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 tag_type rule_calculate_tag(const struct flow *,
114 const struct flow_wildcards *,
116 static void rule_invalidate(const struct rule_dpif *);
118 #define MAX_MIRRORS 32
119 typedef uint32_t mirror_mask_t;
120 #define MIRROR_MASK_C(X) UINT32_C(X)
121 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
123 struct ofproto_dpif *ofproto; /* Owning ofproto. */
124 size_t idx; /* In ofproto's "mirrors" array. */
125 void *aux; /* Key supplied by ofproto's client. */
126 char *name; /* Identifier for log messages. */
128 /* Selection criteria. */
129 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
130 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
131 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
133 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
134 struct ofbundle *out; /* Output port or NULL. */
135 int out_vlan; /* Output VLAN or -1. */
136 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
139 int64_t packet_count; /* Number of packets sent. */
140 int64_t byte_count; /* Number of bytes sent. */
143 static void mirror_destroy(struct ofmirror *);
144 static void update_mirror_stats(struct ofproto_dpif *ofproto,
145 mirror_mask_t mirrors,
146 uint64_t packets, uint64_t bytes);
149 struct ofproto_dpif *ofproto; /* Owning ofproto. */
150 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
151 void *aux; /* Key supplied by ofproto's client. */
152 char *name; /* Identifier for log messages. */
155 struct list ports; /* Contains "struct ofport"s. */
156 enum port_vlan_mode vlan_mode; /* VLAN mode */
157 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
158 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
159 * NULL if all VLANs are trunked. */
160 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
161 struct bond *bond; /* Nonnull iff more than one port. */
162 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
165 bool floodable; /* True if no port has OFPPC_NO_FLOOD set. */
167 /* Port mirroring info. */
168 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
169 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
170 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
173 static void bundle_remove(struct ofport *);
174 static void bundle_update(struct ofbundle *);
175 static void bundle_destroy(struct ofbundle *);
176 static void bundle_del_port(struct ofport_dpif *);
177 static void bundle_run(struct ofbundle *);
178 static void bundle_wait(struct ofbundle *);
179 static struct ofbundle *lookup_input_bundle(struct ofproto_dpif *,
180 uint16_t in_port, bool warn);
182 /* A controller may use OFPP_NONE as the ingress port to indicate that
183 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
184 * when an input bundle is needed for validation (e.g., mirroring or
185 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
186 * any 'port' structs, so care must be taken when dealing with it. */
187 static struct ofbundle ofpp_none_bundle = {
189 .vlan_mode = PORT_VLAN_TRUNK
192 static void stp_run(struct ofproto_dpif *ofproto);
193 static void stp_wait(struct ofproto_dpif *ofproto);
194 static int set_stp_port(struct ofport *,
195 const struct ofproto_port_stp_settings *);
197 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
199 struct action_xlate_ctx {
200 /* action_xlate_ctx_init() initializes these members. */
203 struct ofproto_dpif *ofproto;
205 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
206 * this flow when actions change header fields. */
209 /* The packet corresponding to 'flow', or a null pointer if we are
210 * revalidating without a packet to refer to. */
211 const struct ofpbuf *packet;
213 /* Should OFPP_NORMAL MAC learning and NXAST_LEARN actions execute? We
214 * want to execute them if we are actually processing a packet, or if we
215 * are accounting for packets that the datapath has processed, but not if
216 * we are just revalidating. */
219 /* If nonnull, called just before executing a resubmit action.
221 * This is normally null so the client has to set it manually after
222 * calling action_xlate_ctx_init(). */
223 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *);
225 /* xlate_actions() initializes and uses these members. The client might want
226 * to look at them after it returns. */
228 struct ofpbuf *odp_actions; /* Datapath actions. */
229 tag_type tags; /* Tags associated with actions. */
230 bool may_set_up_flow; /* True ordinarily; false if the actions must
231 * be reassessed for every packet. */
232 bool has_learn; /* Actions include NXAST_LEARN? */
233 bool has_normal; /* Actions output to OFPP_NORMAL? */
234 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
235 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
237 /* xlate_actions() initializes and uses these members, but the client has no
238 * reason to look at them. */
240 int recurse; /* Recursion level, via xlate_table_action. */
241 struct flow base_flow; /* Flow at the last commit. */
242 uint32_t original_priority; /* Priority when packet arrived. */
243 uint8_t table_id; /* OpenFlow table ID where flow was found. */
244 uint32_t sflow_n_outputs; /* Number of output ports. */
245 uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
246 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
247 bool exit; /* No further actions should be processed. */
250 static void action_xlate_ctx_init(struct action_xlate_ctx *,
251 struct ofproto_dpif *, const struct flow *,
252 ovs_be16 initial_tci, const struct ofpbuf *);
253 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
254 const union ofp_action *in, size_t n_in);
256 /* An exact-match instantiation of an OpenFlow flow.
258 * A facet associates a "struct flow", which represents the Open vSwitch
259 * userspace idea of an exact-match flow, with one or more subfacets. Each
260 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
261 * the facet. When the kernel module (or other dpif implementation) and Open
262 * vSwitch userspace agree on the definition of a flow key, there is exactly
263 * one subfacet per facet. If the dpif implementation supports more-specific
264 * flow matching than userspace, however, a facet can have more than one
265 * subfacet, each of which corresponds to some distinction in flow that
266 * userspace simply doesn't understand.
268 * Flow expiration works in terms of subfacets, so a facet must have at least
269 * one subfacet or it will never expire, leaking memory. */
272 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
273 struct list list_node; /* In owning rule's 'facets' list. */
274 struct rule_dpif *rule; /* Owning rule. */
277 struct list subfacets;
278 long long int used; /* Time last used; time created if not used. */
285 * - Do include packets and bytes sent "by hand", e.g. with
288 * - Do include packets and bytes that were obtained from the datapath
289 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
290 * DPIF_FP_ZERO_STATS).
292 * - Do not include packets or bytes that can be obtained from the
293 * datapath for any existing subfacet.
295 uint64_t packet_count; /* Number of packets received. */
296 uint64_t byte_count; /* Number of bytes received. */
298 /* Resubmit statistics. */
299 uint64_t prev_packet_count; /* Number of packets from last stats push. */
300 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
301 long long int prev_used; /* Used time from last stats push. */
304 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
305 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
307 /* Properties of datapath actions.
309 * Every subfacet has its own actions because actions can differ slightly
310 * between splintered and non-splintered subfacets due to the VLAN tag
311 * being initially different (present vs. absent). All of them have these
312 * properties in common so we just store one copy of them here. */
313 bool may_install; /* Reassess actions for every packet? */
314 bool has_learn; /* Actions include NXAST_LEARN? */
315 bool has_normal; /* Actions output to OFPP_NORMAL? */
316 tag_type tags; /* Tags that would require revalidation. */
317 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
320 static struct facet *facet_create(struct rule_dpif *, const struct flow *);
321 static void facet_remove(struct ofproto_dpif *, struct facet *);
322 static void facet_free(struct facet *);
324 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
325 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
326 const struct flow *);
327 static bool facet_revalidate(struct ofproto_dpif *, struct facet *);
329 static bool execute_controller_action(struct ofproto_dpif *,
331 const struct nlattr *odp_actions,
333 struct ofpbuf *packet, bool clone);
335 static void facet_flush_stats(struct ofproto_dpif *, struct facet *);
337 static void facet_update_time(struct ofproto_dpif *, struct facet *,
339 static void facet_reset_counters(struct facet *);
340 static void facet_push_stats(struct facet *);
341 static void facet_account(struct ofproto_dpif *, struct facet *);
343 static bool facet_is_controller_flow(struct facet *);
345 /* A dpif flow and actions associated with a facet.
347 * See also the large comment on struct facet. */
350 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
351 struct list list_node; /* In struct facet's 'facets' list. */
352 struct facet *facet; /* Owning facet. */
356 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
357 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
358 * regenerate the ODP flow key from ->facet->flow. */
359 enum odp_key_fitness key_fitness;
363 long long int used; /* Time last used; time created if not used. */
365 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
366 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
370 * These should be essentially identical for every subfacet in a facet, but
371 * may differ in trivial ways due to VLAN splinters. */
372 size_t actions_len; /* Number of bytes in actions[]. */
373 struct nlattr *actions; /* Datapath actions. */
375 bool installed; /* Installed in datapath? */
377 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
378 * splinters can cause it to differ. This value should be removed when
379 * the VLAN splinters feature is no longer needed. */
380 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
383 static struct subfacet *subfacet_create(struct ofproto_dpif *, struct facet *,
384 enum odp_key_fitness,
385 const struct nlattr *key,
386 size_t key_len, ovs_be16 initial_tci);
387 static struct subfacet *subfacet_find(struct ofproto_dpif *,
388 const struct nlattr *key, size_t key_len);
389 static void subfacet_destroy(struct ofproto_dpif *, struct subfacet *);
390 static void subfacet_destroy__(struct ofproto_dpif *, struct subfacet *);
391 static void subfacet_reset_dp_stats(struct subfacet *,
392 struct dpif_flow_stats *);
393 static void subfacet_update_time(struct ofproto_dpif *, struct subfacet *,
395 static void subfacet_update_stats(struct ofproto_dpif *, struct subfacet *,
396 const struct dpif_flow_stats *);
397 static void subfacet_make_actions(struct ofproto_dpif *, struct subfacet *,
398 const struct ofpbuf *packet);
399 static int subfacet_install(struct ofproto_dpif *, struct subfacet *,
400 const struct nlattr *actions, size_t actions_len,
401 struct dpif_flow_stats *);
402 static void subfacet_uninstall(struct ofproto_dpif *, struct subfacet *);
408 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
409 struct list bundle_node; /* In struct ofbundle's "ports" list. */
410 struct cfm *cfm; /* Connectivity Fault Management, if any. */
411 tag_type tag; /* Tag associated with this port. */
412 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
413 bool may_enable; /* May be enabled in bonds. */
416 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
417 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
418 long long int stp_state_entered;
420 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
422 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
424 * This is deprecated. It is only for compatibility with broken device
425 * drivers in old versions of Linux that do not properly support VLANs when
426 * VLAN devices are not used. When broken device drivers are no longer in
427 * widespread use, we will delete these interfaces. */
428 uint16_t realdev_ofp_port;
432 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
433 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
434 * traffic egressing the 'ofport' with that priority should be marked with. */
435 struct priority_to_dscp {
436 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
437 uint32_t priority; /* Priority of this queue (see struct flow). */
439 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
442 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
444 * This is deprecated. It is only for compatibility with broken device drivers
445 * in old versions of Linux that do not properly support VLANs when VLAN
446 * devices are not used. When broken device drivers are no longer in
447 * widespread use, we will delete these interfaces. */
448 struct vlan_splinter {
449 struct hmap_node realdev_vid_node;
450 struct hmap_node vlandev_node;
451 uint16_t realdev_ofp_port;
452 uint16_t vlandev_ofp_port;
456 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
457 uint32_t realdev, ovs_be16 vlan_tci);
458 static uint16_t vsp_vlandev_to_realdev(const struct ofproto_dpif *,
459 uint16_t vlandev, int *vid);
460 static void vsp_remove(struct ofport_dpif *);
461 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
463 static struct ofport_dpif *
464 ofport_dpif_cast(const struct ofport *ofport)
466 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
467 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
470 static void port_run(struct ofport_dpif *);
471 static void port_wait(struct ofport_dpif *);
472 static int set_cfm(struct ofport *, const struct cfm_settings *);
473 static void ofport_clear_priorities(struct ofport_dpif *);
475 struct dpif_completion {
476 struct list list_node;
477 struct ofoperation *op;
480 /* Extra information about a classifier table.
481 * Currently used just for optimized flow revalidation. */
483 /* If either of these is nonnull, then this table has a form that allows
484 * flows to be tagged to avoid revalidating most flows for the most common
485 * kinds of flow table changes. */
486 struct cls_table *catchall_table; /* Table that wildcards all fields. */
487 struct cls_table *other_table; /* Table with any other wildcard set. */
488 uint32_t basis; /* Keeps each table's tags separate. */
491 struct ofproto_dpif {
492 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
501 struct netflow *netflow;
502 struct dpif_sflow *sflow;
503 struct hmap bundles; /* Contains "struct ofbundle"s. */
504 struct mac_learning *ml;
505 struct ofmirror *mirrors[MAX_MIRRORS];
506 bool has_bonded_bundles;
509 struct timer next_expiration;
513 struct hmap subfacets;
516 struct table_dpif tables[N_TABLES];
517 bool need_revalidate;
518 struct tag_set revalidate_set;
520 /* Support for debugging async flow mods. */
521 struct list completions;
523 bool has_bundle_action; /* True when the first bundle action appears. */
527 long long int stp_last_tick;
529 /* VLAN splinters. */
530 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
531 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
534 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
535 * for debugging the asynchronous flow_mod implementation.) */
538 /* All existing ofproto_dpif instances, indexed by ->up.name. */
539 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
541 static void ofproto_dpif_unixctl_init(void);
543 static struct ofproto_dpif *
544 ofproto_dpif_cast(const struct ofproto *ofproto)
546 assert(ofproto->ofproto_class == &ofproto_dpif_class);
547 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
550 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
552 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
555 /* Packet processing. */
556 static void update_learning_table(struct ofproto_dpif *,
557 const struct flow *, int vlan,
560 #define FLOW_MISS_MAX_BATCH 50
561 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
563 /* Flow expiration. */
564 static int expire(struct ofproto_dpif *);
567 static void send_netflow_active_timeouts(struct ofproto_dpif *);
570 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
572 compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions,
573 const struct flow *, uint32_t odp_port);
574 static void add_mirror_actions(struct action_xlate_ctx *ctx,
575 const struct flow *flow);
576 /* Global variables. */
577 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
579 /* Factory functions. */
582 enumerate_types(struct sset *types)
584 dp_enumerate_types(types);
588 enumerate_names(const char *type, struct sset *names)
590 return dp_enumerate_names(type, names);
594 del(const char *type, const char *name)
599 error = dpif_open(name, type, &dpif);
601 error = dpif_delete(dpif);
607 /* Basic life-cycle. */
609 static struct ofproto *
612 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
617 dealloc(struct ofproto *ofproto_)
619 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
624 construct(struct ofproto *ofproto_, int *n_tablesp)
626 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
627 const char *name = ofproto->up.name;
631 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
633 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
637 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
638 ofproto->n_matches = 0;
640 dpif_flow_flush(ofproto->dpif);
641 dpif_recv_purge(ofproto->dpif);
643 error = dpif_recv_set_mask(ofproto->dpif,
644 ((1u << DPIF_UC_MISS) |
645 (1u << DPIF_UC_ACTION)));
647 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
648 dpif_close(ofproto->dpif);
652 ofproto->netflow = NULL;
653 ofproto->sflow = NULL;
655 hmap_init(&ofproto->bundles);
656 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
657 for (i = 0; i < MAX_MIRRORS; i++) {
658 ofproto->mirrors[i] = NULL;
660 ofproto->has_bonded_bundles = false;
662 timer_set_duration(&ofproto->next_expiration, 1000);
664 hmap_init(&ofproto->facets);
665 hmap_init(&ofproto->subfacets);
667 for (i = 0; i < N_TABLES; i++) {
668 struct table_dpif *table = &ofproto->tables[i];
670 table->catchall_table = NULL;
671 table->other_table = NULL;
672 table->basis = random_uint32();
674 ofproto->need_revalidate = false;
675 tag_set_init(&ofproto->revalidate_set);
677 list_init(&ofproto->completions);
679 ofproto_dpif_unixctl_init();
681 ofproto->has_bundle_action = false;
683 hmap_init(&ofproto->vlandev_map);
684 hmap_init(&ofproto->realdev_vid_map);
686 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
687 hash_string(ofproto->up.name, 0));
689 *n_tablesp = N_TABLES;
694 complete_operations(struct ofproto_dpif *ofproto)
696 struct dpif_completion *c, *next;
698 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
699 ofoperation_complete(c->op, 0);
700 list_remove(&c->list_node);
706 destruct(struct ofproto *ofproto_)
708 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
709 struct rule_dpif *rule, *next_rule;
710 struct classifier *table;
713 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
714 complete_operations(ofproto);
716 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
717 struct cls_cursor cursor;
719 cls_cursor_init(&cursor, table, NULL);
720 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
721 ofproto_rule_destroy(&rule->up);
725 for (i = 0; i < MAX_MIRRORS; i++) {
726 mirror_destroy(ofproto->mirrors[i]);
729 netflow_destroy(ofproto->netflow);
730 dpif_sflow_destroy(ofproto->sflow);
731 hmap_destroy(&ofproto->bundles);
732 mac_learning_destroy(ofproto->ml);
734 hmap_destroy(&ofproto->facets);
735 hmap_destroy(&ofproto->subfacets);
737 hmap_destroy(&ofproto->vlandev_map);
738 hmap_destroy(&ofproto->realdev_vid_map);
740 dpif_close(ofproto->dpif);
744 run_fast(struct ofproto *ofproto_)
746 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
749 /* Handle one or more batches of upcalls, until there's nothing left to do
750 * or until we do a fixed total amount of work.
