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 const struct flow *);
378 static void subfacet_destroy(struct ofproto_dpif *, struct subfacet *);
379 static void subfacet_destroy__(struct ofproto_dpif *, struct subfacet *);
380 static void subfacet_reset_dp_stats(struct subfacet *,
381 struct dpif_flow_stats *);
382 static void subfacet_update_time(struct ofproto_dpif *, struct subfacet *,
384 static void subfacet_update_stats(struct ofproto_dpif *, struct subfacet *,
385 const struct dpif_flow_stats *);
386 static void subfacet_make_actions(struct ofproto_dpif *, struct subfacet *,
387 const struct ofpbuf *packet);
388 static int subfacet_install(struct ofproto_dpif *, struct subfacet *,
389 const struct nlattr *actions, size_t actions_len,
390 struct dpif_flow_stats *);
391 static void subfacet_uninstall(struct ofproto_dpif *, struct subfacet *);
397 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
398 struct list bundle_node; /* In struct ofbundle's "ports" list. */
399 struct cfm *cfm; /* Connectivity Fault Management, if any. */
400 tag_type tag; /* Tag associated with this port. */
401 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
402 bool may_enable; /* May be enabled in bonds. */
405 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
406 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
407 long long int stp_state_entered;
409 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
411 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
413 * This is deprecated. It is only for compatibility with broken device
414 * drivers in old versions of Linux that do not properly support VLANs when
415 * VLAN devices are not used. When broken device drivers are no longer in
416 * widespread use, we will delete these interfaces. */
417 uint16_t realdev_ofp_port;
421 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
422 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
423 * traffic egressing the 'ofport' with that priority should be marked with. */
424 struct priority_to_dscp {
425 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
426 uint32_t priority; /* Priority of this queue (see struct flow). */
428 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
431 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
433 * This is deprecated. It is only for compatibility with broken device drivers
434 * in old versions of Linux that do not properly support VLANs when VLAN
435 * devices are not used. When broken device drivers are no longer in
436 * widespread use, we will delete these interfaces. */
437 struct vlan_splinter {
438 struct hmap_node realdev_vid_node;
439 struct hmap_node vlandev_node;
440 uint16_t realdev_ofp_port;
441 uint16_t vlandev_ofp_port;
445 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
446 uint32_t realdev, ovs_be16 vlan_tci);
447 static uint16_t vsp_vlandev_to_realdev(const struct ofproto_dpif *,
448 uint16_t vlandev, int *vid);
449 static void vsp_remove(struct ofport_dpif *);
450 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
452 static struct ofport_dpif *
453 ofport_dpif_cast(const struct ofport *ofport)
455 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
456 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
459 static void port_run(struct ofport_dpif *);
460 static void port_wait(struct ofport_dpif *);
461 static int set_cfm(struct ofport *, const struct cfm_settings *);
462 static void ofport_clear_priorities(struct ofport_dpif *);
464 struct dpif_completion {
465 struct list list_node;
466 struct ofoperation *op;
469 /* Extra information about a classifier table.
470 * Currently used just for optimized flow revalidation. */
472 /* If either of these is nonnull, then this table has a form that allows
473 * flows to be tagged to avoid revalidating most flows for the most common
474 * kinds of flow table changes. */
475 struct cls_table *catchall_table; /* Table that wildcards all fields. */
476 struct cls_table *other_table; /* Table with any other wildcard set. */
477 uint32_t basis; /* Keeps each table's tags separate. */
480 struct ofproto_dpif {
489 struct netflow *netflow;
490 struct dpif_sflow *sflow;
491 struct hmap bundles; /* Contains "struct ofbundle"s. */
492 struct mac_learning *ml;
493 struct ofmirror *mirrors[MAX_MIRRORS];
494 bool has_bonded_bundles;
497 struct timer next_expiration;
501 struct hmap subfacets;
504 struct table_dpif tables[N_TABLES];
505 bool need_revalidate;
506 struct tag_set revalidate_set;
508 /* Support for debugging async flow mods. */
509 struct list completions;
511 bool has_bundle_action; /* True when the first bundle action appears. */
515 long long int stp_last_tick;
517 /* VLAN splinters. */
518 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
519 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
522 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
523 * for debugging the asynchronous flow_mod implementation.) */
526 static void ofproto_dpif_unixctl_init(void);
528 static struct ofproto_dpif *
529 ofproto_dpif_cast(const struct ofproto *ofproto)
531 assert(ofproto->ofproto_class == &ofproto_dpif_class);
532 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
535 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
537 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
540 /* Packet processing. */
541 static void update_learning_table(struct ofproto_dpif *,
542 const struct flow *, int vlan,
545 #define FLOW_MISS_MAX_BATCH 50
547 static void handle_upcall(struct ofproto_dpif *, struct dpif_upcall *);
548 static void handle_miss_upcalls(struct ofproto_dpif *,
549 struct dpif_upcall *, size_t n);
551 /* Flow expiration. */
552 static int expire(struct ofproto_dpif *);
555 static void send_netflow_active_timeouts(struct ofproto_dpif *);
558 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
560 compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions,
561 const struct flow *, uint32_t odp_port);
562 static void add_mirror_actions(struct action_xlate_ctx *ctx,
563 const struct flow *flow);
564 /* Global variables. */
565 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
567 /* Factory functions. */
570 enumerate_types(struct sset *types)
572 dp_enumerate_types(types);
576 enumerate_names(const char *type, struct sset *names)
578 return dp_enumerate_names(type, names);
582 del(const char *type, const char *name)
587 error = dpif_open(name, type, &dpif);
589 error = dpif_delete(dpif);
595 /* Basic life-cycle. */
597 static struct ofproto *
600 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
605 dealloc(struct ofproto *ofproto_)
607 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
612 construct(struct ofproto *ofproto_, int *n_tablesp)
614 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
615 const char *name = ofproto->up.name;
619 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
621 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
625 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
626 ofproto->n_matches = 0;
628 dpif_flow_flush(ofproto->dpif);
629 dpif_recv_purge(ofproto->dpif);
631 error = dpif_recv_set_mask(ofproto->dpif,
632 ((1u << DPIF_UC_MISS) |
633 (1u << DPIF_UC_ACTION)));
635 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
636 dpif_close(ofproto->dpif);
640 ofproto->netflow = NULL;
641 ofproto->sflow = NULL;
643 hmap_init(&ofproto->bundles);
644 ofproto->ml = mac_learning_create();
645 for (i = 0; i < MAX_MIRRORS; i++) {
646 ofproto->mirrors[i] = NULL;
648 ofproto->has_bonded_bundles = false;
650 timer_set_duration(&ofproto->next_expiration, 1000);
652 hmap_init(&ofproto->facets);
653 hmap_init(&ofproto->subfacets);
655 for (i = 0; i < N_TABLES; i++) {
656 struct table_dpif *table = &ofproto->tables[i];
658 table->catchall_table = NULL;
659 table->other_table = NULL;
660 table->basis = random_uint32();
662 ofproto->need_revalidate = false;
663 tag_set_init(&ofproto->revalidate_set);
665 list_init(&ofproto->completions);
667 ofproto_dpif_unixctl_init();
669 ofproto->has_bundle_action = false;
671 hmap_init(&ofproto->vlandev_map);
672 hmap_init(&ofproto->realdev_vid_map);
674 *n_tablesp = N_TABLES;
679 complete_operations(struct ofproto_dpif *ofproto)
681 struct dpif_completion *c, *next;
683 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
684 ofoperation_complete(c->op, 0);
685 list_remove(&c->list_node);
691 destruct(struct ofproto *ofproto_)
693 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
694 struct rule_dpif *rule, *next_rule;
695 struct classifier *table;
698 complete_operations(ofproto);
700 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
701 struct cls_cursor cursor;
703 cls_cursor_init(&cursor, table, NULL);
704 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
705 ofproto_rule_destroy(&rule->up);
709 for (i = 0; i < MAX_MIRRORS; i++) {
710 mirror_destroy(ofproto->mirrors[i]);
713 netflow_destroy(ofproto->netflow);
714 dpif_sflow_destroy(ofproto->sflow);
715 hmap_destroy(&ofproto->bundles);
716 mac_learning_destroy(ofproto->ml);
718 hmap_destroy(&ofproto->facets);
719 hmap_destroy(&ofproto->subfacets);
721 hmap_destroy(&ofproto->vlandev_map);
722 hmap_destroy(&ofproto->realdev_vid_map);
724 dpif_close(ofproto->dpif);
728 run(struct ofproto *ofproto_)
730 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
731 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
732 struct ofport_dpif *ofport;
733 struct ofbundle *bundle;
738 complete_operations(ofproto);
740 dpif_run(ofproto->dpif);
743 for (i = 0; i < FLOW_MISS_MAX_BATCH; i++) {
744 struct dpif_upcall *upcall = &misses[n_misses];
747 error = dpif_recv(ofproto->dpif, upcall);
749 if (error == ENODEV && n_misses == 0) {
755 if (upcall->type == DPIF_UC_MISS) {
756 /* Handle it later. */
759 handle_upcall(ofproto, upcall);
763 handle_miss_upcalls(ofproto, misses, n_misses);
765 if (timer_expired(&ofproto->next_expiration)) {
766 int delay = expire(ofproto);
767 timer_set_duration(&ofproto->next_expiration, delay);
770 if (ofproto->netflow) {
771 if (netflow_run(ofproto->netflow)) {
772 send_netflow_active_timeouts(ofproto);
775 if (ofproto->sflow) {
776 dpif_sflow_run(ofproto->sflow);
779 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
782 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
787 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
789 /* Now revalidate if there's anything to do. */
790 if (ofproto->need_revalidate
791 || !tag_set_is_empty(&ofproto->revalidate_set)) {
792 struct tag_set revalidate_set = ofproto->revalidate_set;
793 bool revalidate_all = ofproto->need_revalidate;
794 struct facet *facet, *next;
796 /* Clear the revalidation flags. */
797 tag_set_init(&ofproto->revalidate_set);
798 ofproto->need_revalidate = false;
800 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
802 || tag_set_intersects(&revalidate_set, facet->tags)) {
803 facet_revalidate(ofproto, facet);
812 wait(struct ofproto *ofproto_)
814 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
815 struct ofport_dpif *ofport;
816 struct ofbundle *bundle;
818 if (!clogged && !list_is_empty(&ofproto->completions)) {
819 poll_immediate_wake();
822 dpif_wait(ofproto->dpif);
823 dpif_recv_wait(ofproto->dpif);
824 if (ofproto->sflow) {
825 dpif_sflow_wait(ofproto->sflow);
827 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
828 poll_immediate_wake();
830 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
833 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
836 if (ofproto->netflow) {
837 netflow_wait(ofproto->netflow);
839 mac_learning_wait(ofproto->ml);
841 if (ofproto->need_revalidate) {
842 /* Shouldn't happen, but if it does just go around again. */
843 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
844 poll_immediate_wake();
846 timer_wait(&ofproto->next_expiration);
851 flush(struct ofproto *ofproto_)
853 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
854 struct facet *facet, *next_facet;
856 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
857 /* Mark the facet as not installed so that facet_remove() doesn't
858 * bother trying to uninstall it. There is no point in uninstalling it
859 * individually since we are about to blow away all the facets with
860 * dpif_flow_flush(). */
861 struct subfacet *subfacet;
863 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
864 subfacet->installed = false;
865 subfacet->dp_packet_count = 0;
866 subfacet->dp_byte_count = 0;
868 facet_remove(ofproto, facet);
870 dpif_flow_flush(ofproto->dpif);
874 get_features(struct ofproto *ofproto_ OVS_UNUSED,
875 bool *arp_match_ip, uint32_t *actions)
877 *arp_match_ip = true;
878 *actions = ((1u << OFPAT_OUTPUT) |
879 (1u << OFPAT_SET_VLAN_VID) |
880 (1u << OFPAT_SET_VLAN_PCP) |
881 (1u << OFPAT_STRIP_VLAN) |
882 (1u << OFPAT_SET_DL_SRC) |
883 (1u << OFPAT_SET_DL_DST) |
884 (1u << OFPAT_SET_NW_SRC) |
885 (1u << OFPAT_SET_NW_DST) |
886 (1u << OFPAT_SET_NW_TOS) |
887 (1u << OFPAT_SET_TP_SRC) |
888 (1u << OFPAT_SET_TP_DST) |
889 (1u << OFPAT_ENQUEUE));
893 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
895 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
896 struct dpif_dp_stats s;
898 strcpy(ots->name, "classifier");
900 dpif_get_dp_stats(ofproto->dpif, &s);
901 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
902 put_32aligned_be64(&ots->matched_count,
903 htonll(s.n_hit + ofproto->n_matches));
906 static struct ofport *
909 struct ofport_dpif *port = xmalloc(sizeof *port);
914 port_dealloc(struct ofport *port_)
916 struct ofport_dpif *port = ofport_dpif_cast(port_);
921 port_construct(struct ofport *port_)
923 struct ofport_dpif *port = ofport_dpif_cast(port_);
924 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
926 ofproto->need_revalidate = true;
927 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
930 port->tag = tag_create_random();
931 port->may_enable = true;
932 port->stp_port = NULL;
933 port->stp_state = STP_DISABLED;
934 hmap_init(&port->priorities);
935 port->realdev_ofp_port = 0;
936 port->vlandev_vid = 0;
938 if (ofproto->sflow) {
939 dpif_sflow_add_port(ofproto->sflow, port->odp_port,
940 netdev_get_name(port->up.netdev));
947 port_destruct(struct ofport *port_)
949 struct ofport_dpif *port = ofport_dpif_cast(port_);
950 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
952 ofproto->need_revalidate = true;
953 bundle_remove(port_);
954 set_cfm(port_, NULL);
955 if (ofproto->sflow) {
956 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
959 ofport_clear_priorities(port);
960 hmap_destroy(&port->priorities);
964 port_modified(struct ofport *port_)
966 struct ofport_dpif *port = ofport_dpif_cast(port_);
968 if (port->bundle && port->bundle->bond) {
969 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
974 port_reconfigured(struct ofport *port_, ovs_be32 old_config)
976 struct ofport_dpif *port = ofport_dpif_cast(port_);
977 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
978 ovs_be32 changed = old_config ^ port->up.opp.config;
980 if (changed & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
981 OFPPC_NO_FWD | OFPPC_NO_FLOOD)) {
982 ofproto->need_revalidate = true;
984 if (changed & htonl(OFPPC_NO_FLOOD) && port->bundle) {
985 bundle_update(port->bundle);
991 set_sflow(struct ofproto *ofproto_,
992 const struct ofproto_sflow_options *sflow_options)
994 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
995 struct dpif_sflow *ds = ofproto->sflow;
999 struct ofport_dpif *ofport;
1001 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1002 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1003 dpif_sflow_add_port(ds, ofport->odp_port,
1004 netdev_get_name(ofport->up.netdev));
1006 ofproto->need_revalidate = true;
1008 dpif_sflow_set_options(ds, sflow_options);
1011 dpif_sflow_destroy(ds);
1012 ofproto->need_revalidate = true;
1013 ofproto->sflow = NULL;
1020 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1022 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1029 struct ofproto_dpif *ofproto;
1031 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1032 ofproto->need_revalidate = true;
1033 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1036 if (cfm_configure(ofport->cfm, s)) {
1042 cfm_destroy(ofport->cfm);
1048 get_cfm_fault(const struct ofport *ofport_)
1050 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1052 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1056 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1059 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1062 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1069 /* Spanning Tree. */
1072 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1074 struct ofproto_dpif *ofproto = ofproto_;
1075 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1076 struct ofport_dpif *ofport;
1078 ofport = stp_port_get_aux(sp);
1080 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1081 ofproto->up.name, port_num);
1083 struct eth_header *eth = pkt->l2;
1085 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1086 if (eth_addr_is_zero(eth->eth_src)) {
1087 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1088 "with unknown MAC", ofproto->up.name, port_num);
1090 send_packet(ofport, pkt);
1096 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1098 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1100 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1102 /* Only revalidate flows if the configuration changed. */
1103 if (!s != !ofproto->stp) {
1104 ofproto->need_revalidate = true;
1108 if (!ofproto->stp) {
1109 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1110 send_bpdu_cb, ofproto);
