2 * Copyright (c) 2009, 2010, 2011, 2012 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-provider.h"
26 #include "byte-order.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
36 #include "mac-learning.h"
37 #include "multipath.h"
44 #include "ofp-print.h"
45 #include "ofproto-dpif-sflow.h"
46 #include "poll-loop.h"
48 #include "unaligned.h"
50 #include "vlan-bitmap.h"
53 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
55 COVERAGE_DEFINE(ofproto_dpif_ctlr_action);
56 COVERAGE_DEFINE(ofproto_dpif_expired);
57 COVERAGE_DEFINE(ofproto_dpif_no_packet_in);
58 COVERAGE_DEFINE(ofproto_dpif_xlate);
59 COVERAGE_DEFINE(facet_changed_rule);
60 COVERAGE_DEFINE(facet_invalidated);
61 COVERAGE_DEFINE(facet_revalidate);
62 COVERAGE_DEFINE(facet_unexpected);
64 /* Maximum depth of flow table recursion (due to resubmit actions) in a
65 * flow translation. */
66 #define MAX_RESUBMIT_RECURSION 32
68 /* Number of implemented OpenFlow tables. */
69 enum { N_TABLES = 255 };
70 BUILD_ASSERT_DECL(N_TABLES >= 1 && N_TABLES <= 255);
78 long long int used; /* Time last used; time created if not used. */
82 * - Do include packets and bytes from facets that have been deleted or
83 * whose own statistics have been folded into the rule.
85 * - Do include packets and bytes sent "by hand" that were accounted to
86 * the rule without any facet being involved (this is a rare corner
87 * case in rule_execute()).
89 * - Do not include packet or bytes that can be obtained from any facet's
90 * packet_count or byte_count member or that can be obtained from the
91 * datapath by, e.g., dpif_flow_get() for any subfacet.
93 uint64_t packet_count; /* Number of packets received. */
94 uint64_t byte_count; /* Number of bytes received. */
96 tag_type tag; /* Caches rule_calculate_tag() result. */
98 struct list facets; /* List of "struct facet"s. */
101 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
103 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
106 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
107 const struct flow *, uint8_t table);
109 static void flow_push_stats(const struct rule_dpif *, const struct flow *,
110 uint64_t packets, uint64_t bytes,
113 static uint32_t rule_calculate_tag(const struct flow *,
114 const struct flow_wildcards *,
116 static void rule_invalidate(const struct rule_dpif *);
118 #define MAX_MIRRORS 32
119 typedef uint32_t mirror_mask_t;
120 #define MIRROR_MASK_C(X) UINT32_C(X)
121 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
123 struct ofproto_dpif *ofproto; /* Owning ofproto. */
124 size_t idx; /* In ofproto's "mirrors" array. */
125 void *aux; /* Key supplied by ofproto's client. */
126 char *name; /* Identifier for log messages. */
128 /* Selection criteria. */
129 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
130 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
131 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
133 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
134 struct ofbundle *out; /* Output port or NULL. */
135 int out_vlan; /* Output VLAN or -1. */
136 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
139 int64_t packet_count; /* Number of packets sent. */
140 int64_t byte_count; /* Number of bytes sent. */
143 static void mirror_destroy(struct ofmirror *);
144 static void update_mirror_stats(struct ofproto_dpif *ofproto,
145 mirror_mask_t mirrors,
146 uint64_t packets, uint64_t bytes);
149 struct ofproto_dpif *ofproto; /* Owning ofproto. */
150 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
151 void *aux; /* Key supplied by ofproto's client. */
152 char *name; /* Identifier for log messages. */
155 struct list ports; /* Contains "struct ofport"s. */
156 enum port_vlan_mode vlan_mode; /* VLAN mode */
157 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
158 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
159 * NULL if all VLANs are trunked. */
160 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
161 struct bond *bond; /* Nonnull iff more than one port. */
162 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
165 bool floodable; /* True if no port has OFPPC_NO_FLOOD set. */
167 /* Port mirroring info. */
168 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
169 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
170 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
173 static void bundle_remove(struct ofport *);
174 static void bundle_update(struct ofbundle *);
175 static void bundle_destroy(struct ofbundle *);
176 static void bundle_del_port(struct ofport_dpif *);
177 static void bundle_run(struct ofbundle *);
178 static void bundle_wait(struct ofbundle *);
179 static struct ofbundle *lookup_input_bundle(struct ofproto_dpif *,
180 uint16_t in_port, bool warn);
182 /* A controller may use OFPP_NONE as the ingress port to indicate that
183 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
184 * when an input bundle is needed for validation (e.g., mirroring or
185 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
186 * any 'port' structs, so care must be taken when dealing with it. */
187 static struct ofbundle ofpp_none_bundle = {
189 .vlan_mode = PORT_VLAN_TRUNK
192 static void stp_run(struct ofproto_dpif *ofproto);
193 static void stp_wait(struct ofproto_dpif *ofproto);
195 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
197 struct action_xlate_ctx {
198 /* action_xlate_ctx_init() initializes these members. */
201 struct ofproto_dpif *ofproto;
203 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
204 * this flow when actions change header fields. */
207 /* The packet corresponding to 'flow', or a null pointer if we are
208 * revalidating without a packet to refer to. */
209 const struct ofpbuf *packet;
211 /* Should OFPP_NORMAL MAC learning and NXAST_LEARN actions execute? We
212 * want to execute them if we are actually processing a packet, or if we
213 * are accounting for packets that the datapath has processed, but not if
214 * we are just revalidating. */
217 /* Cookie of the currently matching rule, or 0. */
220 /* If nonnull, called just before executing a resubmit action.
222 * This is normally null so the client has to set it manually after
223 * calling action_xlate_ctx_init(). */
224 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *);
226 /* xlate_actions() initializes and uses these members. The client might want
227 * to look at them after it returns. */
229 struct ofpbuf *odp_actions; /* Datapath actions. */
230 tag_type tags; /* Tags associated with actions. */
231 bool may_set_up_flow; /* True ordinarily; false if the actions must
232 * be reassessed for every packet. */
233 bool has_learn; /* Actions include NXAST_LEARN? */
234 bool has_normal; /* Actions output to OFPP_NORMAL? */
235 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
236 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
238 /* xlate_actions() initializes and uses these members, but the client has no
239 * reason to look at them. */
241 int recurse; /* Recursion level, via xlate_table_action. */
242 struct flow base_flow; /* Flow at the last commit. */
243 uint32_t orig_skb_priority; /* Priority when packet arrived. */
244 uint8_t table_id; /* OpenFlow table ID where flow was found. */
245 uint32_t sflow_n_outputs; /* Number of output ports. */
246 uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
247 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
248 bool exit; /* No further actions should be processed. */
251 static void action_xlate_ctx_init(struct action_xlate_ctx *,
252 struct ofproto_dpif *, const struct flow *,
253 ovs_be16 initial_tci, ovs_be64 cookie,
254 const struct ofpbuf *);
255 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
256 const union ofp_action *in, size_t n_in);
258 /* An exact-match instantiation of an OpenFlow flow.
260 * A facet associates a "struct flow", which represents the Open vSwitch
261 * userspace idea of an exact-match flow, with one or more subfacets. Each
262 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
263 * the facet. When the kernel module (or other dpif implementation) and Open
264 * vSwitch userspace agree on the definition of a flow key, there is exactly
265 * one subfacet per facet. If the dpif implementation supports more-specific
266 * flow matching than userspace, however, a facet can have more than one
267 * subfacet, each of which corresponds to some distinction in flow that
268 * userspace simply doesn't understand.
270 * Flow expiration works in terms of subfacets, so a facet must have at least
271 * one subfacet or it will never expire, leaking memory. */
274 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
275 struct list list_node; /* In owning rule's 'facets' list. */
276 struct rule_dpif *rule; /* Owning rule. */
279 struct list subfacets;
280 long long int used; /* Time last used; time created if not used. */
287 * - Do include packets and bytes sent "by hand", e.g. with
290 * - Do include packets and bytes that were obtained from the datapath
291 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
292 * DPIF_FP_ZERO_STATS).
294 * - Do not include packets or bytes that can be obtained from the
295 * datapath for any existing subfacet.
297 uint64_t packet_count; /* Number of packets received. */
298 uint64_t byte_count; /* Number of bytes received. */
300 /* Resubmit statistics. */
301 uint64_t prev_packet_count; /* Number of packets from last stats push. */
302 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
303 long long int prev_used; /* Used time from last stats push. */
306 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
307 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
309 /* Properties of datapath actions.
311 * Every subfacet has its own actions because actions can differ slightly
312 * between splintered and non-splintered subfacets due to the VLAN tag
313 * being initially different (present vs. absent). All of them have these
314 * properties in common so we just store one copy of them here. */
315 bool may_install; /* Reassess actions for every packet? */
316 bool has_learn; /* Actions include NXAST_LEARN? */
317 bool has_normal; /* Actions output to OFPP_NORMAL? */
318 tag_type tags; /* Tags that would require revalidation. */
319 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
322 static struct facet *facet_create(struct rule_dpif *, const struct flow *);
323 static void facet_remove(struct ofproto_dpif *, struct facet *);
324 static void facet_free(struct facet *);
326 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
327 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
328 const struct flow *);
329 static bool facet_revalidate(struct ofproto_dpif *, struct facet *);
331 static void facet_flush_stats(struct ofproto_dpif *, struct facet *);
333 static void facet_update_time(struct ofproto_dpif *, struct facet *,
335 static void facet_reset_counters(struct facet *);
336 static void facet_push_stats(struct facet *);
337 static void facet_account(struct ofproto_dpif *, struct facet *);
339 static bool facet_is_controller_flow(struct facet *);
341 /* A dpif flow and actions associated with a facet.
343 * See also the large comment on struct facet. */
346 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
347 struct list list_node; /* In struct facet's 'facets' list. */
348 struct facet *facet; /* Owning facet. */
352 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
353 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
354 * regenerate the ODP flow key from ->facet->flow. */
355 enum odp_key_fitness key_fitness;
359 long long int used; /* Time last used; time created if not used. */
361 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
362 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
366 * These should be essentially identical for every subfacet in a facet, but
367 * may differ in trivial ways due to VLAN splinters. */
368 size_t actions_len; /* Number of bytes in actions[]. */
369 struct nlattr *actions; /* Datapath actions. */
371 bool installed; /* Installed in datapath? */
373 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
374 * splinters can cause it to differ. This value should be removed when
375 * the VLAN splinters feature is no longer needed. */
376 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
379 static struct subfacet *subfacet_create(struct ofproto_dpif *, struct facet *,
380 enum odp_key_fitness,
381 const struct nlattr *key,
382 size_t key_len, ovs_be16 initial_tci);
383 static struct subfacet *subfacet_find(struct ofproto_dpif *,
384 const struct nlattr *key, size_t key_len);
385 static void subfacet_destroy(struct ofproto_dpif *, struct subfacet *);
386 static void subfacet_destroy__(struct ofproto_dpif *, struct subfacet *);
387 static void subfacet_reset_dp_stats(struct subfacet *,
388 struct dpif_flow_stats *);
389 static void subfacet_update_time(struct ofproto_dpif *, struct subfacet *,
391 static void subfacet_update_stats(struct ofproto_dpif *, struct subfacet *,
392 const struct dpif_flow_stats *);
393 static void subfacet_make_actions(struct ofproto_dpif *, struct subfacet *,
394 const struct ofpbuf *packet);
395 static int subfacet_install(struct ofproto_dpif *, struct subfacet *,
396 const struct nlattr *actions, size_t actions_len,
397 struct dpif_flow_stats *);
398 static void subfacet_uninstall(struct ofproto_dpif *, struct subfacet *);
404 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
405 struct list bundle_node; /* In struct ofbundle's "ports" list. */
406 struct cfm *cfm; /* Connectivity Fault Management, if any. */
407 tag_type tag; /* Tag associated with this port. */
408 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
409 bool may_enable; /* May be enabled in bonds. */
412 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
413 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
414 long long int stp_state_entered;
416 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
418 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
420 * This is deprecated. It is only for compatibility with broken device
421 * drivers in old versions of Linux that do not properly support VLANs when
422 * VLAN devices are not used. When broken device drivers are no longer in
423 * widespread use, we will delete these interfaces. */
424 uint16_t realdev_ofp_port;
428 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
429 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
430 * traffic egressing the 'ofport' with that priority should be marked with. */
431 struct priority_to_dscp {
432 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
433 uint32_t priority; /* Priority of this queue (see struct flow). */
435 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
438 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
440 * This is deprecated. It is only for compatibility with broken device drivers
441 * in old versions of Linux that do not properly support VLANs when VLAN
442 * devices are not used. When broken device drivers are no longer in
443 * widespread use, we will delete these interfaces. */
444 struct vlan_splinter {
445 struct hmap_node realdev_vid_node;
446 struct hmap_node vlandev_node;
447 uint16_t realdev_ofp_port;
448 uint16_t vlandev_ofp_port;
452 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
453 uint32_t realdev, ovs_be16 vlan_tci);
454 static uint16_t vsp_vlandev_to_realdev(const struct ofproto_dpif *,
455 uint16_t vlandev, int *vid);
456 static void vsp_remove(struct ofport_dpif *);
457 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
459 static struct ofport_dpif *
460 ofport_dpif_cast(const struct ofport *ofport)
462 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
463 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
466 static void port_run(struct ofport_dpif *);
467 static void port_wait(struct ofport_dpif *);
468 static int set_cfm(struct ofport *, const struct cfm_settings *);
469 static void ofport_clear_priorities(struct ofport_dpif *);
471 struct dpif_completion {
472 struct list list_node;
473 struct ofoperation *op;
476 /* Extra information about a classifier table.