752 * We do work in batches because it can be much cheaper to set up a number
753 * of flows and fire off their patches all at once. We do multiple batches
754 * because in some cases handling a packet can cause another packet to be
755 * queued almost immediately as part of the return flow. Both
756 * optimizations can make major improvements on some benchmarks and
757 * presumably for real traffic as well. */
759 while (work < FLOW_MISS_MAX_BATCH) {
760 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
770 run(struct ofproto *ofproto_)
772 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
773 struct ofport_dpif *ofport;
774 struct ofbundle *bundle;
778 complete_operations(ofproto);
780 dpif_run(ofproto->dpif);
782 error = run_fast(ofproto_);
787 if (timer_expired(&ofproto->next_expiration)) {
788 int delay = expire(ofproto);
789 timer_set_duration(&ofproto->next_expiration, delay);
792 if (ofproto->netflow) {
793 if (netflow_run(ofproto->netflow)) {
794 send_netflow_active_timeouts(ofproto);
797 if (ofproto->sflow) {
798 dpif_sflow_run(ofproto->sflow);
801 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
804 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
809 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
811 /* Now revalidate if there's anything to do. */
812 if (ofproto->need_revalidate
813 || !tag_set_is_empty(&ofproto->revalidate_set)) {
814 struct tag_set revalidate_set = ofproto->revalidate_set;
815 bool revalidate_all = ofproto->need_revalidate;
816 struct facet *facet, *next;
818 /* Clear the revalidation flags. */
819 tag_set_init(&ofproto->revalidate_set);
820 ofproto->need_revalidate = false;
822 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
824 || tag_set_intersects(&revalidate_set, facet->tags)) {
825 facet_revalidate(ofproto, facet);
834 wait(struct ofproto *ofproto_)
836 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
837 struct ofport_dpif *ofport;
838 struct ofbundle *bundle;
840 if (!clogged && !list_is_empty(&ofproto->completions)) {
841 poll_immediate_wake();
844 dpif_wait(ofproto->dpif);
845 dpif_recv_wait(ofproto->dpif);
846 if (ofproto->sflow) {
847 dpif_sflow_wait(ofproto->sflow);
849 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
850 poll_immediate_wake();
852 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
855 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
858 if (ofproto->netflow) {
859 netflow_wait(ofproto->netflow);
861 mac_learning_wait(ofproto->ml);
863 if (ofproto->need_revalidate) {
864 /* Shouldn't happen, but if it does just go around again. */
865 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
866 poll_immediate_wake();
868 timer_wait(&ofproto->next_expiration);
873 flush(struct ofproto *ofproto_)
875 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
876 struct facet *facet, *next_facet;
878 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
879 /* Mark the facet as not installed so that facet_remove() doesn't
880 * bother trying to uninstall it. There is no point in uninstalling it
881 * individually since we are about to blow away all the facets with
882 * dpif_flow_flush(). */
883 struct subfacet *subfacet;
885 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
886 subfacet->installed = false;
887 subfacet->dp_packet_count = 0;
888 subfacet->dp_byte_count = 0;
890 facet_remove(ofproto, facet);
892 dpif_flow_flush(ofproto->dpif);
896 get_features(struct ofproto *ofproto_ OVS_UNUSED,
897 bool *arp_match_ip, uint32_t *actions)
899 *arp_match_ip = true;
900 *actions = ((1u << OFPAT_OUTPUT) |
901 (1u << OFPAT_SET_VLAN_VID) |
902 (1u << OFPAT_SET_VLAN_PCP) |
903 (1u << OFPAT_STRIP_VLAN) |
904 (1u << OFPAT_SET_DL_SRC) |
905 (1u << OFPAT_SET_DL_DST) |
906 (1u << OFPAT_SET_NW_SRC) |
907 (1u << OFPAT_SET_NW_DST) |
908 (1u << OFPAT_SET_NW_TOS) |
909 (1u << OFPAT_SET_TP_SRC) |
910 (1u << OFPAT_SET_TP_DST) |
911 (1u << OFPAT_ENQUEUE));
915 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
917 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
918 struct dpif_dp_stats s;
920 strcpy(ots->name, "classifier");
922 dpif_get_dp_stats(ofproto->dpif, &s);
923 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
924 put_32aligned_be64(&ots->matched_count,
925 htonll(s.n_hit + ofproto->n_matches));
928 static struct ofport *
931 struct ofport_dpif *port = xmalloc(sizeof *port);
936 port_dealloc(struct ofport *port_)
938 struct ofport_dpif *port = ofport_dpif_cast(port_);
943 port_construct(struct ofport *port_)
945 struct ofport_dpif *port = ofport_dpif_cast(port_);
946 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
948 ofproto->need_revalidate = true;
949 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
952 port->tag = tag_create_random();
953 port->may_enable = true;
954 port->stp_port = NULL;
955 port->stp_state = STP_DISABLED;
956 hmap_init(&port->priorities);
957 port->realdev_ofp_port = 0;
958 port->vlandev_vid = 0;
960 if (ofproto->sflow) {
961 dpif_sflow_add_port(ofproto->sflow, port->odp_port,
962 netdev_get_name(port->up.netdev));
969 port_destruct(struct ofport *port_)
971 struct ofport_dpif *port = ofport_dpif_cast(port_);
972 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
974 ofproto->need_revalidate = true;
975 bundle_remove(port_);
976 set_cfm(port_, NULL);
977 if (ofproto->sflow) {
978 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
981 ofport_clear_priorities(port);
982 hmap_destroy(&port->priorities);
986 port_modified(struct ofport *port_)
988 struct ofport_dpif *port = ofport_dpif_cast(port_);
990 if (port->bundle && port->bundle->bond) {
991 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
996 port_reconfigured(struct ofport *port_, ovs_be32 old_config)
998 struct ofport_dpif *port = ofport_dpif_cast(port_);
999 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1000 ovs_be32 changed = old_config ^ port->up.opp.config;
1002 if (changed & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
1003 OFPPC_NO_FWD | OFPPC_NO_FLOOD)) {
1004 ofproto->need_revalidate = true;
1006 if (changed & htonl(OFPPC_NO_FLOOD) && port->bundle) {
1007 bundle_update(port->bundle);
1013 set_sflow(struct ofproto *ofproto_,
1014 const struct ofproto_sflow_options *sflow_options)
1016 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1017 struct dpif_sflow *ds = ofproto->sflow;
1019 if (sflow_options) {
1021 struct ofport_dpif *ofport;
1023 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1024 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1025 dpif_sflow_add_port(ds, ofport->odp_port,
1026 netdev_get_name(ofport->up.netdev));
1028 ofproto->need_revalidate = true;
1030 dpif_sflow_set_options(ds, sflow_options);
1033 dpif_sflow_destroy(ds);
1034 ofproto->need_revalidate = true;
1035 ofproto->sflow = NULL;
1042 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1044 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1051 struct ofproto_dpif *ofproto;
1053 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1054 ofproto->need_revalidate = true;
1055 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1058 if (cfm_configure(ofport->cfm, s)) {
1064 cfm_destroy(ofport->cfm);
1070 get_cfm_fault(const struct ofport *ofport_)
1072 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1074 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1078 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1081 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1084 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1091 /* Spanning Tree. */
1094 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1096 struct ofproto_dpif *ofproto = ofproto_;
1097 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1098 struct ofport_dpif *ofport;
1100 ofport = stp_port_get_aux(sp);
1102 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1103 ofproto->up.name, port_num);
1105 struct eth_header *eth = pkt->l2;
1107 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1108 if (eth_addr_is_zero(eth->eth_src)) {
1109 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1110 "with unknown MAC", ofproto->up.name, port_num);
1112 send_packet(ofport, pkt);
1118 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1120 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1122 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1124 /* Only revalidate flows if the configuration changed. */
1125 if (!s != !ofproto->stp) {
1126 ofproto->need_revalidate = true;
1130 if (!ofproto->stp) {
1131 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1132 send_bpdu_cb, ofproto);
1133 ofproto->stp_last_tick = time_msec();
1136 stp_set_bridge_id(ofproto->stp, s->system_id);
1137 stp_set_bridge_priority(ofproto->stp, s->priority);
1138 stp_set_hello_time(ofproto->stp, s->hello_time);
1139 stp_set_max_age(ofproto->stp, s->max_age);
1140 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1142 struct ofport *ofport;
1144 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1145 set_stp_port(ofport, NULL);
1148 stp_destroy(ofproto->stp);
1149 ofproto->stp = NULL;
1156 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1158 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1162 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1163 s->designated_root = stp_get_designated_root(ofproto->stp);
1164 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1173 update_stp_port_state(struct ofport_dpif *ofport)
1175 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1176 enum stp_state state;
1178 /* Figure out new state. */
1179 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1183 if (ofport->stp_state != state) {
1187 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1188 netdev_get_name(ofport->up.netdev),
1189 stp_state_name(ofport->stp_state),
1190 stp_state_name(state));
1191 if (stp_learn_in_state(ofport->stp_state)
1192 != stp_learn_in_state(state)) {
1193 /* xxx Learning action flows should also be flushed. */
1194 mac_learning_flush(ofproto->ml);
1196 fwd_change = stp_forward_in_state(ofport->stp_state)
1197 != stp_forward_in_state(state);
1199 ofproto->need_revalidate = true;
1200 ofport->stp_state = state;
1201 ofport->stp_state_entered = time_msec();
1203 if (fwd_change && ofport->bundle) {
1204 bundle_update(ofport->bundle);
1207 /* Update the STP state bits in the OpenFlow port description. */
1208 of_state = (ofport->up.opp.state & htonl(~OFPPS_STP_MASK))
1209 | htonl(state == STP_LISTENING ? OFPPS_STP_LISTEN
1210 : state == STP_LEARNING ? OFPPS_STP_LEARN
1211 : state == STP_FORWARDING ? OFPPS_STP_FORWARD
1212 : state == STP_BLOCKING ? OFPPS_STP_BLOCK
1214 ofproto_port_set_state(&ofport->up, of_state);
1218 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1219 * caller is responsible for assigning STP port numbers and ensuring
1220 * there are no duplicates. */
1222 set_stp_port(struct ofport *ofport_,
1223 const struct ofproto_port_stp_settings *s)
1225 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1226 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1227 struct stp_port *sp = ofport->stp_port;
1229 if (!s || !s->enable) {
1231 ofport->stp_port = NULL;
1232 stp_port_disable(sp);
1233 update_stp_port_state(ofport);
1236 } else if (sp && stp_port_no(sp) != s->port_num
1237 && ofport == stp_port_get_aux(sp)) {
1238 /* The port-id changed, so disable the old one if it's not
1239 * already in use by another port. */
1240 stp_port_disable(sp);
1243 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1244 stp_port_enable(sp);
1246 stp_port_set_aux(sp, ofport);
1247 stp_port_set_priority(sp, s->priority);
1248 stp_port_set_path_cost(sp, s->path_cost);
1250 update_stp_port_state(ofport);
1256 get_stp_port_status(struct ofport *ofport_,
1257 struct ofproto_port_stp_status *s)
1259 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1260 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1261 struct stp_port *sp = ofport->stp_port;
1263 if (!ofproto->stp || !sp) {
1269 s->port_id = stp_port_get_id(sp);
1270 s->state = stp_port_get_state(sp);
1271 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1272 s->role = stp_port_get_role(sp);
1273 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1279 stp_run(struct ofproto_dpif *ofproto)
1282 long long int now = time_msec();
1283 long long int elapsed = now - ofproto->stp_last_tick;
1284 struct stp_port *sp;
1287 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1288 ofproto->stp_last_tick = now;
1290 while (stp_get_changed_port(ofproto->stp, &sp)) {
1291 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1294 update_stp_port_state(ofport);
1301 stp_wait(struct ofproto_dpif *ofproto)
1304 poll_timer_wait(1000);
1308 /* Returns true if STP should process 'flow'. */
1310 stp_should_process_flow(const struct flow *flow)
1312 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1316 stp_process_packet(const struct ofport_dpif *ofport,
1317 const struct ofpbuf *packet)
1319 struct ofpbuf payload = *packet;
1320 struct eth_header *eth = payload.data;
1321 struct stp_port *sp = ofport->stp_port;
1323 /* Sink packets on ports that have STP disabled when the bridge has
1325 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1329 /* Trim off padding on payload. */
1330 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1331 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1334 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1335 stp_received_bpdu(sp, payload.data, payload.size);
1339 static struct priority_to_dscp *
1340 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1342 struct priority_to_dscp *pdscp;
1345 hash = hash_int(priority, 0);
1346 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1347 if (pdscp->priority == priority) {
1355 ofport_clear_priorities(struct ofport_dpif *ofport)
1357 struct priority_to_dscp *pdscp, *next;
1359 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1360 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1366 set_queues(struct ofport *ofport_,
1367 const struct ofproto_port_queue *qdscp_list,
1370 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1371 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1372 struct hmap new = HMAP_INITIALIZER(&new);
1375 for (i = 0; i < n_qdscp; i++) {
1376 struct priority_to_dscp *pdscp;
1380 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1381 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1386 pdscp = get_priority(ofport, priority);
1388 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1390 pdscp = xmalloc(sizeof *pdscp);
1391 pdscp->priority = priority;
1393 ofproto->need_revalidate = true;
1396 if (pdscp->dscp != dscp) {
1398 ofproto->need_revalidate = true;
1401 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1404 if (!hmap_is_empty(&ofport->priorities)) {
1405 ofport_clear_priorities(ofport);
1406 ofproto->need_revalidate = true;
1409 hmap_swap(&new, &ofport->priorities);
1417 /* Expires all MAC learning entries associated with 'bundle' and forces its
1418 * ofproto to revalidate every flow.