1111 ofproto->stp_last_tick = time_msec();
1114 stp_set_bridge_id(ofproto->stp, s->system_id);
1115 stp_set_bridge_priority(ofproto->stp, s->priority);
1116 stp_set_hello_time(ofproto->stp, s->hello_time);
1117 stp_set_max_age(ofproto->stp, s->max_age);
1118 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1120 stp_destroy(ofproto->stp);
1121 ofproto->stp = NULL;
1128 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1130 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1134 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1135 s->designated_root = stp_get_designated_root(ofproto->stp);
1136 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1145 update_stp_port_state(struct ofport_dpif *ofport)
1147 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1148 enum stp_state state;
1150 /* Figure out new state. */
1151 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1155 if (ofport->stp_state != state) {
1159 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1160 netdev_get_name(ofport->up.netdev),
1161 stp_state_name(ofport->stp_state),
1162 stp_state_name(state));
1163 if (stp_learn_in_state(ofport->stp_state)
1164 != stp_learn_in_state(state)) {
1165 /* xxx Learning action flows should also be flushed. */
1166 mac_learning_flush(ofproto->ml);
1168 fwd_change = stp_forward_in_state(ofport->stp_state)
1169 != stp_forward_in_state(state);
1171 ofproto->need_revalidate = true;
1172 ofport->stp_state = state;
1173 ofport->stp_state_entered = time_msec();
1175 if (fwd_change && ofport->bundle) {
1176 bundle_update(ofport->bundle);
1179 /* Update the STP state bits in the OpenFlow port description. */
1180 of_state = (ofport->up.opp.state & htonl(~OFPPS_STP_MASK))
1181 | htonl(state == STP_LISTENING ? OFPPS_STP_LISTEN
1182 : state == STP_LEARNING ? OFPPS_STP_LEARN
1183 : state == STP_FORWARDING ? OFPPS_STP_FORWARD
1184 : state == STP_BLOCKING ? OFPPS_STP_BLOCK
1186 ofproto_port_set_state(&ofport->up, of_state);
1190 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1191 * caller is responsible for assigning STP port numbers and ensuring
1192 * there are no duplicates. */
1194 set_stp_port(struct ofport *ofport_,
1195 const struct ofproto_port_stp_settings *s)
1197 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1198 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1199 struct stp_port *sp = ofport->stp_port;
1201 if (!s || !s->enable) {
1203 ofport->stp_port = NULL;
1204 stp_port_disable(sp);
1205 update_stp_port_state(ofport);
1208 } else if (sp && stp_port_no(sp) != s->port_num
1209 && ofport == stp_port_get_aux(sp)) {
1210 /* The port-id changed, so disable the old one if it's not
1211 * already in use by another port. */
1212 stp_port_disable(sp);
1215 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1216 stp_port_enable(sp);
1218 stp_port_set_aux(sp, ofport);
1219 stp_port_set_priority(sp, s->priority);
1220 stp_port_set_path_cost(sp, s->path_cost);
1222 update_stp_port_state(ofport);
1228 get_stp_port_status(struct ofport *ofport_,
1229 struct ofproto_port_stp_status *s)
1231 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1232 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1233 struct stp_port *sp = ofport->stp_port;
1235 if (!ofproto->stp || !sp) {
1241 s->port_id = stp_port_get_id(sp);
1242 s->state = stp_port_get_state(sp);
1243 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1244 s->role = stp_port_get_role(sp);
1245 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1251 stp_run(struct ofproto_dpif *ofproto)
1254 long long int now = time_msec();
1255 long long int elapsed = now - ofproto->stp_last_tick;
1256 struct stp_port *sp;
1259 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1260 ofproto->stp_last_tick = now;
1262 while (stp_get_changed_port(ofproto->stp, &sp)) {
1263 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1266 update_stp_port_state(ofport);
1273 stp_wait(struct ofproto_dpif *ofproto)
1276 poll_timer_wait(1000);
1280 /* Returns true if STP should process 'flow'. */
1282 stp_should_process_flow(const struct flow *flow)
1284 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1288 stp_process_packet(const struct ofport_dpif *ofport,
1289 const struct ofpbuf *packet)
1291 struct ofpbuf payload = *packet;
1292 struct eth_header *eth = payload.data;
1293 struct stp_port *sp = ofport->stp_port;
1295 /* Sink packets on ports that have STP disabled when the bridge has
1297 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1301 /* Trim off padding on payload. */
1302 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1303 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1306 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1307 stp_received_bpdu(sp, payload.data, payload.size);
1311 static struct priority_to_dscp *
1312 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1314 struct priority_to_dscp *pdscp;
1317 hash = hash_int(priority, 0);
1318 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1319 if (pdscp->priority == priority) {
1327 ofport_clear_priorities(struct ofport_dpif *ofport)
1329 struct priority_to_dscp *pdscp, *next;
1331 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1332 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1338 set_queues(struct ofport *ofport_,
1339 const struct ofproto_port_queue *qdscp_list,
1342 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1343 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1344 struct hmap new = HMAP_INITIALIZER(&new);
1347 for (i = 0; i < n_qdscp; i++) {
1348 struct priority_to_dscp *pdscp;
1352 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1353 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1358 pdscp = get_priority(ofport, priority);
1360 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1362 pdscp = xmalloc(sizeof *pdscp);
1363 pdscp->priority = priority;
1365 ofproto->need_revalidate = true;
1368 if (pdscp->dscp != dscp) {
1370 ofproto->need_revalidate = true;
1373 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1376 if (!hmap_is_empty(&ofport->priorities)) {
1377 ofport_clear_priorities(ofport);
1378 ofproto->need_revalidate = true;
1381 hmap_swap(&new, &ofport->priorities);
1389 /* Expires all MAC learning entries associated with 'port' and forces ofproto
1390 * to revalidate every flow. */
1392 bundle_flush_macs(struct ofbundle *bundle)
1394 struct ofproto_dpif *ofproto = bundle->ofproto;
1395 struct mac_learning *ml = ofproto->ml;
1396 struct mac_entry *mac, *next_mac;
1398 ofproto->need_revalidate = true;
1399 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1400 if (mac->port.p == bundle) {
1401 mac_learning_expire(ml, mac);
1406 static struct ofbundle *
1407 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1409 struct ofbundle *bundle;
1411 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1412 &ofproto->bundles) {
1413 if (bundle->aux == aux) {
1420 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1421 * ones that are found to 'bundles'. */
1423 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1424 void **auxes, size_t n_auxes,
1425 struct hmapx *bundles)
1429 hmapx_init(bundles);
1430 for (i = 0; i < n_auxes; i++) {
1431 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1433 hmapx_add(bundles, bundle);
1439 bundle_update(struct ofbundle *bundle)
1441 struct ofport_dpif *port;
1443 bundle->floodable = true;
1444 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1445 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
1446 bundle->floodable = false;
1453 bundle_del_port(struct ofport_dpif *port)
1455 struct ofbundle *bundle = port->bundle;
1457 bundle->ofproto->need_revalidate = true;
1459 list_remove(&port->bundle_node);
1460 port->bundle = NULL;
1463 lacp_slave_unregister(bundle->lacp, port);
1466 bond_slave_unregister(bundle->bond, port);
1469 bundle_update(bundle);
1473 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1474 struct lacp_slave_settings *lacp,
1475 uint32_t bond_stable_id)
1477 struct ofport_dpif *port;
1479 port = get_ofp_port(bundle->ofproto, ofp_port);
1484 if (port->bundle != bundle) {
1485 bundle->ofproto->need_revalidate = true;
1487 bundle_del_port(port);
1490 port->bundle = bundle;
1491 list_push_back(&bundle->ports, &port->bundle_node);
1492 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
1493 bundle->floodable = false;
1497 port->bundle->ofproto->need_revalidate = true;
1498 lacp_slave_register(bundle->lacp, port, lacp);
1501 port->bond_stable_id = bond_stable_id;
1507 bundle_destroy(struct ofbundle *bundle)
1509 struct ofproto_dpif *ofproto;
1510 struct ofport_dpif *port, *next_port;
1517 ofproto = bundle->ofproto;
1518 for (i = 0; i < MAX_MIRRORS; i++) {
1519 struct ofmirror *m = ofproto->mirrors[i];
1521 if (m->out == bundle) {
1523 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1524 || hmapx_find_and_delete(&m->dsts, bundle)) {
1525 ofproto->need_revalidate = true;
1530 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1531 bundle_del_port(port);
1534 bundle_flush_macs(bundle);
1535 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1537 free(bundle->trunks);
1538 lacp_destroy(bundle->lacp);
1539 bond_destroy(bundle->bond);
1544 bundle_set(struct ofproto *ofproto_, void *aux,
1545 const struct ofproto_bundle_settings *s)
1547 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1548 bool need_flush = false;
1549 struct ofport_dpif *port;
1550 struct ofbundle *bundle;
1551 unsigned long *trunks;
1557 bundle_destroy(bundle_lookup(ofproto, aux));
1561 assert(s->n_slaves == 1 || s->bond != NULL);
1562 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1564 bundle = bundle_lookup(ofproto, aux);
1566 bundle = xmalloc(sizeof *bundle);
1568 bundle->ofproto = ofproto;
1569 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1570 hash_pointer(aux, 0));
1572 bundle->name = NULL;
1574 list_init(&bundle->ports);
1575 bundle->vlan_mode = PORT_VLAN_TRUNK;
1577 bundle->trunks = NULL;
1578 bundle->use_priority_tags = s->use_priority_tags;
1579 bundle->lacp = NULL;
1580 bundle->bond = NULL;
1582 bundle->floodable = true;
1584 bundle->src_mirrors = 0;
1585 bundle->dst_mirrors = 0;
1586 bundle->mirror_out = 0;
1589 if (!bundle->name || strcmp(s->name, bundle->name)) {
1591 bundle->name = xstrdup(s->name);
1596 if (!bundle->lacp) {
1597 ofproto->need_revalidate = true;
1598 bundle->lacp = lacp_create();
1600 lacp_configure(bundle->lacp, s->lacp);
1602 lacp_destroy(bundle->lacp);
1603 bundle->lacp = NULL;
1606 /* Update set of ports. */
1608 for (i = 0; i < s->n_slaves; i++) {
1609 if (!bundle_add_port(bundle, s->slaves[i],
1610 s->lacp ? &s->lacp_slaves[i] : NULL,
1611 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1615 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1616 struct ofport_dpif *next_port;
1618 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1619 for (i = 0; i < s->n_slaves; i++) {
1620 if (s->slaves[i] == port->up.ofp_port) {
1625 bundle_del_port(port);
1629 assert(list_size(&bundle->ports) <= s->n_slaves);
1631 if (list_is_empty(&bundle->ports)) {
1632 bundle_destroy(bundle);
1636 /* Set VLAN tagging mode */
1637 if (s->vlan_mode != bundle->vlan_mode
1638 || s->use_priority_tags != bundle->use_priority_tags) {
1639 bundle->vlan_mode = s->vlan_mode;
1640 bundle->use_priority_tags = s->use_priority_tags;
1645 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1646 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1648 if (vlan != bundle->vlan) {
1649 bundle->vlan = vlan;
1653 /* Get trunked VLANs. */
1654 switch (s->vlan_mode) {
1655 case PORT_VLAN_ACCESS:
1659 case PORT_VLAN_TRUNK:
1660 trunks = (unsigned long *) s->trunks;
1663 case PORT_VLAN_NATIVE_UNTAGGED:
1664 case PORT_VLAN_NATIVE_TAGGED:
1665 if (vlan != 0 && (!s->trunks
1666 || !bitmap_is_set(s->trunks, vlan)
1667 || bitmap_is_set(s->trunks, 0))) {
1668 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1670 trunks = bitmap_clone(s->trunks, 4096);
1672 trunks = bitmap_allocate1(4096);
1674 bitmap_set1(trunks, vlan);
1675 bitmap_set0(trunks, 0);
1677 trunks = (unsigned long *) s->trunks;
1684 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1685 free(bundle->trunks);
1686 if (trunks == s->trunks) {
1687 bundle->trunks = vlan_bitmap_clone(trunks);
1689 bundle->trunks = trunks;
1694 if (trunks != s->trunks) {
1699 if (!list_is_short(&bundle->ports)) {
1700 bundle->ofproto->has_bonded_bundles = true;
1702 if (bond_reconfigure(bundle->bond, s->bond)) {
1703 ofproto->need_revalidate = true;
1706 bundle->bond = bond_create(s->bond);
1707 ofproto->need_revalidate = true;
1710 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1711 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1715 bond_destroy(bundle->bond);
1716 bundle->bond = NULL;
1719 /* If we changed something that would affect MAC learning, un-learn
1720 * everything on this port and force flow revalidation. */
1722 bundle_flush_macs(bundle);
1729 bundle_remove(struct ofport *port_)
1731 struct ofport_dpif *port = ofport_dpif_cast(port_);
1732 struct ofbundle *bundle = port->bundle;
1735 bundle_del_port(port);
1736 if (list_is_empty(&bundle->ports)) {
1737 bundle_destroy(bundle);
1738 } else if (list_is_short(&bundle->ports)) {
1739 bond_destroy(bundle->bond);
1740 bundle->bond = NULL;
1746 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
1748 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1749 struct ofport_dpif *port = port_;
1750 uint8_t ea[ETH_ADDR_LEN];
1753 error = netdev_get_etheraddr(port->up.netdev, ea);
1755 struct ofpbuf packet;
1758 ofpbuf_init(&packet, 0);
1759 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1761 memcpy(packet_pdu, pdu, pdu_size);
1763 send_packet(port, &packet);
1764 ofpbuf_uninit(&packet);
1766 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1767 "%s (%s)", port->bundle->name,
1768 netdev_get_name(port->up.netdev), strerror(error));
1773 bundle_send_learning_packets(struct ofbundle *bundle)
1775 struct ofproto_dpif *ofproto = bundle->ofproto;
1776 int error, n_packets, n_errors;
1777 struct mac_entry *e;
1779 error = n_packets = n_errors = 0;
1780 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1781 if (e->port.p != bundle) {
1782 struct ofpbuf *learning_packet;
1783 struct ofport_dpif *port;
1786 learning_packet = bond_compose_learning_packet(bundle->bond, e->mac,
1789 ret = send_packet(port, learning_packet);
1790 ofpbuf_delete(learning_packet);
1800 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1801 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1802 "packets, last error was: %s",
1803 bundle->name, n_errors, n_packets, strerror(error));
1805 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1806 bundle->name, n_packets);
1811 bundle_run(struct ofbundle *bundle)
1814 lacp_run(bundle->lacp, send_pdu_cb);
1817 struct ofport_dpif *port;
1819 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1820 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1823 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1824 lacp_negotiated(bundle->lacp));
1825 if (bond_should_send_learning_packets(bundle->bond)) {
1826 bundle_send_learning_packets(bundle);
1832 bundle_wait(struct ofbundle *bundle)
1835 lacp_wait(bundle->lacp);
1838 bond_wait(bundle->bond);
1845 mirror_scan(struct ofproto_dpif *ofproto)
1849 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1850 if (!ofproto->mirrors[idx]) {
1857 static struct ofmirror *
1858 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1862 for (i = 0; i < MAX_MIRRORS; i++) {
1863 struct ofmirror *mirror = ofproto->mirrors[i];
1864 if (mirror && mirror->aux == aux) {
1872 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
1874 mirror_update_dups(struct ofproto_dpif *ofproto)
1878 for (i = 0; i < MAX_MIRRORS; i++) {
1879 struct ofmirror *m = ofproto->mirrors[i];
1882 m->dup_mirrors = MIRROR_MASK_C(1) << i;
1886 for (i = 0; i < MAX_MIRRORS; i++) {
1887 struct ofmirror *m1 = ofproto->mirrors[i];
1894 for (j = i + 1; j < MAX_MIRRORS; j++) {
1895 struct ofmirror *m2 = ofproto->mirrors[j];
1897 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
1898 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
1899 m2->dup_mirrors |= m1->dup_mirrors;
1906 mirror_set(struct ofproto *ofproto_, void *aux,
1907 const struct ofproto_mirror_settings *s)
1909 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1910 mirror_mask_t mirror_bit;
1911 struct ofbundle *bundle;
1912 struct ofmirror *mirror;
1913 struct ofbundle *out;
1914 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
1915 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
1918 mirror = mirror_lookup(ofproto, aux);
1920 mirror_destroy(mirror);
1926 idx = mirror_scan(ofproto);
1928 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
1930 ofproto->up.name, MAX_MIRRORS, s->name);
1934 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
1935 mirror->ofproto = ofproto;
1938 mirror->out_vlan = -1;
1939 mirror->name = NULL;
1942 if (!mirror->name || strcmp(s->name, mirror->name)) {
1944 mirror->name = xstrdup(s->name);
1947 /* Get the new configuration. */
1948 if (s->out_bundle) {
1949 out = bundle_lookup(ofproto, s->out_bundle);
1951 mirror_destroy(mirror);
1957 out_vlan = s->out_vlan;
1959 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
1960 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
1962 /* If the configuration has not changed, do nothing. */
1963 if (hmapx_equals(&srcs, &mirror->srcs)
1964 && hmapx_equals(&dsts, &mirror->dsts)
1965 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
1966 && mirror->out == out
1967 && mirror->out_vlan == out_vlan)