477 * Currently used just for optimized flow revalidation. */
479 /* If either of these is nonnull, then this table has a form that allows
480 * flows to be tagged to avoid revalidating most flows for the most common
481 * kinds of flow table changes. */
482 struct cls_table *catchall_table; /* Table that wildcards all fields. */
483 struct cls_table *other_table; /* Table with any other wildcard set. */
484 uint32_t basis; /* Keeps each table's tags separate. */
487 struct ofproto_dpif {
488 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
497 struct netflow *netflow;
498 struct dpif_sflow *sflow;
499 struct hmap bundles; /* Contains "struct ofbundle"s. */
500 struct mac_learning *ml;
501 struct ofmirror *mirrors[MAX_MIRRORS];
502 bool has_bonded_bundles;
505 struct timer next_expiration;
509 struct hmap subfacets;
512 struct table_dpif tables[N_TABLES];
513 bool need_revalidate;
514 struct tag_set revalidate_set;
516 /* Support for debugging async flow mods. */
517 struct list completions;
519 bool has_bundle_action; /* True when the first bundle action appears. */
520 struct netdev_stats stats; /* To account packets generated and consumed in
525 long long int stp_last_tick;
527 /* VLAN splinters. */
528 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
529 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
532 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
533 * for debugging the asynchronous flow_mod implementation.) */
536 /* All existing ofproto_dpif instances, indexed by ->up.name. */
537 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
539 static void ofproto_dpif_unixctl_init(void);
541 static struct ofproto_dpif *
542 ofproto_dpif_cast(const struct ofproto *ofproto)
544 assert(ofproto->ofproto_class == &ofproto_dpif_class);
545 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
548 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
550 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
553 /* Packet processing. */
554 static void update_learning_table(struct ofproto_dpif *,
555 const struct flow *, int vlan,
558 #define FLOW_MISS_MAX_BATCH 50
559 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
561 /* Flow expiration. */
562 static int expire(struct ofproto_dpif *);
565 static void send_netflow_active_timeouts(struct ofproto_dpif *);
568 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
570 compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions,
571 const struct flow *, uint32_t odp_port);
572 static void add_mirror_actions(struct action_xlate_ctx *ctx,
573 const struct flow *flow);
574 /* Global variables. */
575 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
577 /* Factory functions. */
580 enumerate_types(struct sset *types)
582 dp_enumerate_types(types);
586 enumerate_names(const char *type, struct sset *names)
588 return dp_enumerate_names(type, names);
592 del(const char *type, const char *name)
597 error = dpif_open(name, type, &dpif);
599 error = dpif_delete(dpif);
605 /* Basic life-cycle. */
607 static struct ofproto *
610 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
615 dealloc(struct ofproto *ofproto_)
617 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
622 construct(struct ofproto *ofproto_, int *n_tablesp)
624 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
625 const char *name = ofproto->up.name;
629 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
631 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
635 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
636 ofproto->n_matches = 0;
638 dpif_flow_flush(ofproto->dpif);
639 dpif_recv_purge(ofproto->dpif);
641 error = dpif_recv_set_mask(ofproto->dpif,
642 ((1u << DPIF_UC_MISS) |
643 (1u << DPIF_UC_ACTION)));
645 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
646 dpif_close(ofproto->dpif);
650 ofproto->netflow = NULL;
651 ofproto->sflow = NULL;
653 hmap_init(&ofproto->bundles);
654 ofproto->ml = mac_learning_create();
655 for (i = 0; i < MAX_MIRRORS; i++) {
656 ofproto->mirrors[i] = NULL;
658 ofproto->has_bonded_bundles = false;
660 timer_set_duration(&ofproto->next_expiration, 1000);
662 hmap_init(&ofproto->facets);
663 hmap_init(&ofproto->subfacets);
665 for (i = 0; i < N_TABLES; i++) {
666 struct table_dpif *table = &ofproto->tables[i];
668 table->catchall_table = NULL;
669 table->other_table = NULL;
670 table->basis = random_uint32();
672 ofproto->need_revalidate = false;
673 tag_set_init(&ofproto->revalidate_set);
675 list_init(&ofproto->completions);
677 ofproto_dpif_unixctl_init();
679 ofproto->has_bundle_action = false;
681 hmap_init(&ofproto->vlandev_map);
682 hmap_init(&ofproto->realdev_vid_map);
684 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
685 hash_string(ofproto->up.name, 0));
687 *n_tablesp = N_TABLES;
688 memset(&ofproto->stats, 0, sizeof ofproto->stats);
693 complete_operations(struct ofproto_dpif *ofproto)
695 struct dpif_completion *c, *next;
697 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
698 ofoperation_complete(c->op, 0);
699 list_remove(&c->list_node);
705 destruct(struct ofproto *ofproto_)
707 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
708 struct rule_dpif *rule, *next_rule;
709 struct classifier *table;
712 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
713 complete_operations(ofproto);
715 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
716 struct cls_cursor cursor;
718 cls_cursor_init(&cursor, table, NULL);
719 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
720 ofproto_rule_destroy(&rule->up);
724 for (i = 0; i < MAX_MIRRORS; i++) {
725 mirror_destroy(ofproto->mirrors[i]);
728 netflow_destroy(ofproto->netflow);
729 dpif_sflow_destroy(ofproto->sflow);
730 hmap_destroy(&ofproto->bundles);
731 mac_learning_destroy(ofproto->ml);
733 hmap_destroy(&ofproto->facets);
734 hmap_destroy(&ofproto->subfacets);
736 hmap_destroy(&ofproto->vlandev_map);
737 hmap_destroy(&ofproto->realdev_vid_map);
739 dpif_close(ofproto->dpif);
743 run_fast(struct ofproto *ofproto_)
745 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
748 /* Handle one or more batches of upcalls, until there's nothing left to do
749 * or until we do a fixed total amount of work.
751 * We do work in batches because it can be much cheaper to set up a number
752 * of flows and fire off their patches all at once. We do multiple batches
753 * because in some cases handling a packet can cause another packet to be
754 * queued almost immediately as part of the return flow. Both
755 * optimizations can make major improvements on some benchmarks and
756 * presumably for real traffic as well. */
758 while (work < FLOW_MISS_MAX_BATCH) {
759 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
769 run(struct ofproto *ofproto_)
771 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
772 struct ofport_dpif *ofport;
773 struct ofbundle *bundle;
777 complete_operations(ofproto);
779 dpif_run(ofproto->dpif);
781 error = run_fast(ofproto_);
786 if (timer_expired(&ofproto->next_expiration)) {
787 int delay = expire(ofproto);
788 timer_set_duration(&ofproto->next_expiration, delay);
791 if (ofproto->netflow) {
792 if (netflow_run(ofproto->netflow)) {
793 send_netflow_active_timeouts(ofproto);
796 if (ofproto->sflow) {
797 dpif_sflow_run(ofproto->sflow);
800 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
803 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
808 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
810 /* Now revalidate if there's anything to do. */
811 if (ofproto->need_revalidate
812 || !tag_set_is_empty(&ofproto->revalidate_set)) {
813 struct tag_set revalidate_set = ofproto->revalidate_set;
814 bool revalidate_all = ofproto->need_revalidate;
815 struct facet *facet, *next;
817 /* Clear the revalidation flags. */
818 tag_set_init(&ofproto->revalidate_set);
819 ofproto->need_revalidate = false;
821 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
823 || tag_set_intersects(&revalidate_set, facet->tags)) {
824 facet_revalidate(ofproto, facet);
833 wait(struct ofproto *ofproto_)
835 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
836 struct ofport_dpif *ofport;
837 struct ofbundle *bundle;
839 if (!clogged && !list_is_empty(&ofproto->completions)) {
840 poll_immediate_wake();
843 dpif_wait(ofproto->dpif);
844 dpif_recv_wait(ofproto->dpif);
845 if (ofproto->sflow) {
846 dpif_sflow_wait(ofproto->sflow);
848 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
849 poll_immediate_wake();
851 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
854 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
857 if (ofproto->netflow) {
858 netflow_wait(ofproto->netflow);
860 mac_learning_wait(ofproto->ml);
862 if (ofproto->need_revalidate) {
863 /* Shouldn't happen, but if it does just go around again. */
864 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
865 poll_immediate_wake();
867 timer_wait(&ofproto->next_expiration);
872 flush(struct ofproto *ofproto_)
874 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
875 struct facet *facet, *next_facet;
877 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
878 /* Mark the facet as not installed so that facet_remove() doesn't
879 * bother trying to uninstall it. There is no point in uninstalling it
880 * individually since we are about to blow away all the facets with
881 * dpif_flow_flush(). */
882 struct subfacet *subfacet;
884 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
885 subfacet->installed = false;
886 subfacet->dp_packet_count = 0;
887 subfacet->dp_byte_count = 0;
889 facet_remove(ofproto, facet);
891 dpif_flow_flush(ofproto->dpif);
895 get_features(struct ofproto *ofproto_ OVS_UNUSED,
896 bool *arp_match_ip, uint32_t *actions)
898 *arp_match_ip = true;
899 *actions = ((1u << OFPAT_OUTPUT) |
900 (1u << OFPAT_SET_VLAN_VID) |
901 (1u << OFPAT_SET_VLAN_PCP) |
902 (1u << OFPAT_STRIP_VLAN) |
903 (1u << OFPAT_SET_DL_SRC) |
904 (1u << OFPAT_SET_DL_DST) |
905 (1u << OFPAT_SET_NW_SRC) |
906 (1u << OFPAT_SET_NW_DST) |
907 (1u << OFPAT_SET_NW_TOS) |
908 (1u << OFPAT_SET_TP_SRC) |
909 (1u << OFPAT_SET_TP_DST) |
910 (1u << OFPAT_ENQUEUE));
914 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
916 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
917 struct dpif_dp_stats s;
919 strcpy(ots->name, "classifier");
921 dpif_get_dp_stats(ofproto->dpif, &s);
922 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
923 put_32aligned_be64(&ots->matched_count,
924 htonll(s.n_hit + ofproto->n_matches));
927 static struct ofport *
930 struct ofport_dpif *port = xmalloc(sizeof *port);
935 port_dealloc(struct ofport *port_)
937 struct ofport_dpif *port = ofport_dpif_cast(port_);
942 port_construct(struct ofport *port_)
944 struct ofport_dpif *port = ofport_dpif_cast(port_);
945 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
947 ofproto->need_revalidate = true;
948 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
951 port->tag = tag_create_random();
952 port->may_enable = true;
953 port->stp_port = NULL;
954 port->stp_state = STP_DISABLED;
955 hmap_init(&port->priorities);
956 port->realdev_ofp_port = 0;
957 port->vlandev_vid = 0;
959 if (ofproto->sflow) {
960 dpif_sflow_add_port(ofproto->sflow, port_);
967 port_destruct(struct ofport *port_)
969 struct ofport_dpif *port = ofport_dpif_cast(port_);
970 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
972 ofproto->need_revalidate = true;
973 bundle_remove(port_);
974 set_cfm(port_, NULL);
975 if (ofproto->sflow) {
976 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
979 ofport_clear_priorities(port);
980 hmap_destroy(&port->priorities);
984 port_modified(struct ofport *port_)
986 struct ofport_dpif *port = ofport_dpif_cast(port_);
988 if (port->bundle && port->bundle->bond) {
989 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
994 port_reconfigured(struct ofport *port_, ovs_be32 old_config)
996 struct ofport_dpif *port = ofport_dpif_cast(port_);
997 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
998 ovs_be32 changed = old_config ^ port->up.opp.config;
1000 if (changed & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
1001 OFPPC_NO_FWD | OFPPC_NO_FLOOD)) {
1002 ofproto->need_revalidate = true;
1004 if (changed & htonl(OFPPC_NO_FLOOD) && port->bundle) {
1005 bundle_update(port->bundle);
1011 set_sflow(struct ofproto *ofproto_,
1012 const struct ofproto_sflow_options *sflow_options)
1014 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1015 struct dpif_sflow *ds = ofproto->sflow;
1017 if (sflow_options) {
1019 struct ofport_dpif *ofport;
1021 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1022 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1023 dpif_sflow_add_port(ds, &ofport->up);
1025 ofproto->need_revalidate = true;
1027 dpif_sflow_set_options(ds, sflow_options);
1030 dpif_sflow_destroy(ds);
1031 ofproto->need_revalidate = true;
1032 ofproto->sflow = NULL;
1039 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1041 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1048 struct ofproto_dpif *ofproto;
1050 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1051 ofproto->need_revalidate = true;
1052 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1055 if (cfm_configure(ofport->cfm, s)) {
1061 cfm_destroy(ofport->cfm);
1067 get_cfm_fault(const struct ofport *ofport_)
1069 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1071 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1075 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1078 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1081 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1088 /* Spanning Tree. */
1091 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1093 struct ofproto_dpif *ofproto = ofproto_;
1094 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1095 struct ofport_dpif *ofport;
1097 ofport = stp_port_get_aux(sp);
1099 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1100 ofproto->up.name, port_num);
1102 struct eth_header *eth = pkt->l2;
1104 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1105 if (eth_addr_is_zero(eth->eth_src)) {
1106 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1107 "with unknown MAC", ofproto->up.name, port_num);
1109 send_packet(ofport, pkt);
1115 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1117 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1119 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1121 /* Only revalidate flows if the configuration changed. */
1122 if (!s != !ofproto->stp) {
1123 ofproto->need_revalidate = true;
1127 if (!ofproto->stp) {
1128 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1129 send_bpdu_cb, ofproto);
1130 ofproto->stp_last_tick = time_msec();
1133 stp_set_bridge_id(ofproto->stp, s->system_id);
1134 stp_set_bridge_priority(ofproto->stp, s->priority);
1135 stp_set_hello_time(ofproto->stp, s->hello_time);
1136 stp_set_max_age(ofproto->stp, s->max_age);
1137 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1139 stp_destroy(ofproto->stp);
1140 ofproto->stp = NULL;
1147 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1149 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1153 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1154 s->designated_root = stp_get_designated_root(ofproto->stp);
1155 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1164 update_stp_port_state(struct ofport_dpif *ofport)
1166 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1167 enum stp_state state;
1169 /* Figure out new state. */
1170 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1174 if (ofport->stp_state != state) {
1178 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1179 netdev_get_name(ofport->up.netdev),
1180 stp_state_name(ofport->stp_state),
1181 stp_state_name(state));
1182 if (stp_learn_in_state(ofport->stp_state)
1183 != stp_learn_in_state(state)) {
1184 /* xxx Learning action flows should also be flushed. */
1185 mac_learning_flush(ofproto->ml);
1187 fwd_change = stp_forward_in_state(ofport->stp_state)
1188 != stp_forward_in_state(state);
1190 ofproto->need_revalidate = true;
1191 ofport->stp_state = state;
1192 ofport->stp_state_entered = time_msec();
1194 if (fwd_change && ofport->bundle) {
1195 bundle_update(ofport->bundle);
1198 /* Update the STP state bits in the OpenFlow port description. */
1199 of_state = (ofport->up.opp.state & htonl(~OFPPS_STP_MASK))
1200 | htonl(state == STP_LISTENING ? OFPPS_STP_LISTEN
1201 : state == STP_LEARNING ? OFPPS_STP_LEARN
1202 : state == STP_FORWARDING ? OFPPS_STP_FORWARD
1203 : state == STP_BLOCKING ? OFPPS_STP_BLOCK
1205 ofproto_port_set_state(&ofport->up, of_state);
1209 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1210 * caller is responsible for assigning STP port numbers and ensuring
1211 * there are no duplicates. */
1213 set_stp_port(struct ofport *ofport_,
1214 const struct ofproto_port_stp_settings *s)
1216 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1217 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1218 struct stp_port *sp = ofport->stp_port;
1220 if (!s || !s->enable) {
1222 ofport->stp_port = NULL;
1223 stp_port_disable(sp);
1224 update_stp_port_state(ofport);
1227 } else if (sp && stp_port_no(sp) != s->port_num
1228 && ofport == stp_port_get_aux(sp)) {
1229 /* The port-id changed, so disable the old one if it's not
1230 * already in use by another port. */
1231 stp_port_disable(sp);
1234 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1235 stp_port_enable(sp);
1237 stp_port_set_aux(sp, ofport);
1238 stp_port_set_priority(sp, s->priority);
1239 stp_port_set_path_cost(sp, s->path_cost);
1241 update_stp_port_state(ofport);
1247 get_stp_port_status(struct ofport *ofport_,
1248 struct ofproto_port_stp_status *s)
1250 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1251 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1252 struct stp_port *sp = ofport->stp_port;
1254 if (!ofproto->stp || !sp) {
1260 s->port_id = stp_port_get_id(sp);
1261 s->state = stp_port_get_state(sp);
1262 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1263 s->role = stp_port_get_role(sp);
1264 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1270 stp_run(struct ofproto_dpif *ofproto)
1273 long long int now = time_msec();
1274 long long int elapsed = now - ofproto->stp_last_tick;
1275 struct stp_port *sp;
1278 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1279 ofproto->stp_last_tick = now;
1281 while (stp_get_changed_port(ofproto->stp, &sp)) {
1282 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1285 update_stp_port_state(ofport);
1292 stp_wait(struct ofproto_dpif *ofproto)
1295 poll_timer_wait(1000);
1299 /* Returns true if STP should process 'flow'. */
1301 stp_should_process_flow(const struct flow *flow)
1303 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1307 stp_process_packet(const struct ofport_dpif *ofport,
1308 const struct ofpbuf *packet)
1310 struct ofpbuf payload = *packet;
1311 struct eth_header *eth = payload.data;
1312 struct stp_port *sp = ofport->stp_port;
1314 /* Sink packets on ports that have STP disabled when the bridge has
1316 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1320 /* Trim off padding on payload. */
1321 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1322 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1325 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1326 stp_received_bpdu(sp, payload.data, payload.size);
1330 static struct priority_to_dscp *
1331 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1333 struct priority_to_dscp *pdscp;
1336 hash = hash_int(priority, 0);
1337 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1338 if (pdscp->priority == priority) {
1346 ofport_clear_priorities(struct ofport_dpif *ofport)
1348 struct priority_to_dscp *pdscp, *next;
1350 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1351 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1357 set_queues(struct ofport *ofport_,
1358 const struct ofproto_port_queue *qdscp_list,
1361 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1362 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1363 struct hmap new = HMAP_INITIALIZER(&new);
1366 for (i = 0; i < n_qdscp; i++) {
1367 struct priority_to_dscp *pdscp;
1371 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1372 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1377 pdscp = get_priority(ofport, priority);
1379 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1381 pdscp = xmalloc(sizeof *pdscp);
1382 pdscp->priority = priority;
1384 ofproto->need_revalidate = true;
1387 if (pdscp->dscp != dscp) {
1389 ofproto->need_revalidate = true;
1392 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1395 if (!hmap_is_empty(&ofport->priorities)) {
1396 ofport_clear_priorities(ofport);
1397 ofproto->need_revalidate = true;
1400 hmap_swap(&new, &ofport->priorities);
1408 /* Expires all MAC learning entries associated with 'bundle' and forces its
1409 * ofproto to revalidate every flow.