1420 * Normally MAC learning entries are removed only from the ofproto associated
1421 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1422 * are removed from every ofproto. When patch ports and SLB bonds are in use
1423 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1424 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1425 * with the host from which it migrated. */
1427 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1429 struct ofproto_dpif *ofproto = bundle->ofproto;
1430 struct mac_learning *ml = ofproto->ml;
1431 struct mac_entry *mac, *next_mac;
1433 ofproto->need_revalidate = true;
1434 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1435 if (mac->port.p == bundle) {
1437 struct ofproto_dpif *o;
1439 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1441 struct mac_entry *e;
1443 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1446 tag_set_add(&o->revalidate_set, e->tag);
1447 mac_learning_expire(o->ml, e);
1453 mac_learning_expire(ml, mac);
1458 static struct ofbundle *
1459 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1461 struct ofbundle *bundle;
1463 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1464 &ofproto->bundles) {
1465 if (bundle->aux == aux) {
1472 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1473 * ones that are found to 'bundles'. */
1475 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1476 void **auxes, size_t n_auxes,
1477 struct hmapx *bundles)
1481 hmapx_init(bundles);
1482 for (i = 0; i < n_auxes; i++) {
1483 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1485 hmapx_add(bundles, bundle);
1491 bundle_update(struct ofbundle *bundle)
1493 struct ofport_dpif *port;
1495 bundle->floodable = true;
1496 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1497 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
1498 bundle->floodable = false;
1505 bundle_del_port(struct ofport_dpif *port)
1507 struct ofbundle *bundle = port->bundle;
1509 bundle->ofproto->need_revalidate = true;
1511 list_remove(&port->bundle_node);
1512 port->bundle = NULL;
1515 lacp_slave_unregister(bundle->lacp, port);
1518 bond_slave_unregister(bundle->bond, port);
1521 bundle_update(bundle);
1525 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1526 struct lacp_slave_settings *lacp,
1527 uint32_t bond_stable_id)
1529 struct ofport_dpif *port;
1531 port = get_ofp_port(bundle->ofproto, ofp_port);
1536 if (port->bundle != bundle) {
1537 bundle->ofproto->need_revalidate = true;
1539 bundle_del_port(port);
1542 port->bundle = bundle;
1543 list_push_back(&bundle->ports, &port->bundle_node);
1544 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
1545 bundle->floodable = false;
1549 port->bundle->ofproto->need_revalidate = true;
1550 lacp_slave_register(bundle->lacp, port, lacp);
1553 port->bond_stable_id = bond_stable_id;
1559 bundle_destroy(struct ofbundle *bundle)
1561 struct ofproto_dpif *ofproto;
1562 struct ofport_dpif *port, *next_port;
1569 ofproto = bundle->ofproto;
1570 for (i = 0; i < MAX_MIRRORS; i++) {
1571 struct ofmirror *m = ofproto->mirrors[i];
1573 if (m->out == bundle) {
1575 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1576 || hmapx_find_and_delete(&m->dsts, bundle)) {
1577 ofproto->need_revalidate = true;
1582 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1583 bundle_del_port(port);
1586 bundle_flush_macs(bundle, true);
1587 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1589 free(bundle->trunks);
1590 lacp_destroy(bundle->lacp);
1591 bond_destroy(bundle->bond);
1596 bundle_set(struct ofproto *ofproto_, void *aux,
1597 const struct ofproto_bundle_settings *s)
1599 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1600 bool need_flush = false;
1601 struct ofport_dpif *port;
1602 struct ofbundle *bundle;
1603 unsigned long *trunks;
1609 bundle_destroy(bundle_lookup(ofproto, aux));
1613 assert(s->n_slaves == 1 || s->bond != NULL);
1614 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1616 bundle = bundle_lookup(ofproto, aux);
1618 bundle = xmalloc(sizeof *bundle);
1620 bundle->ofproto = ofproto;
1621 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1622 hash_pointer(aux, 0));
1624 bundle->name = NULL;
1626 list_init(&bundle->ports);
1627 bundle->vlan_mode = PORT_VLAN_TRUNK;
1629 bundle->trunks = NULL;
1630 bundle->use_priority_tags = s->use_priority_tags;
1631 bundle->lacp = NULL;
1632 bundle->bond = NULL;
1634 bundle->floodable = true;
1636 bundle->src_mirrors = 0;
1637 bundle->dst_mirrors = 0;
1638 bundle->mirror_out = 0;
1641 if (!bundle->name || strcmp(s->name, bundle->name)) {
1643 bundle->name = xstrdup(s->name);
1648 if (!bundle->lacp) {
1649 ofproto->need_revalidate = true;
1650 bundle->lacp = lacp_create();
1652 lacp_configure(bundle->lacp, s->lacp);
1654 lacp_destroy(bundle->lacp);
1655 bundle->lacp = NULL;
1658 /* Update set of ports. */
1660 for (i = 0; i < s->n_slaves; i++) {
1661 if (!bundle_add_port(bundle, s->slaves[i],
1662 s->lacp ? &s->lacp_slaves[i] : NULL,
1663 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1667 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1668 struct ofport_dpif *next_port;
1670 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1671 for (i = 0; i < s->n_slaves; i++) {
1672 if (s->slaves[i] == port->up.ofp_port) {
1677 bundle_del_port(port);
1681 assert(list_size(&bundle->ports) <= s->n_slaves);
1683 if (list_is_empty(&bundle->ports)) {
1684 bundle_destroy(bundle);
1688 /* Set VLAN tagging mode */
1689 if (s->vlan_mode != bundle->vlan_mode
1690 || s->use_priority_tags != bundle->use_priority_tags) {
1691 bundle->vlan_mode = s->vlan_mode;
1692 bundle->use_priority_tags = s->use_priority_tags;
1697 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1698 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1700 if (vlan != bundle->vlan) {
1701 bundle->vlan = vlan;
1705 /* Get trunked VLANs. */
1706 switch (s->vlan_mode) {
1707 case PORT_VLAN_ACCESS:
1711 case PORT_VLAN_TRUNK:
1712 trunks = (unsigned long *) s->trunks;
1715 case PORT_VLAN_NATIVE_UNTAGGED:
1716 case PORT_VLAN_NATIVE_TAGGED:
1717 if (vlan != 0 && (!s->trunks
1718 || !bitmap_is_set(s->trunks, vlan)
1719 || bitmap_is_set(s->trunks, 0))) {
1720 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1722 trunks = bitmap_clone(s->trunks, 4096);
1724 trunks = bitmap_allocate1(4096);
1726 bitmap_set1(trunks, vlan);
1727 bitmap_set0(trunks, 0);
1729 trunks = (unsigned long *) s->trunks;
1736 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1737 free(bundle->trunks);
1738 if (trunks == s->trunks) {
1739 bundle->trunks = vlan_bitmap_clone(trunks);
1741 bundle->trunks = trunks;
1746 if (trunks != s->trunks) {
1751 if (!list_is_short(&bundle->ports)) {
1752 bundle->ofproto->has_bonded_bundles = true;
1754 if (bond_reconfigure(bundle->bond, s->bond)) {
1755 ofproto->need_revalidate = true;
1758 bundle->bond = bond_create(s->bond);
1759 ofproto->need_revalidate = true;
1762 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1763 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1767 bond_destroy(bundle->bond);
1768 bundle->bond = NULL;
1771 /* If we changed something that would affect MAC learning, un-learn
1772 * everything on this port and force flow revalidation. */
1774 bundle_flush_macs(bundle, false);
1781 bundle_remove(struct ofport *port_)
1783 struct ofport_dpif *port = ofport_dpif_cast(port_);
1784 struct ofbundle *bundle = port->bundle;
1787 bundle_del_port(port);
1788 if (list_is_empty(&bundle->ports)) {
1789 bundle_destroy(bundle);
1790 } else if (list_is_short(&bundle->ports)) {
1791 bond_destroy(bundle->bond);
1792 bundle->bond = NULL;
1798 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
1800 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1801 struct ofport_dpif *port = port_;
1802 uint8_t ea[ETH_ADDR_LEN];
1805 error = netdev_get_etheraddr(port->up.netdev, ea);
1807 struct ofpbuf packet;
1810 ofpbuf_init(&packet, 0);
1811 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1813 memcpy(packet_pdu, pdu, pdu_size);
1815 send_packet(port, &packet);
1816 ofpbuf_uninit(&packet);
1818 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1819 "%s (%s)", port->bundle->name,
1820 netdev_get_name(port->up.netdev), strerror(error));
1825 bundle_send_learning_packets(struct ofbundle *bundle)
1827 struct ofproto_dpif *ofproto = bundle->ofproto;
1828 int error, n_packets, n_errors;
1829 struct mac_entry *e;
1831 error = n_packets = n_errors = 0;
1832 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1833 if (e->port.p != bundle) {
1834 struct ofpbuf *learning_packet;
1835 struct ofport_dpif *port;
1838 learning_packet = bond_compose_learning_packet(bundle->bond, e->mac,
1841 ret = send_packet(port, learning_packet);
1842 ofpbuf_delete(learning_packet);
1852 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1853 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1854 "packets, last error was: %s",
1855 bundle->name, n_errors, n_packets, strerror(error));
1857 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1858 bundle->name, n_packets);
1863 bundle_run(struct ofbundle *bundle)
1866 lacp_run(bundle->lacp, send_pdu_cb);
1869 struct ofport_dpif *port;
1871 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1872 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1875 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1876 lacp_negotiated(bundle->lacp));
1877 if (bond_should_send_learning_packets(bundle->bond)) {
1878 bundle_send_learning_packets(bundle);
1884 bundle_wait(struct ofbundle *bundle)
1887 lacp_wait(bundle->lacp);
1890 bond_wait(bundle->bond);
1897 mirror_scan(struct ofproto_dpif *ofproto)
1901 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1902 if (!ofproto->mirrors[idx]) {
1909 static struct ofmirror *
1910 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1914 for (i = 0; i < MAX_MIRRORS; i++) {
1915 struct ofmirror *mirror = ofproto->mirrors[i];
1916 if (mirror && mirror->aux == aux) {
1924 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
1926 mirror_update_dups(struct ofproto_dpif *ofproto)
1930 for (i = 0; i < MAX_MIRRORS; i++) {
1931 struct ofmirror *m = ofproto->mirrors[i];
1934 m->dup_mirrors = MIRROR_MASK_C(1) << i;
1938 for (i = 0; i < MAX_MIRRORS; i++) {
1939 struct ofmirror *m1 = ofproto->mirrors[i];
1946 for (j = i + 1; j < MAX_MIRRORS; j++) {
1947 struct ofmirror *m2 = ofproto->mirrors[j];
1949 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
1950 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
1951 m2->dup_mirrors |= m1->dup_mirrors;
1958 mirror_set(struct ofproto *ofproto_, void *aux,
1959 const struct ofproto_mirror_settings *s)
1961 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1962 mirror_mask_t mirror_bit;
1963 struct ofbundle *bundle;
1964 struct ofmirror *mirror;
1965 struct ofbundle *out;
1966 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
1967 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
1970 mirror = mirror_lookup(ofproto, aux);
1972 mirror_destroy(mirror);
1978 idx = mirror_scan(ofproto);
1980 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
1982 ofproto->up.name, MAX_MIRRORS, s->name);
1986 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
1987 mirror->ofproto = ofproto;
1990 mirror->out_vlan = -1;
1991 mirror->name = NULL;
1994 if (!mirror->name || strcmp(s->name, mirror->name)) {
1996 mirror->name = xstrdup(s->name);
1999 /* Get the new configuration. */
2000 if (s->out_bundle) {
2001 out = bundle_lookup(ofproto, s->out_bundle);
2003 mirror_destroy(mirror);
2009 out_vlan = s->out_vlan;
2011 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2012 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2014 /* If the configuration has not changed, do nothing. */
2015 if (hmapx_equals(&srcs, &mirror->srcs)
2016 && hmapx_equals(&dsts, &mirror->dsts)
2017 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2018 && mirror->out == out
2019 && mirror->out_vlan == out_vlan)
2021 hmapx_destroy(&srcs);
2022 hmapx_destroy(&dsts);
2026 hmapx_swap(&srcs, &mirror->srcs);
2027 hmapx_destroy(&srcs);
2029 hmapx_swap(&dsts, &mirror->dsts);
2030 hmapx_destroy(&dsts);
2032 free(mirror->vlans);
2033 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2036 mirror->out_vlan = out_vlan;
2038 /* Update bundles. */
2039 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2040 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2041 if (hmapx_contains(&mirror->srcs, bundle)) {
2042 bundle->src_mirrors |= mirror_bit;
2044 bundle->src_mirrors &= ~mirror_bit;
2047 if (hmapx_contains(&mirror->dsts, bundle)) {
2048 bundle->dst_mirrors |= mirror_bit;
2050 bundle->dst_mirrors &= ~mirror_bit;
2053 if (mirror->out == bundle) {
2054 bundle->mirror_out |= mirror_bit;
2056 bundle->mirror_out &= ~mirror_bit;
2060 ofproto->need_revalidate = true;
2061 mac_learning_flush(ofproto->ml);
2062 mirror_update_dups(ofproto);
2068 mirror_destroy(struct ofmirror *mirror)
2070 struct ofproto_dpif *ofproto;
2071 mirror_mask_t mirror_bit;
2072 struct ofbundle *bundle;
2078 ofproto = mirror->ofproto;
2079 ofproto->need_revalidate = true;
2080 mac_learning_flush(ofproto->ml);
2082 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2083 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2084 bundle->src_mirrors &= ~mirror_bit;
2085 bundle->dst_mirrors &= ~mirror_bit;
2086 bundle->mirror_out &= ~mirror_bit;
2089 hmapx_destroy(&mirror->srcs);
2090 hmapx_destroy(&mirror->dsts);
2091 free(mirror->vlans);
2093 ofproto->mirrors[mirror->idx] = NULL;
2097 mirror_update_dups(ofproto);
2101 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2102 uint64_t *packets, uint64_t *bytes)
2104 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2105 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2108 *packets = *bytes = UINT64_MAX;
2112 *packets = mirror->packet_count;
2113 *bytes = mirror->byte_count;
2119 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2121 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2122 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2123 ofproto->need_revalidate = true;
2124 mac_learning_flush(ofproto->ml);
2130 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2132 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2133 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2134 return bundle && bundle->mirror_out != 0;
2138 forward_bpdu_changed(struct ofproto *ofproto_)
2140 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2141 /* Revalidate cached flows whenever forward_bpdu option changes. */
2142 ofproto->need_revalidate = true;
2146 set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time)
2148 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2149 mac_learning_set_idle_time(ofproto->ml, idle_time);
2154 static struct ofport_dpif *
2155 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
2157 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2158 return ofport ? ofport_dpif_cast(ofport) : NULL;
2161 static struct ofport_dpif *
2162 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
2164 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2168 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2169 struct dpif_port *dpif_port)
2171 ofproto_port->name = dpif_port->name;
2172 ofproto_port->type = dpif_port->type;
2173 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2177 port_run(struct ofport_dpif *ofport)
2179 bool enable = netdev_get_carrier(ofport->up.netdev);
2182 cfm_run(ofport->cfm);
2184 if (cfm_should_send_ccm(ofport->cfm)) {
2185 struct ofpbuf packet;
2187 ofpbuf_init(&packet, 0);
2188 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr);
2189 send_packet(ofport, &packet);
2190 ofpbuf_uninit(&packet);
2193 enable = enable && !cfm_get_fault(ofport->cfm)
2194 && cfm_get_opup(ofport->cfm);
2197 if (ofport->bundle) {
2198 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2201 if (ofport->may_enable != enable) {
2202 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2204 if (ofproto->has_bundle_action) {
2205 ofproto->need_revalidate = true;
2209 ofport->may_enable = enable;
2213 port_wait(struct ofport_dpif *ofport)
2216 cfm_wait(ofport->cfm);
2221 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2222 struct ofproto_port *ofproto_port)
2224 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2225 struct dpif_port dpif_port;
2228 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2230 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2236 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2238 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2242 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2244 *ofp_portp = odp_port_to_ofp_port(odp_port);
2250 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2252 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2255 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2257 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2259 /* The caller is going to close ofport->up.netdev. If this is a
2260 * bonded port, then the bond is using that netdev, so remove it
2261 * from the bond. The client will need to reconfigure everything
2262 * after deleting ports, so then the slave will get re-added. */
2263 bundle_remove(&ofport->up);
2269 struct port_dump_state {
2270 struct dpif_port_dump dump;
2275 port_dump_start(const struct ofproto *ofproto_, void **statep)
2277 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2278 struct port_dump_state *state;
2280 *statep = state = xmalloc(sizeof *state);
2281 dpif_port_dump_start(&state->dump, ofproto->dpif);
2282 state->done = false;
2287 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2288 struct ofproto_port *port)
2290 struct port_dump_state *state = state_;
2291 struct dpif_port dpif_port;
2293 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2294 ofproto_port_from_dpif_port(port, &dpif_port);
2297 int error = dpif_port_dump_done(&state->dump);
2299 return error ? error : EOF;
2304 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2306 struct port_dump_state *state = state_;
2309 dpif_port_dump_done(&state->dump);
2316 port_poll(const struct ofproto *ofproto_, char **devnamep)
2318 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2319 return dpif_port_poll(ofproto->dpif, devnamep);
2323 port_poll_wait(const struct ofproto *ofproto_)
2325 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2326 dpif_port_poll_wait(ofproto->dpif);
2330 port_is_lacp_current(const struct ofport *ofport_)
2332 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2333 return (ofport->bundle && ofport->bundle->lacp
2334 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2338 /* Upcall handling. */
2340 /* Flow miss batching.
2342 * Some dpifs implement operations faster when you hand them off in a batch.
2343 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2344 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2345 * more packets, plus possibly installing the flow in the dpif.
2347 * So far we only batch the operations that affect flow setup time the most.
2348 * It's possible to batch more than that, but the benefit might be minimal. */
2350 struct hmap_node hmap_node;
2352 enum odp_key_fitness key_fitness;
2353 const struct nlattr *key;
2355 ovs_be16 initial_tci;
2356 struct list packets;
2359 struct flow_miss_op {
2360 union dpif_op dpif_op;
2361 struct subfacet *subfacet;
2364 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2365 * OpenFlow controller as necessary according to their individual
2368 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
2369 * ownership is transferred to this function. */
2371 send_packet_in_miss(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
2372 const struct flow *flow, bool clone)
2374 struct ofputil_packet_in pin;
2376 pin.packet = packet;
2377 pin.in_port = flow->in_port;
2378 pin.reason = OFPR_NO_MATCH;
2379 pin.buffer_id = 0; /* not yet known */
2380 pin.send_len = 0; /* not used for flow table misses */
2381 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
2382 clone ? NULL : packet);
2385 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_ACTION to each
2386 * OpenFlow controller as necessary according to their individual
2389 * 'send_len' should be the number of bytes of 'packet' to send to the
2390 * controller, as specified in the action that caused the packet to be sent.
2392 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
2393 * Otherwise, ownership is transferred to this function. */
2395 send_packet_in_action(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
2396 uint64_t userdata, const struct flow *flow, bool clone)
2398 struct ofputil_packet_in pin;
2399 struct user_action_cookie cookie;
2401 memcpy(&cookie, &userdata, sizeof(cookie));
2403 pin.packet = packet;
2404 pin.in_port = flow->in_port;
2405 pin.reason = OFPR_ACTION;
2406 pin.buffer_id = 0; /* not yet known */
2407 pin.send_len = cookie.data;
2408 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
2409 clone ? NULL : packet);
2413 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2414 const struct ofpbuf *packet)
2416 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2422 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2424 cfm_process_heartbeat(ofport->cfm, packet);
2427 } else if (ofport->bundle && ofport->bundle->lacp
2428 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2430 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2433 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2435 stp_process_packet(ofport, packet);
2442 static struct flow_miss *
2443 flow_miss_create(struct hmap *todo, const struct flow *flow,
2444 enum odp_key_fitness key_fitness,
2445 const struct nlattr *key, size_t key_len,
2446 ovs_be16 initial_tci)
2448 uint32_t hash = flow_hash(flow, 0);
2449 struct flow_miss *miss;
2451 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2452 if (flow_equal(&miss->flow, flow)) {
2457 miss = xmalloc(sizeof *miss);
2458 hmap_insert(todo, &miss->hmap_node, hash);
2460 miss->key_fitness = key_fitness;
2462 miss->key_len = key_len;
2463 miss->initial_tci = initial_tci;
2464 list_init(&miss->packets);
2469 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2470 struct flow_miss_op *ops, size_t *n_ops)
2472 const struct flow *flow = &miss->flow;
2473 struct ofpbuf *packet, *next_packet;
2474 struct subfacet *subfacet;
2475 struct facet *facet;
2477 facet = facet_lookup_valid(ofproto, flow);
2479 struct rule_dpif *rule;
2481 rule = rule_dpif_lookup(ofproto, flow, 0);
2483 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
2484 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
2486 if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) {
2487 COVERAGE_INC(ofproto_dpif_no_packet_in);
2488 /* XXX install 'drop' flow entry */
2492 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
2496 LIST_FOR_EACH_SAFE (packet, next_packet, list_node,
2498 list_remove(&packet->list_node);
2499 send_packet_in_miss(ofproto, packet, flow, false);
2505 facet = facet_create(rule, flow);
2508 subfacet = subfacet_create(ofproto, facet,
2509 miss->key_fitness, miss->key, miss->key_len,
2512 LIST_FOR_EACH_SAFE (packet, next_packet, list_node, &miss->packets) {
2513 struct dpif_flow_stats stats;
2515 list_remove(&packet->list_node);
2516 ofproto->n_matches++;
2518 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2520 * Extra-special case for fail-open mode.