1969 hmapx_destroy(&srcs);
1970 hmapx_destroy(&dsts);
1974 hmapx_swap(&srcs, &mirror->srcs);
1975 hmapx_destroy(&srcs);
1977 hmapx_swap(&dsts, &mirror->dsts);
1978 hmapx_destroy(&dsts);
1980 free(mirror->vlans);
1981 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
1984 mirror->out_vlan = out_vlan;
1986 /* Update bundles. */
1987 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
1988 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
1989 if (hmapx_contains(&mirror->srcs, bundle)) {
1990 bundle->src_mirrors |= mirror_bit;
1992 bundle->src_mirrors &= ~mirror_bit;
1995 if (hmapx_contains(&mirror->dsts, bundle)) {
1996 bundle->dst_mirrors |= mirror_bit;
1998 bundle->dst_mirrors &= ~mirror_bit;
2001 if (mirror->out == bundle) {
2002 bundle->mirror_out |= mirror_bit;
2004 bundle->mirror_out &= ~mirror_bit;
2008 ofproto->need_revalidate = true;
2009 mac_learning_flush(ofproto->ml);
2010 mirror_update_dups(ofproto);
2016 mirror_destroy(struct ofmirror *mirror)
2018 struct ofproto_dpif *ofproto;
2019 mirror_mask_t mirror_bit;
2020 struct ofbundle *bundle;
2026 ofproto = mirror->ofproto;
2027 ofproto->need_revalidate = true;
2028 mac_learning_flush(ofproto->ml);
2030 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2031 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2032 bundle->src_mirrors &= ~mirror_bit;
2033 bundle->dst_mirrors &= ~mirror_bit;
2034 bundle->mirror_out &= ~mirror_bit;
2037 hmapx_destroy(&mirror->srcs);
2038 hmapx_destroy(&mirror->dsts);
2039 free(mirror->vlans);
2041 ofproto->mirrors[mirror->idx] = NULL;
2045 mirror_update_dups(ofproto);
2049 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2050 uint64_t *packets, uint64_t *bytes)
2052 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2053 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2056 *packets = *bytes = UINT64_MAX;
2060 *packets = mirror->packet_count;
2061 *bytes = mirror->byte_count;
2067 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2069 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2070 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2071 ofproto->need_revalidate = true;
2072 mac_learning_flush(ofproto->ml);
2078 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2080 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2081 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2082 return bundle && bundle->mirror_out != 0;
2086 forward_bpdu_changed(struct ofproto *ofproto_)
2088 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2089 /* Revalidate cached flows whenever forward_bpdu option changes. */
2090 ofproto->need_revalidate = true;
2095 static struct ofport_dpif *
2096 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
2098 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2099 return ofport ? ofport_dpif_cast(ofport) : NULL;
2102 static struct ofport_dpif *
2103 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
2105 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2109 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2110 struct dpif_port *dpif_port)
2112 ofproto_port->name = dpif_port->name;
2113 ofproto_port->type = dpif_port->type;
2114 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2118 port_run(struct ofport_dpif *ofport)
2120 bool enable = netdev_get_carrier(ofport->up.netdev);
2123 cfm_run(ofport->cfm);
2125 if (cfm_should_send_ccm(ofport->cfm)) {
2126 struct ofpbuf packet;
2128 ofpbuf_init(&packet, 0);
2129 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr);
2130 send_packet(ofport, &packet);
2131 ofpbuf_uninit(&packet);
2134 enable = enable && !cfm_get_fault(ofport->cfm)
2135 && cfm_get_opup(ofport->cfm);
2138 if (ofport->bundle) {
2139 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2142 if (ofport->may_enable != enable) {
2143 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2145 if (ofproto->has_bundle_action) {
2146 ofproto->need_revalidate = true;
2150 ofport->may_enable = enable;
2154 port_wait(struct ofport_dpif *ofport)
2157 cfm_wait(ofport->cfm);
2162 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2163 struct ofproto_port *ofproto_port)
2165 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2166 struct dpif_port dpif_port;
2169 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2171 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2177 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2179 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2183 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2185 *ofp_portp = odp_port_to_ofp_port(odp_port);
2191 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2193 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2196 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2198 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2200 /* The caller is going to close ofport->up.netdev. If this is a
2201 * bonded port, then the bond is using that netdev, so remove it
2202 * from the bond. The client will need to reconfigure everything
2203 * after deleting ports, so then the slave will get re-added. */
2204 bundle_remove(&ofport->up);
2210 struct port_dump_state {
2211 struct dpif_port_dump dump;
2216 port_dump_start(const struct ofproto *ofproto_, void **statep)
2218 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2219 struct port_dump_state *state;
2221 *statep = state = xmalloc(sizeof *state);
2222 dpif_port_dump_start(&state->dump, ofproto->dpif);
2223 state->done = false;
2228 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2229 struct ofproto_port *port)
2231 struct port_dump_state *state = state_;
2232 struct dpif_port dpif_port;
2234 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2235 ofproto_port_from_dpif_port(port, &dpif_port);
2238 int error = dpif_port_dump_done(&state->dump);
2240 return error ? error : EOF;
2245 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2247 struct port_dump_state *state = state_;
2250 dpif_port_dump_done(&state->dump);
2257 port_poll(const struct ofproto *ofproto_, char **devnamep)
2259 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2260 return dpif_port_poll(ofproto->dpif, devnamep);
2264 port_poll_wait(const struct ofproto *ofproto_)
2266 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2267 dpif_port_poll_wait(ofproto->dpif);
2271 port_is_lacp_current(const struct ofport *ofport_)
2273 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2274 return (ofport->bundle && ofport->bundle->lacp
2275 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2279 /* Upcall handling. */
2281 /* Flow miss batching.
2283 * Some dpifs implement operations faster when you hand them off in a batch.
2284 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2285 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2286 * more packets, plus possibly installing the flow in the dpif.
2288 * So far we only batch the operations that affect flow setup time the most.
2289 * It's possible to batch more than that, but the benefit might be minimal. */
2291 struct hmap_node hmap_node;
2293 enum odp_key_fitness key_fitness;
2294 const struct nlattr *key;
2296 ovs_be16 initial_tci;
2297 struct list packets;
2300 struct flow_miss_op {
2301 union dpif_op dpif_op;
2302 struct subfacet *subfacet;
2305 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2306 * OpenFlow controller as necessary according to their individual
2309 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
2310 * ownership is transferred to this function. */
2312 send_packet_in_miss(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
2313 const struct flow *flow, bool clone)
2315 struct ofputil_packet_in pin;
2317 pin.packet = packet;
2318 pin.in_port = flow->in_port;
2319 pin.reason = OFPR_NO_MATCH;
2320 pin.buffer_id = 0; /* not yet known */
2321 pin.send_len = 0; /* not used for flow table misses */
2322 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
2323 clone ? NULL : packet);
2326 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_ACTION to each
2327 * OpenFlow controller as necessary according to their individual
2330 * 'send_len' should be the number of bytes of 'packet' to send to the
2331 * controller, as specified in the action that caused the packet to be sent.
2333 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
2334 * Otherwise, ownership is transferred to this function. */
2336 send_packet_in_action(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
2337 uint64_t userdata, const struct flow *flow, bool clone)
2339 struct ofputil_packet_in pin;
2340 struct user_action_cookie cookie;
2342 memcpy(&cookie, &userdata, sizeof(cookie));
2344 pin.packet = packet;
2345 pin.in_port = flow->in_port;
2346 pin.reason = OFPR_ACTION;
2347 pin.buffer_id = 0; /* not yet known */
2348 pin.send_len = cookie.data;
2349 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
2350 clone ? NULL : packet);
2354 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2355 const struct ofpbuf *packet)
2357 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2363 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2365 cfm_process_heartbeat(ofport->cfm, packet);
2368 } else if (ofport->bundle && ofport->bundle->lacp
2369 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2371 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2374 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2376 stp_process_packet(ofport, packet);
2383 static struct flow_miss *
2384 flow_miss_create(struct hmap *todo, const struct flow *flow,
2385 enum odp_key_fitness key_fitness,
2386 const struct nlattr *key, size_t key_len,
2387 ovs_be16 initial_tci)
2389 uint32_t hash = flow_hash(flow, 0);
2390 struct flow_miss *miss;
2392 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2393 if (flow_equal(&miss->flow, flow)) {
2398 miss = xmalloc(sizeof *miss);
2399 hmap_insert(todo, &miss->hmap_node, hash);
2401 miss->key_fitness = key_fitness;
2403 miss->key_len = key_len;
2404 miss->initial_tci = initial_tci;
2405 list_init(&miss->packets);
2410 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2411 struct flow_miss_op *ops, size_t *n_ops)
2413 const struct flow *flow = &miss->flow;
2414 struct ofpbuf *packet, *next_packet;
2415 struct subfacet *subfacet;
2416 struct facet *facet;
2418 facet = facet_lookup_valid(ofproto, flow);
2420 struct rule_dpif *rule;
2422 rule = rule_dpif_lookup(ofproto, flow, 0);
2424 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
2425 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
2427 if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) {
2428 COVERAGE_INC(ofproto_dpif_no_packet_in);
2429 /* XXX install 'drop' flow entry */
2433 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
2437 LIST_FOR_EACH_SAFE (packet, next_packet, list_node,
2439 list_remove(&packet->list_node);
2440 send_packet_in_miss(ofproto, packet, flow, false);
2446 facet = facet_create(rule, flow);
2449 subfacet = subfacet_create(ofproto, facet,
2450 miss->key_fitness, miss->key, miss->key_len,
2453 LIST_FOR_EACH_SAFE (packet, next_packet, list_node, &miss->packets) {
2454 list_remove(&packet->list_node);
2455 ofproto->n_matches++;
2457 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2459 * Extra-special case for fail-open mode.
2461 * We are in fail-open mode and the packet matched the fail-open
2462 * rule, but we are connected to a controller too. We should send
2463 * the packet up to the controller in the hope that it will try to
2464 * set up a flow and thereby allow us to exit fail-open.
2466 * See the top-level comment in fail-open.c for more information.
2468 send_packet_in_miss(ofproto, packet, flow, true);
2471 if (!facet->may_install || !subfacet->actions) {
2472 subfacet_make_actions(ofproto, subfacet, packet);
2474 if (!execute_controller_action(ofproto, &facet->flow,
2476 subfacet->actions_len, packet)) {
2477 struct flow_miss_op *op = &ops[(*n_ops)++];
2478 struct dpif_execute *execute = &op->dpif_op.execute;
2480 op->subfacet = subfacet;
2481 execute->type = DPIF_OP_EXECUTE;
2482 execute->key = miss->key;
2483 execute->key_len = miss->key_len;
2485 = (facet->may_install
2487 : xmemdup(subfacet->actions, subfacet->actions_len));
2488 execute->actions_len = subfacet->actions_len;
2489 execute->packet = packet;
2493 if (facet->may_install && subfacet->key_fitness != ODP_FIT_TOO_LITTLE) {
2494 struct flow_miss_op *op = &ops[(*n_ops)++];
2495 struct dpif_flow_put *put = &op->dpif_op.flow_put;
2497 op->subfacet = subfacet;
2498 put->type = DPIF_OP_FLOW_PUT;
2499 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2500 put->key = miss->key;
2501 put->key_len = miss->key_len;
2502 put->actions = subfacet->actions;
2503 put->actions_len = subfacet->actions_len;
2508 static enum odp_key_fitness
2509 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
2510 const struct nlattr *key, size_t key_len,
2511 struct flow *flow, ovs_be16 *initial_tci)
2513 enum odp_key_fitness fitness;
2517 fitness = odp_flow_key_to_flow(key, key_len, flow);
2518 if (fitness == ODP_FIT_ERROR) {
2521 *initial_tci = flow->vlan_tci;
2523 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
2525 /* Cause the flow to be processed as if it came in on the real device
2526 * with the VLAN device's VLAN ID. */
2527 flow->in_port = realdev;
2528 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
2530 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
2531 if (fitness == ODP_FIT_PERFECT) {
2532 fitness = ODP_FIT_TOO_MUCH;
2540 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
2543 struct dpif_upcall *upcall;
2544 struct flow_miss *miss, *next_miss;
2545 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
2546 union dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
2555 /* Construct the to-do list.
2557 * This just amounts to extracting the flow from each packet and sticking
2558 * the packets that have the same flow in the same "flow_miss" structure so
2559 * that we can process them together. */
2561 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
2562 enum odp_key_fitness fitness;
2563 struct flow_miss *miss;
2564 ovs_be16 initial_tci;
2567 /* Obtain metadata and check userspace/kernel agreement on flow match,
2568 * then set 'flow''s header pointers. */
2569 fitness = ofproto_dpif_extract_flow_key(ofproto,
2570 upcall->key, upcall->key_len,
2571 &flow, &initial_tci);
2572 if (fitness == ODP_FIT_ERROR) {
2575 flow_extract(upcall->packet, flow.priority, flow.tun_id,
2576 flow.in_port, &flow);
2578 /* Handle 802.1ag, LACP, and STP specially. */
2579 if (process_special(ofproto, &flow, upcall->packet)) {
2580 ofpbuf_delete(upcall->packet);
2581 ofproto->n_matches++;
2585 /* Add other packets to a to-do list. */
2586 miss = flow_miss_create(&todo, &flow, fitness,
2587 upcall->key, upcall->key_len, initial_tci);
2588 list_push_back(&miss->packets, &upcall->packet->list_node);
2591 /* Process each element in the to-do list, constructing the set of
2592 * operations to batch. */
2594 HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) {
2595 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
2596 ofpbuf_list_delete(&miss->packets);
2597 hmap_remove(&todo, &miss->hmap_node);
2600 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
2601 hmap_destroy(&todo);
2603 /* Execute batch. */
2604 for (i = 0; i < n_ops; i++) {
2605 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
2607 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
2609 /* Free memory and update facets. */
2610 for (i = 0; i < n_ops; i++) {
2611 struct flow_miss_op *op = &flow_miss_ops[i];
2612 struct dpif_execute *execute;
2613 struct dpif_flow_put *put;
2615 switch (op->dpif_op.type) {
2616 case DPIF_OP_EXECUTE:
2617 execute = &op->dpif_op.execute;
2618 if (op->subfacet->actions != execute->actions) {
2619 free((struct nlattr *) execute->actions);
2621 ofpbuf_delete((struct ofpbuf *) execute->packet);
2624 case DPIF_OP_FLOW_PUT:
2625 put = &op->dpif_op.flow_put;
2627 op->subfacet->installed = true;
2635 handle_userspace_upcall(struct ofproto_dpif *ofproto,
2636 struct dpif_upcall *upcall)
2638 struct user_action_cookie cookie;
2639 enum odp_key_fitness fitness;
2640 ovs_be16 initial_tci;
2643 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
2645 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
2646 upcall->key_len, &flow,
2648 if (fitness == ODP_FIT_ERROR) {
2652 if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
2653 if (ofproto->sflow) {
2654 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
2657 ofpbuf_delete(upcall->packet);
2658 } else if (cookie.type == USER_ACTION_COOKIE_CONTROLLER) {
2659 COVERAGE_INC(ofproto_dpif_ctlr_action);
2660 send_packet_in_action(ofproto, upcall->packet, upcall->userdata,
2663 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
2668 handle_upcall(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall)
2670 switch (upcall->type) {
2671 case DPIF_UC_ACTION:
2672 handle_userspace_upcall(ofproto, upcall);
2676 /* The caller handles these. */
2679 case DPIF_N_UC_TYPES:
2681 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
2686 /* Flow expiration. */
2688 static int subfacet_max_idle(const struct ofproto_dpif *);
2689 static void update_stats(struct ofproto_dpif *);
2690 static void rule_expire(struct rule_dpif *);
2691 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
2693 /* This function is called periodically by run(). Its job is to collect
2694 * updates for the flows that have been installed into the datapath, most
2695 * importantly when they last were used, and then use that information to
2696 * expire flows that have not been used recently.
