpy(packet_pdu, pdu, pdu_size);
1766 send_packet(port, &packet);
1767 ofpbuf_uninit(&packet);
1769 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1770 "%s (%s)", port->bundle->name,
1771 netdev_get_name(port->up.netdev), strerror(error));
1776 bundle_send_learning_packets(struct ofbundle *bundle)
1778 struct ofproto_dpif *ofproto = bundle->ofproto;
1779 int error, n_packets, n_errors;
1780 struct mac_entry *e;
1782 error = n_packets = n_errors = 0;
1783 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1784 if (e->port.p != bundle) {
1785 struct ofpbuf *learning_packet;
1786 struct ofport_dpif *port;
1789 learning_packet = bond_compose_learning_packet(bundle->bond, e->mac,
1792 ret = send_packet(port, learning_packet);
1793 ofpbuf_delete(learning_packet);
1803 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1804 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1805 "packets, last error was: %s",
1806 bundle->name, n_errors, n_packets, strerror(error));
1808 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1809 bundle->name, n_packets);
1814 bundle_run(struct ofbundle *bundle)
1817 lacp_run(bundle->lacp, send_pdu_cb);
1820 struct ofport_dpif *port;
1822 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1823 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1826 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1827 lacp_negotiated(bundle->lacp));
1828 if (bond_should_send_learning_packets(bundle->bond)) {
1829 bundle_send_learning_packets(bundle);
1835 bundle_wait(struct ofbundle *bundle)
1838 lacp_wait(bundle->lacp);
1841 bond_wait(bundle->bond);
1848 mirror_scan(struct ofproto_dpif *ofproto)
1852 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1853 if (!ofproto->mirrors[idx]) {
1860 static struct ofmirror *
1861 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1865 for (i = 0; i < MAX_MIRRORS; i++) {
1866 struct ofmirror *mirror = ofproto->mirrors[i];
1867 if (mirror && mirror->aux == aux) {
1875 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
1877 mirror_update_dups(struct ofproto_dpif *ofproto)
1881 for (i = 0; i < MAX_MIRRORS; i++) {
1882 struct ofmirror *m = ofproto->mirrors[i];
1885 m->dup_mirrors = MIRROR_MASK_C(1) << i;
1889 for (i = 0; i < MAX_MIRRORS; i++) {
1890 struct ofmirror *m1 = ofproto->mirrors[i];
1897 for (j = i + 1; j < MAX_MIRRORS; j++) {
1898 struct ofmirror *m2 = ofproto->mirrors[j];
1900 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
1901 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
1902 m2->dup_mirrors |= m1->dup_mirrors;
1909 mirror_set(struct ofproto *ofproto_, void *aux,
1910 const struct ofproto_mirror_settings *s)
1912 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1913 mirror_mask_t mirror_bit;
1914 struct ofbundle *bundle;
1915 struct ofmirror *mirror;
1916 struct ofbundle *out;
1917 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
1918 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
1921 mirror = mirror_lookup(ofproto, aux);
1923 mirror_destroy(mirror);
1929 idx = mirror_scan(ofproto);
1931 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
1933 ofproto->up.name, MAX_MIRRORS, s->name);
1937 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
1938 mirror->ofproto = ofproto;
1941 mirror->out_vlan = -1;
1942 mirror->name = NULL;
1945 if (!mirror->name || strcmp(s->name, mirror->name)) {
1947 mirror->name = xstrdup(s->name);
1950 /* Get the new configuration. */
1951 if (s->out_bundle) {
1952 out = bundle_lookup(ofproto, s->out_bundle);
1954 mirror_destroy(mirror);
1960 out_vlan = s->out_vlan;
1962 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
1963 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
1965 /* If the configuration has not changed, do nothing. */
1966 if (hmapx_equals(&srcs, &mirror->srcs)
1967 && hmapx_equals(&dsts, &mirror->dsts)
1968 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
1969 && mirror->out == out
1970 && mirror->out_vlan == out_vlan)
1972 hmapx_destroy(&srcs);
1973 hmapx_destroy(&dsts);
1977 hmapx_swap(&srcs, &mirror->srcs);
1978 hmapx_destroy(&srcs);
1980 hmapx_swap(&dsts, &mirror->dsts);
1981 hmapx_destroy(&dsts);
1983 free(mirror->vlans);
1984 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
1987 mirror->out_vlan = out_vlan;
1989 /* Update bundles. */
1990 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
1991 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
1992 if (hmapx_contains(&mirror->srcs, bundle)) {
1993 bundle->src_mirrors |= mirror_bit;
1995 bundle->src_mirrors &= ~mirror_bit;
1998 if (hmapx_contains(&mirror->dsts, bundle)) {
1999 bundle->dst_mirrors |= mirror_bit;
2001 bundle->dst_mirrors &= ~mirror_bit;
2004 if (mirror->out == bundle) {
2005 bundle->mirror_out |= mirror_bit;
2007 bundle->mirror_out &= ~mirror_bit;
2011 ofproto->need_revalidate = true;
2012 mac_learning_flush(ofproto->ml);
2013 mirror_update_dups(ofproto);
2019 mirror_destroy(struct ofmirror *mirror)
2021 struct ofproto_dpif *ofproto;
2022 mirror_mask_t mirror_bit;
2023 struct ofbundle *bundle;
2029 ofproto = mirror->ofproto;
2030 ofproto->need_revalidate = true;
2031 mac_learning_flush(ofproto->ml);
2033 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2034 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2035 bundle->src_mirrors &= ~mirror_bit;
2036 bundle->dst_mirrors &= ~mirror_bit;
2037 bundle->mirror_out &= ~mirror_bit;
2040 hmapx_destroy(&mirror->srcs);
2041 hmapx_destroy(&mirror->dsts);
2042 free(mirror->vlans);
2044 ofproto->mirrors[mirror->idx] = NULL;
2048 mirror_update_dups(ofproto);
2052 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2053 uint64_t *packets, uint64_t *bytes)
2055 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2056 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2059 *packets = *bytes = UINT64_MAX;
2063 *packets = mirror->packet_count;
2064 *bytes = mirror->byte_count;
2070 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2072 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2073 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2074 ofproto->need_revalidate = true;
2075 mac_learning_flush(ofproto->ml);
2081 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2083 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2084 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2085 return bundle && bundle->mirror_out != 0;
2089 forward_bpdu_changed(struct ofproto *ofproto_)
2091 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2092 /* Revalidate cached flows whenever forward_bpdu option changes. */
2093 ofproto->need_revalidate = true;
2098 static struct ofport_dpif *
2099 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
2101 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2102 return ofport ? ofport_dpif_cast(ofport) : NULL;
2105 static struct ofport_dpif *
2106 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
2108 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2112 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2113 struct dpif_port *dpif_port)
2115 ofproto_port->name = dpif_port->name;
2116 ofproto_port->type = dpif_port->type;
2117 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2121 port_run(struct ofport_dpif *ofport)
2123 bool enable = netdev_get_carrier(ofport->up.netdev);
2126 cfm_run(ofport->cfm);
2128 if (cfm_should_send_ccm(ofport->cfm)) {
2129 struct ofpbuf packet;
2131 ofpbuf_init(&packet, 0);
2132 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr);
2133 send_packet(ofport, &packet);
2134 ofpbuf_uninit(&packet);
2137 enable = enable && !cfm_get_fault(ofport->cfm)
2138 && cfm_get_opup(ofport->cfm);
2141 if (ofport->bundle) {
2142 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2145 if (ofport->may_enable != enable) {
2146 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2148 if (ofproto->has_bundle_action) {
2149 ofproto->need_revalidate = true;
2153 ofport->may_enable = enable;
2157 port_wait(struct ofport_dpif *ofport)
2160 cfm_wait(ofport->cfm);
2165 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2166 struct ofproto_port *ofproto_port)
2168 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2169 struct dpif_port dpif_port;
2172 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2174 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2180 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2182 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2186 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2188 *ofp_portp = odp_port_to_ofp_port(odp_port);
2194 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2196 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2199 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2201 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2203 /* The caller is going to close ofport->up.netdev. If this is a
2204 * bonded port, then the bond is using that netdev, so remove it
2205 * from the bond. The client will need to reconfigure everything
2206 * after deleting ports, so then the slave will get re-added. */
2207 bundle_remove(&ofport->up);
2214 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2216 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2219 error = netdev_get_stats(ofport->up.netdev, stats);
2221 if (!error && ofport->odp_port == OVSP_LOCAL) {
2222 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2224 /* ofproto->stats.tx_packets represents packets that we created
2225 * internally and sent to some port (e.g. packets sent with
2226 * send_packet()). Account for them as if they had come from
2227 * OFPP_LOCAL and got forwarded. */
2229 if (stats->rx_packets != UINT64_MAX) {
2230 stats->rx_packets += ofproto->stats.tx_packets;
2233 if (stats->rx_bytes != UINT64_MAX) {
2234 stats->rx_bytes += ofproto->stats.tx_bytes;
2237 /* ofproto->stats.rx_packets represents packets that were received on
2238 * some port and we processed internally and dropped (e.g. STP).
2239 * Account fro them as if they had been forwarded to OFPP_LOCAL. */
2241 if (stats->tx_packets != UINT64_MAX) {
2242 stats->tx_packets += ofproto->stats.rx_packets;
2245 if (stats->tx_bytes != UINT64_MAX) {
2246 stats->tx_bytes += ofproto->stats.rx_bytes;
2253 /* Account packets for LOCAL port. */
2255 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2256 size_t tx_size, size_t rx_size)
2258 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2261 ofproto->stats.rx_packets++;
2262 ofproto->stats.rx_bytes += rx_size;
2265 ofproto->stats.tx_packets++;
2266 ofproto->stats.tx_bytes += tx_size;
2270 struct port_dump_state {
2271 struct dpif_port_dump dump;
2276 port_dump_start(const struct ofproto *ofproto_, void **statep)
2278 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2279 struct port_dump_state *state;
2281 *statep = state = xmalloc(sizeof *state);
2282 dpif_port_dump_start(&state->dump, ofproto->dpif);
2283 state->done = false;
2288 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2289 struct ofproto_port *port)
2291 struct port_dump_state *state = state_;
2292 struct dpif_port dpif_port;
2294 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2295 ofproto_port_from_dpif_port(port, &dpif_port);
2298 int error = dpif_port_dump_done(&state->dump);
2300 return error ? error : EOF;
2305 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2307 struct port_dump_state *state = state_;
2310 dpif_port_dump_done(&state->dump);
2317 port_poll(const struct ofproto *ofproto_, char **devnamep)
2319 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2320 return dpif_port_poll(ofproto->dpif, devnamep);
2324 port_poll_wait(const struct ofproto *ofproto_)
2326 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2327 dpif_port_poll_wait(ofproto->dpif);
2331 port_is_lacp_current(const struct ofport *ofport_)
2333 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2334 return (ofport->bundle && ofport->bundle->lacp
2335 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2339 /* Upcall handling. */
2341 /* Flow miss batching.
2343 * Some dpifs implement operations faster when you hand them off in a batch.
2344 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2345 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2346 * more packets, plus possibly installing the flow in the dpif.
2348 * So far we only batch the operations that affect flow setup time the most.
2349 * It's possible to batch more than that, but the benefit might be minimal. */
2351 struct hmap_node hmap_node;
2353 enum odp_key_fitness key_fitness;
2354 const struct nlattr *key;
2356 ovs_be16 initial_tci;
2357 struct list packets;
2360 struct flow_miss_op {
2361 union dpif_op dpif_op;
2362 struct subfacet *subfacet;
2365 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2366 * OpenFlow controller as necessary according to their individual
2369 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
2370 * ownership is transferred to this function. */
2372 send_packet_in_miss(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
2373 const struct flow *flow, bool clone)
2375 struct ofputil_packet_in pin;
2377 pin.packet = packet;
2378 pin.in_port = flow->in_port;
2379 pin.reason = OFPR_NO_MATCH;
2380 pin.buffer_id = 0; /* not yet known */
2381 pin.send_len = 0; /* not used for flow table misses */
2382 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
2383 clone ? NULL : packet);
2386 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_ACTION to each
2387 * OpenFlow controller as necessary according to their individual
2390 * 'send_len' should be the number of bytes of 'packet' to send to the
2391 * controller, as specified in the action that caused the packet to be sent.