2522 * We are in fail-open mode and the packet matched the fail-open
2523 * rule, but we are connected to a controller too. We should send
2524 * the packet up to the controller in the hope that it will try to
2525 * set up a flow and thereby allow us to exit fail-open.
2527 * See the top-level comment in fail-open.c for more information.
2529 send_packet_in_miss(ofproto, packet, flow, true);
2532 if (!facet->may_install || !subfacet->actions) {
2533 subfacet_make_actions(ofproto, subfacet, packet);
2536 /* Credit statistics to subfacet for this packet. We must do this now
2537 * because execute_controller_action() below may destroy 'packet'. */
2538 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2539 subfacet_update_stats(ofproto, subfacet, &stats);
2541 if (!execute_controller_action(ofproto, &facet->flow,
2543 subfacet->actions_len, packet, true)) {
2544 struct flow_miss_op *op = &ops[(*n_ops)++];
2545 struct dpif_execute *execute = &op->dpif_op.execute;
2547 if (flow->vlan_tci != subfacet->initial_tci) {
2548 /* This packet was received on a VLAN splinter port. We added
2549 * a VLAN to the packet to make the packet resemble the flow,
2550 * but the actions were composed assuming that the packet
2551 * contained no VLAN. So, we must remove the VLAN header from
2552 * the packet before trying to execute the actions. */
2553 eth_pop_vlan(packet);
2556 op->subfacet = subfacet;
2557 execute->type = DPIF_OP_EXECUTE;
2558 execute->key = miss->key;
2559 execute->key_len = miss->key_len;
2561 = (facet->may_install
2563 : xmemdup(subfacet->actions, subfacet->actions_len));
2564 execute->actions_len = subfacet->actions_len;
2565 execute->packet = packet;
2569 if (facet->may_install && subfacet->key_fitness != ODP_FIT_TOO_LITTLE) {
2570 struct flow_miss_op *op = &ops[(*n_ops)++];
2571 struct dpif_flow_put *put = &op->dpif_op.flow_put;
2573 op->subfacet = subfacet;
2574 put->type = DPIF_OP_FLOW_PUT;
2575 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2576 put->key = miss->key;
2577 put->key_len = miss->key_len;
2578 put->actions = subfacet->actions;
2579 put->actions_len = subfacet->actions_len;
2584 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
2585 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
2586 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
2587 * what a flow key should contain.
2589 * This function also includes some logic to help make VLAN splinters
2590 * transparent to the rest of the upcall processing logic. In particular, if
2591 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
2592 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
2593 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
2595 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
2596 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
2597 * (This differs from the value returned in flow->vlan_tci only for packets
2598 * received on VLAN splinters.)
2600 static enum odp_key_fitness
2601 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
2602 const struct nlattr *key, size_t key_len,
2603 struct flow *flow, ovs_be16 *initial_tci,
2604 struct ofpbuf *packet)
2606 enum odp_key_fitness fitness;
2610 fitness = odp_flow_key_to_flow(key, key_len, flow);
2611 if (fitness == ODP_FIT_ERROR) {
2614 *initial_tci = flow->vlan_tci;
2616 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
2618 /* Cause the flow to be processed as if it came in on the real device
2619 * with the VLAN device's VLAN ID. */
2620 flow->in_port = realdev;
2621 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
2623 /* Make the packet resemble the flow, so that it gets sent to an
2624 * OpenFlow controller properly, so that it looks correct for
2625 * sFlow, and so that flow_extract() will get the correct vlan_tci
2626 * if it is called on 'packet'.
2628 * The allocated space inside 'packet' probably also contains
2629 * 'key', that is, both 'packet' and 'key' are probably part of a
2630 * struct dpif_upcall (see the large comment on that structure
2631 * definition), so pushing data on 'packet' is in general not a
2632 * good idea since it could overwrite 'key' or free it as a side
2633 * effect. However, it's OK in this special case because we know
2634 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
2635 * will just overwrite the 4-byte "struct nlattr", which is fine
2636 * since we don't need that header anymore. */
2637 eth_push_vlan(packet, flow->vlan_tci);
2640 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
2641 if (fitness == ODP_FIT_PERFECT) {
2642 fitness = ODP_FIT_TOO_MUCH;
2650 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
2653 struct dpif_upcall *upcall;
2654 struct flow_miss *miss, *next_miss;
2655 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
2656 union dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
2665 /* Construct the to-do list.
2667 * This just amounts to extracting the flow from each packet and sticking
2668 * the packets that have the same flow in the same "flow_miss" structure so
2669 * that we can process them together. */
2671 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
2672 enum odp_key_fitness fitness;
2673 struct flow_miss *miss;
2674 ovs_be16 initial_tci;
2677 /* Obtain metadata and check userspace/kernel agreement on flow match,
2678 * then set 'flow''s header pointers. */
2679 fitness = ofproto_dpif_extract_flow_key(ofproto,
2680 upcall->key, upcall->key_len,
2681 &flow, &initial_tci,
2683 if (fitness == ODP_FIT_ERROR) {
2684 ofpbuf_delete(upcall->packet);
2687 flow_extract(upcall->packet, flow.priority, flow.tun_id,
2688 flow.in_port, &flow);
2690 /* Handle 802.1ag, LACP, and STP specially. */
2691 if (process_special(ofproto, &flow, upcall->packet)) {
2692 ofpbuf_delete(upcall->packet);
2693 ofproto->n_matches++;
2697 /* Add other packets to a to-do list. */
2698 miss = flow_miss_create(&todo, &flow, fitness,
2699 upcall->key, upcall->key_len, initial_tci);
2700 list_push_back(&miss->packets, &upcall->packet->list_node);
2703 /* Process each element in the to-do list, constructing the set of
2704 * operations to batch. */
2706 HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) {
2707 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
2708 ofpbuf_list_delete(&miss->packets);
2709 hmap_remove(&todo, &miss->hmap_node);
2712 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
2713 hmap_destroy(&todo);
2715 /* Execute batch. */
2716 for (i = 0; i < n_ops; i++) {
2717 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
2719 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
2721 /* Free memory and update facets. */
2722 for (i = 0; i < n_ops; i++) {
2723 struct flow_miss_op *op = &flow_miss_ops[i];
2724 struct dpif_execute *execute;
2725 struct dpif_flow_put *put;
2727 switch (op->dpif_op.type) {
2728 case DPIF_OP_EXECUTE:
2729 execute = &op->dpif_op.execute;
2730 if (op->subfacet->actions != execute->actions) {
2731 free((struct nlattr *) execute->actions);
2733 ofpbuf_delete((struct ofpbuf *) execute->packet);
2736 case DPIF_OP_FLOW_PUT:
2737 put = &op->dpif_op.flow_put;
2739 op->subfacet->installed = true;
2747 handle_userspace_upcall(struct ofproto_dpif *ofproto,
2748 struct dpif_upcall *upcall)
2750 struct user_action_cookie cookie;
2751 enum odp_key_fitness fitness;
2752 ovs_be16 initial_tci;
2755 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
2757 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
2758 upcall->key_len, &flow,
2759 &initial_tci, upcall->packet);
2760 if (fitness == ODP_FIT_ERROR) {
2761 ofpbuf_delete(upcall->packet);
2765 if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
2766 if (ofproto->sflow) {
2767 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
2770 ofpbuf_delete(upcall->packet);
2771 } else if (cookie.type == USER_ACTION_COOKIE_CONTROLLER) {
2772 COVERAGE_INC(ofproto_dpif_ctlr_action);
2773 send_packet_in_action(ofproto, upcall->packet, upcall->userdata,
2776 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
2777 ofpbuf_delete(upcall->packet);
2782 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
2784 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
2788 assert (max_batch <= FLOW_MISS_MAX_BATCH);
2791 for (i = 0; i < max_batch; i++) {
2792 struct dpif_upcall *upcall = &misses[n_misses];
2795 error = dpif_recv(ofproto->dpif, upcall);
2800 switch (upcall->type) {
2801 case DPIF_UC_ACTION:
2802 handle_userspace_upcall(ofproto, upcall);
2806 /* Handle it later. */
2810 case DPIF_N_UC_TYPES:
2812 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32,
2818 handle_miss_upcalls(ofproto, misses, n_misses);
2823 /* Flow expiration. */
2825 static int subfacet_max_idle(const struct ofproto_dpif *);
2826 static void update_stats(struct ofproto_dpif *);
2827 static void rule_expire(struct rule_dpif *);
2828 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
2830 /* This function is called periodically by run(). Its job is to collect
2831 * updates for the flows that have been installed into the datapath, most
2832 * importantly when they last were used, and then use that information to
2833 * expire flows that have not been used recently.
2835 * Returns the number of milliseconds after which it should be called again. */
2837 expire(struct ofproto_dpif *ofproto)
2839 struct rule_dpif *rule, *next_rule;
2840 struct classifier *table;
2843 /* Update stats for each flow in the datapath. */
2844 update_stats(ofproto);
2846 /* Expire subfacets that have been idle too long. */
2847 dp_max_idle = subfacet_max_idle(ofproto);
2848 expire_subfacets(ofproto, dp_max_idle);
2850 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
2851 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
2852 struct cls_cursor cursor;
2854 cls_cursor_init(&cursor, table, NULL);
2855 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
2860 /* All outstanding data in existing flows has been accounted, so it's a
2861 * good time to do bond rebalancing. */
2862 if (ofproto->has_bonded_bundles) {
2863 struct ofbundle *bundle;
2865 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2867 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
2872 return MIN(dp_max_idle, 1000);
2875 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
2877 * This function also pushes statistics updates to rules which each facet
2878 * resubmits into. Generally these statistics will be accurate. However, if a
2879 * facet changes the rule it resubmits into at some time in between
2880 * update_stats() runs, it is possible that statistics accrued to the
2881 * old rule will be incorrectly attributed to the new rule. This could be
2882 * avoided by calling update_stats() whenever rules are created or
2883 * deleted. However, the performance impact of making so many calls to the
2884 * datapath do not justify the benefit of having perfectly accurate statistics.
2887 update_stats(struct ofproto_dpif *p)
2889 const struct dpif_flow_stats *stats;
2890 struct dpif_flow_dump dump;
2891 const struct nlattr *key;
2894 dpif_flow_dump_start(&dump, p->dpif);
2895 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
2896 struct subfacet *subfacet;
2898 subfacet = subfacet_find(p, key, key_len);
2899 if (subfacet && subfacet->installed) {
2900 struct facet *facet = subfacet->facet;
2902 if (stats->n_packets >= subfacet->dp_packet_count) {
2903 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
2904 facet->packet_count += extra;
2906 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
2909 if (stats->n_bytes >= subfacet->dp_byte_count) {
2910 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
2912 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
2915 subfacet->dp_packet_count = stats->n_packets;
2916 subfacet->dp_byte_count = stats->n_bytes;
2918 subfacet_update_time(p, subfacet, stats->used);
2919 facet_account(p, facet);
2920 facet_push_stats(facet);
2922 if (!VLOG_DROP_WARN(&rl)) {
2926 odp_flow_key_format(key, key_len, &s);
2927 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
2931 COVERAGE_INC(facet_unexpected);
2932 /* There's a flow in the datapath that we know nothing about, or a
2933 * flow that shouldn't be installed but was anyway. Delete it. */
2934 dpif_flow_del(p->dpif, key, key_len, NULL);
2937 dpif_flow_dump_done(&dump);
2940 /* Calculates and returns the number of milliseconds of idle time after which
2941 * subfacets should expire from the datapath. When a subfacet expires, we fold
2942 * its statistics into its facet, and when a facet's last subfacet expires, we
2943 * fold its statistic into its rule. */
2945 subfacet_max_idle(const struct ofproto_dpif *ofproto)
2948 * Idle time histogram.
2950 * Most of the time a switch has a relatively small number of subfacets.
2951 * When this is the case we might as well keep statistics for all of them
2952 * in userspace and to cache them in the kernel datapath for performance as
2955 * As the number of subfacets increases, the memory required to maintain
2956 * statistics about them in userspace and in the kernel becomes
2957 * significant. However, with a large number of subfacets it is likely
2958 * that only a few of them are "heavy hitters" that consume a large amount
2959 * of bandwidth. At this point, only heavy hitters are worth caching in
2960 * the kernel and maintaining in userspaces; other subfacets we can
2963 * The technique used to compute the idle time is to build a histogram with
2964 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
2965 * that is installed in the kernel gets dropped in the appropriate bucket.
2966 * After the histogram has been built, we compute the cutoff so that only
2967 * the most-recently-used 1% of subfacets (but at least
2968 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
2969 * the most-recently-used bucket of subfacets is kept, so actually an
2970 * arbitrary number of subfacets can be kept in any given expiration run
2971 * (though the next run will delete most of those unless they receive
2974 * This requires a second pass through the subfacets, in addition to the
2975 * pass made by update_stats(), because the former function never looks at
2976 * uninstallable subfacets.
2978 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
2979 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
2980 int buckets[N_BUCKETS] = { 0 };
2981 int total, subtotal, bucket;
2982 struct subfacet *subfacet;
2986 total = hmap_count(&ofproto->subfacets);
2987 if (total <= ofproto->up.flow_eviction_threshold) {
2988 return N_BUCKETS * BUCKET_WIDTH;
2991 /* Build histogram. */
2993 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
2994 long long int idle = now - subfacet->used;
2995 int bucket = (idle <= 0 ? 0
2996 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
2997 : (unsigned int) idle / BUCKET_WIDTH);
3001 /* Find the first bucket whose flows should be expired. */
3002 subtotal = bucket = 0;
3004 subtotal += buckets[bucket++];
3005 } while (bucket < N_BUCKETS &&
3006 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3008 if (VLOG_IS_DBG_ENABLED()) {
3012 ds_put_cstr(&s, "keep");
3013 for (i = 0; i < N_BUCKETS; i++) {
3015 ds_put_cstr(&s, ", drop");
3018 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3021 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3025 return bucket * BUCKET_WIDTH;
3029 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3031 long long int cutoff = time_msec() - dp_max_idle;
3032 struct subfacet *subfacet, *next_subfacet;
3034 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3035 &ofproto->subfacets) {
3036 if (subfacet->used < cutoff) {
3037 subfacet_destroy(ofproto, subfacet);
3042 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3043 * then delete it entirely. */
3045 rule_expire(struct rule_dpif *rule)
3047 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3048 struct facet *facet, *next_facet;
3052 /* Has 'rule' expired? */
3054 if (rule->up.hard_timeout
3055 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3056 reason = OFPRR_HARD_TIMEOUT;
3057 } else if (rule->up.idle_timeout && list_is_empty(&rule->facets)
3058 && now > rule->used + rule->up.idle_timeout * 1000) {
3059 reason = OFPRR_IDLE_TIMEOUT;
3064 COVERAGE_INC(ofproto_dpif_expired);
3066 /* Update stats. (This is a no-op if the rule expired due to an idle
3067 * timeout, because that only happens when the rule has no facets left.) */
3068 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3069 facet_remove(ofproto, facet);
3072 /* Get rid of the rule. */
3073 ofproto_rule_expire(&rule->up, reason);
3078 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3080 * The caller must already have determined that no facet with an identical
3081 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3082 * the ofproto's classifier table.
3084 * The facet will initially have no subfacets. The caller should create (at
3085 * least) one subfacet with subfacet_create(). */
3086 static struct facet *
3087 facet_create(struct rule_dpif *rule, const struct flow *flow)
3089 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3090 struct facet *facet;
3092 facet = xzalloc(sizeof *facet);
3093 facet->used = time_msec();
3094 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
3095 list_push_back(&rule->facets, &facet->list_node);
3097 facet->flow = *flow;
3098 list_init(&facet->subfacets);
3099 netflow_flow_init(&facet->nf_flow);
3100 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3106 facet_free(struct facet *facet)
3111 /* If the 'actions_len' bytes of actions in 'odp_actions' are just a single
3112 * OVS_ACTION_ATTR_USERSPACE action, executes it internally and returns true.
3113 * Otherwise, returns false without doing anything.
3115 * If 'clone' is true, the caller always retains ownership of 'packet'.
3116 * Otherwise, ownership is transferred to this function if it returns true. */
3118 execute_controller_action(struct ofproto_dpif *ofproto,
3119 const struct flow *flow,
3120 const struct nlattr *odp_actions, size_t actions_len,
3121 struct ofpbuf *packet, bool clone)
3124 && odp_actions->nla_type == OVS_ACTION_ATTR_USERSPACE
3125 && NLA_ALIGN(odp_actions->nla_len) == actions_len) {
3126 /* As an optimization, avoid a round-trip from userspace to kernel to
3127 * userspace. This also avoids possibly filling up kernel packet
3128 * buffers along the way.
3130 * This optimization will not accidentally catch sFlow
3131 * OVS_ACTION_ATTR_USERSPACE actions, since those are encapsulated
3132 * inside OVS_ACTION_ATTR_SAMPLE. */
3133 const struct nlattr *nla;
3135 nla = nl_attr_find_nested(odp_actions, OVS_USERSPACE_ATTR_USERDATA);
3136 send_packet_in_action(ofproto, packet, nl_attr_get_u64(nla), flow,
3144 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3145 * 'packet', which arrived on 'in_port'.