2698 * Returns the number of milliseconds after which it should be called again. */
2700 expire(struct ofproto_dpif *ofproto)
2702 struct rule_dpif *rule, *next_rule;
2703 struct classifier *table;
2706 /* Update stats for each flow in the datapath. */
2707 update_stats(ofproto);
2709 /* Expire subfacets that have been idle too long. */
2710 dp_max_idle = subfacet_max_idle(ofproto);
2711 expire_subfacets(ofproto, dp_max_idle);
2713 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
2714 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
2715 struct cls_cursor cursor;
2717 cls_cursor_init(&cursor, table, NULL);
2718 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
2723 /* All outstanding data in existing flows has been accounted, so it's a
2724 * good time to do bond rebalancing. */
2725 if (ofproto->has_bonded_bundles) {
2726 struct ofbundle *bundle;
2728 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2730 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
2735 return MIN(dp_max_idle, 1000);
2738 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
2740 * This function also pushes statistics updates to rules which each facet
2741 * resubmits into. Generally these statistics will be accurate. However, if a
2742 * facet changes the rule it resubmits into at some time in between
2743 * update_stats() runs, it is possible that statistics accrued to the
2744 * old rule will be incorrectly attributed to the new rule. This could be
2745 * avoided by calling update_stats() whenever rules are created or
2746 * deleted. However, the performance impact of making so many calls to the
2747 * datapath do not justify the benefit of having perfectly accurate statistics.
2750 update_stats(struct ofproto_dpif *p)
2752 const struct dpif_flow_stats *stats;
2753 struct dpif_flow_dump dump;
2754 const struct nlattr *key;
2757 dpif_flow_dump_start(&dump, p->dpif);
2758 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
2759 enum odp_key_fitness fitness;
2760 struct subfacet *subfacet;
2763 fitness = odp_flow_key_to_flow(key, key_len, &flow);
2764 if (fitness == ODP_FIT_ERROR) {
2768 subfacet = subfacet_find(p, key, key_len, &flow);
2769 if (subfacet && subfacet->installed) {
2770 struct facet *facet = subfacet->facet;
2772 if (stats->n_packets >= subfacet->dp_packet_count) {
2773 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
2774 facet->packet_count += extra;
2776 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
2779 if (stats->n_bytes >= subfacet->dp_byte_count) {
2780 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
2782 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
2785 subfacet->dp_packet_count = stats->n_packets;
2786 subfacet->dp_byte_count = stats->n_bytes;
2788 subfacet_update_time(p, subfacet, stats->used);
2789 facet_account(p, facet);
2790 facet_push_stats(facet);
2792 /* There's a flow in the datapath that we know nothing about, or a
2793 * flow that shouldn't be installed but was anyway. Delete it. */
2794 COVERAGE_INC(facet_unexpected);
2795 dpif_flow_del(p->dpif, key, key_len, NULL);
2798 dpif_flow_dump_done(&dump);
2801 /* Calculates and returns the number of milliseconds of idle time after which
2802 * subfacets should expire from the datapath. When a subfacet expires, we fold
2803 * its statistics into its facet, and when a facet's last subfacet expires, we
2804 * fold its statistic into its rule. */
2806 subfacet_max_idle(const struct ofproto_dpif *ofproto)
2809 * Idle time histogram.
2811 * Most of the time a switch has a relatively small number of subfacets.
2812 * When this is the case we might as well keep statistics for all of them
2813 * in userspace and to cache them in the kernel datapath for performance as
2816 * As the number of subfacets increases, the memory required to maintain
2817 * statistics about them in userspace and in the kernel becomes
2818 * significant. However, with a large number of subfacets it is likely
2819 * that only a few of them are "heavy hitters" that consume a large amount
2820 * of bandwidth. At this point, only heavy hitters are worth caching in
2821 * the kernel and maintaining in userspaces; other subfacets we can
2824 * The technique used to compute the idle time is to build a histogram with
2825 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
2826 * that is installed in the kernel gets dropped in the appropriate bucket.
2827 * After the histogram has been built, we compute the cutoff so that only
2828 * the most-recently-used 1% of subfacets (but at least
2829 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
2830 * the most-recently-used bucket of subfacets is kept, so actually an
2831 * arbitrary number of subfacets can be kept in any given expiration run
2832 * (though the next run will delete most of those unless they receive
2835 * This requires a second pass through the subfacets, in addition to the
2836 * pass made by update_stats(), because the former function never looks at
2837 * uninstallable subfacets.
2839 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
2840 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
2841 int buckets[N_BUCKETS] = { 0 };
2842 int total, subtotal, bucket;
2843 struct subfacet *subfacet;
2847 total = hmap_count(&ofproto->subfacets);
2848 if (total <= ofproto->up.flow_eviction_threshold) {
2849 return N_BUCKETS * BUCKET_WIDTH;
2852 /* Build histogram. */
2854 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
2855 long long int idle = now - subfacet->used;
2856 int bucket = (idle <= 0 ? 0
2857 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
2858 : (unsigned int) idle / BUCKET_WIDTH);
2862 /* Find the first bucket whose flows should be expired. */
2863 subtotal = bucket = 0;
2865 subtotal += buckets[bucket++];
2866 } while (bucket < N_BUCKETS &&
2867 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
2869 if (VLOG_IS_DBG_ENABLED()) {
2873 ds_put_cstr(&s, "keep");
2874 for (i = 0; i < N_BUCKETS; i++) {
2876 ds_put_cstr(&s, ", drop");
2879 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
2882 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
2886 return bucket * BUCKET_WIDTH;
2890 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
2892 long long int cutoff = time_msec() - dp_max_idle;
2893 struct subfacet *subfacet, *next_subfacet;
2895 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
2896 &ofproto->subfacets) {
2897 if (subfacet->used < cutoff) {
2898 subfacet_destroy(ofproto, subfacet);
2903 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
2904 * then delete it entirely. */
2906 rule_expire(struct rule_dpif *rule)
2908 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2909 struct facet *facet, *next_facet;
2913 /* Has 'rule' expired? */
2915 if (rule->up.hard_timeout
2916 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
2917 reason = OFPRR_HARD_TIMEOUT;
2918 } else if (rule->up.idle_timeout && list_is_empty(&rule->facets)
2919 && now > rule->used + rule->up.idle_timeout * 1000) {
2920 reason = OFPRR_IDLE_TIMEOUT;
2925 COVERAGE_INC(ofproto_dpif_expired);
2927 /* Update stats. (This is a no-op if the rule expired due to an idle
2928 * timeout, because that only happens when the rule has no facets left.) */
2929 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
2930 facet_remove(ofproto, facet);
2933 /* Get rid of the rule. */
2934 ofproto_rule_expire(&rule->up, reason);
2939 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
2941 * The caller must already have determined that no facet with an identical
2942 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2943 * the ofproto's classifier table.
2945 * The facet will initially have no subfacets. The caller should create (at
2946 * least) one subfacet with subfacet_create(). */
2947 static struct facet *
2948 facet_create(struct rule_dpif *rule, const struct flow *flow)
2950 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2951 struct facet *facet;
2953 facet = xzalloc(sizeof *facet);
2954 facet->used = time_msec();
2955 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
2956 list_push_back(&rule->facets, &facet->list_node);
2958 facet->flow = *flow;
2959 list_init(&facet->subfacets);
2960 netflow_flow_init(&facet->nf_flow);
2961 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
2967 facet_free(struct facet *facet)
2973 execute_controller_action(struct ofproto_dpif *ofproto,
2974 const struct flow *flow,
2975 const struct nlattr *odp_actions, size_t actions_len,
2976 struct ofpbuf *packet)
2979 && odp_actions->nla_type == OVS_ACTION_ATTR_USERSPACE
2980 && NLA_ALIGN(odp_actions->nla_len) == actions_len) {
2981 /* As an optimization, avoid a round-trip from userspace to kernel to
2982 * userspace. This also avoids possibly filling up kernel packet
2983 * buffers along the way.
2985 * This optimization will not accidentally catch sFlow
2986 * OVS_ACTION_ATTR_USERSPACE actions, since those are encapsulated
2987 * inside OVS_ACTION_ATTR_SAMPLE. */
2988 const struct nlattr *nla;
2990 nla = nl_attr_find_nested(odp_actions, OVS_USERSPACE_ATTR_USERDATA);
2991 send_packet_in_action(ofproto, packet, nl_attr_get_u64(nla), flow,
2999 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3000 * 'packet', which arrived on 'in_port'.
3002 * Takes ownership of 'packet'. */
3004 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3005 const struct nlattr *odp_actions, size_t actions_len,
3006 struct ofpbuf *packet)
3008 struct odputil_keybuf keybuf;
3012 if (execute_controller_action(ofproto, flow, odp_actions, actions_len,
3017 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3018 odp_flow_key_from_flow(&key, flow);
3020 error = dpif_execute(ofproto->dpif, key.data, key.size,
3021 odp_actions, actions_len, packet);
3023 ofpbuf_delete(packet);
3027 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3029 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3030 * rule's statistics, via subfacet_uninstall().
3032 * - Removes 'facet' from its rule and from ofproto->facets.
3035 facet_remove(struct ofproto_dpif *ofproto, struct facet *facet)
3037 struct subfacet *subfacet, *next_subfacet;
3039 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3040 &facet->subfacets) {
3041 subfacet_destroy__(ofproto, subfacet);
3044 facet_flush_stats(ofproto, facet);
3045 hmap_remove(&ofproto->facets, &facet->hmap_node);
3046 list_remove(&facet->list_node);
3051 facet_account(struct ofproto_dpif *ofproto, struct facet *facet)
3054 struct subfacet *subfacet;
3055 const struct nlattr *a;
3059 if (facet->byte_count <= facet->accounted_bytes) {
3062 n_bytes = facet->byte_count - facet->accounted_bytes;
3063 facet->accounted_bytes = facet->byte_count;
3065 /* Feed information from the active flows back into the learning table to
3066 * ensure that table is always in sync with what is actually flowing
3067 * through the datapath. */
3068 if (facet->has_learn || facet->has_normal) {
3069 struct action_xlate_ctx ctx;
3071 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3072 facet->flow.vlan_tci, NULL);
3073 ctx.may_learn = true;
3074 ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions,
3075 facet->rule->up.n_actions));
3078 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3082 /* This loop feeds byte counters to bond_account() for rebalancing to use
3083 * as a basis. We also need to track the actual VLAN on which the packet
3084 * is going to be sent to ensure that it matches the one passed to
3085 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3088 * We use the actions from an arbitrary subfacet because they should all
3089 * be equally valid for our purpose. */
3090 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3091 struct subfacet, list_node);
3092 vlan_tci = facet->flow.vlan_tci;
3093 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3094 subfacet->actions, subfacet->actions_len) {
3095 const struct ovs_action_push_vlan *vlan;
3096 struct ofport_dpif *port;
3098 switch (nl_attr_type(a)) {
3099 case OVS_ACTION_ATTR_OUTPUT:
3100 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3101 if (port && port->bundle && port->bundle->bond) {
3102 bond_account(port->bundle->bond, &facet->flow,
3103 vlan_tci_to_vid(vlan_tci), n_bytes);
3107 case OVS_ACTION_ATTR_POP_VLAN:
3108 vlan_tci = htons(0);
3111 case OVS_ACTION_ATTR_PUSH_VLAN:
3112 vlan = nl_attr_get(a);
3113 vlan_tci = vlan->vlan_tci;
3119 /* Returns true if the only action for 'facet' is to send to the controller.
3120 * (We don't report NetFlow expiration messages for such facets because they
3121 * are just part of the control logic for the network, not real traffic). */
3123 facet_is_controller_flow(struct facet *facet)
3126 && facet->rule->up.n_actions == 1
3127 && action_outputs_to_port(&facet->rule->up.actions[0],
3128 htons(OFPP_CONTROLLER)));
3131 /* Folds all of 'facet''s statistics into its rule. Also updates the
3132 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3133 * 'facet''s statistics in the datapath should have been zeroed and folded into
3134 * its packet and byte counts before this function is called. */
3136 facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet)
3138 struct subfacet *subfacet;
3140 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3141 assert(!subfacet->dp_byte_count);
3142 assert(!subfacet->dp_packet_count);
3145 facet_push_stats(facet);
3146 facet_account(ofproto, facet);
3148 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3149 struct ofexpired expired;
3150 expired.flow = facet->flow;
3151 expired.packet_count = facet->packet_count;
3152 expired.byte_count = facet->byte_count;
3153 expired.used = facet->used;
3154 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3157 facet->rule->packet_count += facet->packet_count;
3158 facet->rule->byte_count += facet->byte_count;
3160 /* Reset counters to prevent double counting if 'facet' ever gets
3162 facet_reset_counters(facet);
3164 netflow_flow_clear(&facet->nf_flow);
3167 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3168 * Returns it if found, otherwise a null pointer.
3170 * The returned facet might need revalidation; use facet_lookup_valid()
3171 * instead if that is important. */
3172 static struct facet *
3173 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
3175 struct facet *facet;
3177 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
3179 if (flow_equal(flow, &facet->flow)) {
3187 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3188 * Returns it if found, otherwise a null pointer.
3190 * The returned facet is guaranteed to be valid. */
3191 static struct facet *
3192 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
3194 struct facet *facet = facet_find(ofproto, flow);
3196 /* The facet we found might not be valid, since we could be in need of
3197 * revalidation. If it is not valid, don't return it. */
3199 && (ofproto->need_revalidate
3200 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))
3201 && !facet_revalidate(ofproto, facet)) {
3202 COVERAGE_INC(facet_invalidated);
3209 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
3211 * - If the rule found is different from 'facet''s current rule, moves
3212 * 'facet' to the new rule and recompiles its actions.
3214 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3215 * where it is and recompiles its actions anyway.
3217 * - If there is none, destroys 'facet'.
3219 * Returns true if 'facet' still exists, false if it has been destroyed. */
3221 facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet)
3224 struct nlattr *odp_actions;
3227 struct actions *new_actions;
3229 struct action_xlate_ctx ctx;
3230 struct rule_dpif *new_rule;
3231 struct subfacet *subfacet;
3232 bool actions_changed;
3235 COVERAGE_INC(facet_revalidate);
3237 /* Determine the new rule. */
3238 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3240 /* No new rule, so delete the facet. */
3241 facet_remove(ofproto, facet);
3245 /* Calculate new datapath actions.
3247 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3248 * emit a NetFlow expiration and, if so, we need to have the old state
3249 * around to properly compose it. */
3251 /* If the datapath actions changed or the installability changed,
3252 * then we need to talk to the datapath. */
3255 memset(&ctx, 0, sizeof ctx);
3256 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3257 struct ofpbuf *odp_actions;
3258 bool should_install;
3260 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3261 subfacet->initial_tci, NULL);
3262 odp_actions = xlate_actions(&ctx, new_rule->up.actions,
3263 new_rule->up.n_actions);
3264 actions_changed = (subfacet->actions_len != odp_actions->size
3265 || memcmp(subfacet->actions, odp_actions->data,
3266 subfacet->actions_len));
3268 should_install = (ctx.may_set_up_flow
3269 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3270 if (actions_changed || should_install != subfacet->installed) {
3271 if (should_install) {
3272 struct dpif_flow_stats stats;
3274 subfacet_install(ofproto, subfacet,
3275 odp_actions->data, odp_actions->size, &stats);
3276 subfacet_update_stats(ofproto, subfacet, &stats);
3278 subfacet_uninstall(ofproto, subfacet);
3282 new_actions = xcalloc(list_size(&facet->subfacets),
3283 sizeof *new_actions);
3285 new_actions[i].odp_actions = xmemdup(odp_actions->data,
3287 new_actions[i].actions_len = odp_actions->size;
3290 ofpbuf_delete(odp_actions);
3294 facet_flush_stats(ofproto, facet);
3297 /* Update 'facet' now that we've taken care of all the old state. */
3298 facet->tags = ctx.tags;
3299 facet->nf_flow.output_iface = ctx.nf_output_iface;
3300 facet->may_install = ctx.may_set_up_flow;
3301 facet->has_learn = ctx.has_learn;
3302 facet->has_normal = ctx.has_normal;
3303 facet->mirrors = ctx.mirrors;
3306 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3307 if (new_actions[i].odp_actions) {
3308 free(subfacet->actions);
3309 subfacet->actions = new_actions[i].odp_actions;
3310 subfacet->actions_len = new_actions[i].actions_len;
3316 if (facet->rule != new_rule) {
3317 COVERAGE_INC(facet_changed_rule);
3318 list_remove(&facet->list_node);
3319 list_push_back(&new_rule->facets, &facet->list_node);
3320 facet->rule = new_rule;
3321 facet->used = new_rule->up.created;
3322 facet->prev_used = facet->used;
3328 /* Updates 'facet''s used time. Caller is responsible for calling
3329 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3331 facet_update_time(struct ofproto_dpif *ofproto, struct facet *facet,
3334 if (used > facet->used) {
3336 if (used > facet->rule->used) {
3337 facet->rule->used = used;
3339 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3344 facet_reset_counters(struct facet *facet)
3346 facet->packet_count = 0;
3347 facet->byte_count = 0;
3348 facet->prev_packet_count = 0;
3349 facet->prev_byte_count = 0;
3350 facet->accounted_bytes = 0;
3354 facet_push_stats(struct facet *facet)
3356 uint64_t new_packets, new_bytes;
3358 assert(facet->packet_count >= facet->prev_packet_count);
3359 assert(facet->byte_count >= facet->prev_byte_count);
3360 assert(facet->used >= facet->prev_used);
3362 new_packets = facet->packet_count - facet->prev_packet_count;
3363 new_bytes = facet->byte_count - facet->prev_byte_count;
3365 if (new_packets || new_bytes || facet->used > facet->prev_used) {
3366 facet->prev_packet_count = facet->packet_count;
3367 facet->prev_byte_count = facet->byte_count;
3368 facet->prev_used = facet->used;
3370 flow_push_stats(facet->rule, &facet->flow,
3371 new_packets, new_bytes, facet->used);
3373 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
3374 facet->mirrors, new_packets, new_bytes);
3378 struct ofproto_push {
3379 struct action_xlate_ctx ctx;
3386 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
3388 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3391 rule->packet_count += push->packets;
3392 rule->byte_count += push->bytes;
3393 rule->used = MAX(push->used, rule->used);
3397 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3398 * 'rule''s actions and mirrors. */
3400 flow_push_stats(const struct rule_dpif *rule,
3401 const struct flow *flow, uint64_t packets, uint64_t bytes,
3404 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3405 struct ofproto_push push;
3407 push.packets = packets;
3411 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, NULL);
3412 push.ctx.resubmit_hook = push_resubmit;
3413 ofpbuf_delete(xlate_actions(&push.ctx,
3414 rule->up.actions, rule->up.n_actions));
3419 static struct subfacet *
3420 subfacet_find__(struct ofproto_dpif *ofproto,
3421 const struct nlattr *key, size_t key_len, uint32_t key_hash,
3422 const struct flow *flow)
3424 struct subfacet *subfacet;
3426 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
3427 &ofproto->subfacets) {
3429 ? (subfacet->key_len == key_len
3430 && !memcmp(key, subfacet->key, key_len))
3431 : flow_equal(flow, &subfacet->facet->flow)) {
3439 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
3440 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
3441 * there is one, otherwise creates and returns a new subfacet.