1411 * Normally MAC learning entries are removed only from the ofproto associated
1412 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1413 * are removed from every ofproto. When patch ports and SLB bonds are in use
1414 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1415 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1416 * with the host from which it migrated. */
1418 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1420 struct ofproto_dpif *ofproto = bundle->ofproto;
1421 struct mac_learning *ml = ofproto->ml;
1422 struct mac_entry *mac, *next_mac;
1424 ofproto->need_revalidate = true;
1425 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1426 if (mac->port.p == bundle) {
1428 struct ofproto_dpif *o;
1430 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1432 struct mac_entry *e;
1434 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1437 tag_set_add(&o->revalidate_set, e->tag);
1438 mac_learning_expire(o->ml, e);
1444 mac_learning_expire(ml, mac);
1449 static struct ofbundle *
1450 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1452 struct ofbundle *bundle;
1454 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1455 &ofproto->bundles) {
1456 if (bundle->aux == aux) {
1463 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1464 * ones that are found to 'bundles'. */
1466 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1467 void **auxes, size_t n_auxes,
1468 struct hmapx *bundles)
1472 hmapx_init(bundles);
1473 for (i = 0; i < n_auxes; i++) {
1474 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1476 hmapx_add(bundles, bundle);
1482 bundle_update(struct ofbundle *bundle)
1484 struct ofport_dpif *port;
1486 bundle->floodable = true;
1487 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1488 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
1489 bundle->floodable = false;
1496 bundle_del_port(struct ofport_dpif *port)
1498 struct ofbundle *bundle = port->bundle;
1500 bundle->ofproto->need_revalidate = true;
1502 list_remove(&port->bundle_node);
1503 port->bundle = NULL;
1506 lacp_slave_unregister(bundle->lacp, port);
1509 bond_slave_unregister(bundle->bond, port);
1512 bundle_update(bundle);
1516 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1517 struct lacp_slave_settings *lacp,
1518 uint32_t bond_stable_id)
1520 struct ofport_dpif *port;
1522 port = get_ofp_port(bundle->ofproto, ofp_port);
1527 if (port->bundle != bundle) {
1528 bundle->ofproto->need_revalidate = true;
1530 bundle_del_port(port);
1533 port->bundle = bundle;
1534 list_push_back(&bundle->ports, &port->bundle_node);
1535 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
1536 bundle->floodable = false;
1540 port->bundle->ofproto->need_revalidate = true;
1541 lacp_slave_register(bundle->lacp, port, lacp);
1544 port->bond_stable_id = bond_stable_id;
1550 bundle_destroy(struct ofbundle *bundle)
1552 struct ofproto_dpif *ofproto;
1553 struct ofport_dpif *port, *next_port;
1560 ofproto = bundle->ofproto;
1561 for (i = 0; i < MAX_MIRRORS; i++) {
1562 struct ofmirror *m = ofproto->mirrors[i];
1564 if (m->out == bundle) {
1566 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1567 || hmapx_find_and_delete(&m->dsts, bundle)) {
1568 ofproto->need_revalidate = true;
1573 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1574 bundle_del_port(port);
1577 bundle_flush_macs(bundle, true);
1578 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1580 free(bundle->trunks);
1581 lacp_destroy(bundle->lacp);
1582 bond_destroy(bundle->bond);
1587 bundle_set(struct ofproto *ofproto_, void *aux,
1588 const struct ofproto_bundle_settings *s)
1590 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1591 bool need_flush = false;
1592 struct ofport_dpif *port;
1593 struct ofbundle *bundle;
1594 unsigned long *trunks;
1600 bundle_destroy(bundle_lookup(ofproto, aux));
1604 assert(s->n_slaves == 1 || s->bond != NULL);
1605 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1607 bundle = bundle_lookup(ofproto, aux);
1609 bundle = xmalloc(sizeof *bundle);
1611 bundle->ofproto = ofproto;
1612 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1613 hash_pointer(aux, 0));
1615 bundle->name = NULL;
1617 list_init(&bundle->ports);
1618 bundle->vlan_mode = PORT_VLAN_TRUNK;
1620 bundle->trunks = NULL;
1621 bundle->use_priority_tags = s->use_priority_tags;
1622 bundle->lacp = NULL;
1623 bundle->bond = NULL;
1625 bundle->floodable = true;
1627 bundle->src_mirrors = 0;
1628 bundle->dst_mirrors = 0;
1629 bundle->mirror_out = 0;
1632 if (!bundle->name || strcmp(s->name, bundle->name)) {
1634 bundle->name = xstrdup(s->name);
1639 if (!bundle->lacp) {
1640 ofproto->need_revalidate = true;
1641 bundle->lacp = lacp_create();
1643 lacp_configure(bundle->lacp, s->lacp);
1645 lacp_destroy(bundle->lacp);
1646 bundle->lacp = NULL;
1649 /* Update set of ports. */
1651 for (i = 0; i < s->n_slaves; i++) {
1652 if (!bundle_add_port(bundle, s->slaves[i],
1653 s->lacp ? &s->lacp_slaves[i] : NULL,
1654 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1658 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1659 struct ofport_dpif *next_port;
1661 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1662 for (i = 0; i < s->n_slaves; i++) {
1663 if (s->slaves[i] == port->up.ofp_port) {
1668 bundle_del_port(port);
1672 assert(list_size(&bundle->ports) <= s->n_slaves);
1674 if (list_is_empty(&bundle->ports)) {
1675 bundle_destroy(bundle);
1679 /* Set VLAN tagging mode */
1680 if (s->vlan_mode != bundle->vlan_mode
1681 || s->use_priority_tags != bundle->use_priority_tags) {
1682 bundle->vlan_mode = s->vlan_mode;
1683 bundle->use_priority_tags = s->use_priority_tags;
1688 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1689 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1691 if (vlan != bundle->vlan) {
1692 bundle->vlan = vlan;
1696 /* Get trunked VLANs. */
1697 switch (s->vlan_mode) {
1698 case PORT_VLAN_ACCESS:
1702 case PORT_VLAN_TRUNK:
1703 trunks = (unsigned long *) s->trunks;
1706 case PORT_VLAN_NATIVE_UNTAGGED:
1707 case PORT_VLAN_NATIVE_TAGGED:
1708 if (vlan != 0 && (!s->trunks
1709 || !bitmap_is_set(s->trunks, vlan)
1710 || bitmap_is_set(s->trunks, 0))) {
1711 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1713 trunks = bitmap_clone(s->trunks, 4096);
1715 trunks = bitmap_allocate1(4096);
1717 bitmap_set1(trunks, vlan);
1718 bitmap_set0(trunks, 0);
1720 trunks = (unsigned long *) s->trunks;
1727 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1728 free(bundle->trunks);
1729 if (trunks == s->trunks) {
1730 bundle->trunks = vlan_bitmap_clone(trunks);
1732 bundle->trunks = trunks;
1737 if (trunks != s->trunks) {
1742 if (!list_is_short(&bundle->ports)) {
1743 bundle->ofproto->has_bonded_bundles = true;
1745 if (bond_reconfigure(bundle->bond, s->bond)) {
1746 ofproto->need_revalidate = true;
1749 bundle->bond = bond_create(s->bond);
1750 ofproto->need_revalidate = true;
1753 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1754 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1758 bond_destroy(bundle->bond);
1759 bundle->bond = NULL;
1762 /* If we changed something that would affect MAC learning, un-learn
1763 * everything on this port and force flow revalidation. */
1765 bundle_flush_macs(bundle, false);
1772 bundle_remove(struct ofport *port_)
1774 struct ofport_dpif *port = ofport_dpif_cast(port_);
1775 struct ofbundle *bundle = port->bundle;
1778 bundle_del_port(port);
1779 if (list_is_empty(&bundle->ports)) {
1780 bundle_destroy(bundle);
1781 } else if (list_is_short(&bundle->ports)) {
1782 bond_destroy(bundle->bond);
1783 bundle->bond = NULL;
1789 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
1791 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1792 struct ofport_dpif *port = port_;
1793 uint8_t ea[ETH_ADDR_LEN];
1796 error = netdev_get_etheraddr(port->up.netdev, ea);
1798 struct ofpbuf packet;
1801 ofpbuf_init(&packet, 0);
1802 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1804 memcpy(packet_pdu, pdu, pdu_size);
1806 send_packet(port, &packet);
1807 ofpbuf_uninit(&packet);
1809 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1810 "%s (%s)", port->bundle->name,
1811 netdev_get_name(port->up.netdev), strerror(error));
1816 bundle_send_learning_packets(struct ofbundle *bundle)
1818 struct ofproto_dpif *ofproto = bundle->ofproto;
1819 int error, n_packets, n_errors;
1820 struct mac_entry *e;
1822 error = n_packets = n_errors = 0;
1823 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1824 if (e->port.p != bundle) {
1825 struct ofpbuf *learning_packet;
1826 struct ofport_dpif *port;
1830 /* The assignment to "port" is unnecessary but makes "grep"ing for
1831 * struct ofport_dpif more effective. */
1832 learning_packet = bond_compose_learning_packet(bundle->bond,
1836 ret = send_packet(port, learning_packet);
1837 ofpbuf_delete(learning_packet);
1847 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1848 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1849 "packets, last error was: %s",
1850 bundle->name, n_errors, n_packets, strerror(error));
1852 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1853 bundle->name, n_packets);
1858 bundle_run(struct ofbundle *bundle)
1861 lacp_run(bundle->lacp, send_pdu_cb);
1864 struct ofport_dpif *port;
1866 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1867 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1870 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1871 lacp_negotiated(bundle->lacp));
1872 if (bond_should_send_learning_packets(bundle->bond)) {
1873 bundle_send_learning_packets(bundle);
1879 bundle_wait(struct ofbundle *bundle)
1882 lacp_wait(bundle->lacp);
1885 bond_wait(bundle->bond);
1892 mirror_scan(struct ofproto_dpif *ofproto)
1896 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1897 if (!ofproto->mirrors[idx]) {
1904 static struct ofmirror *
1905 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1909 for (i = 0; i < MAX_MIRRORS; i++) {
1910 struct ofmirror *mirror = ofproto->mirrors[i];
1911 if (mirror && mirror->aux == aux) {
1919 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
1921 mirror_update_dups(struct ofproto_dpif *ofproto)
1925 for (i = 0; i < MAX_MIRRORS; i++) {
1926 struct ofmirror *m = ofproto->mirrors[i];
1929 m->dup_mirrors = MIRROR_MASK_C(1) << i;
1933 for (i = 0; i < MAX_MIRRORS; i++) {
1934 struct ofmirror *m1 = ofproto->mirrors[i];
1941 for (j = i + 1; j < MAX_MIRRORS; j++) {
1942 struct ofmirror *m2 = ofproto->mirrors[j];
1944 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
1945 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
1946 m2->dup_mirrors |= m1->dup_mirrors;
1953 mirror_set(struct ofproto *ofproto_, void *aux,
1954 const struct ofproto_mirror_settings *s)
1956 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1957 mirror_mask_t mirror_bit;
1958 struct ofbundle *bundle;
1959 struct ofmirror *mirror;
1960 struct ofbundle *out;
1961 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
1962 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
1965 mirror = mirror_lookup(ofproto, aux);
1967 mirror_destroy(mirror);
1973 idx = mirror_scan(ofproto);
1975 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
1977 ofproto->up.name, MAX_MIRRORS, s->name);
1981 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
1982 mirror->ofproto = ofproto;
1985 mirror->out_vlan = -1;
1986 mirror->name = NULL;
1989 if (!mirror->name || strcmp(s->name, mirror->name)) {
1991 mirror->name = xstrdup(s->name);
1994 /* Get the new configuration. */
1995 if (s->out_bundle) {
1996 out = bundle_lookup(ofproto, s->out_bundle);
1998 mirror_destroy(mirror);
2004 out_vlan = s->out_vlan;
2006 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2007 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2009 /* If the configuration has not changed, do nothing. */
2010 if (hmapx_equals(&srcs, &mirror->srcs)
2011 && hmapx_equals(&dsts, &mirror->dsts)
2012 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2013 && mirror->out == out
2014 && mirror->out_vlan == out_vlan)
2016 hmapx_destroy(&srcs);
2017 hmapx_destroy(&dsts);
2021 hmapx_swap(&srcs, &mirror->srcs);
2022 hmapx_destroy(&srcs);
2024 hmapx_swap(&dsts, &mirror->dsts);
2025 hmapx_destroy(&dsts);
2027 free(mirror->vlans);
2028 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2031 mirror->out_vlan = out_vlan;
2033 /* Update bundles. */
2034 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2035 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2036 if (hmapx_contains(&mirror->srcs, bundle)) {
2037 bundle->src_mirrors |= mirror_bit;
2039 bundle->src_mirrors &= ~mirror_bit;
2042 if (hmapx_contains(&mirror->dsts, bundle)) {
2043 bundle->dst_mirrors |= mirror_bit;
2045 bundle->dst_mirrors &= ~mirror_bit;
2048 if (mirror->out == bundle) {
2049 bundle->mirror_out |= mirror_bit;
2051 bundle->mirror_out &= ~mirror_bit;
2055 ofproto->need_revalidate = true;
2056 mac_learning_flush(ofproto->ml);
2057 mirror_update_dups(ofproto);
2063 mirror_destroy(struct ofmirror *mirror)
2065 struct ofproto_dpif *ofproto;
2066 mirror_mask_t mirror_bit;
2067 struct ofbundle *bundle;
2073 ofproto = mirror->ofproto;
2074 ofproto->need_revalidate = true;
2075 mac_learning_flush(ofproto->ml);
2077 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2078 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2079 bundle->src_mirrors &= ~mirror_bit;
2080 bundle->dst_mirrors &= ~mirror_bit;
2081 bundle->mirror_out &= ~mirror_bit;
2084 hmapx_destroy(&mirror->srcs);
2085 hmapx_destroy(&mirror->dsts);
2086 free(mirror->vlans);
2088 ofproto->mirrors[mirror->idx] = NULL;
2092 mirror_update_dups(ofproto);
2096 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2097 uint64_t *packets, uint64_t *bytes)
2099 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2100 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2103 *packets = *bytes = UINT64_MAX;
2107 *packets = mirror->packet_count;
2108 *bytes = mirror->byte_count;
2114 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2116 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2117 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2118 ofproto->need_revalidate = true;
2119 mac_learning_flush(ofproto->ml);