2393 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
2394 * Otherwise, ownership is transferred to this function. */
2396 send_packet_in_action(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
2397 uint64_t userdata, const struct flow *flow, bool clone)
2399 struct ofputil_packet_in pin;
2400 struct user_action_cookie cookie;
2402 memcpy(&cookie, &userdata, sizeof(cookie));
2404 pin.packet = packet;
2405 pin.in_port = flow->in_port;
2406 pin.reason = OFPR_ACTION;
2407 pin.buffer_id = 0; /* not yet known */
2408 pin.send_len = cookie.data;
2409 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
2410 clone ? NULL : packet);
2414 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2415 const struct ofpbuf *packet)
2417 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2423 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2425 cfm_process_heartbeat(ofport->cfm, packet);
2428 } else if (ofport->bundle && ofport->bundle->lacp
2429 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2431 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2434 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2436 stp_process_packet(ofport, packet);
2443 static struct flow_miss *
2444 flow_miss_create(struct hmap *todo, const struct flow *flow,
2445 enum odp_key_fitness key_fitness,
2446 const struct nlattr *key, size_t key_len,
2447 ovs_be16 initial_tci)
2449 uint32_t hash = flow_hash(flow, 0);
2450 struct flow_miss *miss;
2452 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2453 if (flow_equal(&miss->flow, flow)) {
2458 miss = xmalloc(sizeof *miss);
2459 hmap_insert(todo, &miss->hmap_node, hash);
2461 miss->key_fitness = key_fitness;
2463 miss->key_len = key_len;
2464 miss->initial_tci = initial_tci;
2465 list_init(&miss->packets);
2470 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2471 struct flow_miss_op *ops, size_t *n_ops)
2473 const struct flow *flow = &miss->flow;
2474 struct ofpbuf *packet, *next_packet;
2475 struct subfacet *subfacet;
2476 struct facet *facet;
2478 facet = facet_lookup_valid(ofproto, flow);
2480 struct rule_dpif *rule;
2482 rule = rule_dpif_lookup(ofproto, flow, 0);
2484 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
2485 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
2487 if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) {
2488 COVERAGE_INC(ofproto_dpif_no_packet_in);
2489 /* XXX install 'drop' flow entry */
2493 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
2497 LIST_FOR_EACH_SAFE (packet, next_packet, list_node,
2499 list_remove(&packet->list_node);
2500 send_packet_in_miss(ofproto, packet, flow, false);
2506 facet = facet_create(rule, flow);
2509 subfacet = subfacet_create(ofproto, facet,
2510 miss->key_fitness, miss->key, miss->key_len,
2513 LIST_FOR_EACH_SAFE (packet, next_packet, list_node, &miss->packets) {
2514 list_remove(&packet->list_node);
2515 ofproto->n_matches++;
2517 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2519 * Extra-special case for fail-open mode.
2521 * We are in fail-open mode and the packet matched the fail-open
2522 * rule, but we are connected to a controller too. We should send
2523 * the packet up to the controller in the hope that it will try to
2524 * set up a flow and thereby allow us to exit fail-open.
2526 * See the top-level comment in fail-open.c for more information.
2528 send_packet_in_miss(ofproto, packet, flow, true);
2531 if (!facet->may_install || !subfacet->actions) {
2532 subfacet_make_actions(ofproto, subfacet, packet);
2534 if (!execute_controller_action(ofproto, &facet->flow,
2536 subfacet->actions_len, packet)) {
2537 struct flow_miss_op *op = &ops[(*n_ops)++];
2538 struct dpif_execute *execute = &op->dpif_op.execute;
2540 op->subfacet = subfacet;
2541 execute->type = DPIF_OP_EXECUTE;
2542 execute->key = miss->key;
2543 execute->key_len = miss->key_len;
2545 = (facet->may_install
2547 : xmemdup(subfacet->actions, subfacet->actions_len));
2548 execute->actions_len = subfacet->actions_len;
2549 execute->packet = packet;
2553 if (facet->may_install && subfacet->key_fitness != ODP_FIT_TOO_LITTLE) {
2554 struct flow_miss_op *op = &ops[(*n_ops)++];
2555 struct dpif_flow_put *put = &op->dpif_op.flow_put;
2557 op->subfacet = subfacet;
2558 put->type = DPIF_OP_FLOW_PUT;
2559 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2560 put->key = miss->key;
2561 put->key_len = miss->key_len;
2562 put->actions = subfacet->actions;
2563 put->actions_len = subfacet->actions_len;
2568 static enum odp_key_fitness
2569 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
2570 const struct nlattr *key, size_t key_len,
2571 struct flow *flow, ovs_be16 *initial_tci)
2573 enum odp_key_fitness fitness;
2577 fitness = odp_flow_key_to_flow(key, key_len, flow);
2578 if (fitness == ODP_FIT_ERROR) {
2581 *initial_tci = flow->vlan_tci;
2583 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
2585 /* Cause the flow to be processed as if it came in on the real device
2586 * with the VLAN device's VLAN ID. */
2587 flow->in_port = realdev;
2588 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
2590 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
2591 if (fitness == ODP_FIT_PERFECT) {
2592 fitness = ODP_FIT_TOO_MUCH;
2600 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
2603 struct dpif_upcall *upcall;
2604 struct flow_miss *miss, *next_miss;
2605 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
2606 union dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
2615 /* Construct the to-do list.
2617 * This just amounts to extracting the flow from each packet and sticking
2618 * the packets that have the same flow in the same "flow_miss" structure so
2619 * that we can process them together. */
2621 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
2622 enum odp_key_fitness fitness;
2623 struct flow_miss *miss;
2624 ovs_be16 initial_tci;
2627 /* Obtain metadata and check userspace/kernel agreement on flow match,
2628 * then set 'flow''s header pointers. */
2629 fitness = ofproto_dpif_extract_flow_key(ofproto,
2630 upcall->key, upcall->key_len,
2631 &flow, &initial_tci);
2632 if (fitness == ODP_FIT_ERROR) {
2633 ofpbuf_delete(upcall->packet);
2636 flow_extract(upcall->packet, flow.priority, flow.tun_id,
2637 flow.in_port, &flow);
2639 /* Handle 802.1ag, LACP, and STP specially. */
2640 if (process_special(ofproto, &flow, upcall->packet)) {
2641 ofproto_update_local_port_stats(&ofproto->up,
2642 0, upcall->packet->size);
2643 ofpbuf_delete(upcall->packet);
2644 ofproto->n_matches++;
2648 /* Add other packets to a to-do list. */
2649 miss = flow_miss_create(&todo, &flow, fitness,
2650 upcall->key, upcall->key_len, initial_tci);
2651 list_push_back(&miss->packets, &upcall->packet->list_node);
2654 /* Process each element in the to-do list, constructing the set of
2655 * operations to batch. */
2657 HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) {
2658 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
2659 ofpbuf_list_delete(&miss->packets);
2660 hmap_remove(&todo, &miss->hmap_node);
2663 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
2664 hmap_destroy(&todo);
2666 /* Execute batch. */
2667 for (i = 0; i < n_ops; i++) {
2668 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
2670 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
2672 /* Free memory and update facets. */
2673 for (i = 0; i < n_ops; i++) {
2674 struct flow_miss_op *op = &flow_miss_ops[i];
2675 struct dpif_execute *execute;
2676 struct dpif_flow_put *put;
2678 switch (op->dpif_op.type) {
2679 case DPIF_OP_EXECUTE:
2680 execute = &op->dpif_op.execute;
2681 if (op->subfacet->actions != execute->actions) {
2682 free((struct nlattr *) execute->actions);
2684 ofpbuf_delete((struct ofpbuf *) execute->packet);
2687 case DPIF_OP_FLOW_PUT:
2688 put = &op->dpif_op.flow_put;
2690 op->subfacet->installed = true;
2698 handle_userspace_upcall(struct ofproto_dpif *ofproto,
2699 struct dpif_upcall *upcall)
2701 struct user_action_cookie cookie;
2702 enum odp_key_fitness fitness;
2703 ovs_be16 initial_tci;
2706 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
2708 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
2709 upcall->key_len, &flow,
2711 if (fitness == ODP_FIT_ERROR) {
2712 ofpbuf_delete(upcall->packet);
2716 if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
2717 if (ofproto->sflow) {
2718 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
2721 ofpbuf_delete(upcall->packet);
2722 } else if (cookie.type == USER_ACTION_COOKIE_CONTROLLER) {
2723 COVERAGE_INC(ofproto_dpif_ctlr_action);
2724 send_packet_in_action(ofproto, upcall->packet, upcall->userdata,
2727 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
2728 ofpbuf_delete(upcall->packet);
2733 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
2735 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
2739 assert (max_batch <= FLOW_MISS_MAX_BATCH);
2742 for (i = 0; i < max_batch; i++) {
2743 struct dpif_upcall *upcall = &misses[n_misses];
2746 error = dpif_recv(ofproto->dpif, upcall);
2751 switch (upcall->type) {
2752 case DPIF_UC_ACTION:
2753 handle_userspace_upcall(ofproto, upcall);
2757 /* Handle it later. */
2761 case DPIF_N_UC_TYPES:
2763 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32,
2769 handle_miss_upcalls(ofproto, misses, n_misses);
2774 /* Flow expiration. */
2776 static int subfacet_max_idle(const struct ofproto_dpif *);
2777 static void update_stats(struct ofproto_dpif *);
2778 static void rule_expire(struct rule_dpif *);
2779 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
2781 /* This function is called periodically by run(). Its job is to collect
2782 * updates for the flows that have been installed into the datapath, most
2783 * importantly when they last were used, and then use that information to
2784 * expire flows that have not been used recently.
2786 * Returns the number of milliseconds after which it should be called again. */
2788 expire(struct ofproto_dpif *ofproto)
2790 struct rule_dpif *rule, *next_rule;
2791 struct classifier *table;
2794 /* Update stats for each flow in the datapath. */
2795 update_stats(ofproto);
2797 /* Expire subfacets that have been idle too long. */
2798 dp_max_idle = subfacet_max_idle(ofproto);
2799 expire_subfacets(ofproto, dp_max_idle);
2801 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
2802 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
2803 struct cls_cursor cursor;
2805 cls_cursor_init(&cursor, table, NULL);
2806 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
2811 /* All outstanding data in existing flows has been accounted, so it's a
2812 * good time to do bond rebalancing. */
2813 if (ofproto->has_bonded_bundles) {
2814 struct ofbundle *bundle;
2816 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2818 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
2823 return MIN(dp_max_idle, 1000);
2826 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
2828 * This function also pushes statistics updates to rules which each facet
2829 * resubmits into. Generally these statistics will be accurate. However, if a
2830 * facet changes the rule it resubmits into at some time in between
2831 * update_stats() runs, it is possible that statistics accrued to the
2832 * old rule will be incorrectly attributed to the new rule. This could be
2833 * avoided by calling update_stats() whenever rules are created or
2834 * deleted. However, the performance impact of making so many calls to the
2835 * datapath do not justify the benefit of having perfectly accurate statistics.
2838 update_stats(struct ofproto_dpif *p)
2840 const struct dpif_flow_stats *stats;
2841 struct dpif_flow_dump dump;
2842 const struct nlattr *key;
2845 dpif_flow_dump_start(&dump, p->dpif);
2846 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
2847 struct subfacet *subfacet;
2849 subfacet = subfacet_find(p, key, key_len);
2850 if (subfacet && subfacet->installed) {
2851 struct facet *facet = subfacet->facet;
2853 if (stats->n_packets >= subfacet->dp_packet_count) {
2854 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
2855 facet->packet_count += extra;
2857 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
2860 if (stats->n_bytes >= subfacet->dp_byte_count) {
2861 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
2863 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
2866 subfacet->dp_packet_count = stats->n_packets;
2867 subfacet->dp_byte_count = stats->n_bytes;
2869 subfacet_update_time(p, subfacet, stats->used);
2870 facet_account(p, facet);
2871 facet_push_stats(facet);
2873 if (!VLOG_DROP_WARN(&rl)) {
2877 odp_flow_key_format(key, key_len, &s);
2878 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
2882 COVERAGE_INC(facet_unexpected);
2883 /* There's a flow in the datapath that we know nothing about, or a
2884 * flow that shouldn't be installed but was anyway. Delete it. */
2885 dpif_flow_del(p->dpif, key, key_len, NULL);
2888 dpif_flow_dump_done(&dump);
2891 /* Calculates and returns the number of milliseconds of idle time after which
2892 * subfacets should expire from the datapath. When a subfacet expires, we fold
2893 * its statistics into its facet, and when a facet's last subfacet expires, we
2894 * fold its statistic into its rule. */
2896 subfacet_max_idle(const struct ofproto_dpif *ofproto)
2899 * Idle time histogram.