3443 * If the returned subfacet is new, then subfacet->actions will be NULL, in
3444 * which case the caller must populate the actions with
3445 * subfacet_make_actions(). */
3446 static struct subfacet *
3447 subfacet_create(struct ofproto_dpif *ofproto, struct facet *facet,
3448 enum odp_key_fitness key_fitness,
3449 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
3451 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3452 struct subfacet *subfacet;
3454 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
3456 if (subfacet->facet == facet) {
3460 /* This shouldn't happen. */
3461 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
3462 subfacet_destroy(ofproto, subfacet);
3465 subfacet = xzalloc(sizeof *subfacet);
3466 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
3467 list_push_back(&facet->subfacets, &subfacet->list_node);
3468 subfacet->facet = facet;
3469 subfacet->used = time_msec();
3470 subfacet->key_fitness = key_fitness;
3471 if (key_fitness != ODP_FIT_PERFECT) {
3472 subfacet->key = xmemdup(key, key_len);
3473 subfacet->key_len = key_len;
3475 subfacet->installed = false;
3476 subfacet->initial_tci = initial_tci;
3481 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
3482 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
3483 static struct subfacet *
3484 subfacet_find(struct ofproto_dpif *ofproto,
3485 const struct nlattr *key, size_t key_len,
3486 const struct flow *flow)
3488 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3490 return subfacet_find__(ofproto, key, key_len, key_hash, flow);
3493 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
3494 * its facet within 'ofproto', and frees it. */
3496 subfacet_destroy__(struct ofproto_dpif *ofproto, struct subfacet *subfacet)
3498 subfacet_uninstall(ofproto, subfacet);
3499 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
3500 list_remove(&subfacet->list_node);
3501 free(subfacet->key);
3502 free(subfacet->actions);
3506 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
3507 * last remaining subfacet in its facet destroys the facet too. */
3509 subfacet_destroy(struct ofproto_dpif *ofproto, struct subfacet *subfacet)
3511 struct facet *facet = subfacet->facet;
3513 subfacet_destroy__(ofproto, subfacet);
3514 if (list_is_empty(&facet->subfacets)) {
3515 facet_remove(ofproto, facet);
3519 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
3520 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
3521 * for use as temporary storage. */
3523 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
3526 if (!subfacet->key) {
3527 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
3528 odp_flow_key_from_flow(key, &subfacet->facet->flow);
3530 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
3534 /* Composes the datapath actions for 'subfacet' based on its rule's actions. */
3536 subfacet_make_actions(struct ofproto_dpif *p, struct subfacet *subfacet,
3537 const struct ofpbuf *packet)
3539 struct facet *facet = subfacet->facet;
3540 const struct rule_dpif *rule = facet->rule;
3541 struct ofpbuf *odp_actions;
3542 struct action_xlate_ctx ctx;
3544 action_xlate_ctx_init(&ctx, p, &facet->flow, subfacet->initial_tci,
3546 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3547 facet->tags = ctx.tags;
3548 facet->may_install = ctx.may_set_up_flow;
3549 facet->has_learn = ctx.has_learn;
3550 facet->has_normal = ctx.has_normal;
3551 facet->nf_flow.output_iface = ctx.nf_output_iface;
3552 facet->mirrors = ctx.mirrors;
3554 if (subfacet->actions_len != odp_actions->size
3555 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
3556 free(subfacet->actions);
3557 subfacet->actions_len = odp_actions->size;
3558 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
3561 ofpbuf_delete(odp_actions);
3564 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
3565 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
3566 * in the datapath will be zeroed and 'stats' will be updated with traffic new
3567 * since 'subfacet' was last updated.
3569 * Returns 0 if successful, otherwise a positive errno value. */
3571 subfacet_install(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3572 const struct nlattr *actions, size_t actions_len,
3573 struct dpif_flow_stats *stats)
3575 struct odputil_keybuf keybuf;
3576 enum dpif_flow_put_flags flags;
3580 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3582 flags |= DPIF_FP_ZERO_STATS;
3585 subfacet_get_key(subfacet, &keybuf, &key);
3586 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
3587 actions, actions_len, stats);
3590 subfacet_reset_dp_stats(subfacet, stats);
3596 /* If 'subfacet' is installed in the datapath, uninstalls it. */
3598 subfacet_uninstall(struct ofproto_dpif *p, struct subfacet *subfacet)
3600 if (subfacet->installed) {
3601 struct odputil_keybuf keybuf;
3602 struct dpif_flow_stats stats;
3606 subfacet_get_key(subfacet, &keybuf, &key);
3607 error = dpif_flow_del(p->dpif, key.data, key.size, &stats);
3608 subfacet_reset_dp_stats(subfacet, &stats);
3610 subfacet_update_stats(p, subfacet, &stats);
3612 subfacet->installed = false;
3614 assert(subfacet->dp_packet_count == 0);
3615 assert(subfacet->dp_byte_count == 0);
3619 /* Resets 'subfacet''s datapath statistics counters. This should be called
3620 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
3621 * non-null, it should contain the statistics returned by dpif when 'subfacet'
3622 * was reset in the datapath. 'stats' will be modified to include only
3623 * statistics new since 'subfacet' was last updated. */
3625 subfacet_reset_dp_stats(struct subfacet *subfacet,
3626 struct dpif_flow_stats *stats)
3629 && subfacet->dp_packet_count <= stats->n_packets
3630 && subfacet->dp_byte_count <= stats->n_bytes) {
3631 stats->n_packets -= subfacet->dp_packet_count;
3632 stats->n_bytes -= subfacet->dp_byte_count;
3635 subfacet->dp_packet_count = 0;
3636 subfacet->dp_byte_count = 0;
3639 /* Updates 'subfacet''s used time. The caller is responsible for calling
3640 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
3642 subfacet_update_time(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3645 if (used > subfacet->used) {
3646 subfacet->used = used;
3647 facet_update_time(ofproto, subfacet->facet, used);
3651 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
3653 * Because of the meaning of a subfacet's counters, it only makes sense to do
3654 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
3655 * represents a packet that was sent by hand or if it represents statistics
3656 * that have been cleared out of the datapath. */
3658 subfacet_update_stats(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3659 const struct dpif_flow_stats *stats)
3661 if (stats->n_packets || stats->used > subfacet->used) {
3662 struct facet *facet = subfacet->facet;
3664 subfacet_update_time(ofproto, subfacet, stats->used);
3665 facet->packet_count += stats->n_packets;
3666 facet->byte_count += stats->n_bytes;
3667 facet_push_stats(facet);
3668 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3674 static struct rule_dpif *
3675 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
3678 struct cls_rule *cls_rule;
3679 struct classifier *cls;
3681 if (table_id >= N_TABLES) {
3685 cls = &ofproto->up.tables[table_id];
3686 if (flow->nw_frag & FLOW_NW_FRAG_ANY
3687 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3688 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
3689 * are unavailable. */
3690 struct flow ofpc_normal_flow = *flow;
3691 ofpc_normal_flow.tp_src = htons(0);
3692 ofpc_normal_flow.tp_dst = htons(0);
3693 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
3695 cls_rule = classifier_lookup(cls, flow);
3697 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
3701 complete_operation(struct rule_dpif *rule)
3703 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3705 rule_invalidate(rule);
3707 struct dpif_completion *c = xmalloc(sizeof *c);
3708 c->op = rule->up.pending;
3709 list_push_back(&ofproto->completions, &c->list_node);
3711 ofoperation_complete(rule->up.pending, 0);
3715 static struct rule *
3718 struct rule_dpif *rule = xmalloc(sizeof *rule);
3723 rule_dealloc(struct rule *rule_)
3725 struct rule_dpif *rule = rule_dpif_cast(rule_);
3730 rule_construct(struct rule *rule_)
3732 struct rule_dpif *rule = rule_dpif_cast(rule_);
3733 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3734 struct rule_dpif *victim;
3738 error = validate_actions(rule->up.actions, rule->up.n_actions,
3739 &rule->up.cr.flow, ofproto->max_ports);
3744 rule->used = rule->up.created;
3745 rule->packet_count = 0;
3746 rule->byte_count = 0;
3748 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
3749 if (victim && !list_is_empty(&victim->facets)) {
3750 struct facet *facet;
3752 rule->facets = victim->facets;
3753 list_moved(&rule->facets);
3754 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3755 /* XXX: We're only clearing our local counters here. It's possible
3756 * that quite a few packets are unaccounted for in the datapath
3757 * statistics. These will be accounted to the new rule instead of
3758 * cleared as required. This could be fixed by clearing out the
3759 * datapath statistics for this facet, but currently it doesn't
3761 facet_reset_counters(facet);
3765 /* Must avoid list_moved() in this case. */
3766 list_init(&rule->facets);
3769 table_id = rule->up.table_id;
3770 rule->tag = (victim ? victim->tag
3772 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
3773 ofproto->tables[table_id].basis));
3775 complete_operation(rule);
3780 rule_destruct(struct rule *rule_)
3782 struct rule_dpif *rule = rule_dpif_cast(rule_);
3783 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3784 struct facet *facet, *next_facet;
3786 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3787 facet_revalidate(ofproto, facet);
3790 complete_operation(rule);
3794 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
3796 struct rule_dpif *rule = rule_dpif_cast(rule_);
3797 struct facet *facet;
3799 /* Start from historical data for 'rule' itself that are no longer tracked
3800 * in facets. This counts, for example, facets that have expired. */
3801 *packets = rule->packet_count;
3802 *bytes = rule->byte_count;
3804 /* Add any statistics that are tracked by facets. This includes
3805 * statistical data recently updated by ofproto_update_stats() as well as
3806 * stats for packets that were executed "by hand" via dpif_execute(). */
3807 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3808 *packets += facet->packet_count;
3809 *bytes += facet->byte_count;
3814 rule_execute(struct rule *rule_, const struct flow *flow,
3815 struct ofpbuf *packet)
3817 struct rule_dpif *rule = rule_dpif_cast(rule_);
3818 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3819 struct action_xlate_ctx ctx;
3820 struct ofpbuf *odp_actions;
3823 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, packet);
3824 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3825 size = packet->size;
3826 if (execute_odp_actions(ofproto, flow, odp_actions->data,
3827 odp_actions->size, packet)) {
3828 rule->used = time_msec();
3829 rule->packet_count++;
3830 rule->byte_count += size;
3831 flow_push_stats(rule, flow, 1, size, rule->used);
3833 ofpbuf_delete(odp_actions);
3839 rule_modify_actions(struct rule *rule_)
3841 struct rule_dpif *rule = rule_dpif_cast(rule_);
3842 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3845 error = validate_actions(rule->up.actions, rule->up.n_actions,
3846 &rule->up.cr.flow, ofproto->max_ports);
3848 ofoperation_complete(rule->up.pending, error);
3852 complete_operation(rule);
3855 /* Sends 'packet' out 'ofport'.
3856 * May modify 'packet'.
3857 * Returns 0 if successful, otherwise a positive errno value. */
3859 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
3861 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3862 struct ofpbuf key, odp_actions;
3863 struct odputil_keybuf keybuf;
3868 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
3869 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
3871 if (odp_port != ofport->odp_port) {
3872 eth_pop_vlan(packet);
3873 flow.vlan_tci = htons(0);
3876 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3877 odp_flow_key_from_flow(&key, &flow);
3879 ofpbuf_init(&odp_actions, 32);
3880 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
3882 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
3883 error = dpif_execute(ofproto->dpif,
3885 odp_actions.data, odp_actions.size,
3887 ofpbuf_uninit(&odp_actions);
3890 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
3891 ofproto->up.name, odp_port, strerror(error));
3896 /* OpenFlow to datapath action translation. */
3898 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
3899 struct action_xlate_ctx *ctx);
3900 static void xlate_normal(struct action_xlate_ctx *);
3903 put_userspace_action(const struct ofproto_dpif *ofproto,
3904 struct ofpbuf *odp_actions,
3905 const struct flow *flow,
3906 const struct user_action_cookie *cookie)
3910 pid = dpif_port_get_pid(ofproto->dpif,
3911 ofp_port_to_odp_port(flow->in_port));
3913 return odp_put_userspace_action(pid, cookie, odp_actions);
3916 /* Compose SAMPLE action for sFlow. */
3918 compose_sflow_action(const struct ofproto_dpif *ofproto,
3919 struct ofpbuf *odp_actions,
3920 const struct flow *flow,
3923 uint32_t port_ifindex;
3924 uint32_t probability;
3925 struct user_action_cookie cookie;
3926 size_t sample_offset, actions_offset;
3927 int cookie_offset, n_output;
3929 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
3933 if (odp_port == OVSP_NONE) {
3937 port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
3941 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
3943 /* Number of packets out of UINT_MAX to sample. */
3944 probability = dpif_sflow_get_probability(ofproto->sflow);
3945 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
3947 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
3949 cookie.type = USER_ACTION_COOKIE_SFLOW;
3950 cookie.data = port_ifindex;
3951 cookie.n_output = n_output;
3952 cookie.vlan_tci = 0;
3953 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
3955 nl_msg_end_nested(odp_actions, actions_offset);
3956 nl_msg_end_nested(odp_actions, sample_offset);
3957 return cookie_offset;
3960 /* SAMPLE action must be first action in any given list of actions.
3961 * At this point we do not have all information required to build it. So try to
3962 * build sample action as complete as possible. */
3964 add_sflow_action(struct action_xlate_ctx *ctx)
3966 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
3968 &ctx->flow, OVSP_NONE);
3969 ctx->sflow_odp_port = 0;
3970 ctx->sflow_n_outputs = 0;
3973 /* Fix SAMPLE action according to data collected while composing ODP actions.