2125 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2127 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2128 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2129 return bundle && bundle->mirror_out != 0;
2133 forward_bpdu_changed(struct ofproto *ofproto_)
2135 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2136 /* Revalidate cached flows whenever forward_bpdu option changes. */
2137 ofproto->need_revalidate = true;
2142 static struct ofport_dpif *
2143 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
2145 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2146 return ofport ? ofport_dpif_cast(ofport) : NULL;
2149 static struct ofport_dpif *
2150 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
2152 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2156 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2157 struct dpif_port *dpif_port)
2159 ofproto_port->name = dpif_port->name;
2160 ofproto_port->type = dpif_port->type;
2161 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2165 port_run(struct ofport_dpif *ofport)
2167 bool enable = netdev_get_carrier(ofport->up.netdev);
2170 cfm_run(ofport->cfm);
2172 if (cfm_should_send_ccm(ofport->cfm)) {
2173 struct ofpbuf packet;
2175 ofpbuf_init(&packet, 0);
2176 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr);
2177 send_packet(ofport, &packet);
2178 ofpbuf_uninit(&packet);
2181 enable = enable && !cfm_get_fault(ofport->cfm)
2182 && cfm_get_opup(ofport->cfm);
2185 if (ofport->bundle) {
2186 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2189 if (ofport->may_enable != enable) {
2190 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2192 if (ofproto->has_bundle_action) {
2193 ofproto->need_revalidate = true;
2197 ofport->may_enable = enable;
2201 port_wait(struct ofport_dpif *ofport)
2204 cfm_wait(ofport->cfm);
2209 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2210 struct ofproto_port *ofproto_port)
2212 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2213 struct dpif_port dpif_port;
2216 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2218 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2224 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2226 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2230 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2232 *ofp_portp = odp_port_to_ofp_port(odp_port);
2238 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2240 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2243 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2245 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2247 /* The caller is going to close ofport->up.netdev. If this is a
2248 * bonded port, then the bond is using that netdev, so remove it
2249 * from the bond. The client will need to reconfigure everything
2250 * after deleting ports, so then the slave will get re-added. */
2251 bundle_remove(&ofport->up);
2258 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2260 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2263 error = netdev_get_stats(ofport->up.netdev, stats);
2265 if (!error && ofport->odp_port == OVSP_LOCAL) {
2266 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2268 /* ofproto->stats.tx_packets represents packets that we created
2269 * internally and sent to some port (e.g. packets sent with
2270 * send_packet()). Account for them as if they had come from
2271 * OFPP_LOCAL and got forwarded. */
2273 if (stats->rx_packets != UINT64_MAX) {
2274 stats->rx_packets += ofproto->stats.tx_packets;
2277 if (stats->rx_bytes != UINT64_MAX) {
2278 stats->rx_bytes += ofproto->stats.tx_bytes;
2281 /* ofproto->stats.rx_packets represents packets that were received on
2282 * some port and we processed internally and dropped (e.g. STP).
2283 * Account fro them as if they had been forwarded to OFPP_LOCAL. */
2285 if (stats->tx_packets != UINT64_MAX) {
2286 stats->tx_packets += ofproto->stats.rx_packets;
2289 if (stats->tx_bytes != UINT64_MAX) {
2290 stats->tx_bytes += ofproto->stats.rx_bytes;
2297 /* Account packets for LOCAL port. */
2299 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2300 size_t tx_size, size_t rx_size)
2302 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2305 ofproto->stats.rx_packets++;
2306 ofproto->stats.rx_bytes += rx_size;
2309 ofproto->stats.tx_packets++;
2310 ofproto->stats.tx_bytes += tx_size;
2314 struct port_dump_state {
2315 struct dpif_port_dump dump;
2320 port_dump_start(const struct ofproto *ofproto_, void **statep)
2322 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2323 struct port_dump_state *state;
2325 *statep = state = xmalloc(sizeof *state);
2326 dpif_port_dump_start(&state->dump, ofproto->dpif);
2327 state->done = false;
2332 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2333 struct ofproto_port *port)
2335 struct port_dump_state *state = state_;
2336 struct dpif_port dpif_port;
2338 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2339 ofproto_port_from_dpif_port(port, &dpif_port);
2342 int error = dpif_port_dump_done(&state->dump);
2344 return error ? error : EOF;
2349 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2351 struct port_dump_state *state = state_;
2354 dpif_port_dump_done(&state->dump);
2361 port_poll(const struct ofproto *ofproto_, char **devnamep)
2363 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2364 return dpif_port_poll(ofproto->dpif, devnamep);
2368 port_poll_wait(const struct ofproto *ofproto_)
2370 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2371 dpif_port_poll_wait(ofproto->dpif);
2375 port_is_lacp_current(const struct ofport *ofport_)
2377 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2378 return (ofport->bundle && ofport->bundle->lacp
2379 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2383 /* Upcall handling. */
2385 /* Flow miss batching.
2387 * Some dpifs implement operations faster when you hand them off in a batch.
2388 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2389 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2390 * more packets, plus possibly installing the flow in the dpif.
2392 * So far we only batch the operations that affect flow setup time the most.
2393 * It's possible to batch more than that, but the benefit might be minimal. */
2395 struct hmap_node hmap_node;
2397 enum odp_key_fitness key_fitness;
2398 const struct nlattr *key;
2400 ovs_be16 initial_tci;
2401 struct list packets;
2404 struct flow_miss_op {
2405 union dpif_op dpif_op;
2406 struct subfacet *subfacet;
2409 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2410 * OpenFlow controller as necessary according to their individual
2411 * configurations. */
2413 send_packet_in_miss(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
2414 const struct flow *flow)
2416 struct ofputil_packet_in pin;
2418 pin.packet = packet->data;
2419 pin.packet_len = packet->size;
2420 pin.total_len = packet->size;
2421 pin.reason = OFPR_NO_MATCH;
2426 pin.buffer_id = 0; /* not yet known */
2427 pin.send_len = 0; /* not used for flow table misses */
2429 flow_get_metadata(flow, &pin.fmd);
2431 /* Registers aren't meaningful on a miss. */
2432 memset(pin.fmd.reg_masks, 0, sizeof pin.fmd.reg_masks);
2434 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow);
2438 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2439 const struct ofpbuf *packet)
2441 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2447 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2449 cfm_process_heartbeat(ofport->cfm, packet);
2452 } else if (ofport->bundle && ofport->bundle->lacp
2453 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2455 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2458 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2460 stp_process_packet(ofport, packet);
2467 static struct flow_miss *
2468 flow_miss_create(struct hmap *todo, const struct flow *flow,
2469 enum odp_key_fitness key_fitness,
2470 const struct nlattr *key, size_t key_len,
2471 ovs_be16 initial_tci)
2473 uint32_t hash = flow_hash(flow, 0);
2474 struct flow_miss *miss;
2476 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2477 if (flow_equal(&miss->flow, flow)) {
2482 miss = xmalloc(sizeof *miss);
2483 hmap_insert(todo, &miss->hmap_node, hash);
2485 miss->key_fitness = key_fitness;
2487 miss->key_len = key_len;
2488 miss->initial_tci = initial_tci;
2489 list_init(&miss->packets);
2494 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2495 struct flow_miss_op *ops, size_t *n_ops)
2497 const struct flow *flow = &miss->flow;
2498 struct ofpbuf *packet, *next_packet;
2499 struct subfacet *subfacet;
2500 struct facet *facet;
2502 facet = facet_lookup_valid(ofproto, flow);
2504 struct rule_dpif *rule;
2506 rule = rule_dpif_lookup(ofproto, flow, 0);
2508 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
2509 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
2511 if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) {
2512 COVERAGE_INC(ofproto_dpif_no_packet_in);
2513 /* XXX install 'drop' flow entry */
2517 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
2521 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2522 send_packet_in_miss(ofproto, packet, flow);
2528 facet = facet_create(rule, flow);
2531 subfacet = subfacet_create(ofproto, facet,
2532 miss->key_fitness, miss->key, miss->key_len,
2535 LIST_FOR_EACH_SAFE (packet, next_packet, list_node, &miss->packets) {
2536 struct dpif_flow_stats stats;
2537 struct flow_miss_op *op;
2538 struct dpif_execute *execute;
2540 list_remove(&packet->list_node);
2541 ofproto->n_matches++;
2543 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2545 * Extra-special case for fail-open mode.
2547 * We are in fail-open mode and the packet matched the fail-open
2548 * rule, but we are connected to a controller too. We should send
2549 * the packet up to the controller in the hope that it will try to
2550 * set up a flow and thereby allow us to exit fail-open.
2552 * See the top-level comment in fail-open.c for more information.
2554 send_packet_in_miss(ofproto, packet, flow);
2557 if (!facet->may_install || !subfacet->actions) {
2558 subfacet_make_actions(ofproto, subfacet, packet);
2561 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2562 subfacet_update_stats(ofproto, subfacet, &stats);
2564 if (flow->vlan_tci != subfacet->initial_tci) {
2565 /* This packet was received on a VLAN splinter port. We added
2566 * a VLAN to the packet to make the packet resemble the flow,
2567 * but the actions were composed assuming that the packet
2568 * contained no VLAN. So, we must remove the VLAN header from
2569 * the packet before trying to execute the actions. */
2570 eth_pop_vlan(packet);
2573 op = &ops[(*n_ops)++];
2574 execute = &op->dpif_op.execute;
2575 op->subfacet = subfacet;
2576 execute->type = DPIF_OP_EXECUTE;
2577 execute->key = miss->key;
2578 execute->key_len = miss->key_len;
2579 execute->actions = (facet->may_install
2581 : xmemdup(subfacet->actions,
2582 subfacet->actions_len));
2583 execute->actions_len = subfacet->actions_len;
2584 execute->packet = packet;
2587 if (facet->may_install && subfacet->key_fitness != ODP_FIT_TOO_LITTLE) {
2588 struct flow_miss_op *op = &ops[(*n_ops)++];
2589 struct dpif_flow_put *put = &op->dpif_op.flow_put;
2591 op->subfacet = subfacet;
2592 put->type = DPIF_OP_FLOW_PUT;
2593 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2594 put->key = miss->key;
2595 put->key_len = miss->key_len;
2596 put->actions = subfacet->actions;
2597 put->actions_len = subfacet->actions_len;
2602 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
2603 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
2604 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
2605 * what a flow key should contain.
2607 * This function also includes some logic to help make VLAN splinters
2608 * transparent to the rest of the upcall processing logic. In particular, if
2609 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
2610 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
2611 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
2613 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
2614 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
2615 * (This differs from the value returned in flow->vlan_tci only for packets
2616 * received on VLAN splinters.)
2618 static enum odp_key_fitness
2619 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
2620 const struct nlattr *key, size_t key_len,
2621 struct flow *flow, ovs_be16 *initial_tci,
2622 struct ofpbuf *packet)
2624 enum odp_key_fitness fitness;
2628 fitness = odp_flow_key_to_flow(key, key_len, flow);
2629 if (fitness == ODP_FIT_ERROR) {
2632 *initial_tci = flow->vlan_tci;
2634 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
2636 /* Cause the flow to be processed as if it came in on the real device
2637 * with the VLAN device's VLAN ID. */
2638 flow->in_port = realdev;
2639 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
2641 /* Make the packet resemble the flow, so that it gets sent to an
2642 * OpenFlow controller properly, so that it looks correct for
2643 * sFlow, and so that flow_extract() will get the correct vlan_tci
2644 * if it is called on 'packet'.
2646 * The allocated space inside 'packet' probably also contains
2647 * 'key', that is, both 'packet' and 'key' are probably part of a
2648 * struct dpif_upcall (see the large comment on that structure
2649 * definition), so pushing data on 'packet' is in general not a
2650 * good idea since it could overwrite 'key' or free it as a side
2651 * effect. However, it's OK in this special case because we know
2652 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
2653 * will just overwrite the 4-byte "struct nlattr", which is fine
2654 * since we don't need that header anymore. */
2655 eth_push_vlan(packet, flow->vlan_tci);
2658 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
2659 if (fitness == ODP_FIT_PERFECT) {
2660 fitness = ODP_FIT_TOO_MUCH;
2668 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
2671 struct dpif_upcall *upcall;
2672 struct flow_miss *miss, *next_miss;
2673 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
2674 union dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
2683 /* Construct the to-do list.