2901 * Most of the time a switch has a relatively small number of subfacets.
2902 * When this is the case we might as well keep statistics for all of them
2903 * in userspace and to cache them in the kernel datapath for performance as
2906 * As the number of subfacets increases, the memory required to maintain
2907 * statistics about them in userspace and in the kernel becomes
2908 * significant. However, with a large number of subfacets it is likely
2909 * that only a few of them are "heavy hitters" that consume a large amount
2910 * of bandwidth. At this point, only heavy hitters are worth caching in
2911 * the kernel and maintaining in userspaces; other subfacets we can
2914 * The technique used to compute the idle time is to build a histogram with
2915 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
2916 * that is installed in the kernel gets dropped in the appropriate bucket.
2917 * After the histogram has been built, we compute the cutoff so that only
2918 * the most-recently-used 1% of subfacets (but at least
2919 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
2920 * the most-recently-used bucket of subfacets is kept, so actually an
2921 * arbitrary number of subfacets can be kept in any given expiration run
2922 * (though the next run will delete most of those unless they receive
2925 * This requires a second pass through the subfacets, in addition to the
2926 * pass made by update_stats(), because the former function never looks at
2927 * uninstallable subfacets.
2929 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
2930 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
2931 int buckets[N_BUCKETS] = { 0 };
2932 int total, subtotal, bucket;
2933 struct subfacet *subfacet;
2937 total = hmap_count(&ofproto->subfacets);
2938 if (total <= ofproto->up.flow_eviction_threshold) {
2939 return N_BUCKETS * BUCKET_WIDTH;
2942 /* Build histogram. */
2944 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
2945 long long int idle = now - subfacet->used;
2946 int bucket = (idle <= 0 ? 0
2947 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
2948 : (unsigned int) idle / BUCKET_WIDTH);
2952 /* Find the first bucket whose flows should be expired. */
2953 subtotal = bucket = 0;
2955 subtotal += buckets[bucket++];
2956 } while (bucket < N_BUCKETS &&
2957 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
2959 if (VLOG_IS_DBG_ENABLED()) {
2963 ds_put_cstr(&s, "keep");
2964 for (i = 0; i < N_BUCKETS; i++) {
2966 ds_put_cstr(&s, ", drop");
2969 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
2972 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
2976 return bucket * BUCKET_WIDTH;
2980 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
2982 long long int cutoff = time_msec() - dp_max_idle;
2983 struct subfacet *subfacet, *next_subfacet;
2985 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
2986 &ofproto->subfacets) {
2987 if (subfacet->used < cutoff) {
2988 subfacet_destroy(ofproto, subfacet);
2993 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
2994 * then delete it entirely. */
2996 rule_expire(struct rule_dpif *rule)
2998 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2999 struct facet *facet, *next_facet;
3003 /* Has 'rule' expired? */
3005 if (rule->up.hard_timeout
3006 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3007 reason = OFPRR_HARD_TIMEOUT;
3008 } else if (rule->up.idle_timeout && list_is_empty(&rule->facets)
3009 && now > rule->used + rule->up.idle_timeout * 1000) {
3010 reason = OFPRR_IDLE_TIMEOUT;
3015 COVERAGE_INC(ofproto_dpif_expired);
3017 /* Update stats. (This is a no-op if the rule expired due to an idle
3018 * timeout, because that only happens when the rule has no facets left.) */
3019 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3020 facet_remove(ofproto, facet);
3023 /* Get rid of the rule. */
3024 ofproto_rule_expire(&rule->up, reason);
3029 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3031 * The caller must already have determined that no facet with an identical
3032 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3033 * the ofproto's classifier table.
3035 * The facet will initially have no subfacets. The caller should create (at
3036 * least) one subfacet with subfacet_create(). */
3037 static struct facet *
3038 facet_create(struct rule_dpif *rule, const struct flow *flow)
3040 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3041 struct facet *facet;
3043 facet = xzalloc(sizeof *facet);
3044 facet->used = time_msec();
3045 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
3046 list_push_back(&rule->facets, &facet->list_node);
3048 facet->flow = *flow;
3049 list_init(&facet->subfacets);
3050 netflow_flow_init(&facet->nf_flow);
3051 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3057 facet_free(struct facet *facet)
3063 execute_controller_action(struct ofproto_dpif *ofproto,
3064 const struct flow *flow,
3065 const struct nlattr *odp_actions, size_t actions_len,
3066 struct ofpbuf *packet)
3069 && odp_actions->nla_type == OVS_ACTION_ATTR_USERSPACE
3070 && NLA_ALIGN(odp_actions->nla_len) == actions_len) {
3071 /* As an optimization, avoid a round-trip from userspace to kernel to
3072 * userspace. This also avoids possibly filling up kernel packet
3073 * buffers along the way.
3075 * This optimization will not accidentally catch sFlow
3076 * OVS_ACTION_ATTR_USERSPACE actions, since those are encapsulated
3077 * inside OVS_ACTION_ATTR_SAMPLE. */
3078 const struct nlattr *nla;
3080 nla = nl_attr_find_nested(odp_actions, OVS_USERSPACE_ATTR_USERDATA);
3081 send_packet_in_action(ofproto, packet, nl_attr_get_u64(nla), flow,
3089 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3090 * 'packet', which arrived on 'in_port'.
3092 * Takes ownership of 'packet'. */
3094 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3095 const struct nlattr *odp_actions, size_t actions_len,
3096 struct ofpbuf *packet)
3098 struct odputil_keybuf keybuf;
3102 if (execute_controller_action(ofproto, flow, odp_actions, actions_len,
3107 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3108 odp_flow_key_from_flow(&key, flow);
3110 error = dpif_execute(ofproto->dpif, key.data, key.size,
3111 odp_actions, actions_len, packet);
3113 ofpbuf_delete(packet);
3117 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3119 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3120 * rule's statistics, via subfacet_uninstall().
3122 * - Removes 'facet' from its rule and from ofproto->facets.
3125 facet_remove(struct ofproto_dpif *ofproto, struct facet *facet)
3127 struct subfacet *subfacet, *next_subfacet;
3129 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3130 &facet->subfacets) {
3131 subfacet_destroy__(ofproto, subfacet);
3134 facet_flush_stats(ofproto, facet);
3135 hmap_remove(&ofproto->facets, &facet->hmap_node);
3136 list_remove(&facet->list_node);
3141 facet_account(struct ofproto_dpif *ofproto, struct facet *facet)
3144 struct subfacet *subfacet;
3145 const struct nlattr *a;
3149 if (facet->byte_count <= facet->accounted_bytes) {
3152 n_bytes = facet->byte_count - facet->accounted_bytes;
3153 facet->accounted_bytes = facet->byte_count;
3155 /* Feed information from the active flows back into the learning table to
3156 * ensure that table is always in sync with what is actually flowing
3157 * through the datapath. */
3158 if (facet->has_learn || facet->has_normal) {
3159 struct action_xlate_ctx ctx;
3161 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3162 facet->flow.vlan_tci, NULL);
3163 ctx.may_learn = true;
3164 ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions,
3165 facet->rule->up.n_actions));
3168 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3172 /* This loop feeds byte counters to bond_account() for rebalancing to use
3173 * as a basis. We also need to track the actual VLAN on which the packet
3174 * is going to be sent to ensure that it matches the one passed to
3175 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3178 * We use the actions from an arbitrary subfacet because they should all
3179 * be equally valid for our purpose. */
3180 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3181 struct subfacet, list_node);
3182 vlan_tci = facet->flow.vlan_tci;
3183 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3184 subfacet->actions, subfacet->actions_len) {
3185 const struct ovs_action_push_vlan *vlan;
3186 struct ofport_dpif *port;
3188 switch (nl_attr_type(a)) {
3189 case OVS_ACTION_ATTR_OUTPUT:
3190 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3191 if (port && port->bundle && port->bundle->bond) {
3192 bond_account(port->bundle->bond, &facet->flow,
3193 vlan_tci_to_vid(vlan_tci), n_bytes);
3197 case OVS_ACTION_ATTR_POP_VLAN:
3198 vlan_tci = htons(0);
3201 case OVS_ACTION_ATTR_PUSH_VLAN:
3202 vlan = nl_attr_get(a);
3203 vlan_tci = vlan->vlan_tci;
3209 /* Returns true if the only action for 'facet' is to send to the controller.
3210 * (We don't report NetFlow expiration messages for such facets because they
3211 * are just part of the control logic for the network, not real traffic). */
3213 facet_is_controller_flow(struct facet *facet)
3216 && facet->rule->up.n_actions == 1
3217 && action_outputs_to_port(&facet->rule->up.actions[0],
3218 htons(OFPP_CONTROLLER)));
3221 /* Folds all of 'facet''s statistics into its rule. Also updates the
3222 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3223 * 'facet''s statistics in the datapath should have been zeroed and folded into
3224 * its packet and byte counts before this function is called. */
3226 facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet)
3228 struct subfacet *subfacet;
3230 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3231 assert(!subfacet->dp_byte_count);
3232 assert(!subfacet->dp_packet_count);
3235 facet_push_stats(facet);
3236 facet_account(ofproto, facet);
3238 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3239 struct ofexpired expired;
3240 expired.flow = facet->flow;
3241 expired.packet_count = facet->packet_count;
3242 expired.byte_count = facet->byte_count;
3243 expired.used = facet->used;
3244 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3247 facet->rule->packet_count += facet->packet_count;
3248 facet->rule->byte_count += facet->byte_count;
3250 /* Reset counters to prevent double counting if 'facet' ever gets
3252 facet_reset_counters(facet);
3254 netflow_flow_clear(&facet->nf_flow);
3257 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3258 * Returns it if found, otherwise a null pointer.
3260 * The returned facet might need revalidation; use facet_lookup_valid()
3261 * instead if that is important. */
3262 static struct facet *
3263 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
3265 struct facet *facet;
3267 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
3269 if (flow_equal(flow, &facet->flow)) {
3277 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3278 * Returns it if found, otherwise a null pointer.
3280 * The returned facet is guaranteed to be valid. */
3281 static struct facet *
3282 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
3284 struct facet *facet = facet_find(ofproto, flow);
3286 /* The facet we found might not be valid, since we could be in need of
3287 * revalidation. If it is not valid, don't return it. */
3289 && (ofproto->need_revalidate
3290 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))
3291 && !facet_revalidate(ofproto, facet)) {
3292 COVERAGE_INC(facet_invalidated);
3299 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
3301 * - If the rule found is different from 'facet''s current rule, moves
3302 * 'facet' to the new rule and recompiles its actions.
3304 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3305 * where it is and recompiles its actions anyway.
3307 * - If there is none, destroys 'facet'.
3309 * Returns true if 'facet' still exists, false if it has been destroyed. */
3311 facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet)
3314 struct nlattr *odp_actions;
3317 struct actions *new_actions;
3319 struct action_xlate_ctx ctx;
3320 struct rule_dpif *new_rule;
3321 struct subfacet *subfacet;
3322 bool actions_changed;
3325 COVERAGE_INC(facet_revalidate);
3327 /* Determine the new rule. */
3328 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3330 /* No new rule, so delete the facet. */
3331 facet_remove(ofproto, facet);
3335 /* Calculate new datapath actions.