3147 * Takes ownership of 'packet'. */
3149 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3150 const struct nlattr *odp_actions, size_t actions_len,
3151 struct ofpbuf *packet)
3153 struct odputil_keybuf keybuf;
3157 if (execute_controller_action(ofproto, flow, odp_actions, actions_len,
3162 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3163 odp_flow_key_from_flow(&key, flow);
3165 error = dpif_execute(ofproto->dpif, key.data, key.size,
3166 odp_actions, actions_len, packet);
3168 ofpbuf_delete(packet);
3172 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3174 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3175 * rule's statistics, via subfacet_uninstall().
3177 * - Removes 'facet' from its rule and from ofproto->facets.
3180 facet_remove(struct ofproto_dpif *ofproto, struct facet *facet)
3182 struct subfacet *subfacet, *next_subfacet;
3184 assert(!list_is_empty(&facet->subfacets));
3186 /* First uninstall all of the subfacets to get final statistics. */
3187 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3188 subfacet_uninstall(ofproto, subfacet);
3191 /* Flush the final stats to the rule.
3193 * This might require us to have at least one subfacet around so that we
3194 * can use its actions for accounting in facet_account(), which is why we
3195 * have uninstalled but not yet destroyed the subfacets. */
3196 facet_flush_stats(ofproto, facet);
3198 /* Now we're really all done so destroy everything. */
3199 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3200 &facet->subfacets) {
3201 subfacet_destroy__(ofproto, subfacet);
3203 hmap_remove(&ofproto->facets, &facet->hmap_node);
3204 list_remove(&facet->list_node);
3209 facet_account(struct ofproto_dpif *ofproto, struct facet *facet)
3212 struct subfacet *subfacet;
3213 const struct nlattr *a;
3217 if (facet->byte_count <= facet->accounted_bytes) {
3220 n_bytes = facet->byte_count - facet->accounted_bytes;
3221 facet->accounted_bytes = facet->byte_count;
3223 /* Feed information from the active flows back into the learning table to
3224 * ensure that table is always in sync with what is actually flowing
3225 * through the datapath. */
3226 if (facet->has_learn || facet->has_normal) {
3227 struct action_xlate_ctx ctx;
3229 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3230 facet->flow.vlan_tci, NULL);
3231 ctx.may_learn = true;
3232 ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions,
3233 facet->rule->up.n_actions));
3236 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3240 /* This loop feeds byte counters to bond_account() for rebalancing to use
3241 * as a basis. We also need to track the actual VLAN on which the packet
3242 * is going to be sent to ensure that it matches the one passed to
3243 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3246 * We use the actions from an arbitrary subfacet because they should all
3247 * be equally valid for our purpose. */
3248 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3249 struct subfacet, list_node);
3250 vlan_tci = facet->flow.vlan_tci;
3251 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3252 subfacet->actions, subfacet->actions_len) {
3253 const struct ovs_action_push_vlan *vlan;
3254 struct ofport_dpif *port;
3256 switch (nl_attr_type(a)) {
3257 case OVS_ACTION_ATTR_OUTPUT:
3258 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3259 if (port && port->bundle && port->bundle->bond) {
3260 bond_account(port->bundle->bond, &facet->flow,
3261 vlan_tci_to_vid(vlan_tci), n_bytes);
3265 case OVS_ACTION_ATTR_POP_VLAN:
3266 vlan_tci = htons(0);
3269 case OVS_ACTION_ATTR_PUSH_VLAN:
3270 vlan = nl_attr_get(a);
3271 vlan_tci = vlan->vlan_tci;
3277 /* Returns true if the only action for 'facet' is to send to the controller.
3278 * (We don't report NetFlow expiration messages for such facets because they
3279 * are just part of the control logic for the network, not real traffic). */
3281 facet_is_controller_flow(struct facet *facet)
3284 && facet->rule->up.n_actions == 1
3285 && action_outputs_to_port(&facet->rule->up.actions[0],
3286 htons(OFPP_CONTROLLER)));
3289 /* Folds all of 'facet''s statistics into its rule. Also updates the
3290 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3291 * 'facet''s statistics in the datapath should have been zeroed and folded into
3292 * its packet and byte counts before this function is called. */
3294 facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet)
3296 struct subfacet *subfacet;
3298 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3299 assert(!subfacet->dp_byte_count);
3300 assert(!subfacet->dp_packet_count);
3303 facet_push_stats(facet);
3304 facet_account(ofproto, facet);
3306 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3307 struct ofexpired expired;
3308 expired.flow = facet->flow;
3309 expired.packet_count = facet->packet_count;
3310 expired.byte_count = facet->byte_count;
3311 expired.used = facet->used;
3312 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3315 facet->rule->packet_count += facet->packet_count;
3316 facet->rule->byte_count += facet->byte_count;
3318 /* Reset counters to prevent double counting if 'facet' ever gets
3320 facet_reset_counters(facet);
3322 netflow_flow_clear(&facet->nf_flow);
3325 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3326 * Returns it if found, otherwise a null pointer.
3328 * The returned facet might need revalidation; use facet_lookup_valid()
3329 * instead if that is important. */
3330 static struct facet *
3331 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
3333 struct facet *facet;
3335 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
3337 if (flow_equal(flow, &facet->flow)) {
3345 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3346 * Returns it if found, otherwise a null pointer.
3348 * The returned facet is guaranteed to be valid. */
3349 static struct facet *
3350 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
3352 struct facet *facet = facet_find(ofproto, flow);
3354 /* The facet we found might not be valid, since we could be in need of
3355 * revalidation. If it is not valid, don't return it. */
3357 && (ofproto->need_revalidate
3358 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))
3359 && !facet_revalidate(ofproto, facet)) {
3360 COVERAGE_INC(facet_invalidated);
3367 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
3369 * - If the rule found is different from 'facet''s current rule, moves
3370 * 'facet' to the new rule and recompiles its actions.
3372 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3373 * where it is and recompiles its actions anyway.
3375 * - If there is none, destroys 'facet'.
3377 * Returns true if 'facet' still exists, false if it has been destroyed. */
3379 facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet)
3382 struct nlattr *odp_actions;
3385 struct actions *new_actions;
3387 struct action_xlate_ctx ctx;
3388 struct rule_dpif *new_rule;
3389 struct subfacet *subfacet;
3390 bool actions_changed;
3393 COVERAGE_INC(facet_revalidate);
3395 /* Determine the new rule. */
3396 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3398 /* No new rule, so delete the facet. */
3399 facet_remove(ofproto, facet);
3403 /* Calculate new datapath actions.
3405 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3406 * emit a NetFlow expiration and, if so, we need to have the old state
3407 * around to properly compose it. */
3409 /* If the datapath actions changed or the installability changed,
3410 * then we need to talk to the datapath. */
3413 memset(&ctx, 0, sizeof ctx);
3414 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3415 struct ofpbuf *odp_actions;
3416 bool should_install;
3418 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3419 subfacet->initial_tci, NULL);
3420 odp_actions = xlate_actions(&ctx, new_rule->up.actions,
3421 new_rule->up.n_actions);
3422 actions_changed = (subfacet->actions_len != odp_actions->size
3423 || memcmp(subfacet->actions, odp_actions->data,
3424 subfacet->actions_len));
3426 should_install = (ctx.may_set_up_flow
3427 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3428 if (actions_changed || should_install != subfacet->installed) {
3429 if (should_install) {
3430 struct dpif_flow_stats stats;
3432 subfacet_install(ofproto, subfacet,
3433 odp_actions->data, odp_actions->size, &stats);
3434 subfacet_update_stats(ofproto, subfacet, &stats);
3436 subfacet_uninstall(ofproto, subfacet);
3440 new_actions = xcalloc(list_size(&facet->subfacets),
3441 sizeof *new_actions);
3443 new_actions[i].odp_actions = xmemdup(odp_actions->data,
3445 new_actions[i].actions_len = odp_actions->size;
3448 ofpbuf_delete(odp_actions);
3452 facet_flush_stats(ofproto, facet);
3455 /* Update 'facet' now that we've taken care of all the old state. */
3456 facet->tags = ctx.tags;
3457 facet->nf_flow.output_iface = ctx.nf_output_iface;
3458 facet->may_install = ctx.may_set_up_flow;
3459 facet->has_learn = ctx.has_learn;
3460 facet->has_normal = ctx.has_normal;
3461 facet->mirrors = ctx.mirrors;
3464 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3465 if (new_actions[i].odp_actions) {
3466 free(subfacet->actions);
3467 subfacet->actions = new_actions[i].odp_actions;
3468 subfacet->actions_len = new_actions[i].actions_len;
3474 if (facet->rule != new_rule) {
3475 COVERAGE_INC(facet_changed_rule);
3476 list_remove(&facet->list_node);
3477 list_push_back(&new_rule->facets, &facet->list_node);
3478 facet->rule = new_rule;
3479 facet->used = new_rule->up.created;
3480 facet->prev_used = facet->used;
3486 /* Updates 'facet''s used time. Caller is responsible for calling
3487 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3489 facet_update_time(struct ofproto_dpif *ofproto, struct facet *facet,
3492 if (used > facet->used) {
3494 if (used > facet->rule->used) {
3495 facet->rule->used = used;
3497 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3502 facet_reset_counters(struct facet *facet)
3504 facet->packet_count = 0;
3505 facet->byte_count = 0;
3506 facet->prev_packet_count = 0;
3507 facet->prev_byte_count = 0;
3508 facet->accounted_bytes = 0;
3512 facet_push_stats(struct facet *facet)
3514 uint64_t new_packets, new_bytes;
3516 assert(facet->packet_count >= facet->prev_packet_count);
3517 assert(facet->byte_count >= facet->prev_byte_count);
3518 assert(facet->used >= facet->prev_used);
3520 new_packets = facet->packet_count - facet->prev_packet_count;
3521 new_bytes = facet->byte_count - facet->prev_byte_count;
3523 if (new_packets || new_bytes || facet->used > facet->prev_used) {
3524 facet->prev_packet_count = facet->packet_count;
3525 facet->prev_byte_count = facet->byte_count;
3526 facet->prev_used = facet->used;
3528 flow_push_stats(facet->rule, &facet->flow,
3529 new_packets, new_bytes, facet->used);
3531 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
3532 facet->mirrors, new_packets, new_bytes);
3536 struct ofproto_push {
3537 struct action_xlate_ctx ctx;
3544 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
3546 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3549 rule->packet_count += push->packets;
3550 rule->byte_count += push->bytes;
3551 rule->used = MAX(push->used, rule->used);
3555 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3556 * 'rule''s actions and mirrors. */
3558 flow_push_stats(const struct rule_dpif *rule,
3559 const struct flow *flow, uint64_t packets, uint64_t bytes,
3562 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3563 struct ofproto_push push;
3565 push.packets = packets;
3569 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, NULL);
3570 push.ctx.resubmit_hook = push_resubmit;
3571 ofpbuf_delete(xlate_actions(&push.ctx,
3572 rule->up.actions, rule->up.n_actions));
3577 static struct subfacet *
3578 subfacet_find__(struct ofproto_dpif *ofproto,
3579 const struct nlattr *key, size_t key_len, uint32_t key_hash,
3580 const struct flow *flow)
3582 struct subfacet *subfacet;
3584 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
3585 &ofproto->subfacets) {
3587 ? (subfacet->key_len == key_len
3588 && !memcmp(key, subfacet->key, key_len))
3589 : flow_equal(flow, &subfacet->facet->flow)) {
3597 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
3598 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
3599 * there is one, otherwise creates and returns a new subfacet.
3601 * If the returned subfacet is new, then subfacet->actions will be NULL, in
3602 * which case the caller must populate the actions with
3603 * subfacet_make_actions(). */
3604 static struct subfacet *
3605 subfacet_create(struct ofproto_dpif *ofproto, struct facet *facet,
3606 enum odp_key_fitness key_fitness,
3607 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
3609 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3610 struct subfacet *subfacet;
3612 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
3614 if (subfacet->facet == facet) {
3618 /* This shouldn't happen. */
3619 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
3620 subfacet_destroy(ofproto, subfacet);
3623 subfacet = xzalloc(sizeof *subfacet);
3624 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
3625 list_push_back(&facet->subfacets, &subfacet->list_node);
3626 subfacet->facet = facet;
3627 subfacet->used = time_msec();
3628 subfacet->key_fitness = key_fitness;
3629 if (key_fitness != ODP_FIT_PERFECT) {
3630 subfacet->key = xmemdup(key, key_len);
3631 subfacet->key_len = key_len;
3633 subfacet->installed = false;
3634 subfacet->initial_tci = initial_tci;
3639 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
3640 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
3641 static struct subfacet *
3642 subfacet_find(struct ofproto_dpif *ofproto,
3643 const struct nlattr *key, size_t key_len)
3645 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3646 enum odp_key_fitness fitness;
3649 fitness = odp_flow_key_to_flow(key, key_len, &flow);
3650 if (fitness == ODP_FIT_ERROR) {
3654 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
3657 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
3658 * its facet within 'ofproto', and frees it. */
3660 subfacet_destroy__(struct ofproto_dpif *ofproto, struct subfacet *subfacet)
3662 subfacet_uninstall(ofproto, subfacet);
3663 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
3664 list_remove(&subfacet->list_node);
3665 free(subfacet->key);
3666 free(subfacet->actions);
3670 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
3671 * last remaining subfacet in its facet destroys the facet too. */
3673 subfacet_destroy(struct ofproto_dpif *ofproto, struct subfacet *subfacet)
3675 struct facet *facet = subfacet->facet;
3677 if (list_is_singleton(&facet->subfacets)) {
3678 /* facet_remove() needs at least one subfacet (it will remove it). */
3679 facet_remove(ofproto, facet);
3681 subfacet_destroy__(ofproto, subfacet);
3685 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
3686 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
3687 * for use as temporary storage. */
3689 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
3692 if (!subfacet->key) {
3693 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
3694 odp_flow_key_from_flow(key, &subfacet->facet->flow);
3696 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
3700 /* Composes the datapath actions for 'subfacet' based on its rule's actions. */
3702 subfacet_make_actions(struct ofproto_dpif *p, struct subfacet *subfacet,
3703 const struct ofpbuf *packet)
3705 struct facet *facet = subfacet->facet;
3706 const struct rule_dpif *rule = facet->rule;
3707 struct ofpbuf *odp_actions;
3708 struct action_xlate_ctx ctx;
3710 action_xlate_ctx_init(&ctx, p, &facet->flow, subfacet->initial_tci,
3712 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3713 facet->tags = ctx.tags;
3714 facet->may_install = ctx.may_set_up_flow;
3715 facet->has_learn = ctx.has_learn;
3716 facet->has_normal = ctx.has_normal;
3717 facet->nf_flow.output_iface = ctx.nf_output_iface;
3718 facet->mirrors = ctx.mirrors;
3720 if (subfacet->actions_len != odp_actions->size
3721 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
3722 free(subfacet->actions);
3723 subfacet->actions_len = odp_actions->size;
3724 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
3727 ofpbuf_delete(odp_actions);
3730 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
3731 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
3732 * in the datapath will be zeroed and 'stats' will be updated with traffic new
3733 * since 'subfacet' was last updated.