2685 * This just amounts to extracting the flow from each packet and sticking
2686 * the packets that have the same flow in the same "flow_miss" structure so
2687 * that we can process them together. */
2689 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
2690 enum odp_key_fitness fitness;
2691 struct flow_miss *miss;
2692 ovs_be16 initial_tci;
2695 /* Obtain metadata and check userspace/kernel agreement on flow match,
2696 * then set 'flow''s header pointers. */
2697 fitness = ofproto_dpif_extract_flow_key(ofproto,
2698 upcall->key, upcall->key_len,
2699 &flow, &initial_tci,
2701 if (fitness == ODP_FIT_ERROR) {
2702 ofpbuf_delete(upcall->packet);
2705 flow_extract(upcall->packet, flow.skb_priority, flow.tun_id,
2706 flow.in_port, &flow);
2708 /* Handle 802.1ag, LACP, and STP specially. */
2709 if (process_special(ofproto, &flow, upcall->packet)) {
2710 ofproto_update_local_port_stats(&ofproto->up,
2711 0, upcall->packet->size);
2712 ofpbuf_delete(upcall->packet);
2713 ofproto->n_matches++;
2717 /* Add other packets to a to-do list. */
2718 miss = flow_miss_create(&todo, &flow, fitness,
2719 upcall->key, upcall->key_len, initial_tci);
2720 list_push_back(&miss->packets, &upcall->packet->list_node);
2723 /* Process each element in the to-do list, constructing the set of
2724 * operations to batch. */
2726 HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) {
2727 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
2728 ofpbuf_list_delete(&miss->packets);
2729 hmap_remove(&todo, &miss->hmap_node);
2732 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
2733 hmap_destroy(&todo);
2735 /* Execute batch. */
2736 for (i = 0; i < n_ops; i++) {
2737 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
2739 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
2741 /* Free memory and update facets. */
2742 for (i = 0; i < n_ops; i++) {
2743 struct flow_miss_op *op = &flow_miss_ops[i];
2744 struct dpif_execute *execute;
2745 struct dpif_flow_put *put;
2747 switch (op->dpif_op.type) {
2748 case DPIF_OP_EXECUTE:
2749 execute = &op->dpif_op.execute;
2750 if (op->subfacet->actions != execute->actions) {
2751 free((struct nlattr *) execute->actions);
2753 ofpbuf_delete((struct ofpbuf *) execute->packet);
2756 case DPIF_OP_FLOW_PUT:
2757 put = &op->dpif_op.flow_put;
2759 op->subfacet->installed = true;
2767 handle_userspace_upcall(struct ofproto_dpif *ofproto,
2768 struct dpif_upcall *upcall)
2770 struct user_action_cookie cookie;
2771 enum odp_key_fitness fitness;
2772 ovs_be16 initial_tci;
2775 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
2777 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
2778 upcall->key_len, &flow,
2779 &initial_tci, upcall->packet);
2780 if (fitness == ODP_FIT_ERROR) {
2781 ofpbuf_delete(upcall->packet);
2785 if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
2786 if (ofproto->sflow) {
2787 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
2791 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
2793 ofpbuf_delete(upcall->packet);
2797 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
2799 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
2803 assert (max_batch <= FLOW_MISS_MAX_BATCH);
2806 for (i = 0; i < max_batch; i++) {
2807 struct dpif_upcall *upcall = &misses[n_misses];
2810 error = dpif_recv(ofproto->dpif, upcall);
2815 switch (upcall->type) {
2816 case DPIF_UC_ACTION:
2817 handle_userspace_upcall(ofproto, upcall);
2821 /* Handle it later. */
2825 case DPIF_N_UC_TYPES:
2827 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32,
2833 handle_miss_upcalls(ofproto, misses, n_misses);
2838 /* Flow expiration. */
2840 static int subfacet_max_idle(const struct ofproto_dpif *);
2841 static void update_stats(struct ofproto_dpif *);
2842 static void rule_expire(struct rule_dpif *);
2843 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
2845 /* This function is called periodically by run(). Its job is to collect
2846 * updates for the flows that have been installed into the datapath, most
2847 * importantly when they last were used, and then use that information to
2848 * expire flows that have not been used recently.
2850 * Returns the number of milliseconds after which it should be called again. */
2852 expire(struct ofproto_dpif *ofproto)
2854 struct rule_dpif *rule, *next_rule;
2855 struct classifier *table;
2858 /* Update stats for each flow in the datapath. */
2859 update_stats(ofproto);
2861 /* Expire subfacets that have been idle too long. */
2862 dp_max_idle = subfacet_max_idle(ofproto);
2863 expire_subfacets(ofproto, dp_max_idle);
2865 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
2866 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
2867 struct cls_cursor cursor;
2869 cls_cursor_init(&cursor, table, NULL);
2870 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
2875 /* All outstanding data in existing flows has been accounted, so it's a
2876 * good time to do bond rebalancing. */
2877 if (ofproto->has_bonded_bundles) {
2878 struct ofbundle *bundle;
2880 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2882 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
2887 return MIN(dp_max_idle, 1000);
2890 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
2892 * This function also pushes statistics updates to rules which each facet
2893 * resubmits into. Generally these statistics will be accurate. However, if a
2894 * facet changes the rule it resubmits into at some time in between
2895 * update_stats() runs, it is possible that statistics accrued to the
2896 * old rule will be incorrectly attributed to the new rule. This could be
2897 * avoided by calling update_stats() whenever rules are created or
2898 * deleted. However, the performance impact of making so many calls to the
2899 * datapath do not justify the benefit of having perfectly accurate statistics.
2902 update_stats(struct ofproto_dpif *p)
2904 const struct dpif_flow_stats *stats;
2905 struct dpif_flow_dump dump;
2906 const struct nlattr *key;
2909 dpif_flow_dump_start(&dump, p->dpif);
2910 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
2911 struct subfacet *subfacet;
2913 subfacet = subfacet_find(p, key, key_len);
2914 if (subfacet && subfacet->installed) {
2915 struct facet *facet = subfacet->facet;
2917 if (stats->n_packets >= subfacet->dp_packet_count) {
2918 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
2919 facet->packet_count += extra;
2921 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
2924 if (stats->n_bytes >= subfacet->dp_byte_count) {
2925 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
2927 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
2930 subfacet->dp_packet_count = stats->n_packets;
2931 subfacet->dp_byte_count = stats->n_bytes;
2933 subfacet_update_time(p, subfacet, stats->used);
2934 facet_account(p, facet);
2935 facet_push_stats(facet);
2937 if (!VLOG_DROP_WARN(&rl)) {
2941 odp_flow_key_format(key, key_len, &s);
2942 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
2946 COVERAGE_INC(facet_unexpected);
2947 /* There's a flow in the datapath that we know nothing about, or a
2948 * flow that shouldn't be installed but was anyway. Delete it. */
2949 dpif_flow_del(p->dpif, key, key_len, NULL);
2952 dpif_flow_dump_done(&dump);
2955 /* Calculates and returns the number of milliseconds of idle time after which
2956 * subfacets should expire from the datapath. When a subfacet expires, we fold
2957 * its statistics into its facet, and when a facet's last subfacet expires, we
2958 * fold its statistic into its rule. */
2960 subfacet_max_idle(const struct ofproto_dpif *ofproto)
2963 * Idle time histogram.
2965 * Most of the time a switch has a relatively small number of subfacets.
2966 * When this is the case we might as well keep statistics for all of them
2967 * in userspace and to cache them in the kernel datapath for performance as
2970 * As the number of subfacets increases, the memory required to maintain
2971 * statistics about them in userspace and in the kernel becomes
2972 * significant. However, with a large number of subfacets it is likely
2973 * that only a few of them are "heavy hitters" that consume a large amount
2974 * of bandwidth. At this point, only heavy hitters are worth caching in
2975 * the kernel and maintaining in userspaces; other subfacets we can
2978 * The technique used to compute the idle time is to build a histogram with
2979 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
2980 * that is installed in the kernel gets dropped in the appropriate bucket.
2981 * After the histogram has been built, we compute the cutoff so that only
2982 * the most-recently-used 1% of subfacets (but at least
2983 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
2984 * the most-recently-used bucket of subfacets is kept, so actually an
2985 * arbitrary number of subfacets can be kept in any given expiration run
2986 * (though the next run will delete most of those unless they receive
2989 * This requires a second pass through the subfacets, in addition to the
2990 * pass made by update_stats(), because the former function never looks at
2991 * uninstallable subfacets.
2993 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
2994 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
2995 int buckets[N_BUCKETS] = { 0 };
2996 int total, subtotal, bucket;
2997 struct subfacet *subfacet;
3001 total = hmap_count(&ofproto->subfacets);
3002 if (total <= ofproto->up.flow_eviction_threshold) {
3003 return N_BUCKETS * BUCKET_WIDTH;
3006 /* Build histogram. */
3008 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3009 long long int idle = now - subfacet->used;
3010 int bucket = (idle <= 0 ? 0
3011 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3012 : (unsigned int) idle / BUCKET_WIDTH);
3016 /* Find the first bucket whose flows should be expired. */
3017 subtotal = bucket = 0;
3019 subtotal += buckets[bucket++];
3020 } while (bucket < N_BUCKETS &&
3021 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3023 if (VLOG_IS_DBG_ENABLED()) {
3027 ds_put_cstr(&s, "keep");
3028 for (i = 0; i < N_BUCKETS; i++) {
3030 ds_put_cstr(&s, ", drop");
3033 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3036 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3040 return bucket * BUCKET_WIDTH;
3044 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3046 long long int cutoff = time_msec() - dp_max_idle;
3047 struct subfacet *subfacet, *next_subfacet;
3049 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3050 &ofproto->subfacets) {
3051 if (subfacet->used < cutoff) {
3052 subfacet_destroy(ofproto, subfacet);
3057 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3058 * then delete it entirely. */
3060 rule_expire(struct rule_dpif *rule)
3062 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3063 struct facet *facet, *next_facet;
3067 /* Has 'rule' expired? */
3069 if (rule->up.hard_timeout
3070 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3071 reason = OFPRR_HARD_TIMEOUT;
3072 } else if (rule->up.idle_timeout && list_is_empty(&rule->facets)
3073 && now > rule->used + rule->up.idle_timeout * 1000) {
3074 reason = OFPRR_IDLE_TIMEOUT;
3079 COVERAGE_INC(ofproto_dpif_expired);
3081 /* Update stats. (This is a no-op if the rule expired due to an idle
3082 * timeout, because that only happens when the rule has no facets left.) */
3083 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3084 facet_remove(ofproto, facet);
3087 /* Get rid of the rule. */
3088 ofproto_rule_expire(&rule->up, reason);
3093 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3095 * The caller must already have determined that no facet with an identical
3096 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3097 * the ofproto's classifier table.
3099 * The facet will initially have no subfacets. The caller should create (at
3100 * least) one subfacet with subfacet_create(). */
3101 static struct facet *
3102 facet_create(struct rule_dpif *rule, const struct flow *flow)
3104 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3105 struct facet *facet;
3107 facet = xzalloc(sizeof *facet);
3108 facet->used = time_msec();
3109 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
3110 list_push_back(&rule->facets, &facet->list_node);
3112 facet->flow = *flow;
3113 list_init(&facet->subfacets);
3114 netflow_flow_init(&facet->nf_flow);
3115 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3121 facet_free(struct facet *facet)
3126 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3127 * 'packet', which arrived on 'in_port'.
3129 * Takes ownership of 'packet'. */
3131 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3132 const struct nlattr *odp_actions, size_t actions_len,
3133 struct ofpbuf *packet)
3135 struct odputil_keybuf keybuf;
3139 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3140 odp_flow_key_from_flow(&key, flow);
3142 error = dpif_execute(ofproto->dpif, key.data, key.size,
3143 odp_actions, actions_len, packet);
3145 ofpbuf_delete(packet);
3149 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3151 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3152 * rule's statistics, via subfacet_uninstall().
3154 * - Removes 'facet' from its rule and from ofproto->facets.
3157 facet_remove(struct ofproto_dpif *ofproto, struct facet *facet)
3159 struct subfacet *subfacet, *next_subfacet;
3161 assert(!list_is_empty(&facet->subfacets));
3163 /* First uninstall all of the subfacets to get final statistics. */
3164 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3165 subfacet_uninstall(ofproto, subfacet);
3168 /* Flush the final stats to the rule.
3170 * This might require us to have at least one subfacet around so that we
3171 * can use its actions for accounting in facet_account(), which is why we
3172 * have uninstalled but not yet destroyed the subfacets. */
3173 facet_flush_stats(ofproto, facet);
3175 /* Now we're really all done so destroy everything. */
3176 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3177 &facet->subfacets) {
3178 subfacet_destroy__(ofproto, subfacet);
3180 hmap_remove(&ofproto->facets, &facet->hmap_node);
3181 list_remove(&facet->list_node);
3186 facet_account(struct ofproto_dpif *ofproto, struct facet *facet)
3189 struct subfacet *subfacet;
3190 const struct nlattr *a;
3194 if (facet->byte_count <= facet->accounted_bytes) {
3197 n_bytes = facet->byte_count - facet->accounted_bytes;
3198 facet->accounted_bytes = facet->byte_count;
3200 /* Feed information from the active flows back into the learning table to
3201 * ensure that table is always in sync with what is actually flowing
3202 * through the datapath. */
3203 if (facet->has_learn || facet->has_normal) {
3204 struct action_xlate_ctx ctx;
3206 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3207 facet->flow.vlan_tci,
3208 facet->rule->up.flow_cookie, NULL);
3209 ctx.may_learn = true;
3210 ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions,
3211 facet->rule->up.n_actions));
3214 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3218 /* This loop feeds byte counters to bond_account() for rebalancing to use
3219 * as a basis. We also need to track the actual VLAN on which the packet
3220 * is going to be sent to ensure that it matches the one passed to
3221 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3224 * We use the actions from an arbitrary subfacet because they should all
3225 * be equally valid for our purpose. */
3226 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3227 struct subfacet, list_node);
3228 vlan_tci = facet->flow.vlan_tci;
3229 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3230 subfacet->actions, subfacet->actions_len) {
3231 const struct ovs_action_push_vlan *vlan;
3232 struct ofport_dpif *port;
3234 switch (nl_attr_type(a)) {
3235 case OVS_ACTION_ATTR_OUTPUT:
3236 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3237 if (port && port->bundle && port->bundle->bond) {
3238 bond_account(port->bundle->bond, &facet->flow,
3239 vlan_tci_to_vid(vlan_tci), n_bytes);
3243 case OVS_ACTION_ATTR_POP_VLAN:
3244 vlan_tci = htons(0);
3247 case OVS_ACTION_ATTR_PUSH_VLAN:
3248 vlan = nl_attr_get(a);
3249 vlan_tci = vlan->vlan_tci;
3255 /* Returns true if the only action for 'facet' is to send to the controller.