3337 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3338 * emit a NetFlow expiration and, if so, we need to have the old state
3339 * around to properly compose it. */
3341 /* If the datapath actions changed or the installability changed,
3342 * then we need to talk to the datapath. */
3345 memset(&ctx, 0, sizeof ctx);
3346 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3347 struct ofpbuf *odp_actions;
3348 bool should_install;
3350 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3351 subfacet->initial_tci, NULL);
3352 odp_actions = xlate_actions(&ctx, new_rule->up.actions,
3353 new_rule->up.n_actions);
3354 actions_changed = (subfacet->actions_len != odp_actions->size
3355 || memcmp(subfacet->actions, odp_actions->data,
3356 subfacet->actions_len));
3358 should_install = (ctx.may_set_up_flow
3359 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3360 if (actions_changed || should_install != subfacet->installed) {
3361 if (should_install) {
3362 struct dpif_flow_stats stats;
3364 subfacet_install(ofproto, subfacet,
3365 odp_actions->data, odp_actions->size, &stats);
3366 subfacet_update_stats(ofproto, subfacet, &stats);
3368 subfacet_uninstall(ofproto, subfacet);
3372 new_actions = xcalloc(list_size(&facet->subfacets),
3373 sizeof *new_actions);
3375 new_actions[i].odp_actions = xmemdup(odp_actions->data,
3377 new_actions[i].actions_len = odp_actions->size;
3380 ofpbuf_delete(odp_actions);
3384 facet_flush_stats(ofproto, facet);
3387 /* Update 'facet' now that we've taken care of all the old state. */
3388 facet->tags = ctx.tags;
3389 facet->nf_flow.output_iface = ctx.nf_output_iface;
3390 facet->may_install = ctx.may_set_up_flow;
3391 facet->has_learn = ctx.has_learn;
3392 facet->has_normal = ctx.has_normal;
3393 facet->mirrors = ctx.mirrors;
3396 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3397 if (new_actions[i].odp_actions) {
3398 free(subfacet->actions);
3399 subfacet->actions = new_actions[i].odp_actions;
3400 subfacet->actions_len = new_actions[i].actions_len;
3406 if (facet->rule != new_rule) {
3407 COVERAGE_INC(facet_changed_rule);
3408 list_remove(&facet->list_node);
3409 list_push_back(&new_rule->facets, &facet->list_node);
3410 facet->rule = new_rule;
3411 facet->used = new_rule->up.created;
3412 facet->prev_used = facet->used;
3418 /* Updates 'facet''s used time. Caller is responsible for calling
3419 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3421 facet_update_time(struct ofproto_dpif *ofproto, struct facet *facet,
3424 if (used > facet->used) {
3426 if (used > facet->rule->used) {
3427 facet->rule->used = used;
3429 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3434 facet_reset_counters(struct facet *facet)
3436 facet->packet_count = 0;
3437 facet->byte_count = 0;
3438 facet->prev_packet_count = 0;
3439 facet->prev_byte_count = 0;
3440 facet->accounted_bytes = 0;
3444 facet_push_stats(struct facet *facet)
3446 uint64_t new_packets, new_bytes;
3448 assert(facet->packet_count >= facet->prev_packet_count);
3449 assert(facet->byte_count >= facet->prev_byte_count);
3450 assert(facet->used >= facet->prev_used);
3452 new_packets = facet->packet_count - facet->prev_packet_count;
3453 new_bytes = facet->byte_count - facet->prev_byte_count;
3455 if (new_packets || new_bytes || facet->used > facet->prev_used) {
3456 facet->prev_packet_count = facet->packet_count;
3457 facet->prev_byte_count = facet->byte_count;
3458 facet->prev_used = facet->used;
3460 flow_push_stats(facet->rule, &facet->flow,
3461 new_packets, new_bytes, facet->used);
3463 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
3464 facet->mirrors, new_packets, new_bytes);
3468 struct ofproto_push {
3469 struct action_xlate_ctx ctx;
3476 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
3478 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3481 rule->packet_count += push->packets;
3482 rule->byte_count += push->bytes;
3483 rule->used = MAX(push->used, rule->used);
3487 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3488 * 'rule''s actions and mirrors. */
3490 flow_push_stats(const struct rule_dpif *rule,
3491 const struct flow *flow, uint64_t packets, uint64_t bytes,
3494 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3495 struct ofproto_push push;
3497 push.packets = packets;
3501 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, NULL);
3502 push.ctx.resubmit_hook = push_resubmit;
3503 ofpbuf_delete(xlate_actions(&push.ctx,
3504 rule->up.actions, rule->up.n_actions));
3509 static struct subfacet *
3510 subfacet_find__(struct ofproto_dpif *ofproto,
3511 const struct nlattr *key, size_t key_len, uint32_t key_hash,
3512 const struct flow *flow)
3514 struct subfacet *subfacet;
3516 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
3517 &ofproto->subfacets) {
3519 ? (subfacet->key_len == key_len
3520 && !memcmp(key, subfacet->key, key_len))
3521 : flow_equal(flow, &subfacet->facet->flow)) {
3529 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
3530 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
3531 * there is one, otherwise creates and returns a new subfacet.
3533 * If the returned subfacet is new, then subfacet->actions will be NULL, in
3534 * which case the caller must populate the actions with
3535 * subfacet_make_actions(). */
3536 static struct subfacet *
3537 subfacet_create(struct ofproto_dpif *ofproto, struct facet *facet,
3538 enum odp_key_fitness key_fitness,
3539 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
3541 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3542 struct subfacet *subfacet;
3544 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
3546 if (subfacet->facet == facet) {
3550 /* This shouldn't happen. */
3551 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
3552 subfacet_destroy(ofproto, subfacet);
3555 subfacet = xzalloc(sizeof *subfacet);
3556 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
3557 list_push_back(&facet->subfacets, &subfacet->list_node);
3558 subfacet->facet = facet;
3559 subfacet->used = time_msec();
3560 subfacet->key_fitness = key_fitness;
3561 if (key_fitness != ODP_FIT_PERFECT) {
3562 subfacet->key = xmemdup(key, key_len);
3563 subfacet->key_len = key_len;
3565 subfacet->installed = false;
3566 subfacet->initial_tci = initial_tci;
3571 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
3572 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
3573 static struct subfacet *
3574 subfacet_find(struct ofproto_dpif *ofproto,
3575 const struct nlattr *key, size_t key_len)
3577 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3578 enum odp_key_fitness fitness;
3581 fitness = odp_flow_key_to_flow(key, key_len, &flow);
3582 if (fitness == ODP_FIT_ERROR) {
3586 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
3589 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
3590 * its facet within 'ofproto', and frees it. */
3592 subfacet_destroy__(struct ofproto_dpif *ofproto, struct subfacet *subfacet)
3594 subfacet_uninstall(ofproto, subfacet);
3595 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
3596 list_remove(&subfacet->list_node);
3597 free(subfacet->key);
3598 free(subfacet->actions);
3602 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
3603 * last remaining subfacet in its facet destroys the facet too. */
3605 subfacet_destroy(struct ofproto_dpif *ofproto, struct subfacet *subfacet)
3607 struct facet *facet = subfacet->facet;
3609 subfacet_destroy__(ofproto, subfacet);
3610 if (list_is_empty(&facet->subfacets)) {
3611 facet_remove(ofproto, facet);
3615 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
3616 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
3617 * for use as temporary storage. */
3619 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
3622 if (!subfacet->key) {
3623 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
3624 odp_flow_key_from_flow(key, &subfacet->facet->flow);
3626 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
3630 /* Composes the datapath actions for 'subfacet' based on its rule's actions. */
3632 subfacet_make_actions(struct ofproto_dpif *p, struct subfacet *subfacet,
3633 const struct ofpbuf *packet)
3635 struct facet *facet = subfacet->facet;
3636 const struct rule_dpif *rule = facet->rule;
3637 struct ofpbuf *odp_actions;
3638 struct action_xlate_ctx ctx;
3640 action_xlate_ctx_init(&ctx, p, &facet->flow, subfacet->initial_tci,
3642 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3643 facet->tags = ctx.tags;
3644 facet->may_install = ctx.may_set_up_flow;
3645 facet->has_learn = ctx.has_learn;
3646 facet->has_normal = ctx.has_normal;
3647 facet->nf_flow.output_iface = ctx.nf_output_iface;
3648 facet->mirrors = ctx.mirrors;
3650 if (subfacet->actions_len != odp_actions->size
3651 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
3652 free(subfacet->actions);
3653 subfacet->actions_len = odp_actions->size;
3654 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
3657 ofpbuf_delete(odp_actions);
3660 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
3661 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
3662 * in the datapath will be zeroed and 'stats' will be updated with traffic new
3663 * since 'subfacet' was last updated.
3665 * Returns 0 if successful, otherwise a positive errno value. */
3667 subfacet_install(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3668 const struct nlattr *actions, size_t actions_len,
3669 struct dpif_flow_stats *stats)
3671 struct odputil_keybuf keybuf;
3672 enum dpif_flow_put_flags flags;
3676 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3678 flags |= DPIF_FP_ZERO_STATS;
3681 subfacet_get_key(subfacet, &keybuf, &key);
3682 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
3683 actions, actions_len, stats);
3686 subfacet_reset_dp_stats(subfacet, stats);
3692 /* If 'subfacet' is installed in the datapath, uninstalls it. */
3694 subfacet_uninstall(struct ofproto_dpif *p, struct subfacet *subfacet)
3696 if (subfacet->installed) {
3697 struct odputil_keybuf keybuf;
3698 struct dpif_flow_stats stats;
3702 subfacet_get_key(subfacet, &keybuf, &key);
3703 error = dpif_flow_del(p->dpif, key.data, key.size, &stats);
3704 subfacet_reset_dp_stats(subfacet, &stats);
3706 subfacet_update_stats(p, subfacet, &stats);
3708 subfacet->installed = false;
3710 assert(subfacet->dp_packet_count == 0);
3711 assert(subfacet->dp_byte_count == 0);
3715 /* Resets 'subfacet''s datapath statistics counters. This should be called
3716 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
3717 * non-null, it should contain the statistics returned by dpif when 'subfacet'
3718 * was reset in the datapath. 'stats' will be modified to include only
3719 * statistics new since 'subfacet' was last updated. */
3721 subfacet_reset_dp_stats(struct subfacet *subfacet,
3722 struct dpif_flow_stats *stats)
3725 && subfacet->dp_packet_count <= stats->n_packets
3726 && subfacet->dp_byte_count <= stats->n_bytes) {
3727 stats->n_packets -= subfacet->dp_packet_count;
3728 stats->n_bytes -= subfacet->dp_byte_count;
3731 subfacet->dp_packet_count = 0;
3732 subfacet->dp_byte_count = 0;
3735 /* Updates 'subfacet''s used time. The caller is responsible for calling
3736 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
3738 subfacet_update_time(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3741 if (used > subfacet->used) {
3742 subfacet->used = used;
3743 facet_update_time(ofproto, subfacet->facet, used);
3747 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
3749 * Because of the meaning of a subfacet's counters, it only makes sense to do
3750 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
3751 * represents a packet that was sent by hand or if it represents statistics
3752 * that have been cleared out of the datapath. */
3754 subfacet_update_stats(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3755 const struct dpif_flow_stats *stats)
3757 if (stats->n_packets || stats->used > subfacet->used) {
3758 struct facet *facet = subfacet->facet;
3760 subfacet_update_time(ofproto, subfacet, stats->used);
3761 facet->packet_count += stats->n_packets;
3762 facet->byte_count += stats->n_bytes;
3763 facet_push_stats(facet);
3764 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3770 static struct rule_dpif *
3771 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
3774 struct cls_rule *cls_rule;
3775 struct classifier *cls;
3777 if (table_id >= N_TABLES) {
3781 cls = &ofproto->up.tables[table_id];
3782 if (flow->nw_frag & FLOW_NW_FRAG_ANY
3783 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3784 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
3785 * are unavailable. */
3786 struct flow ofpc_normal_flow = *flow;
3787 ofpc_normal_flow.tp_src = htons(0);
3788 ofpc_normal_flow.tp_dst = htons(0);
3789 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
3791 cls_rule = classifier_lookup(cls, flow);
3793 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
3797 complete_operation(struct rule_dpif *rule)
3799 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3801 rule_invalidate(rule);
3803 struct dpif_completion *c = xmalloc(sizeof *c);
3804 c->op = rule->up.pending;
3805 list_push_back(&ofproto->completions, &c->list_node);
3807 ofoperation_complete(rule->up.pending, 0);
3811 static struct rule *
3814 struct rule_dpif *rule = xmalloc(sizeof *rule);
3819 rule_dealloc(struct rule *rule_)
3821 struct rule_dpif *rule = rule_dpif_cast(rule_);
3826 rule_construct(struct rule *rule_)
3828 struct rule_dpif *rule = rule_dpif_cast(rule_);
3829 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3830 struct rule_dpif *victim;
3834 error = validate_actions(rule->up.actions, rule->up.n_actions,
3835 &rule->up.cr.flow, ofproto->max_ports);
3840 rule->used = rule->up.created;
3841 rule->packet_count = 0;
3842 rule->byte_count = 0;
3844 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
3845 if (victim && !list_is_empty(&victim->facets)) {
3846 struct facet *facet;
3848 rule->facets = victim->facets;
3849 list_moved(&rule->facets);
3850 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3851 /* XXX: We're only clearing our local counters here. It's possible
3852 * that quite a few packets are unaccounted for in the datapath
3853 * statistics. These will be accounted to the new rule instead of
3854 * cleared as required. This could be fixed by clearing out the
3855 * datapath statistics for this facet, but currently it doesn't
3857 facet_reset_counters(facet);
3861 /* Must avoid list_moved() in this case. */
3862 list_init(&rule->facets);
3865 table_id = rule->up.table_id;
3866 rule->tag = (victim ? victim->tag
3868 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
3869 ofproto->tables[table_id].basis));
3871 complete_operation(rule);
3876 rule_destruct(struct rule *rule_)
3878 struct rule_dpif *rule = rule_dpif_cast(rule_);
3879 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3880 struct facet *facet, *next_facet;
3882 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3883 facet_revalidate(ofproto, facet);
3886 complete_operation(rule);
3890 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
3892 struct rule_dpif *rule = rule_dpif_cast(rule_);
3893 struct facet *facet;
3895 /* Start from historical data for 'rule' itself that are no longer tracked
3896 * in facets. This counts, for example, facets that have expired. */
3897 *packets = rule->packet_count;
3898 *bytes = rule->byte_count;
3900 /* Add any statistics that are tracked by facets. This includes
3901 * statistical data recently updated by ofproto_update_stats() as well as
3902 * stats for packets that were executed "by hand" via dpif_execute(). */
3903 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3904 *packets += facet->packet_count;
3905 *bytes += facet->byte_count;
3910 rule_execute(struct rule *rule_, const struct flow *flow,
3911 struct ofpbuf *packet)
3913 struct rule_dpif *rule = rule_dpif_cast(rule_);
3914 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3915 struct action_xlate_ctx ctx;
3916 struct ofpbuf *odp_actions;
3919 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, packet);
3920 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3921 size = packet->size;
3922 if (execute_odp_actions(ofproto, flow, odp_actions->data,
3923 odp_actions->size, packet)) {
3924 rule->used = time_msec();
3925 rule->packet_count++;
3926 rule->byte_count += size;
3927 flow_push_stats(rule, flow, 1, size, rule->used);
3929 ofpbuf_delete(odp_actions);
3935 rule_modify_actions(struct rule *rule_)
3937 struct rule_dpif *rule = rule_dpif_cast(rule_);
3938 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3941 error = validate_actions(rule->up.actions, rule->up.n_actions,
3942 &rule->up.cr.flow, ofproto->max_ports);
3944 ofoperation_complete(rule->up.pending, error);
3948 complete_operation(rule);
3951 /* Sends 'packet' out 'ofport'.