3735 * Returns 0 if successful, otherwise a positive errno value. */
3737 subfacet_install(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3738 const struct nlattr *actions, size_t actions_len,
3739 struct dpif_flow_stats *stats)
3741 struct odputil_keybuf keybuf;
3742 enum dpif_flow_put_flags flags;
3746 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3748 flags |= DPIF_FP_ZERO_STATS;
3751 subfacet_get_key(subfacet, &keybuf, &key);
3752 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
3753 actions, actions_len, stats);
3756 subfacet_reset_dp_stats(subfacet, stats);
3762 /* If 'subfacet' is installed in the datapath, uninstalls it. */
3764 subfacet_uninstall(struct ofproto_dpif *p, struct subfacet *subfacet)
3766 if (subfacet->installed) {
3767 struct odputil_keybuf keybuf;
3768 struct dpif_flow_stats stats;
3772 subfacet_get_key(subfacet, &keybuf, &key);
3773 error = dpif_flow_del(p->dpif, key.data, key.size, &stats);
3774 subfacet_reset_dp_stats(subfacet, &stats);
3776 subfacet_update_stats(p, subfacet, &stats);
3778 subfacet->installed = false;
3780 assert(subfacet->dp_packet_count == 0);
3781 assert(subfacet->dp_byte_count == 0);
3785 /* Resets 'subfacet''s datapath statistics counters. This should be called
3786 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
3787 * non-null, it should contain the statistics returned by dpif when 'subfacet'
3788 * was reset in the datapath. 'stats' will be modified to include only
3789 * statistics new since 'subfacet' was last updated. */
3791 subfacet_reset_dp_stats(struct subfacet *subfacet,
3792 struct dpif_flow_stats *stats)
3795 && subfacet->dp_packet_count <= stats->n_packets
3796 && subfacet->dp_byte_count <= stats->n_bytes) {
3797 stats->n_packets -= subfacet->dp_packet_count;
3798 stats->n_bytes -= subfacet->dp_byte_count;
3801 subfacet->dp_packet_count = 0;
3802 subfacet->dp_byte_count = 0;
3805 /* Updates 'subfacet''s used time. The caller is responsible for calling
3806 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
3808 subfacet_update_time(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3811 if (used > subfacet->used) {
3812 subfacet->used = used;
3813 facet_update_time(ofproto, subfacet->facet, used);
3817 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
3819 * Because of the meaning of a subfacet's counters, it only makes sense to do
3820 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
3821 * represents a packet that was sent by hand or if it represents statistics
3822 * that have been cleared out of the datapath. */
3824 subfacet_update_stats(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3825 const struct dpif_flow_stats *stats)
3827 if (stats->n_packets || stats->used > subfacet->used) {
3828 struct facet *facet = subfacet->facet;
3830 subfacet_update_time(ofproto, subfacet, stats->used);
3831 facet->packet_count += stats->n_packets;
3832 facet->byte_count += stats->n_bytes;
3833 facet_push_stats(facet);
3834 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3840 static struct rule_dpif *
3841 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
3844 struct cls_rule *cls_rule;
3845 struct classifier *cls;
3847 if (table_id >= N_TABLES) {
3851 cls = &ofproto->up.tables[table_id];
3852 if (flow->nw_frag & FLOW_NW_FRAG_ANY
3853 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3854 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
3855 * are unavailable. */
3856 struct flow ofpc_normal_flow = *flow;
3857 ofpc_normal_flow.tp_src = htons(0);
3858 ofpc_normal_flow.tp_dst = htons(0);
3859 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
3861 cls_rule = classifier_lookup(cls, flow);
3863 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
3867 complete_operation(struct rule_dpif *rule)
3869 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3871 rule_invalidate(rule);
3873 struct dpif_completion *c = xmalloc(sizeof *c);
3874 c->op = rule->up.pending;
3875 list_push_back(&ofproto->completions, &c->list_node);
3877 ofoperation_complete(rule->up.pending, 0);
3881 static struct rule *
3884 struct rule_dpif *rule = xmalloc(sizeof *rule);
3889 rule_dealloc(struct rule *rule_)
3891 struct rule_dpif *rule = rule_dpif_cast(rule_);
3896 rule_construct(struct rule *rule_)
3898 struct rule_dpif *rule = rule_dpif_cast(rule_);
3899 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3900 struct rule_dpif *victim;
3904 error = validate_actions(rule->up.actions, rule->up.n_actions,
3905 &rule->up.cr.flow, ofproto->max_ports);
3910 rule->used = rule->up.created;
3911 rule->packet_count = 0;
3912 rule->byte_count = 0;
3914 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
3915 if (victim && !list_is_empty(&victim->facets)) {
3916 struct facet *facet;
3918 rule->facets = victim->facets;
3919 list_moved(&rule->facets);
3920 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3921 /* XXX: We're only clearing our local counters here. It's possible
3922 * that quite a few packets are unaccounted for in the datapath
3923 * statistics. These will be accounted to the new rule instead of
3924 * cleared as required. This could be fixed by clearing out the
3925 * datapath statistics for this facet, but currently it doesn't
3927 facet_reset_counters(facet);
3931 /* Must avoid list_moved() in this case. */
3932 list_init(&rule->facets);
3935 table_id = rule->up.table_id;
3936 rule->tag = (victim ? victim->tag
3938 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
3939 ofproto->tables[table_id].basis));
3941 complete_operation(rule);
3946 rule_destruct(struct rule *rule_)
3948 struct rule_dpif *rule = rule_dpif_cast(rule_);
3949 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3950 struct facet *facet, *next_facet;
3952 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3953 facet_revalidate(ofproto, facet);
3956 complete_operation(rule);
3960 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
3962 struct rule_dpif *rule = rule_dpif_cast(rule_);
3963 struct facet *facet;
3965 /* Start from historical data for 'rule' itself that are no longer tracked
3966 * in facets. This counts, for example, facets that have expired. */
3967 *packets = rule->packet_count;
3968 *bytes = rule->byte_count;
3970 /* Add any statistics that are tracked by facets. This includes
3971 * statistical data recently updated by ofproto_update_stats() as well as
3972 * stats for packets that were executed "by hand" via dpif_execute(). */
3973 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3974 *packets += facet->packet_count;
3975 *bytes += facet->byte_count;
3980 rule_execute(struct rule *rule_, const struct flow *flow,
3981 struct ofpbuf *packet)
3983 struct rule_dpif *rule = rule_dpif_cast(rule_);
3984 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3985 struct action_xlate_ctx ctx;
3986 struct ofpbuf *odp_actions;
3989 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, packet);
3990 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3991 size = packet->size;
3992 if (execute_odp_actions(ofproto, flow, odp_actions->data,
3993 odp_actions->size, packet)) {
3994 rule->used = time_msec();
3995 rule->packet_count++;
3996 rule->byte_count += size;
3997 flow_push_stats(rule, flow, 1, size, rule->used);
3999 ofpbuf_delete(odp_actions);
4005 rule_modify_actions(struct rule *rule_)
4007 struct rule_dpif *rule = rule_dpif_cast(rule_);
4008 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4011 error = validate_actions(rule->up.actions, rule->up.n_actions,
4012 &rule->up.cr.flow, ofproto->max_ports);
4014 ofoperation_complete(rule->up.pending, error);
4018 complete_operation(rule);
4021 /* Sends 'packet' out 'ofport'.
4022 * May modify 'packet'.
4023 * Returns 0 if successful, otherwise a positive errno value. */
4025 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4027 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4028 struct ofpbuf key, odp_actions;
4029 struct odputil_keybuf keybuf;
4034 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
4035 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4037 if (odp_port != ofport->odp_port) {
4038 eth_pop_vlan(packet);
4039 flow.vlan_tci = htons(0);
4042 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4043 odp_flow_key_from_flow(&key, &flow);
4045 ofpbuf_init(&odp_actions, 32);
4046 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4048 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4049 error = dpif_execute(ofproto->dpif,
4051 odp_actions.data, odp_actions.size,
4053 ofpbuf_uninit(&odp_actions);
4056 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4057 ofproto->up.name, odp_port, strerror(error));
4062 /* OpenFlow to datapath action translation. */
4064 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
4065 struct action_xlate_ctx *ctx);
4066 static void xlate_normal(struct action_xlate_ctx *);
4069 put_userspace_action(const struct ofproto_dpif *ofproto,
4070 struct ofpbuf *odp_actions,
4071 const struct flow *flow,
4072 const struct user_action_cookie *cookie)
4076 pid = dpif_port_get_pid(ofproto->dpif,
4077 ofp_port_to_odp_port(flow->in_port));
4079 return odp_put_userspace_action(pid, cookie, odp_actions);
4082 /* Compose SAMPLE action for sFlow. */
4084 compose_sflow_action(const struct ofproto_dpif *ofproto,
4085 struct ofpbuf *odp_actions,
4086 const struct flow *flow,
4089 uint32_t port_ifindex;
4090 uint32_t probability;
4091 struct user_action_cookie cookie;
4092 size_t sample_offset, actions_offset;
4093 int cookie_offset, n_output;
4095 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4099 if (odp_port == OVSP_NONE) {
4103 port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4107 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4109 /* Number of packets out of UINT_MAX to sample. */
4110 probability = dpif_sflow_get_probability(ofproto->sflow);
4111 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4113 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4115 cookie.type = USER_ACTION_COOKIE_SFLOW;
4116 cookie.data = port_ifindex;
4117 cookie.n_output = n_output;
4118 cookie.vlan_tci = 0;
4119 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4121 nl_msg_end_nested(odp_actions, actions_offset);
4122 nl_msg_end_nested(odp_actions, sample_offset);
4123 return cookie_offset;
4126 /* SAMPLE action must be first action in any given list of actions.
4127 * At this point we do not have all information required to build it. So try to
4128 * build sample action as complete as possible. */
4130 add_sflow_action(struct action_xlate_ctx *ctx)
4132 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4134 &ctx->flow, OVSP_NONE);
4135 ctx->sflow_odp_port = 0;
4136 ctx->sflow_n_outputs = 0;
4139 /* Fix SAMPLE action according to data collected while composing ODP actions.
4140 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4141 * USERSPACE action's user-cookie which is required for sflow. */
4143 fix_sflow_action(struct action_xlate_ctx *ctx)
4145 const struct flow *base = &ctx->base_flow;
4146 struct user_action_cookie *cookie;
4148 if (!ctx->user_cookie_offset) {
4152 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4154 assert(cookie != NULL);
4155 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4157 if (ctx->sflow_n_outputs) {
4158 cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow,
4159 ctx->sflow_odp_port);
4161 if (ctx->sflow_n_outputs >= 255) {
4162 cookie->n_output = 255;
4164 cookie->n_output = ctx->sflow_n_outputs;
4166 cookie->vlan_tci = base->vlan_tci;
4170 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4173 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4174 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4175 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4176 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4180 struct priority_to_dscp *pdscp;
4182 if (ofport->up.opp.config & htonl(OFPPC_NO_FWD)
4183 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4187 pdscp = get_priority(ofport, ctx->flow.priority);
4189 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4190 ctx->flow.nw_tos |= pdscp->dscp;
4193 /* We may not have an ofport record for this port, but it doesn't hurt
4194 * to allow forwarding to it anyhow. Maybe such a port will appear
4195 * later and we're pre-populating the flow table. */
4198 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4199 ctx->flow.vlan_tci);
4200 if (out_port != odp_port) {
4201 ctx->flow.vlan_tci = htons(0);
4203 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4204 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4206 ctx->sflow_odp_port = odp_port;
4207 ctx->sflow_n_outputs++;
4208 ctx->nf_output_iface = ofp_port;
4209 ctx->flow.vlan_tci = flow_vlan_tci;
4210 ctx->flow.nw_tos = flow_nw_tos;
4214 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4216 compose_output_action__(ctx, ofp_port, true);
4220 xlate_table_action(struct action_xlate_ctx *ctx,
4221 uint16_t in_port, uint8_t table_id)
4223 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4224 struct ofproto_dpif *ofproto = ctx->ofproto;
4225 struct rule_dpif *rule;
4226 uint16_t old_in_port;
4227 uint8_t old_table_id;
4229 old_table_id = ctx->table_id;
4230 ctx->table_id = table_id;
4232 /* Look up a flow with 'in_port' as the input port. */
4233 old_in_port = ctx->flow.in_port;
4234 ctx->flow.in_port = in_port;
4235 rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
4238 if (table_id > 0 && table_id < N_TABLES) {
4239 struct table_dpif *table = &ofproto->tables[table_id];
4240 if (table->other_table) {
4243 : rule_calculate_tag(&ctx->flow,
4244 &table->other_table->wc,
4249 /* Restore the original input port. Otherwise OFPP_NORMAL and
4250 * OFPP_IN_PORT will have surprising behavior. */
4251 ctx->flow.in_port = old_in_port;
4253 if (ctx->resubmit_hook) {
4254 ctx->resubmit_hook(ctx, rule);
4259 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
4263 ctx->table_id = old_table_id;
4265 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4267 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
4268 MAX_RESUBMIT_RECURSION);
4273 xlate_resubmit_table(struct action_xlate_ctx *ctx,
4274 const struct nx_action_resubmit *nar)
4279 in_port = (nar->in_port == htons(OFPP_IN_PORT)
4281 : ntohs(nar->in_port));
4282 table_id = nar->table == 255 ? ctx->table_id : nar->table;
4284 xlate_table_action(ctx, in_port, table_id);
4288 flood_packets(struct action_xlate_ctx *ctx, bool all)
4290 struct ofport_dpif *ofport;
4292 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
4293 uint16_t ofp_port = ofport->up.ofp_port;
4295 if (ofp_port == ctx->flow.in_port) {
4300 compose_output_action__(ctx, ofp_port, false);
4301 } else if (!(ofport->up.opp.config & htonl(OFPPC_NO_FLOOD))) {
4302 compose_output_action(ctx, ofp_port);
4306 ctx->nf_output_iface = NF_OUT_FLOOD;
4310 compose_controller_action(struct action_xlate_ctx *ctx, int len)
4312 struct user_action_cookie cookie;
4314 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4315 cookie.type = USER_ACTION_COOKIE_CONTROLLER;
4317 cookie.n_output = 0;
4318 cookie.vlan_tci = 0;
4319 put_userspace_action(ctx->ofproto, ctx->odp_actions, &ctx->flow, &cookie);
4323 xlate_output_action__(struct action_xlate_ctx *ctx,
4324 uint16_t port, uint16_t max_len)
4326 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
4328 ctx->nf_output_iface = NF_OUT_DROP;
4332 compose_output_action(ctx, ctx->flow.in_port);
4335 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
4341 flood_packets(ctx, false);
4344 flood_packets(ctx, true);
4346 case OFPP_CONTROLLER:
4347 compose_controller_action(ctx, max_len);
4353 if (port != ctx->flow.in_port) {
4354 compose_output_action(ctx, port);
4359 if (prev_nf_output_iface == NF_OUT_FLOOD) {
4360 ctx->nf_output_iface = NF_OUT_FLOOD;
4361 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
4362 ctx->nf_output_iface = prev_nf_output_iface;
4363 } else if (prev_nf_output_iface != NF_OUT_DROP &&
4364 ctx->nf_output_iface != NF_OUT_FLOOD) {
4365 ctx->nf_output_iface = NF_OUT_MULTI;
4370 xlate_output_reg_action(struct action_xlate_ctx *ctx,
4371 const struct nx_action_output_reg *naor)
4375 ofp_port = nxm_read_field_bits(naor->src, naor->ofs_nbits, &ctx->flow);
4377 if (ofp_port <= UINT16_MAX) {
4378 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
4383 xlate_output_action(struct action_xlate_ctx *ctx,
4384 const struct ofp_action_output *oao)
4386 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
4390 xlate_enqueue_action(struct action_xlate_ctx *ctx,
4391 const struct ofp_action_enqueue *oae)
4394 uint32_t flow_priority, priority;
4397 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
4400 /* Fall back to ordinary output action. */
4401 xlate_output_action__(ctx, ntohs(oae->port), 0);
4405 /* Figure out datapath output port. */
4406 ofp_port = ntohs(oae->port);
4407 if (ofp_port == OFPP_IN_PORT) {
4408 ofp_port = ctx->flow.in_port;
4409 } else if (ofp_port == ctx->flow.in_port) {
4413 /* Add datapath actions. */
4414 flow_priority = ctx->flow.priority;
4415 ctx->flow.priority = priority;
4416 compose_output_action(ctx, ofp_port);
4417 ctx->flow.priority = flow_priority;
4419 /* Update NetFlow output port. */
4420 if (ctx->nf_output_iface == NF_OUT_DROP) {
4421 ctx->nf_output_iface = ofp_port;
4422 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
4423 ctx->nf_output_iface = NF_OUT_MULTI;
4428 xlate_set_queue_action(struct action_xlate_ctx *ctx,
4429 const struct nx_action_set_queue *nasq)
4434 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
4437 /* Couldn't translate queue to a priority, so ignore. A warning
4438 * has already been logged. */
4442 ctx->flow.priority = priority;
4445 struct xlate_reg_state {
4451 xlate_autopath(struct action_xlate_ctx *ctx,
4452 const struct nx_action_autopath *naa)
4454 uint16_t ofp_port = ntohl(naa->id);
4455 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
4457 if (!port || !port->bundle) {
4458 ofp_port = OFPP_NONE;
4459 } else if (port->bundle->bond) {
4460 /* Autopath does not support VLAN hashing. */
4461 struct ofport_dpif *slave = bond_choose_output_slave(
4462 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
4464 ofp_port = slave->up.ofp_port;
4467 autopath_execute(naa, &ctx->flow, ofp_port);
4471 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
4473 struct ofproto_dpif *ofproto = ofproto_;
4474 struct ofport_dpif *port;
4484 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
4487 port = get_ofp_port(ofproto, ofp_port);
4488 return port ? port->may_enable : false;
4493 xlate_learn_action(struct action_xlate_ctx *ctx,
4494 const struct nx_action_learn *learn)
4496 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
4497 struct ofputil_flow_mod fm;
4500 learn_execute(learn, &ctx->flow, &fm);
4502 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
4503 if (error && !VLOG_DROP_WARN(&rl)) {
4504 char *msg = ofputil_error_to_string(error);
4505 VLOG_WARN("learning action failed to modify flow table (%s)", msg);
4513 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
4515 if (port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
4516 ? htonl(OFPPC_NO_RECV_STP)
4517 : htonl(OFPPC_NO_RECV))) {
4521 /* Only drop packets here if both forwarding and learning are
4522 * disabled. If just learning is enabled, we need to have
4523 * OFPP_NORMAL and the learning action have a look at the packet
4524 * before we can drop it. */
4525 if (!stp_forward_in_state(port->stp_state)
4526 && !stp_learn_in_state(port->stp_state)) {
4534 do_xlate_actions(const union ofp_action *in, size_t n_in,
4535 struct action_xlate_ctx *ctx)
4537 const struct ofport_dpif *port;
4538 const union ofp_action *ia;
4541 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
4542 if (port && !may_receive(port, ctx)) {
4543 /* Drop this flow. */