3256 * (We don't report NetFlow expiration messages for such facets because they
3257 * are just part of the control logic for the network, not real traffic). */
3259 facet_is_controller_flow(struct facet *facet)
3262 && facet->rule->up.n_actions == 1
3263 && action_outputs_to_port(&facet->rule->up.actions[0],
3264 htons(OFPP_CONTROLLER)));
3267 /* Folds all of 'facet''s statistics into its rule. Also updates the
3268 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3269 * 'facet''s statistics in the datapath should have been zeroed and folded into
3270 * its packet and byte counts before this function is called. */
3272 facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet)
3274 struct subfacet *subfacet;
3276 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3277 assert(!subfacet->dp_byte_count);
3278 assert(!subfacet->dp_packet_count);
3281 facet_push_stats(facet);
3282 facet_account(ofproto, facet);
3284 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3285 struct ofexpired expired;
3286 expired.flow = facet->flow;
3287 expired.packet_count = facet->packet_count;
3288 expired.byte_count = facet->byte_count;
3289 expired.used = facet->used;
3290 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3293 facet->rule->packet_count += facet->packet_count;
3294 facet->rule->byte_count += facet->byte_count;
3296 /* Reset counters to prevent double counting if 'facet' ever gets
3298 facet_reset_counters(facet);
3300 netflow_flow_clear(&facet->nf_flow);
3303 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3304 * Returns it if found, otherwise a null pointer.
3306 * The returned facet might need revalidation; use facet_lookup_valid()
3307 * instead if that is important. */
3308 static struct facet *
3309 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
3311 struct facet *facet;
3313 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
3315 if (flow_equal(flow, &facet->flow)) {
3323 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3324 * Returns it if found, otherwise a null pointer.
3326 * The returned facet is guaranteed to be valid. */
3327 static struct facet *
3328 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
3330 struct facet *facet = facet_find(ofproto, flow);
3332 /* The facet we found might not be valid, since we could be in need of
3333 * revalidation. If it is not valid, don't return it. */
3335 && (ofproto->need_revalidate
3336 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))
3337 && !facet_revalidate(ofproto, facet)) {
3338 COVERAGE_INC(facet_invalidated);
3345 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
3347 * - If the rule found is different from 'facet''s current rule, moves
3348 * 'facet' to the new rule and recompiles its actions.
3350 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3351 * where it is and recompiles its actions anyway.
3353 * - If there is none, destroys 'facet'.
3355 * Returns true if 'facet' still exists, false if it has been destroyed. */
3357 facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet)
3360 struct nlattr *odp_actions;
3363 struct actions *new_actions;
3365 struct action_xlate_ctx ctx;
3366 struct rule_dpif *new_rule;
3367 struct subfacet *subfacet;
3368 bool actions_changed;
3371 COVERAGE_INC(facet_revalidate);
3373 /* Determine the new rule. */
3374 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3376 /* No new rule, so delete the facet. */
3377 facet_remove(ofproto, facet);
3381 /* Calculate new datapath actions.
3383 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3384 * emit a NetFlow expiration and, if so, we need to have the old state
3385 * around to properly compose it. */
3387 /* If the datapath actions changed or the installability changed,
3388 * then we need to talk to the datapath. */
3391 memset(&ctx, 0, sizeof ctx);
3392 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3393 struct ofpbuf *odp_actions;
3394 bool should_install;
3396 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3397 subfacet->initial_tci, new_rule->up.flow_cookie,
3399 odp_actions = xlate_actions(&ctx, new_rule->up.actions,
3400 new_rule->up.n_actions);
3401 actions_changed = (subfacet->actions_len != odp_actions->size
3402 || memcmp(subfacet->actions, odp_actions->data,
3403 subfacet->actions_len));
3405 should_install = (ctx.may_set_up_flow
3406 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3407 if (actions_changed || should_install != subfacet->installed) {
3408 if (should_install) {
3409 struct dpif_flow_stats stats;
3411 subfacet_install(ofproto, subfacet,
3412 odp_actions->data, odp_actions->size, &stats);
3413 subfacet_update_stats(ofproto, subfacet, &stats);
3415 subfacet_uninstall(ofproto, subfacet);
3419 new_actions = xcalloc(list_size(&facet->subfacets),
3420 sizeof *new_actions);
3422 new_actions[i].odp_actions = xmemdup(odp_actions->data,
3424 new_actions[i].actions_len = odp_actions->size;
3427 ofpbuf_delete(odp_actions);
3431 facet_flush_stats(ofproto, facet);
3434 /* Update 'facet' now that we've taken care of all the old state. */
3435 facet->tags = ctx.tags;
3436 facet->nf_flow.output_iface = ctx.nf_output_iface;
3437 facet->may_install = ctx.may_set_up_flow;
3438 facet->has_learn = ctx.has_learn;
3439 facet->has_normal = ctx.has_normal;
3440 facet->mirrors = ctx.mirrors;
3443 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3444 if (new_actions[i].odp_actions) {
3445 free(subfacet->actions);
3446 subfacet->actions = new_actions[i].odp_actions;
3447 subfacet->actions_len = new_actions[i].actions_len;
3453 if (facet->rule != new_rule) {
3454 COVERAGE_INC(facet_changed_rule);
3455 list_remove(&facet->list_node);
3456 list_push_back(&new_rule->facets, &facet->list_node);
3457 facet->rule = new_rule;
3458 facet->used = new_rule->up.created;
3459 facet->prev_used = facet->used;
3465 /* Updates 'facet''s used time. Caller is responsible for calling
3466 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3468 facet_update_time(struct ofproto_dpif *ofproto, struct facet *facet,
3471 if (used > facet->used) {
3473 if (used > facet->rule->used) {
3474 facet->rule->used = used;
3476 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3481 facet_reset_counters(struct facet *facet)
3483 facet->packet_count = 0;
3484 facet->byte_count = 0;
3485 facet->prev_packet_count = 0;
3486 facet->prev_byte_count = 0;
3487 facet->accounted_bytes = 0;
3491 facet_push_stats(struct facet *facet)
3493 uint64_t new_packets, new_bytes;
3495 assert(facet->packet_count >= facet->prev_packet_count);
3496 assert(facet->byte_count >= facet->prev_byte_count);
3497 assert(facet->used >= facet->prev_used);
3499 new_packets = facet->packet_count - facet->prev_packet_count;
3500 new_bytes = facet->byte_count - facet->prev_byte_count;
3502 if (new_packets || new_bytes || facet->used > facet->prev_used) {
3503 facet->prev_packet_count = facet->packet_count;
3504 facet->prev_byte_count = facet->byte_count;
3505 facet->prev_used = facet->used;
3507 flow_push_stats(facet->rule, &facet->flow,
3508 new_packets, new_bytes, facet->used);
3510 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
3511 facet->mirrors, new_packets, new_bytes);
3515 struct ofproto_push {
3516 struct action_xlate_ctx ctx;
3523 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
3525 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3528 rule->packet_count += push->packets;
3529 rule->byte_count += push->bytes;
3530 rule->used = MAX(push->used, rule->used);
3534 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3535 * 'rule''s actions and mirrors. */
3537 flow_push_stats(const struct rule_dpif *rule,
3538 const struct flow *flow, uint64_t packets, uint64_t bytes,
3541 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3542 struct ofproto_push push;
3544 push.packets = packets;
3548 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci,
3549 rule->up.flow_cookie, NULL);
3550 push.ctx.resubmit_hook = push_resubmit;
3551 ofpbuf_delete(xlate_actions(&push.ctx,
3552 rule->up.actions, rule->up.n_actions));
3557 static struct subfacet *
3558 subfacet_find__(struct ofproto_dpif *ofproto,
3559 const struct nlattr *key, size_t key_len, uint32_t key_hash,
3560 const struct flow *flow)
3562 struct subfacet *subfacet;
3564 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
3565 &ofproto->subfacets) {
3567 ? (subfacet->key_len == key_len
3568 && !memcmp(key, subfacet->key, key_len))
3569 : flow_equal(flow, &subfacet->facet->flow)) {
3577 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
3578 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
3579 * there is one, otherwise creates and returns a new subfacet.
3581 * If the returned subfacet is new, then subfacet->actions will be NULL, in
3582 * which case the caller must populate the actions with
3583 * subfacet_make_actions(). */
3584 static struct subfacet *
3585 subfacet_create(struct ofproto_dpif *ofproto, struct facet *facet,
3586 enum odp_key_fitness key_fitness,
3587 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
3589 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3590 struct subfacet *subfacet;
3592 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
3594 if (subfacet->facet == facet) {
3598 /* This shouldn't happen. */
3599 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
3600 subfacet_destroy(ofproto, subfacet);
3603 subfacet = xzalloc(sizeof *subfacet);
3604 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
3605 list_push_back(&facet->subfacets, &subfacet->list_node);
3606 subfacet->facet = facet;
3607 subfacet->used = time_msec();
3608 subfacet->key_fitness = key_fitness;
3609 if (key_fitness != ODP_FIT_PERFECT) {
3610 subfacet->key = xmemdup(key, key_len);
3611 subfacet->key_len = key_len;
3613 subfacet->installed = false;
3614 subfacet->initial_tci = initial_tci;
3619 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
3620 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
3621 static struct subfacet *
3622 subfacet_find(struct ofproto_dpif *ofproto,
3623 const struct nlattr *key, size_t key_len)
3625 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3626 enum odp_key_fitness fitness;
3629 fitness = odp_flow_key_to_flow(key, key_len, &flow);
3630 if (fitness == ODP_FIT_ERROR) {
3634 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
3637 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
3638 * its facet within 'ofproto', and frees it. */
3640 subfacet_destroy__(struct ofproto_dpif *ofproto, struct subfacet *subfacet)
3642 subfacet_uninstall(ofproto, subfacet);
3643 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
3644 list_remove(&subfacet->list_node);
3645 free(subfacet->key);
3646 free(subfacet->actions);
3650 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
3651 * last remaining subfacet in its facet destroys the facet too. */
3653 subfacet_destroy(struct ofproto_dpif *ofproto, struct subfacet *subfacet)
3655 struct facet *facet = subfacet->facet;
3657 if (list_is_singleton(&facet->subfacets)) {
3658 /* facet_remove() needs at least one subfacet (it will remove it). */
3659 facet_remove(ofproto, facet);
3661 subfacet_destroy__(ofproto, subfacet);
3665 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
3666 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
3667 * for use as temporary storage. */
3669 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
3672 if (!subfacet->key) {
3673 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
3674 odp_flow_key_from_flow(key, &subfacet->facet->flow);
3676 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
3680 /* Composes the datapath actions for 'subfacet' based on its rule's actions. */
3682 subfacet_make_actions(struct ofproto_dpif *p, struct subfacet *subfacet,
3683 const struct ofpbuf *packet)
3685 struct facet *facet = subfacet->facet;
3686 const struct rule_dpif *rule = facet->rule;
3687 struct ofpbuf *odp_actions;
3688 struct action_xlate_ctx ctx;
3690 action_xlate_ctx_init(&ctx, p, &facet->flow, subfacet->initial_tci,
3691 rule->up.flow_cookie, packet);
3692 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3693 facet->tags = ctx.tags;
3694 facet->may_install = ctx.may_set_up_flow;
3695 facet->has_learn = ctx.has_learn;
3696 facet->has_normal = ctx.has_normal;
3697 facet->nf_flow.output_iface = ctx.nf_output_iface;
3698 facet->mirrors = ctx.mirrors;
3700 if (subfacet->actions_len != odp_actions->size
3701 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
3702 free(subfacet->actions);
3703 subfacet->actions_len = odp_actions->size;
3704 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
3707 ofpbuf_delete(odp_actions);
3710 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
3711 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
3712 * in the datapath will be zeroed and 'stats' will be updated with traffic new
3713 * since 'subfacet' was last updated.