3952 * May modify 'packet'.
3953 * Returns 0 if successful, otherwise a positive errno value. */
3955 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
3957 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3958 struct ofpbuf key, odp_actions;
3959 struct odputil_keybuf keybuf;
3964 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
3965 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
3967 if (odp_port != ofport->odp_port) {
3968 eth_pop_vlan(packet);
3969 flow.vlan_tci = htons(0);
3972 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3973 odp_flow_key_from_flow(&key, &flow);
3975 ofpbuf_init(&odp_actions, 32);
3976 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
3978 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
3979 error = dpif_execute(ofproto->dpif,
3981 odp_actions.data, odp_actions.size,
3983 ofpbuf_uninit(&odp_actions);
3986 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
3987 ofproto->up.name, odp_port, strerror(error));
3989 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
3993 /* OpenFlow to datapath action translation. */
3995 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
3996 struct action_xlate_ctx *ctx);
3997 static void xlate_normal(struct action_xlate_ctx *);
4000 put_userspace_action(const struct ofproto_dpif *ofproto,
4001 struct ofpbuf *odp_actions,
4002 const struct flow *flow,
4003 const struct user_action_cookie *cookie)
4007 pid = dpif_port_get_pid(ofproto->dpif,
4008 ofp_port_to_odp_port(flow->in_port));
4010 return odp_put_userspace_action(pid, cookie, odp_actions);
4013 /* Compose SAMPLE action for sFlow. */
4015 compose_sflow_action(const struct ofproto_dpif *ofproto,
4016 struct ofpbuf *odp_actions,
4017 const struct flow *flow,
4020 uint32_t port_ifindex;
4021 uint32_t probability;
4022 struct user_action_cookie cookie;
4023 size_t sample_offset, actions_offset;
4024 int cookie_offset, n_output;
4026 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4030 if (odp_port == OVSP_NONE) {
4034 port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4038 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4040 /* Number of packets out of UINT_MAX to sample. */
4041 probability = dpif_sflow_get_probability(ofproto->sflow);
4042 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4044 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4046 cookie.type = USER_ACTION_COOKIE_SFLOW;
4047 cookie.data = port_ifindex;
4048 cookie.n_output = n_output;
4049 cookie.vlan_tci = 0;
4050 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4052 nl_msg_end_nested(odp_actions, actions_offset);
4053 nl_msg_end_nested(odp_actions, sample_offset);
4054 return cookie_offset;
4057 /* SAMPLE action must be first action in any given list of actions.
4058 * At this point we do not have all information required to build it. So try to
4059 * build sample action as complete as possible. */
4061 add_sflow_action(struct action_xlate_ctx *ctx)
4063 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4065 &ctx->flow, OVSP_NONE);
4066 ctx->sflow_odp_port = 0;
4067 ctx->sflow_n_outputs = 0;
4070 /* Fix SAMPLE action according to data collected while composing ODP actions.
4071 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4072 * USERSPACE action's user-cookie which is required for sflow. */
4074 fix_sflow_action(struct action_xlate_ctx *ctx)
4076 const struct flow *base = &ctx->base_flow;
4077 struct user_action_cookie *cookie;
4079 if (!ctx->user_cookie_offset) {
4083 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4085 assert(cookie != NULL);
4086 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4088 if (ctx->sflow_n_outputs) {
4089 cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow,
4090 ctx->sflow_odp_port);
4092 if (ctx->sflow_n_outputs >= 255) {
4093 cookie->n_output = 255;
4095 cookie->n_output = ctx->sflow_n_outputs;
4097 cookie->vlan_tci = base->vlan_tci;
4101 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4104 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4105 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4106 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4107 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4111 struct priority_to_dscp *pdscp;
4113 if (ofport->up.opp.config & htonl(OFPPC_NO_FWD)
4114 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4118 pdscp = get_priority(ofport, ctx->flow.priority);
4120 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4121 ctx->flow.nw_tos |= pdscp->dscp;
4124 /* We may not have an ofport record for this port, but it doesn't hurt
4125 * to allow forwarding to it anyhow. Maybe such a port will appear
4126 * later and we're pre-populating the flow table. */
4129 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4130 ctx->flow.vlan_tci);
4131 if (out_port != odp_port) {
4132 ctx->flow.vlan_tci = htons(0);
4134 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4135 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4137 ctx->sflow_odp_port = odp_port;
4138 ctx->sflow_n_outputs++;
4139 ctx->nf_output_iface = ofp_port;
4140 ctx->flow.vlan_tci = flow_vlan_tci;
4141 ctx->flow.nw_tos = flow_nw_tos;
4145 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4147 compose_output_action__(ctx, ofp_port, true);
4151 xlate_table_action(struct action_xlate_ctx *ctx,
4152 uint16_t in_port, uint8_t table_id)
4154 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4155 struct ofproto_dpif *ofproto = ctx->ofproto;
4156 struct rule_dpif *rule;
4157 uint16_t old_in_port;
4158 uint8_t old_table_id;
4160 old_table_id = ctx->table_id;
4161 ctx->table_id = table_id;
4163 /* Look up a flow with 'in_port' as the input port. */
4164 old_in_port = ctx->flow.in_port;
4165 ctx->flow.in_port = in_port;
4166 rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
4169 if (table_id > 0 && table_id < N_TABLES) {
4170 struct table_dpif *table = &ofproto->tables[table_id];
4171 if (table->other_table) {
4174 : rule_calculate_tag(&ctx->flow,
4175 &table->other_table->wc,
4180 /* Restore the original input port. Otherwise OFPP_NORMAL and
4181 * OFPP_IN_PORT will have surprising behavior. */
4182 ctx->flow.in_port = old_in_port;
4184 if (ctx->resubmit_hook) {
4185 ctx->resubmit_hook(ctx, rule);
4190 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
4194 ctx->table_id = old_table_id;
4196 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4198 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
4199 MAX_RESUBMIT_RECURSION);
4204 xlate_resubmit_table(struct action_xlate_ctx *ctx,
4205 const struct nx_action_resubmit *nar)
4210 in_port = (nar->in_port == htons(OFPP_IN_PORT)
4212 : ntohs(nar->in_port));
4213 table_id = nar->table == 255 ? ctx->table_id : nar->table;
4215 xlate_table_action(ctx, in_port, table_id);
4219 flood_packets(struct action_xlate_ctx *ctx, bool all)
4221 struct ofport_dpif *ofport;
4223 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
4224 uint16_t ofp_port = ofport->up.ofp_port;
4226 if (ofp_port == ctx->flow.in_port) {
4231 compose_output_action__(ctx, ofp_port, false);
4232 } else if (!(ofport->up.opp.config & htonl(OFPPC_NO_FLOOD))) {
4233 compose_output_action(ctx, ofp_port);
4237 ctx->nf_output_iface = NF_OUT_FLOOD;
4241 compose_controller_action(struct action_xlate_ctx *ctx, int len)
4243 struct user_action_cookie cookie;
4245 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4246 cookie.type = USER_ACTION_COOKIE_CONTROLLER;
4248 cookie.n_output = 0;
4249 cookie.vlan_tci = 0;
4250 put_userspace_action(ctx->ofproto, ctx->odp_actions, &ctx->flow, &cookie);
4254 xlate_output_action__(struct action_xlate_ctx *ctx,
4255 uint16_t port, uint16_t max_len)
4257 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
4259 ctx->nf_output_iface = NF_OUT_DROP;
4263 compose_output_action(ctx, ctx->flow.in_port);
4266 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
4272 flood_packets(ctx, false);
4275 flood_packets(ctx, true);
4277 case OFPP_CONTROLLER:
4278 compose_controller_action(ctx, max_len);
4281 compose_output_action(ctx, OFPP_LOCAL);
4286 if (port != ctx->flow.in_port) {
4287 compose_output_action(ctx, port);
4292 if (prev_nf_output_iface == NF_OUT_FLOOD) {
4293 ctx->nf_output_iface = NF_OUT_FLOOD;
4294 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
4295 ctx->nf_output_iface = prev_nf_output_iface;
4296 } else if (prev_nf_output_iface != NF_OUT_DROP &&
4297 ctx->nf_output_iface != NF_OUT_FLOOD) {
4298 ctx->nf_output_iface = NF_OUT_MULTI;
4303 xlate_output_reg_action(struct action_xlate_ctx *ctx,
4304 const struct nx_action_output_reg *naor)
4308 ofp_port = nxm_read_field_bits(naor->src, naor->ofs_nbits, &ctx->flow);
4310 if (ofp_port <= UINT16_MAX) {
4311 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
4316 xlate_output_action(struct action_xlate_ctx *ctx,
4317 const struct ofp_action_output *oao)
4319 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
4323 xlate_enqueue_action(struct action_xlate_ctx *ctx,
4324 const struct ofp_action_enqueue *oae)
4327 uint32_t flow_priority, priority;
4330 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
4333 /* Fall back to ordinary output action. */
4334 xlate_output_action__(ctx, ntohs(oae->port), 0);
4338 /* Figure out datapath output port. */
4339 ofp_port = ntohs(oae->port);
4340 if (ofp_port == OFPP_IN_PORT) {
4341 ofp_port = ctx->flow.in_port;
4342 } else if (ofp_port == ctx->flow.in_port) {
4346 /* Add datapath actions. */
4347 flow_priority = ctx->flow.priority;
4348 ctx->flow.priority = priority;
4349 compose_output_action(ctx, ofp_port);
4350 ctx->flow.priority = flow_priority;
4352 /* Update NetFlow output port. */
4353 if (ctx->nf_output_iface == NF_OUT_DROP) {
4354 ctx->nf_output_iface = ofp_port;
4355 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
4356 ctx->nf_output_iface = NF_OUT_MULTI;
4361 xlate_set_queue_action(struct action_xlate_ctx *ctx,
4362 const struct nx_action_set_queue *nasq)
4367 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
4370 /* Couldn't translate queue to a priority, so ignore. A warning
4371 * has already been logged. */
4375 ctx->flow.priority = priority;
4378 struct xlate_reg_state {
4384 xlate_autopath(struct action_xlate_ctx *ctx,
4385 const struct nx_action_autopath *naa)
4387 uint16_t ofp_port = ntohl(naa->id);
4388 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
4390 if (!port || !port->bundle) {
4391 ofp_port = OFPP_NONE;
4392 } else if (port->bundle->bond) {
4393 /* Autopath does not support VLAN hashing. */
4394 struct ofport_dpif *slave = bond_choose_output_slave(
4395 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
4397 ofp_port = slave->up.ofp_port;
4400 autopath_execute(naa, &ctx->flow, ofp_port);
4404 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
4406 struct ofproto_dpif *ofproto = ofproto_;
4407 struct ofport_dpif *port;
4417 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
4420 port = get_ofp_port(ofproto, ofp_port);
4421 return port ? port->may_enable : false;
4426 xlate_learn_action(struct action_xlate_ctx *ctx,
4427 const struct nx_action_learn *learn)
4429 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
4430 struct ofputil_flow_mod fm;
4433 learn_execute(learn, &ctx->flow, &fm);
4435 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
4436 if (error && !VLOG_DROP_WARN(&rl)) {
4437 char *msg = ofputil_error_to_string(error);
4438 VLOG_WARN("learning action failed to modify flow table (%s)", msg);
4446 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
4448 if (port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
4449 ? htonl(OFPPC_NO_RECV_STP)
4450 : htonl(OFPPC_NO_RECV))) {
4454 /* Only drop packets here if both forwarding and learning are