4547 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
4548 const struct ofp_action_dl_addr *oada;
4549 const struct nx_action_resubmit *nar;
4550 const struct nx_action_set_tunnel *nast;
4551 const struct nx_action_set_queue *nasq;
4552 const struct nx_action_multipath *nam;
4553 const struct nx_action_autopath *naa;
4554 const struct nx_action_bundle *nab;
4555 const struct nx_action_output_reg *naor;
4556 enum ofputil_action_code code;
4563 code = ofputil_decode_action_unsafe(ia);
4565 case OFPUTIL_OFPAT_OUTPUT:
4566 xlate_output_action(ctx, &ia->output);
4569 case OFPUTIL_OFPAT_SET_VLAN_VID:
4570 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
4571 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
4574 case OFPUTIL_OFPAT_SET_VLAN_PCP:
4575 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
4576 ctx->flow.vlan_tci |= htons(
4577 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
4580 case OFPUTIL_OFPAT_STRIP_VLAN:
4581 ctx->flow.vlan_tci = htons(0);
4584 case OFPUTIL_OFPAT_SET_DL_SRC:
4585 oada = ((struct ofp_action_dl_addr *) ia);
4586 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
4589 case OFPUTIL_OFPAT_SET_DL_DST:
4590 oada = ((struct ofp_action_dl_addr *) ia);
4591 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
4594 case OFPUTIL_OFPAT_SET_NW_SRC:
4595 ctx->flow.nw_src = ia->nw_addr.nw_addr;
4598 case OFPUTIL_OFPAT_SET_NW_DST:
4599 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
4602 case OFPUTIL_OFPAT_SET_NW_TOS:
4603 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4604 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
4607 case OFPUTIL_OFPAT_SET_TP_SRC:
4608 ctx->flow.tp_src = ia->tp_port.tp_port;
4611 case OFPUTIL_OFPAT_SET_TP_DST:
4612 ctx->flow.tp_dst = ia->tp_port.tp_port;
4615 case OFPUTIL_OFPAT_ENQUEUE:
4616 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
4619 case OFPUTIL_NXAST_RESUBMIT:
4620 nar = (const struct nx_action_resubmit *) ia;
4621 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
4624 case OFPUTIL_NXAST_RESUBMIT_TABLE:
4625 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
4628 case OFPUTIL_NXAST_SET_TUNNEL:
4629 nast = (const struct nx_action_set_tunnel *) ia;
4630 tun_id = htonll(ntohl(nast->tun_id));
4631 ctx->flow.tun_id = tun_id;
4634 case OFPUTIL_NXAST_SET_QUEUE:
4635 nasq = (const struct nx_action_set_queue *) ia;
4636 xlate_set_queue_action(ctx, nasq);
4639 case OFPUTIL_NXAST_POP_QUEUE:
4640 ctx->flow.priority = ctx->original_priority;
4643 case OFPUTIL_NXAST_REG_MOVE:
4644 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
4648 case OFPUTIL_NXAST_REG_LOAD:
4649 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
4653 case OFPUTIL_NXAST_NOTE:
4654 /* Nothing to do. */
4657 case OFPUTIL_NXAST_SET_TUNNEL64:
4658 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
4659 ctx->flow.tun_id = tun_id;
4662 case OFPUTIL_NXAST_MULTIPATH:
4663 nam = (const struct nx_action_multipath *) ia;
4664 multipath_execute(nam, &ctx->flow);
4667 case OFPUTIL_NXAST_AUTOPATH:
4668 naa = (const struct nx_action_autopath *) ia;
4669 xlate_autopath(ctx, naa);
4672 case OFPUTIL_NXAST_BUNDLE:
4673 ctx->ofproto->has_bundle_action = true;
4674 nab = (const struct nx_action_bundle *) ia;
4675 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
4680 case OFPUTIL_NXAST_BUNDLE_LOAD:
4681 ctx->ofproto->has_bundle_action = true;
4682 nab = (const struct nx_action_bundle *) ia;
4683 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
4687 case OFPUTIL_NXAST_OUTPUT_REG:
4688 naor = (const struct nx_action_output_reg *) ia;
4689 xlate_output_reg_action(ctx, naor);
4692 case OFPUTIL_NXAST_LEARN:
4693 ctx->has_learn = true;
4694 if (ctx->may_learn) {
4695 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
4699 case OFPUTIL_NXAST_EXIT:
4705 /* We've let OFPP_NORMAL and the learning action look at the packet,
4706 * so drop it now if forwarding is disabled. */
4707 if (port && !stp_forward_in_state(port->stp_state)) {
4708 ofpbuf_clear(ctx->odp_actions);
4709 add_sflow_action(ctx);
4714 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
4715 struct ofproto_dpif *ofproto, const struct flow *flow,
4716 ovs_be16 initial_tci, const struct ofpbuf *packet)
4718 ctx->ofproto = ofproto;
4720 ctx->base_flow = ctx->flow;
4721 ctx->base_flow.tun_id = 0;
4722 ctx->base_flow.vlan_tci = initial_tci;
4723 ctx->packet = packet;
4724 ctx->may_learn = packet != NULL;
4725 ctx->resubmit_hook = NULL;
4728 static struct ofpbuf *
4729 xlate_actions(struct action_xlate_ctx *ctx,
4730 const union ofp_action *in, size_t n_in)
4732 struct flow orig_flow = ctx->flow;
4734 COVERAGE_INC(ofproto_dpif_xlate);
4736 ctx->odp_actions = ofpbuf_new(512);
4737 ofpbuf_reserve(ctx->odp_actions, NL_A_U32_SIZE);
4739 ctx->may_set_up_flow = true;
4740 ctx->has_learn = false;
4741 ctx->has_normal = false;
4742 ctx->nf_output_iface = NF_OUT_DROP;
4745 ctx->original_priority = ctx->flow.priority;
4749 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
4750 switch (ctx->ofproto->up.frag_handling) {
4751 case OFPC_FRAG_NORMAL:
4752 /* We must pretend that transport ports are unavailable. */
4753 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
4754 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
4757 case OFPC_FRAG_DROP:
4758 return ctx->odp_actions;
4760 case OFPC_FRAG_REASM:
4763 case OFPC_FRAG_NX_MATCH:
4764 /* Nothing to do. */
4769 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
4770 ctx->may_set_up_flow = false;
4771 return ctx->odp_actions;
4773 add_sflow_action(ctx);
4774 do_xlate_actions(in, n_in, ctx);
4776 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
4777 ctx->odp_actions->data,
4778 ctx->odp_actions->size)) {
4779 ctx->may_set_up_flow = false;
4781 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
4783 compose_output_action(ctx, OFPP_LOCAL);
4786 add_mirror_actions(ctx, &orig_flow);
4787 fix_sflow_action(ctx);
4790 return ctx->odp_actions;
4793 /* OFPP_NORMAL implementation. */
4795 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
4797 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
4798 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
4799 * the bundle on which the packet was received, returns the VLAN to which the
4802 * Both 'vid' and the return value are in the range 0...4095. */
4804 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
4806 switch (in_bundle->vlan_mode) {
4807 case PORT_VLAN_ACCESS:
4808 return in_bundle->vlan;
4811 case PORT_VLAN_TRUNK:
4814 case PORT_VLAN_NATIVE_UNTAGGED:
4815 case PORT_VLAN_NATIVE_TAGGED:
4816 return vid ? vid : in_bundle->vlan;
4823 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
4824 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
4827 * 'vid' should be the VID obtained from the 802.1Q header that was received as
4828 * part of a packet (specify 0 if there was no 802.1Q header), in the range
4831 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
4833 /* Allow any VID on the OFPP_NONE port. */
4834 if (in_bundle == &ofpp_none_bundle) {
4838 switch (in_bundle->vlan_mode) {
4839 case PORT_VLAN_ACCESS:
4842 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4843 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
4844 "packet received on port %s configured as VLAN "
4845 "%"PRIu16" access port",
4846 in_bundle->ofproto->up.name, vid,
4847 in_bundle->name, in_bundle->vlan);
4853 case PORT_VLAN_NATIVE_UNTAGGED:
4854 case PORT_VLAN_NATIVE_TAGGED:
4856 /* Port must always carry its native VLAN. */
4860 case PORT_VLAN_TRUNK:
4861 if (!ofbundle_includes_vlan(in_bundle, vid)) {
4863 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4864 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
4865 "received on port %s not configured for trunking "
4867 in_bundle->ofproto->up.name, vid,
4868 in_bundle->name, vid);
4880 /* Given 'vlan', the VLAN that a packet belongs to, and
4881 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
4882 * that should be included in the 802.1Q header. (If the return value is 0,
4883 * then the 802.1Q header should only be included in the packet if there is a
4886 * Both 'vlan' and the return value are in the range 0...4095. */
4888 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
4890 switch (out_bundle->vlan_mode) {
4891 case PORT_VLAN_ACCESS:
4894 case PORT_VLAN_TRUNK:
4895 case PORT_VLAN_NATIVE_TAGGED:
4898 case PORT_VLAN_NATIVE_UNTAGGED:
4899 return vlan == out_bundle->vlan ? 0 : vlan;
4907 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
4910 struct ofport_dpif *port;
4912 ovs_be16 tci, old_tci;
4914 vid = output_vlan_to_vid(out_bundle, vlan);
4915 if (!out_bundle->bond) {
4916 port = ofbundle_get_a_port(out_bundle);
4918 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
4921 /* No slaves enabled, so drop packet. */
4926 old_tci = ctx->flow.vlan_tci;
4928 if (tci || out_bundle->use_priority_tags) {
4929 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
4931 tci |= htons(VLAN_CFI);
4934 ctx->flow.vlan_tci = tci;
4936 compose_output_action(ctx, port->up.ofp_port);
4937 ctx->flow.vlan_tci = old_tci;
4941 mirror_mask_ffs(mirror_mask_t mask)
4943 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
4948 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
4950 return (bundle->vlan_mode != PORT_VLAN_ACCESS
4951 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
4955 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
4957 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
4960 /* Returns an arbitrary interface within 'bundle'. */
4961 static struct ofport_dpif *
4962 ofbundle_get_a_port(const struct ofbundle *bundle)
4964 return CONTAINER_OF(list_front(&bundle->ports),
4965 struct ofport_dpif, bundle_node);
4969 vlan_is_mirrored(const struct ofmirror *m, int vlan)
4971 return !m->vlans || bitmap_is_set(m->vlans, vlan);
4974 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
4975 * to a VLAN. In general most packets may be mirrored but we want to drop
4976 * protocols that may confuse switches. */
4978 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
4980 /* If you change this function's behavior, please update corresponding
4981 * documentation in vswitch.xml at the same time. */
4982 if (dst[0] != 0x01) {
4983 /* All the currently banned MACs happen to start with 01 currently, so
4984 * this is a quick way to eliminate most of the good ones. */
4986 if (eth_addr_is_reserved(dst)) {
4987 /* Drop STP, IEEE pause frames, and other reserved protocols
4988 * (01-80-c2-00-00-0x). */
4992 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
4994 if ((dst[3] & 0xfe) == 0xcc &&
4995 (dst[4] & 0xfe) == 0xcc &&
4996 (dst[5] & 0xfe) == 0xcc) {
4997 /* Drop the following protocols plus others following the same
5000 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
5001 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
5002 STP Uplink Fast (01-00-0c-cd-cd-cd) */
5006 if (!(dst[3] | dst[4] | dst[5])) {
5007 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
5016 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5018 struct ofproto_dpif *ofproto = ctx->ofproto;
5019 mirror_mask_t mirrors;
5020 struct ofbundle *in_bundle;
5023 const struct nlattr *a;
5026 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5027 ctx->packet != NULL);
5031 mirrors = in_bundle->src_mirrors;
5033 /* Drop frames on bundles reserved for mirroring. */
5034 if (in_bundle->mirror_out) {
5035 if (ctx->packet != NULL) {
5036 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5037 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5038 "%s, which is reserved exclusively for mirroring",
5039 ctx->ofproto->up.name, in_bundle->name);
5045 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5046 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5049 vlan = input_vid_to_vlan(in_bundle, vid);
5051 /* Look at the output ports to check for destination selections. */
5053 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5054 ctx->odp_actions->size) {
5055 enum ovs_action_attr type = nl_attr_type(a);
5056 struct ofport_dpif *ofport;
5058 if (type != OVS_ACTION_ATTR_OUTPUT) {
5062 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5063 if (ofport && ofport->bundle) {
5064 mirrors |= ofport->bundle->dst_mirrors;
5072 /* Restore the original packet before adding the mirror actions. */
5073 ctx->flow = *orig_flow;
5078 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5080 if (!vlan_is_mirrored(m, vlan)) {
5081 mirrors &= mirrors - 1;
5085 mirrors &= ~m->dup_mirrors;
5086 ctx->mirrors |= m->dup_mirrors;
5088 output_normal(ctx, m->out, vlan);
5089 } else if (eth_dst_may_rspan(orig_flow->dl_dst)
5090 && vlan != m->out_vlan) {
5091 struct ofbundle *bundle;
5093 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5094 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5095 && !bundle->mirror_out) {
5096 output_normal(ctx, bundle, m->out_vlan);
5104 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5105 uint64_t packets, uint64_t bytes)
5111 for (; mirrors; mirrors &= mirrors - 1) {
5114 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5117 /* In normal circumstances 'm' will not be NULL. However,
5118 * if mirrors are reconfigured, we can temporarily get out
5119 * of sync in facet_revalidate(). We could "correct" the
5120 * mirror list before reaching here, but doing that would
5121 * not properly account the traffic stats we've currently
5122 * accumulated for previous mirror configuration. */
5126 m->packet_count += packets;
5127 m->byte_count += bytes;
5131 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
5132 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
5133 * indicate this; newer upstream kernels use gratuitous ARP requests. */
5135 is_gratuitous_arp(const struct flow *flow)
5137 return (flow->dl_type == htons(ETH_TYPE_ARP)
5138 && eth_addr_is_broadcast(flow->dl_dst)
5139 && (flow->nw_proto == ARP_OP_REPLY
5140 || (flow->nw_proto == ARP_OP_REQUEST
5141 && flow->nw_src == flow->nw_dst)));
5145 update_learning_table(struct ofproto_dpif *ofproto,
5146 const struct flow *flow, int vlan,
5147 struct ofbundle *in_bundle)
5149 struct mac_entry *mac;
5151 /* Don't learn the OFPP_NONE port. */
5152 if (in_bundle == &ofpp_none_bundle) {
5156 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
5160 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
5161 if (is_gratuitous_arp(flow)) {
5162 /* We don't want to learn from gratuitous ARP packets that are
5163 * reflected back over bond slaves so we lock the learning table. */
5164 if (!in_bundle->bond) {
5165 mac_entry_set_grat_arp_lock(mac);
5166 } else if (mac_entry_is_grat_arp_locked(mac)) {
5171 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
5172 /* The log messages here could actually be useful in debugging,
5173 * so keep the rate limit relatively high. */
5174 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5175 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
5176 "on port %s in VLAN %d",
5177 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
5178 in_bundle->name, vlan);
5180 mac->port.p = in_bundle;
5181 tag_set_add(&ofproto->revalidate_set,
5182 mac_learning_changed(ofproto->ml, mac));
5186 static struct ofbundle *
5187 lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn)
5189 struct ofport_dpif *ofport;
5191 /* Special-case OFPP_NONE, which a controller may use as the ingress
5192 * port for traffic that it is sourcing. */
5193 if (in_port == OFPP_NONE) {
5194 return &ofpp_none_bundle;
5197 /* Find the port and bundle for the received packet. */
5198 ofport = get_ofp_port(ofproto, in_port);
5199 if (ofport && ofport->bundle) {
5200 return ofport->bundle;
5203 /* Odd. A few possible reasons here:
5205 * - We deleted a port but there are still a few packets queued up
5208 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
5209 * we don't know about.
5211 * - The ofproto client didn't configure the port as part of a bundle.
5214 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5216 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
5217 "port %"PRIu16, ofproto->up.name, in_port);
5222 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
5223 * dropped. Returns true if they may be forwarded, false if they should be
5226 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
5227 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
5229 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
5230 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
5231 * checked by input_vid_is_valid().
5233 * May also add tags to '*tags', although the current implementation only does
5234 * so in one special case.
5237 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
5238 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
5240 struct ofbundle *in_bundle = in_port->bundle;
5242 /* Drop frames for reserved multicast addresses
5243 * only if forward_bpdu option is absent. */
5244 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
5248 if (in_bundle->bond) {
5249 struct mac_entry *mac;
5251 switch (bond_check_admissibility(in_bundle->bond, in_port,
5252 flow->dl_dst, tags)) {
5259 case BV_DROP_IF_MOVED:
5260 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
5261 if (mac && mac->port.p != in_bundle &&
5262 (!is_gratuitous_arp(flow)
5263 || mac_entry_is_grat_arp_locked(mac))) {
5274 xlate_normal(struct action_xlate_ctx *ctx)
5276 struct ofport_dpif *in_port;
5277 struct ofbundle *in_bundle;
5278 struct mac_entry *mac;
5282 ctx->has_normal = true;
5284 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
5285 ctx->packet != NULL);
5290 /* We know 'in_port' exists unless it is "ofpp_none_bundle",
5291 * since lookup_input_bundle() succeeded. */
5292 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5294 /* Drop malformed frames. */
5295 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
5296 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
5297 if (ctx->packet != NULL) {
5298 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5299 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
5300 "VLAN tag received on port %s",
5301 ctx->ofproto->up.name, in_bundle->name);
5306 /* Drop frames on bundles reserved for mirroring. */
5307 if (in_bundle->mirror_out) {
5308 if (ctx->packet != NULL) {
5309 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5310 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5311 "%s, which is reserved exclusively for mirroring",
5312 ctx->ofproto->up.name, in_bundle->name);
5318 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
5319 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5322 vlan = input_vid_to_vlan(in_bundle, vid);
5324 /* Check other admissibility requirements. */
5326 !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
5330 /* Learn source MAC. */
5331 if (ctx->may_learn) {
5332 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
5335 /* Determine output bundle. */
5336 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
5339 if (mac->port.p != in_bundle) {
5340 output_normal(ctx, mac->port.p, vlan);
5342 } else if (!ctx->packet && !eth_addr_is_multicast(ctx->flow.dl_dst)) {
5343 /* If we are revalidating but don't have a learning entry then eject
5344 * the flow. Installing a flow that floods packets opens up a window
5345 * of time where we could learn from a packet reflected on a bond and
5346 * blackhole packets before the learning table is updated to reflect
5347 * the correct port. */
5348 ctx->may_set_up_flow = false;
5351 struct ofbundle *bundle;
5353 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
5354 if (bundle != in_bundle
5355 && ofbundle_includes_vlan(bundle, vlan)
5356 && bundle->floodable
5357 && !bundle->mirror_out) {
5358 output_normal(ctx, bundle, vlan);
5361 ctx->nf_output_iface = NF_OUT_FLOOD;
5365 /* Optimized flow revalidation.