3715 * Returns 0 if successful, otherwise a positive errno value. */
3717 subfacet_install(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3718 const struct nlattr *actions, size_t actions_len,
3719 struct dpif_flow_stats *stats)
3721 struct odputil_keybuf keybuf;
3722 enum dpif_flow_put_flags flags;
3726 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3728 flags |= DPIF_FP_ZERO_STATS;
3731 subfacet_get_key(subfacet, &keybuf, &key);
3732 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
3733 actions, actions_len, stats);
3736 subfacet_reset_dp_stats(subfacet, stats);
3742 /* If 'subfacet' is installed in the datapath, uninstalls it. */
3744 subfacet_uninstall(struct ofproto_dpif *p, struct subfacet *subfacet)
3746 if (subfacet->installed) {
3747 struct odputil_keybuf keybuf;
3748 struct dpif_flow_stats stats;
3752 subfacet_get_key(subfacet, &keybuf, &key);
3753 error = dpif_flow_del(p->dpif, key.data, key.size, &stats);
3754 subfacet_reset_dp_stats(subfacet, &stats);
3756 subfacet_update_stats(p, subfacet, &stats);
3758 subfacet->installed = false;
3760 assert(subfacet->dp_packet_count == 0);
3761 assert(subfacet->dp_byte_count == 0);
3765 /* Resets 'subfacet''s datapath statistics counters. This should be called
3766 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
3767 * non-null, it should contain the statistics returned by dpif when 'subfacet'
3768 * was reset in the datapath. 'stats' will be modified to include only
3769 * statistics new since 'subfacet' was last updated. */
3771 subfacet_reset_dp_stats(struct subfacet *subfacet,
3772 struct dpif_flow_stats *stats)
3775 && subfacet->dp_packet_count <= stats->n_packets
3776 && subfacet->dp_byte_count <= stats->n_bytes) {
3777 stats->n_packets -= subfacet->dp_packet_count;
3778 stats->n_bytes -= subfacet->dp_byte_count;
3781 subfacet->dp_packet_count = 0;
3782 subfacet->dp_byte_count = 0;
3785 /* Updates 'subfacet''s used time. The caller is responsible for calling
3786 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
3788 subfacet_update_time(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3791 if (used > subfacet->used) {
3792 subfacet->used = used;
3793 facet_update_time(ofproto, subfacet->facet, used);
3797 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
3799 * Because of the meaning of a subfacet's counters, it only makes sense to do
3800 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
3801 * represents a packet that was sent by hand or if it represents statistics
3802 * that have been cleared out of the datapath. */
3804 subfacet_update_stats(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3805 const struct dpif_flow_stats *stats)
3807 if (stats->n_packets || stats->used > subfacet->used) {
3808 struct facet *facet = subfacet->facet;
3810 subfacet_update_time(ofproto, subfacet, stats->used);
3811 facet->packet_count += stats->n_packets;
3812 facet->byte_count += stats->n_bytes;
3813 facet_push_stats(facet);
3814 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3820 static struct rule_dpif *
3821 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
3824 struct cls_rule *cls_rule;
3825 struct classifier *cls;
3827 if (table_id >= N_TABLES) {
3831 cls = &ofproto->up.tables[table_id];
3832 if (flow->nw_frag & FLOW_NW_FRAG_ANY
3833 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3834 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
3835 * are unavailable. */
3836 struct flow ofpc_normal_flow = *flow;
3837 ofpc_normal_flow.tp_src = htons(0);
3838 ofpc_normal_flow.tp_dst = htons(0);
3839 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
3841 cls_rule = classifier_lookup(cls, flow);
3843 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
3847 complete_operation(struct rule_dpif *rule)
3849 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3851 rule_invalidate(rule);
3853 struct dpif_completion *c = xmalloc(sizeof *c);
3854 c->op = rule->up.pending;
3855 list_push_back(&ofproto->completions, &c->list_node);
3857 ofoperation_complete(rule->up.pending, 0);
3861 static struct rule *
3864 struct rule_dpif *rule = xmalloc(sizeof *rule);
3869 rule_dealloc(struct rule *rule_)
3871 struct rule_dpif *rule = rule_dpif_cast(rule_);
3876 rule_construct(struct rule *rule_)
3878 struct rule_dpif *rule = rule_dpif_cast(rule_);
3879 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3880 struct rule_dpif *victim;
3884 error = validate_actions(rule->up.actions, rule->up.n_actions,
3885 &rule->up.cr.flow, ofproto->max_ports);
3890 rule->used = rule->up.created;
3891 rule->packet_count = 0;
3892 rule->byte_count = 0;
3894 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
3895 if (victim && !list_is_empty(&victim->facets)) {
3896 struct facet *facet;
3898 rule->facets = victim->facets;
3899 list_moved(&rule->facets);
3900 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3901 /* XXX: We're only clearing our local counters here. It's possible
3902 * that quite a few packets are unaccounted for in the datapath
3903 * statistics. These will be accounted to the new rule instead of
3904 * cleared as required. This could be fixed by clearing out the
3905 * datapath statistics for this facet, but currently it doesn't
3907 facet_reset_counters(facet);
3911 /* Must avoid list_moved() in this case. */
3912 list_init(&rule->facets);
3915 table_id = rule->up.table_id;
3916 rule->tag = (victim ? victim->tag
3918 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
3919 ofproto->tables[table_id].basis));
3921 complete_operation(rule);
3926 rule_destruct(struct rule *rule_)
3928 struct rule_dpif *rule = rule_dpif_cast(rule_);
3929 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3930 struct facet *facet, *next_facet;
3932 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3933 facet_revalidate(ofproto, facet);
3936 complete_operation(rule);
3940 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
3942 struct rule_dpif *rule = rule_dpif_cast(rule_);
3943 struct facet *facet;
3945 /* Start from historical data for 'rule' itself that are no longer tracked
3946 * in facets. This counts, for example, facets that have expired. */
3947 *packets = rule->packet_count;
3948 *bytes = rule->byte_count;
3950 /* Add any statistics that are tracked by facets. This includes
3951 * statistical data recently updated by ofproto_update_stats() as well as
3952 * stats for packets that were executed "by hand" via dpif_execute(). */
3953 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3954 *packets += facet->packet_count;
3955 *bytes += facet->byte_count;
3960 rule_execute(struct rule *rule_, const struct flow *flow,
3961 struct ofpbuf *packet)
3963 struct rule_dpif *rule = rule_dpif_cast(rule_);
3964 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3965 struct action_xlate_ctx ctx;
3966 struct ofpbuf *odp_actions;
3969 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
3970 rule->up.flow_cookie, packet);
3971 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3972 size = packet->size;
3973 if (execute_odp_actions(ofproto, flow, odp_actions->data,
3974 odp_actions->size, packet)) {
3975 rule->used = time_msec();
3976 rule->packet_count++;
3977 rule->byte_count += size;
3978 flow_push_stats(rule, flow, 1, size, rule->used);
3980 ofpbuf_delete(odp_actions);
3986 rule_modify_actions(struct rule *rule_)
3988 struct rule_dpif *rule = rule_dpif_cast(rule_);
3989 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3992 error = validate_actions(rule->up.actions, rule->up.n_actions,
3993 &rule->up.cr.flow, ofproto->max_ports);
3995 ofoperation_complete(rule->up.pending, error);
3999 complete_operation(rule);
4002 /* Sends 'packet' out 'ofport'.
4003 * May modify 'packet'.
4004 * Returns 0 if successful, otherwise a positive errno value. */
4006 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4008 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4009 struct ofpbuf key, odp_actions;
4010 struct odputil_keybuf keybuf;
4015 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
4016 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4018 if (odp_port != ofport->odp_port) {
4019 eth_pop_vlan(packet);
4020 flow.vlan_tci = htons(0);
4023 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4024 odp_flow_key_from_flow(&key, &flow);
4026 ofpbuf_init(&odp_actions, 32);
4027 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4029 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4030 error = dpif_execute(ofproto->dpif,
4032 odp_actions.data, odp_actions.size,
4034 ofpbuf_uninit(&odp_actions);
4037 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4038 ofproto->up.name, odp_port, strerror(error));
4040 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
4044 /* OpenFlow to datapath action translation. */
4046 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
4047 struct action_xlate_ctx *ctx);
4048 static void xlate_normal(struct action_xlate_ctx *);
4051 put_userspace_action(const struct ofproto_dpif *ofproto,
4052 struct ofpbuf *odp_actions,
4053 const struct flow *flow,
4054 const struct user_action_cookie *cookie)
4058 pid = dpif_port_get_pid(ofproto->dpif,
4059 ofp_port_to_odp_port(flow->in_port));
4061 return odp_put_userspace_action(pid, cookie, odp_actions);
4064 /* Compose SAMPLE action for sFlow. */
4066 compose_sflow_action(const struct ofproto_dpif *ofproto,
4067 struct ofpbuf *odp_actions,
4068 const struct flow *flow,
4071 uint32_t port_ifindex;
4072 uint32_t probability;
4073 struct user_action_cookie cookie;
4074 size_t sample_offset, actions_offset;
4075 int cookie_offset, n_output;
4077 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4081 if (odp_port == OVSP_NONE) {
4085 port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4089 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4091 /* Number of packets out of UINT_MAX to sample. */
4092 probability = dpif_sflow_get_probability(ofproto->sflow);
4093 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4095 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4097 cookie.type = USER_ACTION_COOKIE_SFLOW;
4098 cookie.data = port_ifindex;
4099 cookie.n_output = n_output;
4100 cookie.vlan_tci = 0;
4101 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4103 nl_msg_end_nested(odp_actions, actions_offset);
4104 nl_msg_end_nested(odp_actions, sample_offset);
4105 return cookie_offset;
4108 /* SAMPLE action must be first action in any given list of actions.
4109 * At this point we do not have all information required to build it. So try to
4110 * build sample action as complete as possible. */
4112 add_sflow_action(struct action_xlate_ctx *ctx)
4114 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4116 &ctx->flow, OVSP_NONE);
4117 ctx->sflow_odp_port = 0;
4118 ctx->sflow_n_outputs = 0;
4121 /* Fix SAMPLE action according to data collected while composing ODP actions.
4122 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4123 * USERSPACE action's user-cookie which is required for sflow. */
4125 fix_sflow_action(struct action_xlate_ctx *ctx)
4127 const struct flow *base = &ctx->base_flow;
4128 struct user_action_cookie *cookie;
4130 if (!ctx->user_cookie_offset) {
4134 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4136 assert(cookie != NULL);
4137 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4139 if (ctx->sflow_n_outputs) {
4140 cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow,
4141 ctx->sflow_odp_port);
4143 if (ctx->sflow_n_outputs >= 255) {
4144 cookie->n_output = 255;
4146 cookie->n_output = ctx->sflow_n_outputs;
4148 cookie->vlan_tci = base->vlan_tci;
4152 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4155 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4156 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4157 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4158 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4162 struct priority_to_dscp *pdscp;
4164 if (ofport->up.opp.config & htonl(OFPPC_NO_FWD)
4165 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4169 pdscp = get_priority(ofport, ctx->flow.skb_priority);
4171 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4172 ctx->flow.nw_tos |= pdscp->dscp;
4175 /* We may not have an ofport record for this port, but it doesn't hurt
4176 * to allow forwarding to it anyhow. Maybe such a port will appear
4177 * later and we're pre-populating the flow table. */
4180 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4181 ctx->flow.vlan_tci);
4182 if (out_port != odp_port) {
4183 ctx->flow.vlan_tci = htons(0);
4185 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4186 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4188 ctx->sflow_odp_port = odp_port;
4189 ctx->sflow_n_outputs++;
4190 ctx->nf_output_iface = ofp_port;
4191 ctx->flow.vlan_tci = flow_vlan_tci;
4192 ctx->flow.nw_tos = flow_nw_tos;
4196 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4198 compose_output_action__(ctx, ofp_port, true);
4202 xlate_table_action(struct action_xlate_ctx *ctx,
4203 uint16_t in_port, uint8_t table_id)
4205 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4206 struct ofproto_dpif *ofproto = ctx->ofproto;
4207 struct rule_dpif *rule;
4208 uint16_t old_in_port;
4209 uint8_t old_table_id;
4211 old_table_id = ctx->table_id;
4212 ctx->table_id = table_id;
4214 /* Look up a flow with 'in_port' as the input port. */
4215 old_in_port = ctx->flow.in_port;
4216 ctx->flow.in_port = in_port;
4217 rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
4220 if (table_id > 0 && table_id < N_TABLES) {
4221 struct table_dpif *table = &ofproto->tables[table_id];
4222 if (table->other_table) {
4225 : rule_calculate_tag(&ctx->flow,
4226 &table->other_table->wc,
4231 /* Restore the original input port. Otherwise OFPP_NORMAL and
4232 * OFPP_IN_PORT will have surprising behavior. */
4233 ctx->flow.in_port = old_in_port;
4235 if (ctx->resubmit_hook) {
4236 ctx->resubmit_hook(ctx, rule);
4240 ovs_be64 old_cookie = ctx->cookie;
4243 ctx->cookie = rule->up.flow_cookie;
4244 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
4245 ctx->cookie = old_cookie;
4249 ctx->table_id = old_table_id;
4251 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4253 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
4254 MAX_RESUBMIT_RECURSION);
4259 xlate_resubmit_table(struct action_xlate_ctx *ctx,
4260 const struct nx_action_resubmit *nar)
4265 in_port = (nar->in_port == htons(OFPP_IN_PORT)
4267 : ntohs(nar->in_port));
4268 table_id = nar->table == 255 ? ctx->table_id : nar->table;
4270 xlate_table_action(ctx, in_port, table_id);
4274 flood_packets(struct action_xlate_ctx *ctx, bool all)
4276 struct ofport_dpif *ofport;
4278 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
4279 uint16_t ofp_port = ofport->up.ofp_port;
4281 if (ofp_port == ctx->flow.in_port) {
4286 compose_output_action__(ctx, ofp_port, false);
4287 } else if (!(ofport->up.opp.config & htonl(OFPPC_NO_FLOOD))) {
4288 compose_output_action(ctx, ofp_port);
4292 ctx->nf_output_iface = NF_OUT_FLOOD;
4296 execute_controller_action(struct action_xlate_ctx *ctx, int len)
4298 struct ofputil_packet_in pin;
4299 struct ofpbuf *packet;
4301 ctx->may_set_up_flow = false;
4306 packet = ofpbuf_clone(ctx->packet);
4308 if (packet->l2 && packet->l3) {
4309 struct eth_header *eh;
4311 eth_pop_vlan(packet);
4313 assert(eh->eth_type == ctx->flow.dl_type);
4314 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
4315 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
4317 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
4318 eth_push_vlan(packet, ctx->flow.vlan_tci);
4322 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4323 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
4324 ctx->flow.nw_tos, ctx->flow.nw_ttl);
4328 if (ctx->flow.nw_proto == IPPROTO_TCP) {
4329 packet_set_tcp_port(packet, ctx->flow.tp_src,
4331 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
4332 packet_set_udp_port(packet, ctx->flow.tp_src,
4339 pin.packet = packet->data;
4340 pin.packet_len = packet->size;
4341 pin.reason = OFPR_ACTION;
4342 pin.table_id = ctx->table_id;
4343 pin.cookie = ctx->cookie;
4347 pin.total_len = packet->size;
4348 flow_get_metadata(&ctx->flow, &pin.fmd);
4350 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin, &ctx->flow);
4351 ofpbuf_delete(packet);
4355 xlate_output_action__(struct action_xlate_ctx *ctx,
4356 uint16_t port, uint16_t max_len)
4358 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
4360 ctx->nf_output_iface = NF_OUT_DROP;
4364 compose_output_action(ctx, ctx->flow.in_port);
4367 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
4373 flood_packets(ctx, false);
4376 flood_packets(ctx, true);
4378 case OFPP_CONTROLLER:
4379 execute_controller_action(ctx, max_len);
4382 compose_output_action(ctx, OFPP_LOCAL);
4387 if (port != ctx->flow.in_port) {