4455 * disabled. If just learning is enabled, we need to have
4456 * OFPP_NORMAL and the learning action have a look at the packet
4457 * before we can drop it. */
4458 if (!stp_forward_in_state(port->stp_state)
4459 && !stp_learn_in_state(port->stp_state)) {
4467 do_xlate_actions(const union ofp_action *in, size_t n_in,
4468 struct action_xlate_ctx *ctx)
4470 const struct ofport_dpif *port;
4471 const union ofp_action *ia;
4474 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
4475 if (port && !may_receive(port, ctx)) {
4476 /* Drop this flow. */
4480 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
4481 const struct ofp_action_dl_addr *oada;
4482 const struct nx_action_resubmit *nar;
4483 const struct nx_action_set_tunnel *nast;
4484 const struct nx_action_set_queue *nasq;
4485 const struct nx_action_multipath *nam;
4486 const struct nx_action_autopath *naa;
4487 const struct nx_action_bundle *nab;
4488 const struct nx_action_output_reg *naor;
4489 enum ofputil_action_code code;
4496 code = ofputil_decode_action_unsafe(ia);
4498 case OFPUTIL_OFPAT_OUTPUT:
4499 xlate_output_action(ctx, &ia->output);
4502 case OFPUTIL_OFPAT_SET_VLAN_VID:
4503 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
4504 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
4507 case OFPUTIL_OFPAT_SET_VLAN_PCP:
4508 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
4509 ctx->flow.vlan_tci |= htons(
4510 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
4513 case OFPUTIL_OFPAT_STRIP_VLAN:
4514 ctx->flow.vlan_tci = htons(0);
4517 case OFPUTIL_OFPAT_SET_DL_SRC:
4518 oada = ((struct ofp_action_dl_addr *) ia);
4519 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
4522 case OFPUTIL_OFPAT_SET_DL_DST:
4523 oada = ((struct ofp_action_dl_addr *) ia);
4524 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
4527 case OFPUTIL_OFPAT_SET_NW_SRC:
4528 ctx->flow.nw_src = ia->nw_addr.nw_addr;
4531 case OFPUTIL_OFPAT_SET_NW_DST:
4532 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
4535 case OFPUTIL_OFPAT_SET_NW_TOS:
4536 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4537 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
4540 case OFPUTIL_OFPAT_SET_TP_SRC:
4541 ctx->flow.tp_src = ia->tp_port.tp_port;
4544 case OFPUTIL_OFPAT_SET_TP_DST:
4545 ctx->flow.tp_dst = ia->tp_port.tp_port;
4548 case OFPUTIL_OFPAT_ENQUEUE:
4549 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
4552 case OFPUTIL_NXAST_RESUBMIT:
4553 nar = (const struct nx_action_resubmit *) ia;
4554 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
4557 case OFPUTIL_NXAST_RESUBMIT_TABLE:
4558 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
4561 case OFPUTIL_NXAST_SET_TUNNEL:
4562 nast = (const struct nx_action_set_tunnel *) ia;
4563 tun_id = htonll(ntohl(nast->tun_id));
4564 ctx->flow.tun_id = tun_id;
4567 case OFPUTIL_NXAST_SET_QUEUE:
4568 nasq = (const struct nx_action_set_queue *) ia;
4569 xlate_set_queue_action(ctx, nasq);
4572 case OFPUTIL_NXAST_POP_QUEUE:
4573 ctx->flow.priority = ctx->original_priority;
4576 case OFPUTIL_NXAST_REG_MOVE:
4577 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
4581 case OFPUTIL_NXAST_REG_LOAD:
4582 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
4586 case OFPUTIL_NXAST_NOTE:
4587 /* Nothing to do. */
4590 case OFPUTIL_NXAST_SET_TUNNEL64:
4591 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
4592 ctx->flow.tun_id = tun_id;
4595 case OFPUTIL_NXAST_MULTIPATH:
4596 nam = (const struct nx_action_multipath *) ia;
4597 multipath_execute(nam, &ctx->flow);
4600 case OFPUTIL_NXAST_AUTOPATH:
4601 naa = (const struct nx_action_autopath *) ia;
4602 xlate_autopath(ctx, naa);
4605 case OFPUTIL_NXAST_BUNDLE:
4606 ctx->ofproto->has_bundle_action = true;
4607 nab = (const struct nx_action_bundle *) ia;
4608 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
4613 case OFPUTIL_NXAST_BUNDLE_LOAD:
4614 ctx->ofproto->has_bundle_action = true;
4615 nab = (const struct nx_action_bundle *) ia;
4616 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
4620 case OFPUTIL_NXAST_OUTPUT_REG:
4621 naor = (const struct nx_action_output_reg *) ia;
4622 xlate_output_reg_action(ctx, naor);
4625 case OFPUTIL_NXAST_LEARN:
4626 ctx->has_learn = true;
4627 if (ctx->may_learn) {
4628 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
4632 case OFPUTIL_NXAST_EXIT:
4638 /* We've let OFPP_NORMAL and the learning action look at the packet,
4639 * so drop it now if forwarding is disabled. */
4640 if (port && !stp_forward_in_state(port->stp_state)) {
4641 ofpbuf_clear(ctx->odp_actions);
4642 add_sflow_action(ctx);
4647 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
4648 struct ofproto_dpif *ofproto, const struct flow *flow,
4649 ovs_be16 initial_tci, const struct ofpbuf *packet)
4651 ctx->ofproto = ofproto;
4653 ctx->base_flow = ctx->flow;
4654 ctx->base_flow.tun_id = 0;
4655 ctx->base_flow.vlan_tci = initial_tci;
4656 ctx->packet = packet;
4657 ctx->may_learn = packet != NULL;
4658 ctx->resubmit_hook = NULL;
4661 static struct ofpbuf *
4662 xlate_actions(struct action_xlate_ctx *ctx,
4663 const union ofp_action *in, size_t n_in)
4665 struct flow orig_flow = ctx->flow;
4667 COVERAGE_INC(ofproto_dpif_xlate);
4669 ctx->odp_actions = ofpbuf_new(512);
4670 ofpbuf_reserve(ctx->odp_actions, NL_A_U32_SIZE);
4672 ctx->may_set_up_flow = true;
4673 ctx->has_learn = false;
4674 ctx->has_normal = false;
4675 ctx->nf_output_iface = NF_OUT_DROP;
4678 ctx->original_priority = ctx->flow.priority;
4682 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
4683 switch (ctx->ofproto->up.frag_handling) {
4684 case OFPC_FRAG_NORMAL:
4685 /* We must pretend that transport ports are unavailable. */
4686 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
4687 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
4690 case OFPC_FRAG_DROP:
4691 return ctx->odp_actions;
4693 case OFPC_FRAG_REASM:
4696 case OFPC_FRAG_NX_MATCH:
4697 /* Nothing to do. */
4702 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
4703 ctx->may_set_up_flow = false;
4704 return ctx->odp_actions;
4706 add_sflow_action(ctx);
4707 do_xlate_actions(in, n_in, ctx);
4709 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
4710 ctx->odp_actions->data,
4711 ctx->odp_actions->size)) {
4712 ctx->may_set_up_flow = false;
4714 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
4716 compose_output_action(ctx, OFPP_LOCAL);
4719 add_mirror_actions(ctx, &orig_flow);
4720 fix_sflow_action(ctx);
4723 return ctx->odp_actions;
4726 /* OFPP_NORMAL implementation. */
4728 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
4730 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
4731 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
4732 * the bundle on which the packet was received, returns the VLAN to which the
4735 * Both 'vid' and the return value are in the range 0...4095. */
4737 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
4739 switch (in_bundle->vlan_mode) {
4740 case PORT_VLAN_ACCESS:
4741 return in_bundle->vlan;
4744 case PORT_VLAN_TRUNK:
4747 case PORT_VLAN_NATIVE_UNTAGGED:
4748 case PORT_VLAN_NATIVE_TAGGED:
4749 return vid ? vid : in_bundle->vlan;
4756 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
4757 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
4760 * 'vid' should be the VID obtained from the 802.1Q header that was received as
4761 * part of a packet (specify 0 if there was no 802.1Q header), in the range
4764 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
4766 switch (in_bundle->vlan_mode) {
4767 case PORT_VLAN_ACCESS:
4770 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4771 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
4772 "packet received on port %s configured as VLAN "
4773 "%"PRIu16" access port",
4774 in_bundle->ofproto->up.name, vid,
4775 in_bundle->name, in_bundle->vlan);
4781 case PORT_VLAN_NATIVE_UNTAGGED:
4782 case PORT_VLAN_NATIVE_TAGGED:
4784 /* Port must always carry its native VLAN. */
4788 case PORT_VLAN_TRUNK:
4789 if (!ofbundle_includes_vlan(in_bundle, vid)) {
4791 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4792 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
4793 "received on port %s not configured for trunking "
4795 in_bundle->ofproto->up.name, vid,
4796 in_bundle->name, vid);
4808 /* Given 'vlan', the VLAN that a packet belongs to, and
4809 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
4810 * that should be included in the 802.1Q header. (If the return value is 0,
4811 * then the 802.1Q header should only be included in the packet if there is a
4814 * Both 'vlan' and the return value are in the range 0...4095. */
4816 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
4818 switch (out_bundle->vlan_mode) {
4819 case PORT_VLAN_ACCESS:
4822 case PORT_VLAN_TRUNK:
4823 case PORT_VLAN_NATIVE_TAGGED:
4826 case PORT_VLAN_NATIVE_UNTAGGED:
4827 return vlan == out_bundle->vlan ? 0 : vlan;
4835 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
4838 struct ofport_dpif *port;
4840 ovs_be16 tci, old_tci;
4842 vid = output_vlan_to_vid(out_bundle, vlan);
4843 if (!out_bundle->bond) {
4844 port = ofbundle_get_a_port(out_bundle);
4846 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
4849 /* No slaves enabled, so drop packet. */
4854 old_tci = ctx->flow.vlan_tci;
4856 if (tci || out_bundle->use_priority_tags) {
4857 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
4859 tci |= htons(VLAN_CFI);
4862 ctx->flow.vlan_tci = tci;
4864 compose_output_action(ctx, port->up.ofp_port);
4865 ctx->flow.vlan_tci = old_tci;
4869 mirror_mask_ffs(mirror_mask_t mask)
4871 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
4876 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
4878 return (bundle->vlan_mode != PORT_VLAN_ACCESS
4879 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
4883 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
4885 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
4888 /* Returns an arbitrary interface within 'bundle'. */
4889 static struct ofport_dpif *
4890 ofbundle_get_a_port(const struct ofbundle *bundle)
4892 return CONTAINER_OF(list_front(&bundle->ports),
4893 struct ofport_dpif, bundle_node);
4897 vlan_is_mirrored(const struct ofmirror *m, int vlan)
4899 return !m->vlans || bitmap_is_set(m->vlans, vlan);
4902 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
4903 * to a VLAN. In general most packets may be mirrored but we want to drop
4904 * protocols that may confuse switches. */
4906 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
4908 /* If you change this function's behavior, please update corresponding
4909 * documentation in vswitch.xml at the same time. */
4910 if (dst[0] != 0x01) {
4911 /* All the currently banned MACs happen to start with 01 currently, so
4912 * this is a quick way to eliminate most of the good ones. */
4914 if (eth_addr_is_reserved(dst)) {
4915 /* Drop STP, IEEE pause frames, and other reserved protocols
4916 * (01-80-c2-00-00-0x). */
4920 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
4922 if ((dst[3] & 0xfe) == 0xcc &&
4923 (dst[4] & 0xfe) == 0xcc &&
4924 (dst[5] & 0xfe) == 0xcc) {
4925 /* Drop the following protocols plus others following the same
4928 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
4929 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
4930 STP Uplink Fast (01-00-0c-cd-cd-cd) */
4934 if (!(dst[3] | dst[4] | dst[5])) {
4935 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
4944 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
4946 struct ofproto_dpif *ofproto = ctx->ofproto;
4947 mirror_mask_t mirrors;
4948 struct ofport_dpif *in_port;
4949 struct ofbundle *in_bundle;
4952 const struct nlattr *a;
4955 /* Obtain in_port from orig_flow.in_port.