3974 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
3975 * USERSPACE action's user-cookie which is required for sflow. */
3977 fix_sflow_action(struct action_xlate_ctx *ctx)
3979 const struct flow *base = &ctx->base_flow;
3980 struct user_action_cookie *cookie;
3982 if (!ctx->user_cookie_offset) {
3986 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
3988 assert(cookie != NULL);
3989 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
3991 if (ctx->sflow_n_outputs) {
3992 cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow,
3993 ctx->sflow_odp_port);
3995 if (ctx->sflow_n_outputs >= 255) {
3996 cookie->n_output = 255;
3998 cookie->n_output = ctx->sflow_n_outputs;
4000 cookie->vlan_tci = base->vlan_tci;
4004 commit_set_action(struct ofpbuf *odp_actions, enum ovs_key_attr key_type,
4005 const void *key, size_t key_size)
4007 size_t offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SET);
4008 nl_msg_put_unspec(odp_actions, key_type, key, key_size);
4009 nl_msg_end_nested(odp_actions, offset);
4013 commit_set_tun_id_action(const struct flow *flow, struct flow *base,
4014 struct ofpbuf *odp_actions)
4016 if (base->tun_id == flow->tun_id) {
4019 base->tun_id = flow->tun_id;
4021 commit_set_action(odp_actions, OVS_KEY_ATTR_TUN_ID,
4022 &base->tun_id, sizeof(base->tun_id));
4026 commit_set_ether_addr_action(const struct flow *flow, struct flow *base,
4027 struct ofpbuf *odp_actions)
4029 struct ovs_key_ethernet eth_key;
4031 if (eth_addr_equals(base->dl_src, flow->dl_src) &&
4032 eth_addr_equals(base->dl_dst, flow->dl_dst)) {
4036 memcpy(base->dl_src, flow->dl_src, ETH_ADDR_LEN);
4037 memcpy(base->dl_dst, flow->dl_dst, ETH_ADDR_LEN);
4039 memcpy(eth_key.eth_src, base->dl_src, ETH_ADDR_LEN);
4040 memcpy(eth_key.eth_dst, base->dl_dst, ETH_ADDR_LEN);
4042 commit_set_action(odp_actions, OVS_KEY_ATTR_ETHERNET,
4043 ð_key, sizeof(eth_key));
4047 commit_vlan_action(const struct flow *flow, struct flow *base,
4048 struct ofpbuf *odp_actions)
4050 if (base->vlan_tci == flow->vlan_tci) {
4054 if (base->vlan_tci & htons(VLAN_CFI)) {
4055 nl_msg_put_flag(odp_actions, OVS_ACTION_ATTR_POP_VLAN);
4058 if (flow->vlan_tci & htons(VLAN_CFI)) {
4059 struct ovs_action_push_vlan vlan;
4061 vlan.vlan_tpid = htons(ETH_TYPE_VLAN);
4062 vlan.vlan_tci = flow->vlan_tci;
4063 nl_msg_put_unspec(odp_actions, OVS_ACTION_ATTR_PUSH_VLAN,
4064 &vlan, sizeof vlan);
4066 base->vlan_tci = flow->vlan_tci;
4070 commit_set_nw_action(const struct flow *flow, struct flow *base,
4071 struct ofpbuf *odp_actions)
4073 struct ovs_key_ipv4 ipv4_key;
4075 if (base->dl_type != htons(ETH_TYPE_IP) ||
4076 !base->nw_src || !base->nw_dst) {
4080 if (base->nw_src == flow->nw_src &&
4081 base->nw_dst == flow->nw_dst &&
4082 base->nw_tos == flow->nw_tos &&
4083 base->nw_ttl == flow->nw_ttl &&
4084 base->nw_frag == flow->nw_frag) {
4088 ipv4_key.ipv4_src = base->nw_src = flow->nw_src;
4089 ipv4_key.ipv4_dst = base->nw_dst = flow->nw_dst;
4090 ipv4_key.ipv4_tos = base->nw_tos = flow->nw_tos;
4091 ipv4_key.ipv4_ttl = base->nw_ttl = flow->nw_ttl;
4092 ipv4_key.ipv4_proto = base->nw_proto;
4093 ipv4_key.ipv4_frag = (base->nw_frag == 0 ? OVS_FRAG_TYPE_NONE
4094 : base->nw_frag == FLOW_NW_FRAG_ANY
4095 ? OVS_FRAG_TYPE_FIRST : OVS_FRAG_TYPE_LATER);
4097 commit_set_action(odp_actions, OVS_KEY_ATTR_IPV4,
4098 &ipv4_key, sizeof(ipv4_key));
4102 commit_set_port_action(const struct flow *flow, struct flow *base,
4103 struct ofpbuf *odp_actions)
4105 if (!base->tp_src || !base->tp_dst) {
4109 if (base->tp_src == flow->tp_src &&
4110 base->tp_dst == flow->tp_dst) {
4114 if (flow->nw_proto == IPPROTO_TCP) {
4115 struct ovs_key_tcp port_key;
4117 port_key.tcp_src = base->tp_src = flow->tp_src;
4118 port_key.tcp_dst = base->tp_dst = flow->tp_dst;
4120 commit_set_action(odp_actions, OVS_KEY_ATTR_TCP,
4121 &port_key, sizeof(port_key));
4123 } else if (flow->nw_proto == IPPROTO_UDP) {
4124 struct ovs_key_udp port_key;
4126 port_key.udp_src = base->tp_src = flow->tp_src;
4127 port_key.udp_dst = base->tp_dst = flow->tp_dst;
4129 commit_set_action(odp_actions, OVS_KEY_ATTR_UDP,
4130 &port_key, sizeof(port_key));
4135 commit_set_priority_action(const struct flow *flow, struct flow *base,
4136 struct ofpbuf *odp_actions)
4138 if (base->priority == flow->priority) {
4141 base->priority = flow->priority;
4143 commit_set_action(odp_actions, OVS_KEY_ATTR_PRIORITY,
4144 &base->priority, sizeof(base->priority));
4148 commit_odp_actions(struct action_xlate_ctx *ctx)
4150 const struct flow *flow = &ctx->flow;
4151 struct flow *base = &ctx->base_flow;
4152 struct ofpbuf *odp_actions = ctx->odp_actions;
4154 commit_set_tun_id_action(flow, base, odp_actions);
4155 commit_set_ether_addr_action(flow, base, odp_actions);
4156 commit_vlan_action(flow, base, odp_actions);
4157 commit_set_nw_action(flow, base, odp_actions);
4158 commit_set_port_action(flow, base, odp_actions);
4159 commit_set_priority_action(flow, base, odp_actions);
4163 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4166 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4167 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4168 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4169 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4173 struct priority_to_dscp *pdscp;
4175 if (ofport->up.opp.config & htonl(OFPPC_NO_FWD)
4176 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4180 pdscp = get_priority(ofport, ctx->flow.priority);
4182 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4183 ctx->flow.nw_tos |= pdscp->dscp;
4186 /* We may not have an ofport record for this port, but it doesn't hurt
4187 * to allow forwarding to it anyhow. Maybe such a port will appear
4188 * later and we're pre-populating the flow table. */
4191 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4192 ctx->flow.vlan_tci);
4193 if (out_port != odp_port) {
4194 ctx->flow.vlan_tci = htons(0);
4196 commit_odp_actions(ctx);
4197 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4199 ctx->sflow_odp_port = odp_port;
4200 ctx->sflow_n_outputs++;
4201 ctx->nf_output_iface = ofp_port;
4202 ctx->flow.vlan_tci = flow_vlan_tci;
4203 ctx->flow.nw_tos = flow_nw_tos;
4207 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4209 compose_output_action__(ctx, ofp_port, true);
4213 xlate_table_action(struct action_xlate_ctx *ctx,
4214 uint16_t in_port, uint8_t table_id)
4216 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4217 struct ofproto_dpif *ofproto = ctx->ofproto;
4218 struct rule_dpif *rule;
4219 uint16_t old_in_port;
4220 uint8_t old_table_id;
4222 old_table_id = ctx->table_id;
4223 ctx->table_id = table_id;
4225 /* Look up a flow with 'in_port' as the input port. */
4226 old_in_port = ctx->flow.in_port;
4227 ctx->flow.in_port = in_port;
4228 rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
4231 if (table_id > 0 && table_id < N_TABLES) {
4232 struct table_dpif *table = &ofproto->tables[table_id];
4233 if (table->other_table) {
4236 : rule_calculate_tag(&ctx->flow,
4237 &table->other_table->wc,
4242 /* Restore the original input port. Otherwise OFPP_NORMAL and
4243 * OFPP_IN_PORT will have surprising behavior. */
4244 ctx->flow.in_port = old_in_port;
4246 if (ctx->resubmit_hook) {
4247 ctx->resubmit_hook(ctx, rule);
4252 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
4256 ctx->table_id = old_table_id;
4258 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4260 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
4261 MAX_RESUBMIT_RECURSION);
4266 xlate_resubmit_table(struct action_xlate_ctx *ctx,
4267 const struct nx_action_resubmit *nar)
4272 in_port = (nar->in_port == htons(OFPP_IN_PORT)
4274 : ntohs(nar->in_port));
4275 table_id = nar->table == 255 ? ctx->table_id : nar->table;
4277 xlate_table_action(ctx, in_port, table_id);
4281 flood_packets(struct action_xlate_ctx *ctx, bool all)
4283 struct ofport_dpif *ofport;
4285 commit_odp_actions(ctx);
4286 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
4287 uint16_t ofp_port = ofport->up.ofp_port;
4289 if (ofp_port == ctx->flow.in_port) {
4294 compose_output_action__(ctx, ofp_port, false);
4295 } else if (!(ofport->up.opp.config & htonl(OFPPC_NO_FLOOD))) {
4296 compose_output_action(ctx, ofp_port);
4300 ctx->nf_output_iface = NF_OUT_FLOOD;
4304 compose_controller_action(struct action_xlate_ctx *ctx, int len)
4306 struct user_action_cookie cookie;
4308 commit_odp_actions(ctx);
4309 cookie.type = USER_ACTION_COOKIE_CONTROLLER;
4311 cookie.n_output = 0;
4312 cookie.vlan_tci = 0;
4313 put_userspace_action(ctx->ofproto, ctx->odp_actions, &ctx->flow, &cookie);
4317 xlate_output_action__(struct action_xlate_ctx *ctx,
4318 uint16_t port, uint16_t max_len)
4320 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
4322 ctx->nf_output_iface = NF_OUT_DROP;
4326 compose_output_action(ctx, ctx->flow.in_port);
4329 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
4335 flood_packets(ctx, false);
4338 flood_packets(ctx, true);
4340 case OFPP_CONTROLLER:
4341 compose_controller_action(ctx, max_len);
4344 compose_output_action(ctx, OFPP_LOCAL);
4349 if (port != ctx->flow.in_port) {
4350 compose_output_action(ctx, port);
4355 if (prev_nf_output_iface == NF_OUT_FLOOD) {
4356 ctx->nf_output_iface = NF_OUT_FLOOD;
4357 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
4358 ctx->nf_output_iface = prev_nf_output_iface;
4359 } else if (prev_nf_output_iface != NF_OUT_DROP &&
4360 ctx->nf_output_iface != NF_OUT_FLOOD) {
4361 ctx->nf_output_iface = NF_OUT_MULTI;
4366 xlate_output_reg_action(struct action_xlate_ctx *ctx,
4367 const struct nx_action_output_reg *naor)
4371 ofp_port = nxm_read_field_bits(naor->src, naor->ofs_nbits, &ctx->flow);
4373 if (ofp_port <= UINT16_MAX) {
4374 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
4379 xlate_output_action(struct action_xlate_ctx *ctx,
4380 const struct ofp_action_output *oao)
4382 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
4386 xlate_enqueue_action(struct action_xlate_ctx *ctx,
4387 const struct ofp_action_enqueue *oae)
4390 uint32_t flow_priority, priority;
4393 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
4396 /* Fall back to ordinary output action. */
4397 xlate_output_action__(ctx, ntohs(oae->port), 0);
4401 /* Figure out datapath output port. */
4402 ofp_port = ntohs(oae->port);
4403 if (ofp_port == OFPP_IN_PORT) {
4404 ofp_port = ctx->flow.in_port;
4405 } else if (ofp_port == ctx->flow.in_port) {
4409 /* Add datapath actions. */
4410 flow_priority = ctx->flow.priority;
4411 ctx->flow.priority = priority;
4412 compose_output_action(ctx, ofp_port);
4413 ctx->flow.priority = flow_priority;
4415 /* Update NetFlow output port. */
4416 if (ctx->nf_output_iface == NF_OUT_DROP) {
4417 ctx->nf_output_iface = ofp_port;
4418 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
4419 ctx->nf_output_iface = NF_OUT_MULTI;
4424 xlate_set_queue_action(struct action_xlate_ctx *ctx,
4425 const struct nx_action_set_queue *nasq)
4430 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
4433 /* Couldn't translate queue to a priority, so ignore. A warning
4434 * has already been logged. */
4438 ctx->flow.priority = priority;
4441 struct xlate_reg_state {
4447 xlate_autopath(struct action_xlate_ctx *ctx,
4448 const struct nx_action_autopath *naa)
4450 uint16_t ofp_port = ntohl(naa->id);
4451 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
4453 if (!port || !port->bundle) {
4454 ofp_port = OFPP_NONE;
4455 } else if (port->bundle->bond) {
4456 /* Autopath does not support VLAN hashing. */
4457 struct ofport_dpif *slave = bond_choose_output_slave(
4458 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
4460 ofp_port = slave->up.ofp_port;
4463 autopath_execute(naa, &ctx->flow, ofp_port);
4467 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
4469 struct ofproto_dpif *ofproto = ofproto_;
4470 struct ofport_dpif *port;
4480 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
4483 port = get_ofp_port(ofproto, ofp_port);
4484 return port ? port->may_enable : false;
4489 xlate_learn_action(struct action_xlate_ctx *ctx,
4490 const struct nx_action_learn *learn)
4492 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
4493 struct ofputil_flow_mod fm;
4496 learn_execute(learn, &ctx->flow, &fm);
4498 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
4499 if (error && !VLOG_DROP_WARN(&rl)) {
4500 char *msg = ofputil_error_to_string(error);
4501 VLOG_WARN("learning action failed to modify flow table (%s)", msg);
4509 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
4511 if (port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
4512 ? htonl(OFPPC_NO_RECV_STP)
4513 : htonl(OFPPC_NO_RECV))) {
4517 /* Only drop packets here if both forwarding and learning are
4518 * disabled. If just learning is enabled, we need to have
4519 * OFPP_NORMAL and the learning action have a look at the packet
4520 * before we can drop it. */
4521 if (!stp_forward_in_state(port->stp_state)
4522 && !stp_learn_in_state(port->stp_state)) {
4530 do_xlate_actions(const union ofp_action *in, size_t n_in,
4531 struct action_xlate_ctx *ctx)
4533 const struct ofport_dpif *port;
4534 const union ofp_action *ia;
4537 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
4538 if (port && !may_receive(port, ctx)) {
4539 /* Drop this flow. */
4543 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
4544 const struct ofp_action_dl_addr *oada;
4545 const struct nx_action_resubmit *nar;
4546 const struct nx_action_set_tunnel *nast;
4547 const struct nx_action_set_queue *nasq;
4548 const struct nx_action_multipath *nam;
4549 const struct nx_action_autopath *naa;
4550 const struct nx_action_bundle *nab;
4551 const struct nx_action_output_reg *naor;
4552 enum ofputil_action_code code;
4559 code = ofputil_decode_action_unsafe(ia);
4561 case OFPUTIL_OFPAT_OUTPUT:
4562 xlate_output_action(ctx, &ia->output);
4565 case OFPUTIL_OFPAT_SET_VLAN_VID:
4566 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
4567 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
4570 case OFPUTIL_OFPAT_SET_VLAN_PCP:
4571 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
4572 ctx->flow.vlan_tci |= htons(
4573 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
4576 case OFPUTIL_OFPAT_STRIP_VLAN:
4577 ctx->flow.vlan_tci = htons(0);
4580 case OFPUTIL_OFPAT_SET_DL_SRC:
4581 oada = ((struct ofp_action_dl_addr *) ia);
4582 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
4585 case OFPUTIL_OFPAT_SET_DL_DST:
4586 oada = ((struct ofp_action_dl_addr *) ia);
4587 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
4590 case OFPUTIL_OFPAT_SET_NW_SRC:
4591 ctx->flow.nw_src = ia->nw_addr.nw_addr;
4594 case OFPUTIL_OFPAT_SET_NW_DST:
4595 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
4598 case OFPUTIL_OFPAT_SET_NW_TOS:
4599 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4600 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
4603 case OFPUTIL_OFPAT_SET_TP_SRC:
4604 ctx->flow.tp_src = ia->tp_port.tp_port;
4607 case OFPUTIL_OFPAT_SET_TP_DST:
4608 ctx->flow.tp_dst = ia->tp_port.tp_port;
4611 case OFPUTIL_OFPAT_ENQUEUE:
4612 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
4615 case OFPUTIL_NXAST_RESUBMIT:
4616 nar = (const struct nx_action_resubmit *) ia;
4617 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
4620 case OFPUTIL_NXAST_RESUBMIT_TABLE:
4621 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
4624 case OFPUTIL_NXAST_SET_TUNNEL:
4625 nast = (const struct nx_action_set_tunnel *) ia;
4626 tun_id = htonll(ntohl(nast->tun_id));
4627 ctx->flow.tun_id = tun_id;
4630 case OFPUTIL_NXAST_SET_QUEUE:
4631 nasq = (const struct nx_action_set_queue *) ia;
4632 xlate_set_queue_action(ctx, nasq);
4635 case OFPUTIL_NXAST_POP_QUEUE:
4636 ctx->flow.priority = ctx->original_priority;
4639 case OFPUTIL_NXAST_REG_MOVE:
4640 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
4644 case OFPUTIL_NXAST_REG_LOAD:
4645 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
4649 case OFPUTIL_NXAST_NOTE:
4650 /* Nothing to do. */
4653 case OFPUTIL_NXAST_SET_TUNNEL64:
4654 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
4655 ctx->flow.tun_id = tun_id;
4658 case OFPUTIL_NXAST_MULTIPATH:
4659 nam = (const struct nx_action_multipath *) ia;
4660 multipath_execute(nam, &ctx->flow);
4663 case OFPUTIL_NXAST_AUTOPATH:
4664 naa = (const struct nx_action_autopath *) ia;
4665 xlate_autopath(ctx, naa);
4668 case OFPUTIL_NXAST_BUNDLE:
4669 ctx->ofproto->has_bundle_action = true;