5367 * It's a difficult problem, in general, to tell which facets need to have
5368 * their actions recalculated whenever the OpenFlow flow table changes. We
5369 * don't try to solve that general problem: for most kinds of OpenFlow flow
5370 * table changes, we recalculate the actions for every facet. This is
5371 * relatively expensive, but it's good enough if the OpenFlow flow table
5372 * doesn't change very often.
5374 * However, we can expect one particular kind of OpenFlow flow table change to
5375 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5376 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5377 * table, we add a special case that applies to flow tables in which every rule
5378 * has the same form (that is, the same wildcards), except that the table is
5379 * also allowed to have a single "catch-all" flow that matches all packets. We
5380 * optimize this case by tagging all of the facets that resubmit into the table
5381 * and invalidating the same tag whenever a flow changes in that table. The
5382 * end result is that we revalidate just the facets that need it (and sometimes
5383 * a few more, but not all of the facets or even all of the facets that
5384 * resubmit to the table modified by MAC learning). */
5386 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
5387 * into an OpenFlow table with the given 'basis'. */
5389 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
5392 if (flow_wildcards_is_catchall(wc)) {
5395 struct flow tag_flow = *flow;
5396 flow_zero_wildcards(&tag_flow, wc);
5397 return tag_create_deterministic(flow_hash(&tag_flow, secret));
5401 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5402 * taggability of that table.
5404 * This function must be called after *each* change to a flow table. If you
5405 * skip calling it on some changes then the pointer comparisons at the end can
5406 * be invalid if you get unlucky. For example, if a flow removal causes a
5407 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5408 * different wildcards to be created with the same address, then this function
5409 * will incorrectly skip revalidation. */
5411 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5413 struct table_dpif *table = &ofproto->tables[table_id];
5414 const struct classifier *cls = &ofproto->up.tables[table_id];
5415 struct cls_table *catchall, *other;
5416 struct cls_table *t;
5418 catchall = other = NULL;
5420 switch (hmap_count(&cls->tables)) {
5422 /* We could tag this OpenFlow table but it would make the logic a
5423 * little harder and it's a corner case that doesn't seem worth it
5429 HMAP_FOR_EACH (t, hmap_node, &cls->tables) {
5430 if (cls_table_is_catchall(t)) {
5432 } else if (!other) {
5435 /* Indicate that we can't tag this by setting both tables to
5436 * NULL. (We know that 'catchall' is already NULL.) */
5443 /* Can't tag this table. */
5447 if (table->catchall_table != catchall || table->other_table != other) {
5448 table->catchall_table = catchall;
5449 table->other_table = other;
5450 ofproto->need_revalidate = true;
5454 /* Given 'rule' that has changed in some way (either it is a rule being
5455 * inserted, a rule being deleted, or a rule whose actions are being
5456 * modified), marks facets for revalidation to ensure that packets will be
5457 * forwarded correctly according to the new state of the flow table.
5459 * This function must be called after *each* change to a flow table. See
5460 * the comment on table_update_taggable() for more information. */
5462 rule_invalidate(const struct rule_dpif *rule)
5464 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5466 table_update_taggable(ofproto, rule->up.table_id);
5468 if (!ofproto->need_revalidate) {
5469 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5471 if (table->other_table && rule->tag) {
5472 tag_set_add(&ofproto->revalidate_set, rule->tag);
5474 ofproto->need_revalidate = true;
5480 set_frag_handling(struct ofproto *ofproto_,
5481 enum ofp_config_flags frag_handling)
5483 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5485 if (frag_handling != OFPC_FRAG_REASM) {
5486 ofproto->need_revalidate = true;
5494 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5495 const struct flow *flow,
5496 const union ofp_action *ofp_actions, size_t n_ofp_actions)
5498 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5501 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
5502 return ofp_mkerr_nicira(OFPET_BAD_REQUEST, NXBRC_BAD_IN_PORT);
5505 error = validate_actions(ofp_actions, n_ofp_actions, flow,
5506 ofproto->max_ports);
5508 struct odputil_keybuf keybuf;
5509 struct action_xlate_ctx ctx;
5510 struct ofpbuf *odp_actions;
5513 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5514 odp_flow_key_from_flow(&key, flow);
5516 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, packet);
5517 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
5518 dpif_execute(ofproto->dpif, key.data, key.size,
5519 odp_actions->data, odp_actions->size, packet);
5520 ofpbuf_delete(odp_actions);
5528 set_netflow(struct ofproto *ofproto_,
5529 const struct netflow_options *netflow_options)
5531 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5533 if (netflow_options) {
5534 if (!ofproto->netflow) {
5535 ofproto->netflow = netflow_create();
5537 return netflow_set_options(ofproto->netflow, netflow_options);
5539 netflow_destroy(ofproto->netflow);
5540 ofproto->netflow = NULL;
5546 get_netflow_ids(const struct ofproto *ofproto_,
5547 uint8_t *engine_type, uint8_t *engine_id)
5549 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5551 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
5555 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5557 if (!facet_is_controller_flow(facet) &&
5558 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5559 struct subfacet *subfacet;
5560 struct ofexpired expired;
5562 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5563 if (subfacet->installed) {
5564 struct dpif_flow_stats stats;
5566 subfacet_install(ofproto, subfacet, subfacet->actions,
5567 subfacet->actions_len, &stats);
5568 subfacet_update_stats(ofproto, subfacet, &stats);
5572 expired.flow = facet->flow;
5573 expired.packet_count = facet->packet_count;
5574 expired.byte_count = facet->byte_count;
5575 expired.used = facet->used;
5576 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5581 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5583 struct facet *facet;
5585 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5586 send_active_timeout(ofproto, facet);
5590 static struct ofproto_dpif *
5591 ofproto_dpif_lookup(const char *name)
5593 struct ofproto_dpif *ofproto;
5595 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5596 hash_string(name, 0), &all_ofproto_dpifs) {
5597 if (!strcmp(ofproto->up.name, name)) {
5605 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn,
5606 const char *args, void *aux OVS_UNUSED)
5608 struct ofproto_dpif *ofproto;
5610 if (args[0] != '\0') {
5611 ofproto = ofproto_dpif_lookup(args);
5613 unixctl_command_reply(conn, 501, "no such bridge");
5616 mac_learning_flush(ofproto->ml);
5617 ofproto->need_revalidate = true;
5619 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5620 mac_learning_flush(ofproto->ml);
5621 ofproto->need_revalidate = true;
5625 unixctl_command_reply(conn, 200, "table successfully flushed");
5629 ofproto_unixctl_fdb_show(struct unixctl_conn *conn,
5630 const char *args, void *aux OVS_UNUSED)
5632 struct ds ds = DS_EMPTY_INITIALIZER;
5633 const struct ofproto_dpif *ofproto;
5634 const struct mac_entry *e;
5636 ofproto = ofproto_dpif_lookup(args);
5638 unixctl_command_reply(conn, 501, "no such bridge");
5642 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5643 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5644 struct ofbundle *bundle = e->port.p;
5645 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
5646 ofbundle_get_a_port(bundle)->odp_port,
5647 e->vlan, ETH_ADDR_ARGS(e->mac),
5648 mac_entry_age(ofproto->ml, e));
5650 unixctl_command_reply(conn, 200, ds_cstr(&ds));
5654 struct ofproto_trace {
5655 struct action_xlate_ctx ctx;
5661 trace_format_rule(struct ds *result, uint8_t table_id, int level,
5662 const struct rule_dpif *rule)
5664 ds_put_char_multiple(result, '\t', level);
5666 ds_put_cstr(result, "No match\n");
5670 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5671 table_id, ntohll(rule->up.flow_cookie));
5672 cls_rule_format(&rule->up.cr, result);
5673 ds_put_char(result, '\n');
5675 ds_put_char_multiple(result, '\t', level);
5676 ds_put_cstr(result, "OpenFlow ");
5677 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
5678 ds_put_char(result, '\n');
5682 trace_format_flow(struct ds *result, int level, const char *title,
5683 struct ofproto_trace *trace)
5685 ds_put_char_multiple(result, '\t', level);
5686 ds_put_format(result, "%s: ", title);
5687 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
5688 ds_put_cstr(result, "unchanged");
5690 flow_format(result, &trace->ctx.flow);
5691 trace->flow = trace->ctx.flow;
5693 ds_put_char(result, '\n');
5697 trace_format_regs(struct ds *result, int level, const char *title,
5698 struct ofproto_trace *trace)
5702 ds_put_char_multiple(result, '\t', level);
5703 ds_put_format(result, "%s:", title);
5704 for (i = 0; i < FLOW_N_REGS; i++) {
5705 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5707 ds_put_char(result, '\n');
5711 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
5713 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
5714 struct ds *result = trace->result;
5716 ds_put_char(result, '\n');
5717 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
5718 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
5719 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
5723 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
5724 void *aux OVS_UNUSED)
5726 char *dpname, *arg1, *arg2, *arg3, *arg4;
5727 char *args = xstrdup(args_);
5728 char *save_ptr = NULL;
5729 struct ofproto_dpif *ofproto;
5730 struct ofpbuf odp_key;
5731 struct ofpbuf *packet;
5732 struct rule_dpif *rule;
5733 ovs_be16 initial_tci;
5739 ofpbuf_init(&odp_key, 0);
5742 dpname = strtok_r(args, " ", &save_ptr);
5744 unixctl_command_reply(conn, 501, "Bad command syntax");
5748 ofproto = ofproto_dpif_lookup(dpname);
5750 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
5754 arg1 = strtok_r(NULL, " ", &save_ptr);
5755 arg2 = strtok_r(NULL, " ", &save_ptr);
5756 arg3 = strtok_r(NULL, " ", &save_ptr);
5757 arg4 = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
5758 if (dpname && arg1 && (!arg2 || !strcmp(arg2, "-generate")) && !arg3) {
5759 /* ofproto/trace dpname flow [-generate] */
5762 /* Convert string to datapath key. */
5763 ofpbuf_init(&odp_key, 0);
5764 error = odp_flow_key_from_string(arg1, NULL, &odp_key);
5766 unixctl_command_reply(conn, 501, "Bad flow syntax");
5770 /* Convert odp_key to flow. */
5771 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
5772 odp_key.size, &flow,
5773 &initial_tci, NULL);
5774 if (error == ODP_FIT_ERROR) {
5775 unixctl_command_reply(conn, 501, "Invalid flow");
5779 /* Generate a packet, if requested. */
5781 packet = ofpbuf_new(0);
5782 flow_compose(packet, &flow);
5784 } else if (dpname && arg1 && arg2 && arg3 && arg4) {
5785 /* ofproto/trace dpname priority tun_id in_port packet */
5790 priority = atoi(arg1);
5791 tun_id = htonll(strtoull(arg2, NULL, 0));
5792 in_port = ofp_port_to_odp_port(atoi(arg3));
5794 packet = ofpbuf_new(strlen(args) / 2);
5795 arg4 = ofpbuf_put_hex(packet, arg4, NULL);
5796 arg4 += strspn(arg4, " ");
5797 if (*arg4 != '\0') {
5798 unixctl_command_reply(conn, 501, "Trailing garbage in command");
5801 if (packet->size < ETH_HEADER_LEN) {
5802 unixctl_command_reply(conn, 501,
5803 "Packet data too short for Ethernet");
5807 ds_put_cstr(&result, "Packet: ");
5808 s = ofp_packet_to_string(packet->data, packet->size, packet->size);
5809 ds_put_cstr(&result, s);
5812 flow_extract(packet, priority, tun_id, in_port, &flow);
5813 initial_tci = flow.vlan_tci;
5815 unixctl_command_reply(conn, 501, "Bad command syntax");
5819 ds_put_cstr(&result, "Flow: ");
5820 flow_format(&result, &flow);
5821 ds_put_char(&result, '\n');
5823 rule = rule_dpif_lookup(ofproto, &flow, 0);
5824 trace_format_rule(&result, 0, 0, rule);
5826 struct ofproto_trace trace;
5827 struct ofpbuf *odp_actions;
5829 trace.result = &result;
5831 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, initial_tci, packet);
5832 trace.ctx.resubmit_hook = trace_resubmit;
5833 odp_actions = xlate_actions(&trace.ctx,
5834 rule->up.actions, rule->up.n_actions);
5836 ds_put_char(&result, '\n');
5837 trace_format_flow(&result, 0, "Final flow", &trace);
5838 ds_put_cstr(&result, "Datapath actions: ");
5839 format_odp_actions(&result, odp_actions->data, odp_actions->size);
5840 ofpbuf_delete(odp_actions);
5842 if (!trace.ctx.may_set_up_flow) {
5844 ds_put_cstr(&result, "\nThis flow is not cachable.");
5846 ds_put_cstr(&result, "\nThe datapath actions are incomplete--"
5847 "for complete actions, please supply a packet.");
5852 unixctl_command_reply(conn, 200, ds_cstr(&result));
5855 ds_destroy(&result);
5856 ofpbuf_delete(packet);
5857 ofpbuf_uninit(&odp_key);
5862 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED,
5863 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
5866 unixctl_command_reply(conn, 200, NULL);
5870 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED,
5871 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
5874 unixctl_command_reply(conn, 200, NULL);
5878 ofproto_dpif_unixctl_init(void)
5880 static bool registered;
5886 unixctl_command_register("ofproto/trace",
5887 "bridge {tun_id in_port packet | odp_flow [-generate]}",
5888 ofproto_unixctl_trace, NULL);
5889 unixctl_command_register("fdb/flush", "[bridge]",
5890 ofproto_unixctl_fdb_flush, NULL);
5891 unixctl_command_register("fdb/show", "bridge", ofproto_unixctl_fdb_show,
5893 unixctl_command_register("ofproto/clog", "", ofproto_dpif_clog, NULL);
5894 unixctl_command_register("ofproto/unclog", "", ofproto_dpif_unclog, NULL);
5897 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5899 * This is deprecated. It is only for compatibility with broken device drivers
5900 * in old versions of Linux that do not properly support VLANs when VLAN
5901 * devices are not used. When broken device drivers are no longer in
5902 * widespread use, we will delete these interfaces. */
5905 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
5907 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
5908 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
5910 if (realdev_ofp_port == ofport->realdev_ofp_port
5911 && vid == ofport->vlandev_vid) {
5915 ofproto->need_revalidate = true;
5917 if (ofport->realdev_ofp_port) {
5920 if (realdev_ofp_port && ofport->bundle) {
5921 /* vlandevs are enslaved to their realdevs, so they are not allowed to
5922 * themselves be part of a bundle. */
5923 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
5926 ofport->realdev_ofp_port = realdev_ofp_port;
5927 ofport->vlandev_vid = vid;
5929 if (realdev_ofp_port) {
5930 vsp_add(ofport, realdev_ofp_port, vid);
5937 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
5939 return hash_2words(realdev_ofp_port, vid);
5943 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
5944 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
5946 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
5947 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
5948 int vid = vlan_tci_to_vid(vlan_tci);
5949 const struct vlan_splinter *vsp;
5951 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
5952 hash_realdev_vid(realdev_ofp_port, vid),
5953 &ofproto->realdev_vid_map) {
5954 if (vsp->realdev_ofp_port == realdev_ofp_port
5955 && vsp->vid == vid) {
5956 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
5960 return realdev_odp_port;
5963 static struct vlan_splinter *
5964 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
5966 struct vlan_splinter *vsp;
5968 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
5969 &ofproto->vlandev_map) {
5970 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
5979 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
5980 uint16_t vlandev_ofp_port, int *vid)
5982 if (!hmap_is_empty(&ofproto->vlandev_map)) {
5983 const struct vlan_splinter *vsp;
5985 vsp = vlandev_find(ofproto, vlandev_ofp_port);
5990 return vsp->realdev_ofp_port;
5997 vsp_remove(struct ofport_dpif *port)
5999 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6000 struct vlan_splinter *vsp;
6002 vsp = vlandev_find(ofproto, port->up.ofp_port);
6004 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6005 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6008 port->realdev_ofp_port = 0;
6010 VLOG_ERR("missing vlan device record");
6015 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
6017 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6019 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6020 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6021 == realdev_ofp_port)) {
6022 struct vlan_splinter *vsp;
6024 vsp = xmalloc(sizeof *vsp);
6025 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6026 hash_int(port->up.ofp_port, 0));
6027 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6028 hash_realdev_vid(realdev_ofp_port, vid));
6029 vsp->realdev_ofp_port = realdev_ofp_port;
6030 vsp->vlandev_ofp_port = port->up.ofp_port;
6033 port->realdev_ofp_port = realdev_ofp_port;
6035 VLOG_ERR("duplicate vlan device record");
6039 const struct ofproto_class ofproto_dpif_class = {
6067 port_is_lacp_current,
6068 NULL, /* rule_choose_table */
6075 rule_modify_actions,
6083 get_cfm_remote_mpids,
6087 get_stp_port_status,
6094 is_mirror_output_bundle,
6095 forward_bpdu_changed,