4957 * lookup_input_bundle() also ensures that in_port belongs to a bundle. */
4958 in_port = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
4959 ctx->packet != NULL);
4963 in_bundle = in_port->bundle;
4964 mirrors = in_bundle->src_mirrors;
4966 /* Drop frames on bundles reserved for mirroring. */
4967 if (in_bundle->mirror_out) {
4968 if (ctx->packet != NULL) {
4969 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4970 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
4971 "%s, which is reserved exclusively for mirroring",
4972 ctx->ofproto->up.name, in_bundle->name);
4978 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
4979 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
4982 vlan = input_vid_to_vlan(in_bundle, vid);
4984 /* Look at the output ports to check for destination selections. */
4986 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
4987 ctx->odp_actions->size) {
4988 enum ovs_action_attr type = nl_attr_type(a);
4989 struct ofport_dpif *ofport;
4991 if (type != OVS_ACTION_ATTR_OUTPUT) {
4995 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
4996 if (ofport && ofport->bundle) {
4997 mirrors |= ofport->bundle->dst_mirrors;
5005 /* Restore the original packet before adding the mirror actions. */
5006 ctx->flow = *orig_flow;
5011 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5013 if (!vlan_is_mirrored(m, vlan)) {
5014 mirrors &= mirrors - 1;
5018 mirrors &= ~m->dup_mirrors;
5019 ctx->mirrors |= m->dup_mirrors;
5021 output_normal(ctx, m->out, vlan);
5022 } else if (eth_dst_may_rspan(orig_flow->dl_dst)
5023 && vlan != m->out_vlan) {
5024 struct ofbundle *bundle;
5026 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5027 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5028 && !bundle->mirror_out) {
5029 output_normal(ctx, bundle, m->out_vlan);
5037 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5038 uint64_t packets, uint64_t bytes)
5044 for (; mirrors; mirrors &= mirrors - 1) {
5047 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5050 /* In normal circumstances 'm' will not be NULL. However,
5051 * if mirrors are reconfigured, we can temporarily get out
5052 * of sync in facet_revalidate(). We could "correct" the
5053 * mirror list before reaching here, but doing that would
5054 * not properly account the traffic stats we've currently
5055 * accumulated for previous mirror configuration. */
5059 m->packet_count += packets;
5060 m->byte_count += bytes;
5064 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
5065 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
5066 * indicate this; newer upstream kernels use gratuitous ARP requests. */
5068 is_gratuitous_arp(const struct flow *flow)
5070 return (flow->dl_type == htons(ETH_TYPE_ARP)
5071 && eth_addr_is_broadcast(flow->dl_dst)
5072 && (flow->nw_proto == ARP_OP_REPLY
5073 || (flow->nw_proto == ARP_OP_REQUEST
5074 && flow->nw_src == flow->nw_dst)));
5078 update_learning_table(struct ofproto_dpif *ofproto,
5079 const struct flow *flow, int vlan,
5080 struct ofbundle *in_bundle)
5082 struct mac_entry *mac;
5084 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
5088 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
5089 if (is_gratuitous_arp(flow)) {
5090 /* We don't want to learn from gratuitous ARP packets that are
5091 * reflected back over bond slaves so we lock the learning table. */
5092 if (!in_bundle->bond) {
5093 mac_entry_set_grat_arp_lock(mac);
5094 } else if (mac_entry_is_grat_arp_locked(mac)) {
5099 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
5100 /* The log messages here could actually be useful in debugging,
5101 * so keep the rate limit relatively high. */
5102 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5103 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
5104 "on port %s in VLAN %d",
5105 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
5106 in_bundle->name, vlan);
5108 mac->port.p = in_bundle;
5109 tag_set_add(&ofproto->revalidate_set,
5110 mac_learning_changed(ofproto->ml, mac));
5114 static struct ofport_dpif *
5115 lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn)
5117 struct ofport_dpif *ofport;
5119 /* Find the port and bundle for the received packet. */
5120 ofport = get_ofp_port(ofproto, in_port);
5121 if (ofport && ofport->bundle) {
5125 /* Odd. A few possible reasons here:
5127 * - We deleted a port but there are still a few packets queued up
5130 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
5131 * we don't know about.
5133 * - The ofproto client didn't configure the port as part of a bundle.
5136 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5138 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
5139 "port %"PRIu16, ofproto->up.name, in_port);
5144 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
5145 * dropped. Returns true if they may be forwarded, false if they should be
5148 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
5149 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
5151 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
5152 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
5153 * checked by input_vid_is_valid().
5155 * May also add tags to '*tags', although the current implementation only does
5156 * so in one special case.
5159 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
5160 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
5162 struct ofbundle *in_bundle = in_port->bundle;
5164 /* Drop frames for reserved multicast addresses
5165 * only if forward_bpdu option is absent. */
5166 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
5170 if (in_bundle->bond) {
5171 struct mac_entry *mac;
5173 switch (bond_check_admissibility(in_bundle->bond, in_port,
5174 flow->dl_dst, tags)) {
5181 case BV_DROP_IF_MOVED:
5182 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
5183 if (mac && mac->port.p != in_bundle &&
5184 (!is_gratuitous_arp(flow)
5185 || mac_entry_is_grat_arp_locked(mac))) {
5196 xlate_normal(struct action_xlate_ctx *ctx)
5198 struct ofport_dpif *in_port;
5199 struct ofbundle *in_bundle;
5200 struct mac_entry *mac;
5204 ctx->has_normal = true;
5206 /* Obtain in_port from ctx->flow.in_port.
5208 * lookup_input_bundle() also ensures that in_port belongs to a bundle. */
5209 in_port = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
5210 ctx->packet != NULL);
5214 in_bundle = in_port->bundle;
5216 /* Drop malformed frames. */
5217 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
5218 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
5219 if (ctx->packet != NULL) {
5220 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5221 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
5222 "VLAN tag received on port %s",
5223 ctx->ofproto->up.name, in_bundle->name);
5228 /* Drop frames on bundles reserved for mirroring. */
5229 if (in_bundle->mirror_out) {
5230 if (ctx->packet != NULL) {
5231 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5232 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5233 "%s, which is reserved exclusively for mirroring",
5234 ctx->ofproto->up.name, in_bundle->name);
5240 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
5241 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5244 vlan = input_vid_to_vlan(in_bundle, vid);
5246 /* Check other admissibility requirements. */
5247 if (!is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
5251 /* Learn source MAC. */
5252 if (ctx->may_learn) {
5253 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
5256 /* Determine output bundle. */
5257 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
5260 if (mac->port.p != in_bundle) {
5261 output_normal(ctx, mac->port.p, vlan);
5263 } else if (!ctx->packet && !eth_addr_is_multicast(ctx->flow.dl_dst)) {
5264 /* If we are revalidating but don't have a learning entry then eject
5265 * the flow. Installing a flow that floods packets opens up a window
5266 * of time where we could learn from a packet reflected on a bond and
5267 * blackhole packets before the learning table is updated to reflect
5268 * the correct port. */
5269 ctx->may_set_up_flow = false;
5272 struct ofbundle *bundle;
5274 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
5275 if (bundle != in_bundle
5276 && ofbundle_includes_vlan(bundle, vlan)
5277 && bundle->floodable
5278 && !bundle->mirror_out) {
5279 output_normal(ctx, bundle, vlan);
5282 ctx->nf_output_iface = NF_OUT_FLOOD;
5286 /* Optimized flow revalidation.
5288 * It's a difficult problem, in general, to tell which facets need to have
5289 * their actions recalculated whenever the OpenFlow flow table changes. We
5290 * don't try to solve that general problem: for most kinds of OpenFlow flow
5291 * table changes, we recalculate the actions for every facet. This is
5292 * relatively expensive, but it's good enough if the OpenFlow flow table
5293 * doesn't change very often.
5295 * However, we can expect one particular kind of OpenFlow flow table change to
5296 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5297 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5298 * table, we add a special case that applies to flow tables in which every rule
5299 * has the same form (that is, the same wildcards), except that the table is
5300 * also allowed to have a single "catch-all" flow that matches all packets. We
5301 * optimize this case by tagging all of the facets that resubmit into the table
5302 * and invalidating the same tag whenever a flow changes in that table. The
5303 * end result is that we revalidate just the facets that need it (and sometimes
5304 * a few more, but not all of the facets or even all of the facets that
5305 * resubmit to the table modified by MAC learning). */
5307 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
5308 * into an OpenFlow table with the given 'basis'. */
5310 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
5313 if (flow_wildcards_is_catchall(wc)) {
5316 struct flow tag_flow = *flow;
5317 flow_zero_wildcards(&tag_flow, wc);
5318 return tag_create_deterministic(flow_hash(&tag_flow, secret));
5322 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5323 * taggability of that table.
5325 * This function must be called after *each* change to a flow table. If you
5326 * skip calling it on some changes then the pointer comparisons at the end can
5327 * be invalid if you get unlucky. For example, if a flow removal causes a
5328 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5329 * different wildcards to be created with the same address, then this function
5330 * will incorrectly skip revalidation. */
5332 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5334 struct table_dpif *table = &ofproto->tables[table_id];
5335 const struct classifier *cls = &ofproto->up.tables[table_id];
5336 struct cls_table *catchall, *other;
5337 struct cls_table *t;
5339 catchall = other = NULL;
5341 switch (hmap_count(&cls->tables)) {
5343 /* We could tag this OpenFlow table but it would make the logic a
5344 * little harder and it's a corner case that doesn't seem worth it
5350 HMAP_FOR_EACH (t, hmap_node, &cls->tables) {
5351 if (cls_table_is_catchall(t)) {
5353 } else if (!other) {
5356 /* Indicate that we can't tag this by setting both tables to
5357 * NULL. (We know that 'catchall' is already NULL.) */
5364 /* Can't tag this table. */
5368 if (table->catchall_table != catchall || table->other_table != other) {
5369 table->catchall_table = catchall;
5370 table->other_table = other;
5371 ofproto->need_revalidate = true;
5375 /* Given 'rule' that has changed in some way (either it is a rule being
5376 * inserted, a rule being deleted, or a rule whose actions are being
5377 * modified), marks facets for revalidation to ensure that packets will be
5378 * forwarded correctly according to the new state of the flow table.