4670 nab = (const struct nx_action_bundle *) ia;
4671 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
4676 case OFPUTIL_NXAST_BUNDLE_LOAD:
4677 ctx->ofproto->has_bundle_action = true;
4678 nab = (const struct nx_action_bundle *) ia;
4679 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
4683 case OFPUTIL_NXAST_OUTPUT_REG:
4684 naor = (const struct nx_action_output_reg *) ia;
4685 xlate_output_reg_action(ctx, naor);
4688 case OFPUTIL_NXAST_LEARN:
4689 ctx->has_learn = true;
4690 if (ctx->may_learn) {
4691 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
4695 case OFPUTIL_NXAST_EXIT:
4701 /* We've let OFPP_NORMAL and the learning action look at the packet,
4702 * so drop it now if forwarding is disabled. */
4703 if (port && !stp_forward_in_state(port->stp_state)) {
4704 ofpbuf_clear(ctx->odp_actions);
4705 add_sflow_action(ctx);
4710 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
4711 struct ofproto_dpif *ofproto, const struct flow *flow,
4712 ovs_be16 initial_tci, const struct ofpbuf *packet)
4714 ctx->ofproto = ofproto;
4716 ctx->base_flow = ctx->flow;
4717 ctx->base_flow.tun_id = 0;
4718 ctx->base_flow.vlan_tci = initial_tci;
4719 ctx->packet = packet;
4720 ctx->may_learn = packet != NULL;
4721 ctx->resubmit_hook = NULL;
4724 static struct ofpbuf *
4725 xlate_actions(struct action_xlate_ctx *ctx,
4726 const union ofp_action *in, size_t n_in)
4728 struct flow orig_flow = ctx->flow;
4730 COVERAGE_INC(ofproto_dpif_xlate);
4732 ctx->odp_actions = ofpbuf_new(512);
4733 ofpbuf_reserve(ctx->odp_actions, NL_A_U32_SIZE);
4735 ctx->may_set_up_flow = true;
4736 ctx->has_learn = false;
4737 ctx->has_normal = false;
4738 ctx->nf_output_iface = NF_OUT_DROP;
4741 ctx->original_priority = ctx->flow.priority;
4745 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
4746 switch (ctx->ofproto->up.frag_handling) {
4747 case OFPC_FRAG_NORMAL:
4748 /* We must pretend that transport ports are unavailable. */
4749 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
4750 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
4753 case OFPC_FRAG_DROP:
4754 return ctx->odp_actions;
4756 case OFPC_FRAG_REASM:
4759 case OFPC_FRAG_NX_MATCH:
4760 /* Nothing to do. */
4765 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
4766 ctx->may_set_up_flow = false;
4767 return ctx->odp_actions;
4769 add_sflow_action(ctx);
4770 do_xlate_actions(in, n_in, ctx);
4772 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
4773 ctx->odp_actions->data,
4774 ctx->odp_actions->size)) {
4775 ctx->may_set_up_flow = false;
4777 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
4779 compose_output_action(ctx, OFPP_LOCAL);
4782 add_mirror_actions(ctx, &orig_flow);
4783 fix_sflow_action(ctx);
4786 return ctx->odp_actions;
4789 /* OFPP_NORMAL implementation. */
4791 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
4793 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
4794 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
4795 * the bundle on which the packet was received, returns the VLAN to which the
4798 * Both 'vid' and the return value are in the range 0...4095. */
4800 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
4802 switch (in_bundle->vlan_mode) {
4803 case PORT_VLAN_ACCESS:
4804 return in_bundle->vlan;
4807 case PORT_VLAN_TRUNK:
4810 case PORT_VLAN_NATIVE_UNTAGGED:
4811 case PORT_VLAN_NATIVE_TAGGED:
4812 return vid ? vid : in_bundle->vlan;
4819 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
4820 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
4823 * 'vid' should be the VID obtained from the 802.1Q header that was received as
4824 * part of a packet (specify 0 if there was no 802.1Q header), in the range
4827 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
4829 switch (in_bundle->vlan_mode) {
4830 case PORT_VLAN_ACCESS:
4833 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4834 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
4835 "packet received on port %s configured as VLAN "
4836 "%"PRIu16" access port",
4837 in_bundle->ofproto->up.name, vid,
4838 in_bundle->name, in_bundle->vlan);
4844 case PORT_VLAN_NATIVE_UNTAGGED:
4845 case PORT_VLAN_NATIVE_TAGGED:
4847 /* Port must always carry its native VLAN. */
4851 case PORT_VLAN_TRUNK:
4852 if (!ofbundle_includes_vlan(in_bundle, vid)) {
4854 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4855 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
4856 "received on port %s not configured for trunking "
4858 in_bundle->ofproto->up.name, vid,
4859 in_bundle->name, vid);
4871 /* Given 'vlan', the VLAN that a packet belongs to, and
4872 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
4873 * that should be included in the 802.1Q header. (If the return value is 0,
4874 * then the 802.1Q header should only be included in the packet if there is a
4877 * Both 'vlan' and the return value are in the range 0...4095. */
4879 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
4881 switch (out_bundle->vlan_mode) {
4882 case PORT_VLAN_ACCESS:
4885 case PORT_VLAN_TRUNK:
4886 case PORT_VLAN_NATIVE_TAGGED:
4889 case PORT_VLAN_NATIVE_UNTAGGED:
4890 return vlan == out_bundle->vlan ? 0 : vlan;
4898 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
4901 struct ofport_dpif *port;
4903 ovs_be16 tci, old_tci;
4905 vid = output_vlan_to_vid(out_bundle, vlan);
4906 if (!out_bundle->bond) {
4907 port = ofbundle_get_a_port(out_bundle);
4909 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
4912 /* No slaves enabled, so drop packet. */
4917 old_tci = ctx->flow.vlan_tci;
4919 if (tci || out_bundle->use_priority_tags) {
4920 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
4922 tci |= htons(VLAN_CFI);
4925 ctx->flow.vlan_tci = tci;
4927 compose_output_action(ctx, port->up.ofp_port);
4928 ctx->flow.vlan_tci = old_tci;
4932 mirror_mask_ffs(mirror_mask_t mask)
4934 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
4939 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
4941 return (bundle->vlan_mode != PORT_VLAN_ACCESS
4942 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
4946 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
4948 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
4951 /* Returns an arbitrary interface within 'bundle'. */
4952 static struct ofport_dpif *
4953 ofbundle_get_a_port(const struct ofbundle *bundle)
4955 return CONTAINER_OF(list_front(&bundle->ports),
4956 struct ofport_dpif, bundle_node);
4960 vlan_is_mirrored(const struct ofmirror *m, int vlan)
4962 return !m->vlans || bitmap_is_set(m->vlans, vlan);
4965 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
4966 * to a VLAN. In general most packets may be mirrored but we want to drop
4967 * protocols that may confuse switches. */
4969 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
4971 /* If you change this function's behavior, please update corresponding
4972 * documentation in vswitch.xml at the same time. */
4973 if (dst[0] != 0x01) {
4974 /* All the currently banned MACs happen to start with 01 currently, so
4975 * this is a quick way to eliminate most of the good ones. */
4977 if (eth_addr_is_reserved(dst)) {
4978 /* Drop STP, IEEE pause frames, and other reserved protocols
4979 * (01-80-c2-00-00-0x). */
4983 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
4985 if ((dst[3] & 0xfe) == 0xcc &&
4986 (dst[4] & 0xfe) == 0xcc &&
4987 (dst[5] & 0xfe) == 0xcc) {
4988 /* Drop the following protocols plus others following the same
4991 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
4992 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
4993 STP Uplink Fast (01-00-0c-cd-cd-cd) */
4997 if (!(dst[3] | dst[4] | dst[5])) {
4998 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
5007 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5009 struct ofproto_dpif *ofproto = ctx->ofproto;
5010 mirror_mask_t mirrors;
5011 struct ofport_dpif *in_port;
5012 struct ofbundle *in_bundle;
5015 const struct nlattr *a;
5018 /* Obtain in_port from orig_flow.in_port.
5020 * lookup_input_bundle() also ensures that in_port belongs to a bundle. */
5021 in_port = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5022 ctx->packet != NULL);
5026 in_bundle = in_port->bundle;
5027 mirrors = in_bundle->src_mirrors;
5029 /* Drop frames on bundles reserved for mirroring. */
5030 if (in_bundle->mirror_out) {
5031 if (ctx->packet != NULL) {
5032 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5033 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5034 "%s, which is reserved exclusively for mirroring",
5035 ctx->ofproto->up.name, in_bundle->name);
5041 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5042 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5045 vlan = input_vid_to_vlan(in_bundle, vid);
5047 /* Look at the output ports to check for destination selections. */
5049 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5050 ctx->odp_actions->size) {
5051 enum ovs_action_attr type = nl_attr_type(a);
5052 struct ofport_dpif *ofport;
5054 if (type != OVS_ACTION_ATTR_OUTPUT) {
5058 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5059 mirrors |= ofport ? ofport->bundle->dst_mirrors : 0;
5066 /* Restore the original packet before adding the mirror actions. */
5067 ctx->flow = *orig_flow;
5072 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5074 if (!vlan_is_mirrored(m, vlan)) {
5075 mirrors &= mirrors - 1;
5079 mirrors &= ~m->dup_mirrors;
5080 ctx->mirrors |= m->dup_mirrors;
5082 output_normal(ctx, m->out, vlan);
5083 } else if (eth_dst_may_rspan(orig_flow->dl_dst)
5084 && vlan != m->out_vlan) {
5085 struct ofbundle *bundle;
5087 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5088 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5089 && !bundle->mirror_out) {
5090 output_normal(ctx, bundle, m->out_vlan);
5098 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5099 uint64_t packets, uint64_t bytes)
5105 for (; mirrors; mirrors &= mirrors - 1) {
5108 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5111 /* In normal circumstances 'm' will not be NULL. However,
5112 * if mirrors are reconfigured, we can temporarily get out
5113 * of sync in facet_revalidate(). We could "correct" the
5114 * mirror list before reaching here, but doing that would
5115 * not properly account the traffic stats we've currently
5116 * accumulated for previous mirror configuration. */
5120 m->packet_count += packets;
5121 m->byte_count += bytes;
5125 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
5126 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
5127 * indicate this; newer upstream kernels use gratuitous ARP requests. */
5129 is_gratuitous_arp(const struct flow *flow)
5131 return (flow->dl_type == htons(ETH_TYPE_ARP)
5132 && eth_addr_is_broadcast(flow->dl_dst)
5133 && (flow->nw_proto == ARP_OP_REPLY
5134 || (flow->nw_proto == ARP_OP_REQUEST
5135 && flow->nw_src == flow->nw_dst)));
5139 update_learning_table(struct ofproto_dpif *ofproto,
5140 const struct flow *flow, int vlan,
5141 struct ofbundle *in_bundle)
5143 struct mac_entry *mac;
5145 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
5149 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
5150 if (is_gratuitous_arp(flow)) {
5151 /* We don't want to learn from gratuitous ARP packets that are
5152 * reflected back over bond slaves so we lock the learning table. */
5153 if (!in_bundle->bond) {
5154 mac_entry_set_grat_arp_lock(mac);
5155 } else if (mac_entry_is_grat_arp_locked(mac)) {
5160 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
5161 /* The log messages here could actually be useful in debugging,
5162 * so keep the rate limit relatively high. */
5163 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5164 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
5165 "on port %s in VLAN %d",
5166 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
5167 in_bundle->name, vlan);
5169 mac->port.p = in_bundle;
5170 tag_set_add(&ofproto->revalidate_set,
5171 mac_learning_changed(ofproto->ml, mac));
5175 static struct ofport_dpif *
5176 lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn)
5178 struct ofport_dpif *ofport;
5180 /* Find the port and bundle for the received packet. */
5181 ofport = get_ofp_port(ofproto, in_port);
5182 if (ofport && ofport->bundle) {
5186 /* Odd. A few possible reasons here:
5188 * - We deleted a port but there are still a few packets queued up
5191 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
5192 * we don't know about.
5194 * - The ofproto client didn't configure the port as part of a bundle.
5197 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5199 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
5200 "port %"PRIu16, ofproto->up.name, in_port);
5205 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
5206 * dropped. Returns true if they may be forwarded, false if they should be
5209 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
5210 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
5212 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
5213 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
5214 * checked by input_vid_is_valid().
5216 * May also add tags to '*tags', although the current implementation only does
5217 * so in one special case.
5220 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
5221 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
5223 struct ofbundle *in_bundle = in_port->bundle;
5225 /* Drop frames for reserved multicast addresses
5226 * only if forward_bpdu option is absent. */
5227 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
5231 if (in_bundle->bond) {
5232 struct mac_entry *mac;
5234 switch (bond_check_admissibility(in_bundle->bond, in_port,
5235 flow->dl_dst, tags)) {
5242 case BV_DROP_IF_MOVED:
5243 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
5244 if (mac && mac->port.p != in_bundle &&
5245 (!is_gratuitous_arp(flow)
5246 || mac_entry_is_grat_arp_locked(mac))) {
5257 xlate_normal(struct action_xlate_ctx *ctx)
5259 struct ofport_dpif *in_port;
5260 struct ofbundle *in_bundle;
5261 struct mac_entry *mac;
5265 ctx->has_normal = true;
5267 /* Obtain in_port from ctx->flow.in_port.
5269 * lookup_input_bundle() also ensures that in_port belongs to a bundle. */
5270 in_port = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
5271 ctx->packet != NULL);
5275 in_bundle = in_port->bundle;
5277 /* Drop malformed frames. */
5278 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
5279 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
5280 if (ctx->packet != NULL) {
5281 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5282 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
5283 "VLAN tag received on port %s",
5284 ctx->ofproto->up.name, in_bundle->name);
5289 /* Drop frames on bundles reserved for mirroring. */
5290 if (in_bundle->mirror_out) {
5291 if (ctx->packet != NULL) {
5292 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5293 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5294 "%s, which is reserved exclusively for mirroring",
5295 ctx->ofproto->up.name, in_bundle->name);
5301 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
5302 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5305 vlan = input_vid_to_vlan(in_bundle, vid);
5307 /* Check other admissibility requirements. */
5308 if (!is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
5312 /* Learn source MAC. */
5313 if (ctx->may_learn) {
5314 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
5317 /* Determine output bundle. */
5318 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
5321 if (mac->port.p != in_bundle) {
5322 output_normal(ctx, mac->port.p, vlan);
5324 } else if (!ctx->packet && !eth_addr_is_multicast(ctx->flow.dl_dst)) {
5325 /* If we are revalidating but don't have a learning entry then eject
5326 * the flow. Installing a flow that floods packets opens up a window
5327 * of time where we could learn from a packet reflected on a bond and
5328 * blackhole packets before the learning table is updated to reflect
5329 * the correct port. */
5330 ctx->may_set_up_flow = false;
5333 struct ofbundle *bundle;
5335 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
5336 if (bundle != in_bundle
5337 && ofbundle_includes_vlan(bundle, vlan)
5338 && bundle->floodable
5339 && !bundle->mirror_out) {
5340 output_normal(ctx, bundle, vlan);
5343 ctx->nf_output_iface = NF_OUT_FLOOD;
5347 /* Optimized flow revalidation.
5349 * It's a difficult problem, in general, to tell which facets need to have
5350 * their actions recalculated whenever the OpenFlow flow table changes. We
5351 * don't try to solve that general problem: for most kinds of OpenFlow flow
5352 * table changes, we recalculate the actions for every facet. This is
5353 * relatively expensive, but it's good enough if the OpenFlow flow table
5354 * doesn't change very often.
5356 * However, we can expect one particular kind of OpenFlow flow table change to
5357 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5358 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5359 * table, we add a special case that applies to flow tables in which every rule
5360 * has the same form (that is, the same wildcards), except that the table is
5361 * also allowed to have a single "catch-all" flow that matches all packets. We
5362 * optimize this case by tagging all of the facets that resubmit into the table
5363 * and invalidating the same tag whenever a flow changes in that table. The
5364 * end result is that we revalidate just the facets that need it (and sometimes
5365 * a few more, but not all of the facets or even all of the facets that
5366 * resubmit to the table modified by MAC learning). */
5368 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
5369 * into an OpenFlow table with the given 'basis'. */
5371 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
5374 if (flow_wildcards_is_catchall(wc)) {
5377 struct flow tag_flow = *flow;
5378 flow_zero_wildcards(&tag_flow, wc);
5379 return tag_create_deterministic(flow_hash(&tag_flow, secret));
5383 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5384 * taggability of that table.
5386 * This function must be called after *each* change to a flow table. If you
5387 * skip calling it on some changes then the pointer comparisons at the end can
5388 * be invalid if you get unlucky. For example, if a flow removal causes a
5389 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5390 * different wildcards to be created with the same address, then this function
5391 * will incorrectly skip revalidation. */
5393 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5395 struct table_dpif *table = &ofproto->tables[table_id];
5396 const struct classifier *cls = &ofproto->up.tables[table_id];
5397 struct cls_table *catchall, *other;
5398 struct cls_table *t;
5400 catchall = other = NULL;
5402 switch (hmap_count(&cls->tables)) {
5404 /* We could tag this OpenFlow table but it would make the logic a
5405 * little harder and it's a corner case that doesn't seem worth it
5411 HMAP_FOR_EACH (t, hmap_node, &cls->tables) {
5412 if (cls_table_is_catchall(t)) {
5414 } else if (!other) {
5417 /* Indicate that we can't tag this by setting both tables to
5418 * NULL. (We know that 'catchall' is already NULL.) */
5425 /* Can't tag this table. */
5429 if (table->catchall_table != catchall || table->other_table != other) {
5430 table->catchall_table = catchall;
5431 table->other_table = other;
5432 ofproto->need_revalidate = true;
5436 /* Given 'rule' that has changed in some way (either it is a rule being
5437 * inserted, a rule being deleted, or a rule whose actions are being
5438 * modified), marks facets for revalidation to ensure that packets will be
5439 * forwarded correctly according to the new state of the flow table.
5441 * This function must be called after *each* change to a flow table. See
5442 * the comment on table_update_taggable() for more information. */
5444 rule_invalidate(const struct rule_dpif *rule)
5446 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5448 table_update_taggable(ofproto, rule->up.table_id);
5450 if (!ofproto->need_revalidate) {
5451 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5453 if (table->other_table && rule->tag) {
5454 tag_set_add(&ofproto->revalidate_set, rule->tag);
5456 ofproto->need_revalidate = true;
5462 set_frag_handling(struct ofproto *ofproto_,
5463 enum ofp_config_flags frag_handling)
5465 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5467 if (frag_handling != OFPC_FRAG_REASM) {
5468 ofproto->need_revalidate = true;
5476 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5477 const struct flow *flow,
5478 const union ofp_action *ofp_actions, size_t n_ofp_actions)
5480 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5483 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
5484 return ofp_mkerr_nicira(OFPET_BAD_REQUEST, NXBRC_BAD_IN_PORT);
5487 error = validate_actions(ofp_actions, n_ofp_actions, flow,
5488 ofproto->max_ports);
5490 struct odputil_keybuf keybuf;
5491 struct action_xlate_ctx ctx;
5492 struct ofpbuf *odp_actions;
5495 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5496 odp_flow_key_from_flow(&key, flow);
5498 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, packet);
5499 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
5500 dpif_execute(ofproto->dpif, key.data, key.size,
5501 odp_actions->data, odp_actions->size, packet);
5502 ofpbuf_delete(odp_actions);
5510 set_netflow(struct ofproto *ofproto_,
5511 const struct netflow_options *netflow_options)
5513 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5515 if (netflow_options) {
5516 if (!ofproto->netflow) {
5517 ofproto->netflow = netflow_create();
5519 return netflow_set_options(ofproto->netflow, netflow_options);
5521 netflow_destroy(ofproto->netflow);
5522 ofproto->netflow = NULL;
5528 get_netflow_ids(const struct ofproto *ofproto_,
5529 uint8_t *engine_type, uint8_t *engine_id)
5531 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5533 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
5537 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5539 if (!facet_is_controller_flow(facet) &&
5540 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5541 struct subfacet *subfacet;
5542 struct ofexpired expired;
5544 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5545 if (subfacet->installed) {
5546 struct dpif_flow_stats stats;
5548 subfacet_install(ofproto, subfacet, subfacet->actions,
5549 subfacet->actions_len, &stats);
5550 subfacet_update_stats(ofproto, subfacet, &stats);
5554 expired.flow = facet->flow;
5555 expired.packet_count = facet->packet_count;
5556 expired.byte_count = facet->byte_count;
5557 expired.used = facet->used;
5558 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5563 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5565 struct facet *facet;
5567 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5568 send_active_timeout(ofproto, facet);
5572 static struct ofproto_dpif *
5573 ofproto_dpif_lookup(const char *name)
5575 struct ofproto *ofproto = ofproto_lookup(name);
5576 return (ofproto && ofproto->ofproto_class == &ofproto_dpif_class
5577 ? ofproto_dpif_cast(ofproto)
5582 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn,
5583 const char *args, void *aux OVS_UNUSED)
5585 const struct ofproto_dpif *ofproto;
5587 ofproto = ofproto_dpif_lookup(args);
5589 unixctl_command_reply(conn, 501, "no such bridge");
5592 mac_learning_flush(ofproto->ml);
5594 unixctl_command_reply(conn, 200, "table successfully flushed");
5598 ofproto_unixctl_fdb_show(struct unixctl_conn *conn,
5599 const char *args, void *aux OVS_UNUSED)
5601 struct ds ds = DS_EMPTY_INITIALIZER;
5602 const struct ofproto_dpif *ofproto;
5603 const struct mac_entry *e;
5605 ofproto = ofproto_dpif_lookup(args);
5607 unixctl_command_reply(conn, 501, "no such bridge");
5611 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5612 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5613 struct ofbundle *bundle = e->port.p;
5614 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
5615 ofbundle_get_a_port(bundle)->odp_port,
5616 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
5618 unixctl_command_reply(conn, 200, ds_cstr(&ds));
5622 struct ofproto_trace {
5623 struct action_xlate_ctx ctx;
5629 trace_format_rule(struct ds *result, uint8_t table_id, int level,
5630 const struct rule_dpif *rule)
5632 ds_put_char_multiple(result, '\t', level);
5634 ds_put_cstr(result, "No match\n");
5638 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5639 table_id, ntohll(rule->up.flow_cookie));
5640 cls_rule_format(&rule->up.cr, result);
5641 ds_put_char(result, '\n');
5643 ds_put_char_multiple(result, '\t', level);
5644 ds_put_cstr(result, "OpenFlow ");
5645 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
5646 ds_put_char(result, '\n');
5650 trace_format_flow(struct ds *result, int level, const char *title,
5651 struct ofproto_trace *trace)
5653 ds_put_char_multiple(result, '\t', level);
5654 ds_put_format(result, "%s: ", title);
5655 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
5656 ds_put_cstr(result, "unchanged");
5658 flow_format(result, &trace->ctx.flow);
5659 trace->flow = trace->ctx.flow;
5661 ds_put_char(result, '\n');
5665 trace_format_regs(struct ds *result, int level, const char *title,
5666 struct ofproto_trace *trace)
5670 ds_put_char_multiple(result, '\t', level);
5671 ds_put_format(result, "%s:", title);
5672 for (i = 0; i < FLOW_N_REGS; i++) {
5673 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5675 ds_put_char(result, '\n');
5679 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
5681 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
5682 struct ds *result = trace->result;
5684 ds_put_char(result, '\n');
5685 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
5686 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
5687 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
5691 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
5692 void *aux OVS_UNUSED)
5694 char *dpname, *arg1, *arg2, *arg3, *arg4;
5695 char *args = xstrdup(args_);
5696 char *save_ptr = NULL;
5697 struct ofproto_dpif *ofproto;
5698 struct ofpbuf odp_key;
5699 struct ofpbuf *packet;
5700 struct rule_dpif *rule;
5701 ovs_be16 initial_tci;
5707 ofpbuf_init(&odp_key, 0);
5710 dpname = strtok_r(args, " ", &save_ptr);
5712 unixctl_command_reply(conn, 501, "Bad command syntax");
5716 ofproto = ofproto_dpif_lookup(dpname);
5718 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
5722 arg1 = strtok_r(NULL, " ", &save_ptr);
5723 arg2 = strtok_r(NULL, " ", &save_ptr);
5724 arg3 = strtok_r(NULL, " ", &save_ptr);
5725 arg4 = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
5726 if (dpname && arg1 && (!arg2 || !strcmp(arg2, "-generate")) && !arg3) {
5727 /* ofproto/trace dpname flow [-generate] */
5730 /* Convert string to datapath key. */
5731 ofpbuf_init(&odp_key, 0);
5732 error = odp_flow_key_from_string(arg1, NULL, &odp_key);
5734 unixctl_command_reply(conn, 501, "Bad flow syntax");
5738 /* Convert odp_key to flow. */
5739 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
5740 odp_key.size, &flow,
5742 if (error == ODP_FIT_ERROR) {
5743 unixctl_command_reply(conn, 501, "Invalid flow");
5747 /* Generate a packet, if requested. */
5749 packet = ofpbuf_new(0);
5750 flow_compose(packet, &flow);
5752 } else if (dpname && arg1 && arg2 && arg3 && arg4) {
5753 /* ofproto/trace dpname priority tun_id in_port packet */
5758 priority = atoi(arg1);
5759 tun_id = htonll(strtoull(arg2, NULL, 0));
5760 in_port = ofp_port_to_odp_port(atoi(arg3));
5762 packet = ofpbuf_new(strlen(args) / 2);
5763 arg4 = ofpbuf_put_hex(packet, arg4, NULL);
5764 arg4 += strspn(arg4, " ");
5765 if (*arg4 != '\0') {
5766 unixctl_command_reply(conn, 501, "Trailing garbage in command");
5769 if (packet->size < ETH_HEADER_LEN) {
5770 unixctl_command_reply(conn, 501,
5771 "Packet data too short for Ethernet");
5775 ds_put_cstr(&result, "Packet: ");
5776 s = ofp_packet_to_string(packet->data, packet->size, packet->size);
5777 ds_put_cstr(&result, s);
5780 flow_extract(packet, priority, tun_id, in_port, &flow);
5781 initial_tci = flow.vlan_tci;
5783 unixctl_command_reply(conn, 501, "Bad command syntax");
5787 ds_put_cstr(&result, "Flow: ");
5788 flow_format(&result, &flow);
5789 ds_put_char(&result, '\n');
5791 rule = rule_dpif_lookup(ofproto, &flow, 0);
5792 trace_format_rule(&result, 0, 0, rule);
5794 struct ofproto_trace trace;
5795 struct ofpbuf *odp_actions;
5797 trace.result = &result;
5799 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, initial_tci, packet);
5800 trace.ctx.resubmit_hook = trace_resubmit;
5801 odp_actions = xlate_actions(&trace.ctx,
5802 rule->up.actions, rule->up.n_actions);
5804 ds_put_char(&result, '\n');
5805 trace_format_flow(&result, 0, "Final flow", &trace);
5806 ds_put_cstr(&result, "Datapath actions: ");
5807 format_odp_actions(&result, odp_actions->data, odp_actions->size);
5808 ofpbuf_delete(odp_actions);
5810 if (!trace.ctx.may_set_up_flow) {
5812 ds_put_cstr(&result, "\nThis flow is not cachable.");
5814 ds_put_cstr(&result, "\nThe datapath actions are incomplete--"
5815 "for complete actions, please supply a packet.");
5820 unixctl_command_reply(conn, 200, ds_cstr(&result));
5823 ds_destroy(&result);
5824 ofpbuf_delete(packet);
5825 ofpbuf_uninit(&odp_key);
5830 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED,
5831 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
5834 unixctl_command_reply(conn, 200, NULL);
5838 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED,
5839 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
5842 unixctl_command_reply(conn, 200, NULL);
5846 ofproto_dpif_unixctl_init(void)
5848 static bool registered;
5854 unixctl_command_register("ofproto/trace",
5855 "bridge {tun_id in_port packet | odp_flow [-generate]}",
5856 ofproto_unixctl_trace, NULL);
5857 unixctl_command_register("fdb/flush", "bridge", ofproto_unixctl_fdb_flush,
5859 unixctl_command_register("fdb/show", "bridge", ofproto_unixctl_fdb_show,
5861 unixctl_command_register("ofproto/clog", "", ofproto_dpif_clog, NULL);
5862 unixctl_command_register("ofproto/unclog", "", ofproto_dpif_unclog, NULL);
5865 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5867 * This is deprecated. It is only for compatibility with broken device drivers
5868 * in old versions of Linux that do not properly support VLANs when VLAN
5869 * devices are not used. When broken device drivers are no longer in
5870 * widespread use, we will delete these interfaces. */
5873 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
5875 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
5876 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
5878 if (realdev_ofp_port == ofport->realdev_ofp_port
5879 && vid == ofport->vlandev_vid) {
5883 ofproto->need_revalidate = true;
5885 if (ofport->realdev_ofp_port) {
5888 if (realdev_ofp_port && ofport->bundle) {
5889 /* vlandevs are enslaved to their realdevs, so they are not allowed to
5890 * themselves be part of a bundle. */
5891 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
5894 ofport->realdev_ofp_port = realdev_ofp_port;
5895 ofport->vlandev_vid = vid;
5897 if (realdev_ofp_port) {
5898 vsp_add(ofport, realdev_ofp_port, vid);
5905 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
5907 return hash_2words(realdev_ofp_port, vid);
5911 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
5912 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
5914 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
5915 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
5916 int vid = vlan_tci_to_vid(vlan_tci);
5917 const struct vlan_splinter *vsp;
5919 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
5920 hash_realdev_vid(realdev_ofp_port, vid),
5921 &ofproto->realdev_vid_map) {
5922 if (vsp->realdev_ofp_port == realdev_ofp_port
5923 && vsp->vid == vid) {
5924 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
5928 return realdev_odp_port;
5931 static struct vlan_splinter *
5932 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
5934 struct vlan_splinter *vsp;
5936 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
5937 &ofproto->vlandev_map) {
5938 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
5947 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
5948 uint16_t vlandev_ofp_port, int *vid)
5950 if (!hmap_is_empty(&ofproto->vlandev_map)) {
5951 const struct vlan_splinter *vsp;
5953 vsp = vlandev_find(ofproto, vlandev_ofp_port);
5958 return vsp->realdev_ofp_port;
5965 vsp_remove(struct ofport_dpif *port)
5967 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
5968 struct vlan_splinter *vsp;
5970 vsp = vlandev_find(ofproto, port->up.ofp_port);
5972 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
5973 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
5976 port->realdev_ofp_port = 0;
5978 VLOG_ERR("missing vlan device record");
5983 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
5985 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
5987 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
5988 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
5989 == realdev_ofp_port)) {
5990 struct vlan_splinter *vsp;
5992 vsp = xmalloc(sizeof *vsp);
5993 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
5994 hash_int(port->up.ofp_port, 0));
5995 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
5996 hash_realdev_vid(realdev_ofp_port, vid));
5997 vsp->realdev_ofp_port = realdev_ofp_port;
5998 vsp->vlandev_ofp_port = port->up.ofp_port;
6001 port->realdev_ofp_port = realdev_ofp_port;
6003 VLOG_ERR("duplicate vlan device record");
6007 const struct ofproto_class ofproto_dpif_class = {
6034 port_is_lacp_current,
6035 NULL, /* rule_choose_table */
6042 rule_modify_actions,
6050 get_cfm_remote_mpids,
6054 get_stp_port_status,
6061 is_mirror_output_bundle,
6062 forward_bpdu_changed,