5380 * This function must be called after *each* change to a flow table. See
5381 * the comment on table_update_taggable() for more information. */
5383 rule_invalidate(const struct rule_dpif *rule)
5385 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5387 table_update_taggable(ofproto, rule->up.table_id);
5389 if (!ofproto->need_revalidate) {
5390 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5392 if (table->other_table && rule->tag) {
5393 tag_set_add(&ofproto->revalidate_set, rule->tag);
5395 ofproto->need_revalidate = true;
5401 set_frag_handling(struct ofproto *ofproto_,
5402 enum ofp_config_flags frag_handling)
5404 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5406 if (frag_handling != OFPC_FRAG_REASM) {
5407 ofproto->need_revalidate = true;
5415 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5416 const struct flow *flow,
5417 const union ofp_action *ofp_actions, size_t n_ofp_actions)
5419 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5422 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
5423 return ofp_mkerr_nicira(OFPET_BAD_REQUEST, NXBRC_BAD_IN_PORT);
5426 error = validate_actions(ofp_actions, n_ofp_actions, flow,
5427 ofproto->max_ports);
5429 struct odputil_keybuf keybuf;
5430 struct action_xlate_ctx ctx;
5431 struct ofpbuf *odp_actions;
5434 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5435 odp_flow_key_from_flow(&key, flow);
5437 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, packet);
5438 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
5439 dpif_execute(ofproto->dpif, key.data, key.size,
5440 odp_actions->data, odp_actions->size, packet);
5441 ofpbuf_delete(odp_actions);
5449 set_netflow(struct ofproto *ofproto_,
5450 const struct netflow_options *netflow_options)
5452 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5454 if (netflow_options) {
5455 if (!ofproto->netflow) {
5456 ofproto->netflow = netflow_create();
5458 return netflow_set_options(ofproto->netflow, netflow_options);
5460 netflow_destroy(ofproto->netflow);
5461 ofproto->netflow = NULL;
5467 get_netflow_ids(const struct ofproto *ofproto_,
5468 uint8_t *engine_type, uint8_t *engine_id)
5470 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5472 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
5476 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5478 if (!facet_is_controller_flow(facet) &&
5479 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5480 struct subfacet *subfacet;
5481 struct ofexpired expired;
5483 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5484 if (subfacet->installed) {
5485 struct dpif_flow_stats stats;
5487 subfacet_install(ofproto, subfacet, subfacet->actions,
5488 subfacet->actions_len, &stats);
5489 subfacet_update_stats(ofproto, subfacet, &stats);
5493 expired.flow = facet->flow;
5494 expired.packet_count = facet->packet_count;
5495 expired.byte_count = facet->byte_count;
5496 expired.used = facet->used;
5497 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5502 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5504 struct facet *facet;
5506 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5507 send_active_timeout(ofproto, facet);
5511 static struct ofproto_dpif *
5512 ofproto_dpif_lookup(const char *name)
5514 struct ofproto *ofproto = ofproto_lookup(name);
5515 return (ofproto && ofproto->ofproto_class == &ofproto_dpif_class
5516 ? ofproto_dpif_cast(ofproto)
5521 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn,
5522 const char *args, void *aux OVS_UNUSED)
5524 const struct ofproto_dpif *ofproto;
5526 ofproto = ofproto_dpif_lookup(args);
5528 unixctl_command_reply(conn, 501, "no such bridge");
5531 mac_learning_flush(ofproto->ml);
5533 unixctl_command_reply(conn, 200, "table successfully flushed");
5537 ofproto_unixctl_fdb_show(struct unixctl_conn *conn,
5538 const char *args, void *aux OVS_UNUSED)
5540 struct ds ds = DS_EMPTY_INITIALIZER;
5541 const struct ofproto_dpif *ofproto;
5542 const struct mac_entry *e;
5544 ofproto = ofproto_dpif_lookup(args);
5546 unixctl_command_reply(conn, 501, "no such bridge");
5550 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5551 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5552 struct ofbundle *bundle = e->port.p;
5553 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
5554 ofbundle_get_a_port(bundle)->odp_port,
5555 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
5557 unixctl_command_reply(conn, 200, ds_cstr(&ds));
5561 struct ofproto_trace {
5562 struct action_xlate_ctx ctx;
5568 trace_format_rule(struct ds *result, uint8_t table_id, int level,
5569 const struct rule_dpif *rule)
5571 ds_put_char_multiple(result, '\t', level);
5573 ds_put_cstr(result, "No match\n");
5577 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5578 table_id, ntohll(rule->up.flow_cookie));
5579 cls_rule_format(&rule->up.cr, result);
5580 ds_put_char(result, '\n');
5582 ds_put_char_multiple(result, '\t', level);
5583 ds_put_cstr(result, "OpenFlow ");
5584 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
5585 ds_put_char(result, '\n');
5589 trace_format_flow(struct ds *result, int level, const char *title,
5590 struct ofproto_trace *trace)
5592 ds_put_char_multiple(result, '\t', level);
5593 ds_put_format(result, "%s: ", title);
5594 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
5595 ds_put_cstr(result, "unchanged");
5597 flow_format(result, &trace->ctx.flow);
5598 trace->flow = trace->ctx.flow;
5600 ds_put_char(result, '\n');
5604 trace_format_regs(struct ds *result, int level, const char *title,
5605 struct ofproto_trace *trace)
5609 ds_put_char_multiple(result, '\t', level);
5610 ds_put_format(result, "%s:", title);
5611 for (i = 0; i < FLOW_N_REGS; i++) {
5612 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5614 ds_put_char(result, '\n');
5618 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
5620 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
5621 struct ds *result = trace->result;
5623 ds_put_char(result, '\n');
5624 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
5625 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
5626 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
5630 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
5631 void *aux OVS_UNUSED)
5633 char *dpname, *arg1, *arg2, *arg3, *arg4;
5634 char *args = xstrdup(args_);
5635 char *save_ptr = NULL;
5636 struct ofproto_dpif *ofproto;
5637 struct ofpbuf odp_key;
5638 struct ofpbuf *packet;
5639 struct rule_dpif *rule;
5640 ovs_be16 initial_tci;
5646 ofpbuf_init(&odp_key, 0);
5649 dpname = strtok_r(args, " ", &save_ptr);
5651 unixctl_command_reply(conn, 501, "Bad command syntax");
5655 ofproto = ofproto_dpif_lookup(dpname);
5657 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
5661 arg1 = strtok_r(NULL, " ", &save_ptr);
5662 arg2 = strtok_r(NULL, " ", &save_ptr);
5663 arg3 = strtok_r(NULL, " ", &save_ptr);
5664 arg4 = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
5665 if (dpname && arg1 && (!arg2 || !strcmp(arg2, "-generate")) && !arg3) {
5666 /* ofproto/trace dpname flow [-generate] */
5669 /* Convert string to datapath key. */
5670 ofpbuf_init(&odp_key, 0);
5671 error = odp_flow_key_from_string(arg1, NULL, &odp_key);
5673 unixctl_command_reply(conn, 501, "Bad flow syntax");
5677 /* Convert odp_key to flow. */
5678 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
5679 odp_key.size, &flow,
5681 if (error == ODP_FIT_ERROR) {
5682 unixctl_command_reply(conn, 501, "Invalid flow");
5686 /* Generate a packet, if requested. */
5688 packet = ofpbuf_new(0);
5689 flow_compose(packet, &flow);
5691 } else if (dpname && arg1 && arg2 && arg3 && arg4) {
5692 /* ofproto/trace dpname priority tun_id in_port packet */
5697 priority = atoi(arg1);
5698 tun_id = htonll(strtoull(arg2, NULL, 0));
5699 in_port = ofp_port_to_odp_port(atoi(arg3));
5701 packet = ofpbuf_new(strlen(args) / 2);
5702 arg4 = ofpbuf_put_hex(packet, arg4, NULL);
5703 arg4 += strspn(arg4, " ");
5704 if (*arg4 != '\0') {
5705 unixctl_command_reply(conn, 501, "Trailing garbage in command");
5708 if (packet->size < ETH_HEADER_LEN) {
5709 unixctl_command_reply(conn, 501,
5710 "Packet data too short for Ethernet");
5714 ds_put_cstr(&result, "Packet: ");
5715 s = ofp_packet_to_string(packet->data, packet->size, packet->size);
5716 ds_put_cstr(&result, s);
5719 flow_extract(packet, priority, tun_id, in_port, &flow);
5720 initial_tci = flow.vlan_tci;
5722 unixctl_command_reply(conn, 501, "Bad command syntax");
5726 ds_put_cstr(&result, "Flow: ");
5727 flow_format(&result, &flow);
5728 ds_put_char(&result, '\n');
5730 rule = rule_dpif_lookup(ofproto, &flow, 0);
5731 trace_format_rule(&result, 0, 0, rule);
5733 struct ofproto_trace trace;
5734 struct ofpbuf *odp_actions;
5736 trace.result = &result;
5738 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, initial_tci, packet);
5739 trace.ctx.resubmit_hook = trace_resubmit;
5740 odp_actions = xlate_actions(&trace.ctx,
5741 rule->up.actions, rule->up.n_actions);
5743 ds_put_char(&result, '\n');
5744 trace_format_flow(&result, 0, "Final flow", &trace);
5745 ds_put_cstr(&result, "Datapath actions: ");
5746 format_odp_actions(&result, odp_actions->data, odp_actions->size);
5747 ofpbuf_delete(odp_actions);
5749 if (!trace.ctx.may_set_up_flow) {
5751 ds_put_cstr(&result, "\nThis flow is not cachable.");
5753 ds_put_cstr(&result, "\nThe datapath actions are incomplete--"
5754 "for complete actions, please supply a packet.");
5759 unixctl_command_reply(conn, 200, ds_cstr(&result));
5762 ds_destroy(&result);
5763 ofpbuf_delete(packet);
5764 ofpbuf_uninit(&odp_key);
5769 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED,
5770 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
5773 unixctl_command_reply(conn, 200, NULL);
5777 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED,
5778 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
5781 unixctl_command_reply(conn, 200, NULL);
5785 ofproto_dpif_unixctl_init(void)
5787 static bool registered;
5793 unixctl_command_register("ofproto/trace",
5794 "bridge {tun_id in_port packet | odp_flow [-generate]}",
5795 ofproto_unixctl_trace, NULL);
5796 unixctl_command_register("fdb/flush", "bridge", ofproto_unixctl_fdb_flush,
5798 unixctl_command_register("fdb/show", "bridge", ofproto_unixctl_fdb_show,
5800 unixctl_command_register("ofproto/clog", "", ofproto_dpif_clog, NULL);
5801 unixctl_command_register("ofproto/unclog", "", ofproto_dpif_unclog, NULL);
5804 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5806 * This is deprecated. It is only for compatibility with broken device drivers
5807 * in old versions of Linux that do not properly support VLANs when VLAN
5808 * devices are not used. When broken device drivers are no longer in
5809 * widespread use, we will delete these interfaces. */
5812 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
5814 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
5815 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
5817 if (realdev_ofp_port == ofport->realdev_ofp_port
5818 && vid == ofport->vlandev_vid) {
5822 ofproto->need_revalidate = true;
5824 if (ofport->realdev_ofp_port) {
5827 if (realdev_ofp_port && ofport->bundle) {
5828 /* vlandevs are enslaved to their realdevs, so they are not allowed to
5829 * themselves be part of a bundle. */
5830 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
5833 ofport->realdev_ofp_port = realdev_ofp_port;
5834 ofport->vlandev_vid = vid;
5836 if (realdev_ofp_port) {
5837 vsp_add(ofport, realdev_ofp_port, vid);
5844 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
5846 return hash_2words(realdev_ofp_port, vid);
5850 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
5851 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
5853 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
5854 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
5855 int vid = vlan_tci_to_vid(vlan_tci);
5856 const struct vlan_splinter *vsp;
5858 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
5859 hash_realdev_vid(realdev_ofp_port, vid),
5860 &ofproto->realdev_vid_map) {
5861 if (vsp->realdev_ofp_port == realdev_ofp_port
5862 && vsp->vid == vid) {
5863 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
5867 return realdev_odp_port;
5870 static struct vlan_splinter *
5871 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
5873 struct vlan_splinter *vsp;
5875 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
5876 &ofproto->vlandev_map) {
5877 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
5886 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
5887 uint16_t vlandev_ofp_port, int *vid)
5889 if (!hmap_is_empty(&ofproto->vlandev_map)) {
5890 const struct vlan_splinter *vsp;
5892 vsp = vlandev_find(ofproto, vlandev_ofp_port);
5897 return vsp->realdev_ofp_port;
5904 vsp_remove(struct ofport_dpif *port)
5906 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
5907 struct vlan_splinter *vsp;
5909 vsp = vlandev_find(ofproto, port->up.ofp_port);
5911 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
5912 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
5915 port->realdev_ofp_port = 0;
5917 VLOG_ERR("missing vlan device record");
5922 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
5924 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
5926 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
5927 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
5928 == realdev_ofp_port)) {
5929 struct vlan_splinter *vsp;
5931 vsp = xmalloc(sizeof *vsp);
5932 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
5933 hash_int(port->up.ofp_port, 0));
5934 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
5935 hash_realdev_vid(realdev_ofp_port, vid));
5936 vsp->realdev_ofp_port = realdev_ofp_port;
5937 vsp->vlandev_ofp_port = port->up.ofp_port;
5940 port->realdev_ofp_port = realdev_ofp_port;
5942 VLOG_ERR("duplicate vlan device record");
5946 const struct ofproto_class ofproto_dpif_class = {
5975 port_is_lacp_current,
5976 NULL, /* rule_choose_table */
5983 rule_modify_actions,
5991 get_cfm_remote_mpids,
5995 get_stp_port_status,
6002 is_mirror_output_bundle,
6003 forward_bpdu_changed,