4388 compose_output_action(ctx, port);
4393 if (prev_nf_output_iface == NF_OUT_FLOOD) {
4394 ctx->nf_output_iface = NF_OUT_FLOOD;
4395 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
4396 ctx->nf_output_iface = prev_nf_output_iface;
4397 } else if (prev_nf_output_iface != NF_OUT_DROP &&
4398 ctx->nf_output_iface != NF_OUT_FLOOD) {
4399 ctx->nf_output_iface = NF_OUT_MULTI;
4404 xlate_output_reg_action(struct action_xlate_ctx *ctx,
4405 const struct nx_action_output_reg *naor)
4409 ofp_port = nxm_read_field_bits(naor->src, naor->ofs_nbits, &ctx->flow);
4411 if (ofp_port <= UINT16_MAX) {
4412 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
4417 xlate_output_action(struct action_xlate_ctx *ctx,
4418 const struct ofp_action_output *oao)
4420 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
4424 xlate_enqueue_action(struct action_xlate_ctx *ctx,
4425 const struct ofp_action_enqueue *oae)
4428 uint32_t flow_priority, priority;
4431 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
4434 /* Fall back to ordinary output action. */
4435 xlate_output_action__(ctx, ntohs(oae->port), 0);
4439 /* Figure out datapath output port. */
4440 ofp_port = ntohs(oae->port);
4441 if (ofp_port == OFPP_IN_PORT) {
4442 ofp_port = ctx->flow.in_port;
4443 } else if (ofp_port == ctx->flow.in_port) {
4447 /* Add datapath actions. */
4448 flow_priority = ctx->flow.skb_priority;
4449 ctx->flow.skb_priority = priority;
4450 compose_output_action(ctx, ofp_port);
4451 ctx->flow.skb_priority = flow_priority;
4453 /* Update NetFlow output port. */
4454 if (ctx->nf_output_iface == NF_OUT_DROP) {
4455 ctx->nf_output_iface = ofp_port;
4456 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
4457 ctx->nf_output_iface = NF_OUT_MULTI;
4462 xlate_set_queue_action(struct action_xlate_ctx *ctx,
4463 const struct nx_action_set_queue *nasq)
4468 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
4471 /* Couldn't translate queue to a priority, so ignore. A warning
4472 * has already been logged. */
4476 ctx->flow.skb_priority = priority;
4479 struct xlate_reg_state {
4485 xlate_autopath(struct action_xlate_ctx *ctx,
4486 const struct nx_action_autopath *naa)
4488 uint16_t ofp_port = ntohl(naa->id);
4489 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
4491 if (!port || !port->bundle) {
4492 ofp_port = OFPP_NONE;
4493 } else if (port->bundle->bond) {
4494 /* Autopath does not support VLAN hashing. */
4495 struct ofport_dpif *slave = bond_choose_output_slave(
4496 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
4498 ofp_port = slave->up.ofp_port;
4501 autopath_execute(naa, &ctx->flow, ofp_port);
4505 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
4507 struct ofproto_dpif *ofproto = ofproto_;
4508 struct ofport_dpif *port;
4518 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
4521 port = get_ofp_port(ofproto, ofp_port);
4522 return port ? port->may_enable : false;
4527 xlate_learn_action(struct action_xlate_ctx *ctx,
4528 const struct nx_action_learn *learn)
4530 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
4531 struct ofputil_flow_mod fm;
4534 learn_execute(learn, &ctx->flow, &fm);
4536 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
4537 if (error && !VLOG_DROP_WARN(&rl)) {
4538 char *msg = ofputil_error_to_string(error);
4539 VLOG_WARN("learning action failed to modify flow table (%s)", msg);
4547 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
4549 if (port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
4550 ? htonl(OFPPC_NO_RECV_STP)
4551 : htonl(OFPPC_NO_RECV))) {
4555 /* Only drop packets here if both forwarding and learning are
4556 * disabled. If just learning is enabled, we need to have
4557 * OFPP_NORMAL and the learning action have a look at the packet
4558 * before we can drop it. */
4559 if (!stp_forward_in_state(port->stp_state)
4560 && !stp_learn_in_state(port->stp_state)) {
4568 do_xlate_actions(const union ofp_action *in, size_t n_in,
4569 struct action_xlate_ctx *ctx)
4571 const struct ofport_dpif *port;
4572 const union ofp_action *ia;
4575 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
4576 if (port && !may_receive(port, ctx)) {
4577 /* Drop this flow. */
4581 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
4582 const struct ofp_action_dl_addr *oada;
4583 const struct nx_action_resubmit *nar;
4584 const struct nx_action_set_tunnel *nast;
4585 const struct nx_action_set_queue *nasq;
4586 const struct nx_action_multipath *nam;
4587 const struct nx_action_autopath *naa;
4588 const struct nx_action_bundle *nab;
4589 const struct nx_action_output_reg *naor;
4590 enum ofputil_action_code code;
4597 code = ofputil_decode_action_unsafe(ia);
4599 case OFPUTIL_OFPAT_OUTPUT:
4600 xlate_output_action(ctx, &ia->output);
4603 case OFPUTIL_OFPAT_SET_VLAN_VID:
4604 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
4605 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
4608 case OFPUTIL_OFPAT_SET_VLAN_PCP:
4609 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
4610 ctx->flow.vlan_tci |= htons(
4611 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
4614 case OFPUTIL_OFPAT_STRIP_VLAN:
4615 ctx->flow.vlan_tci = htons(0);
4618 case OFPUTIL_OFPAT_SET_DL_SRC:
4619 oada = ((struct ofp_action_dl_addr *) ia);
4620 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
4623 case OFPUTIL_OFPAT_SET_DL_DST:
4624 oada = ((struct ofp_action_dl_addr *) ia);
4625 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
4628 case OFPUTIL_OFPAT_SET_NW_SRC:
4629 ctx->flow.nw_src = ia->nw_addr.nw_addr;
4632 case OFPUTIL_OFPAT_SET_NW_DST:
4633 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
4636 case OFPUTIL_OFPAT_SET_NW_TOS:
4637 /* OpenFlow 1.0 only supports IPv4. */
4638 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4639 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4640 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
4644 case OFPUTIL_OFPAT_SET_TP_SRC:
4645 ctx->flow.tp_src = ia->tp_port.tp_port;
4648 case OFPUTIL_OFPAT_SET_TP_DST:
4649 ctx->flow.tp_dst = ia->tp_port.tp_port;
4652 case OFPUTIL_OFPAT_ENQUEUE:
4653 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
4656 case OFPUTIL_NXAST_RESUBMIT:
4657 nar = (const struct nx_action_resubmit *) ia;
4658 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
4661 case OFPUTIL_NXAST_RESUBMIT_TABLE:
4662 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
4665 case OFPUTIL_NXAST_SET_TUNNEL:
4666 nast = (const struct nx_action_set_tunnel *) ia;
4667 tun_id = htonll(ntohl(nast->tun_id));
4668 ctx->flow.tun_id = tun_id;
4671 case OFPUTIL_NXAST_SET_QUEUE:
4672 nasq = (const struct nx_action_set_queue *) ia;
4673 xlate_set_queue_action(ctx, nasq);
4676 case OFPUTIL_NXAST_POP_QUEUE:
4677 ctx->flow.skb_priority = ctx->orig_skb_priority;
4680 case OFPUTIL_NXAST_REG_MOVE:
4681 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
4685 case OFPUTIL_NXAST_REG_LOAD:
4686 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
4690 case OFPUTIL_NXAST_NOTE:
4691 /* Nothing to do. */
4694 case OFPUTIL_NXAST_SET_TUNNEL64:
4695 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
4696 ctx->flow.tun_id = tun_id;
4699 case OFPUTIL_NXAST_MULTIPATH:
4700 nam = (const struct nx_action_multipath *) ia;
4701 multipath_execute(nam, &ctx->flow);
4704 case OFPUTIL_NXAST_AUTOPATH:
4705 naa = (const struct nx_action_autopath *) ia;
4706 xlate_autopath(ctx, naa);
4709 case OFPUTIL_NXAST_BUNDLE:
4710 ctx->ofproto->has_bundle_action = true;
4711 nab = (const struct nx_action_bundle *) ia;
4712 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
4717 case OFPUTIL_NXAST_BUNDLE_LOAD:
4718 ctx->ofproto->has_bundle_action = true;
4719 nab = (const struct nx_action_bundle *) ia;
4720 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
4724 case OFPUTIL_NXAST_OUTPUT_REG:
4725 naor = (const struct nx_action_output_reg *) ia;
4726 xlate_output_reg_action(ctx, naor);
4729 case OFPUTIL_NXAST_LEARN:
4730 ctx->has_learn = true;
4731 if (ctx->may_learn) {
4732 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
4736 case OFPUTIL_NXAST_EXIT:
4742 /* We've let OFPP_NORMAL and the learning action look at the packet,
4743 * so drop it now if forwarding is disabled. */
4744 if (port && !stp_forward_in_state(port->stp_state)) {
4745 ofpbuf_clear(ctx->odp_actions);
4746 add_sflow_action(ctx);
4751 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
4752 struct ofproto_dpif *ofproto, const struct flow *flow,
4753 ovs_be16 initial_tci, ovs_be64 cookie,
4754 const struct ofpbuf *packet)
4756 ctx->ofproto = ofproto;
4758 ctx->base_flow = ctx->flow;
4759 ctx->base_flow.tun_id = 0;
4760 ctx->base_flow.vlan_tci = initial_tci;
4761 ctx->cookie = cookie;
4762 ctx->packet = packet;
4763 ctx->may_learn = packet != NULL;
4764 ctx->resubmit_hook = NULL;
4767 static struct ofpbuf *
4768 xlate_actions(struct action_xlate_ctx *ctx,
4769 const union ofp_action *in, size_t n_in)
4771 struct flow orig_flow = ctx->flow;
4773 COVERAGE_INC(ofproto_dpif_xlate);
4775 ctx->odp_actions = ofpbuf_new(512);
4776 ofpbuf_reserve(ctx->odp_actions, NL_A_U32_SIZE);
4778 ctx->may_set_up_flow = true;
4779 ctx->has_learn = false;
4780 ctx->has_normal = false;
4781 ctx->nf_output_iface = NF_OUT_DROP;
4784 ctx->orig_skb_priority = ctx->flow.skb_priority;
4788 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
4789 switch (ctx->ofproto->up.frag_handling) {
4790 case OFPC_FRAG_NORMAL:
4791 /* We must pretend that transport ports are unavailable. */
4792 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
4793 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
4796 case OFPC_FRAG_DROP:
4797 return ctx->odp_actions;
4799 case OFPC_FRAG_REASM:
4802 case OFPC_FRAG_NX_MATCH:
4803 /* Nothing to do. */
4808 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
4809 ctx->may_set_up_flow = false;
4810 return ctx->odp_actions;
4812 add_sflow_action(ctx);
4813 do_xlate_actions(in, n_in, ctx);
4815 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
4816 ctx->odp_actions->data,
4817 ctx->odp_actions->size)) {
4818 ctx->may_set_up_flow = false;
4820 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
4822 compose_output_action(ctx, OFPP_LOCAL);
4825 add_mirror_actions(ctx, &orig_flow);
4826 fix_sflow_action(ctx);
4829 return ctx->odp_actions;
4832 /* OFPP_NORMAL implementation. */
4834 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
4836 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
4837 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
4838 * the bundle on which the packet was received, returns the VLAN to which the
4841 * Both 'vid' and the return value are in the range 0...4095. */
4843 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
4845 switch (in_bundle->vlan_mode) {
4846 case PORT_VLAN_ACCESS:
4847 return in_bundle->vlan;
4850 case PORT_VLAN_TRUNK:
4853 case PORT_VLAN_NATIVE_UNTAGGED:
4854 case PORT_VLAN_NATIVE_TAGGED:
4855 return vid ? vid : in_bundle->vlan;
4862 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
4863 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
4866 * 'vid' should be the VID obtained from the 802.1Q header that was received as
4867 * part of a packet (specify 0 if there was no 802.1Q header), in the range
4870 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
4872 /* Allow any VID on the OFPP_NONE port. */
4873 if (in_bundle == &ofpp_none_bundle) {
4877 switch (in_bundle->vlan_mode) {
4878 case PORT_VLAN_ACCESS:
4881 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4882 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
4883 "packet received on port %s configured as VLAN "
4884 "%"PRIu16" access port",
4885 in_bundle->ofproto->up.name, vid,
4886 in_bundle->name, in_bundle->vlan);
4892 case PORT_VLAN_NATIVE_UNTAGGED:
4893 case PORT_VLAN_NATIVE_TAGGED:
4895 /* Port must always carry its native VLAN. */
4899 case PORT_VLAN_TRUNK:
4900 if (!ofbundle_includes_vlan(in_bundle, vid)) {
4902 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4903 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
4904 "received on port %s not configured for trunking "
4906 in_bundle->ofproto->up.name, vid,
4907 in_bundle->name, vid);
4919 /* Given 'vlan', the VLAN that a packet belongs to, and
4920 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
4921 * that should be included in the 802.1Q header. (If the return value is 0,
4922 * then the 802.1Q header should only be included in the packet if there is a
4925 * Both 'vlan' and the return value are in the range 0...4095. */
4927 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
4929 switch (out_bundle->vlan_mode) {
4930 case PORT_VLAN_ACCESS:
4933 case PORT_VLAN_TRUNK:
4934 case PORT_VLAN_NATIVE_TAGGED:
4937 case PORT_VLAN_NATIVE_UNTAGGED:
4938 return vlan == out_bundle->vlan ? 0 : vlan;
4946 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
4949 struct ofport_dpif *port;
4951 ovs_be16 tci, old_tci;
4953 vid = output_vlan_to_vid(out_bundle, vlan);
4954 if (!out_bundle->bond) {
4955 port = ofbundle_get_a_port(out_bundle);
4957 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
4960 /* No slaves enabled, so drop packet. */
4965 old_tci = ctx->flow.vlan_tci;
4967 if (tci || out_bundle->use_priority_tags) {
4968 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
4970 tci |= htons(VLAN_CFI);
4973 ctx->flow.vlan_tci = tci;
4975 compose_output_action(ctx, port->up.ofp_port);
4976 ctx->flow.vlan_tci = old_tci;
4980 mirror_mask_ffs(mirror_mask_t mask)
4982 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
4987 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
4989 return (bundle->vlan_mode != PORT_VLAN_ACCESS
4990 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
4994 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
4996 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
4999 /* Returns an arbitrary interface within 'bundle'. */
5000 static struct ofport_dpif *
5001 ofbundle_get_a_port(const struct ofbundle *bundle)
5003 return CONTAINER_OF(list_front(&bundle->ports),
5004 struct ofport_dpif, bundle_node);
5008 vlan_is_mirrored(const struct ofmirror *m, int vlan)
5010 return !m->vlans || bitmap_is_set(m->vlans, vlan);
5013 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
5014 * to a VLAN. In general most packets may be mirrored but we want to drop
5015 * protocols that may confuse switches. */
5017 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
5019 /* If you change this function's behavior, please update corresponding
5020 * documentation in vswitch.xml at the same time. */
5021 if (dst[0] != 0x01) {
5022 /* All the currently banned MACs happen to start with 01 currently, so
5023 * this is a quick way to eliminate most of the good ones. */
5025 if (eth_addr_is_reserved(dst)) {
5026 /* Drop STP, IEEE pause frames, and other reserved protocols
5027 * (01-80-c2-00-00-0x). */
5031 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
5033 if ((dst[3] & 0xfe) == 0xcc &&
5034 (dst[4] & 0xfe) == 0xcc &&
5035 (dst[5] & 0xfe) == 0xcc) {
5036 /* Drop the following protocols plus others following the same
5039 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
5040 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
5041 STP Uplink Fast (01-00-0c-cd-cd-cd) */
5045 if (!(dst[3] | dst[4] | dst[5])) {
5046 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
5055 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5057 struct ofproto_dpif *ofproto = ctx->ofproto;
5058 mirror_mask_t mirrors;
5059 struct ofbundle *in_bundle;
5062 const struct nlattr *a;
5065 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5066 ctx->packet != NULL);
5070 mirrors = in_bundle->src_mirrors;
5072 /* Drop frames on bundles reserved for mirroring. */
5073 if (in_bundle->mirror_out) {
5074 if (ctx->packet != NULL) {
5075 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5076 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5077 "%s, which is reserved exclusively for mirroring",
5078 ctx->ofproto->up.name, in_bundle->name);
5084 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5085 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5088 vlan = input_vid_to_vlan(in_bundle, vid);
5090 /* Look at the output ports to check for destination selections. */
5092 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5093 ctx->odp_actions->size) {
5094 enum ovs_action_attr type = nl_attr_type(a);
5095 struct ofport_dpif *ofport;
5097 if (type != OVS_ACTION_ATTR_OUTPUT) {
5101 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5102 if (ofport && ofport->bundle) {
5103 mirrors |= ofport->bundle->dst_mirrors;
5111 /* Restore the original packet before adding the mirror actions. */
5112 ctx->flow = *orig_flow;
5117 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5119 if (!vlan_is_mirrored(m, vlan)) {
5120 mirrors &= mirrors - 1;
5124 mirrors &= ~m->dup_mirrors;
5125 ctx->mirrors |= m->dup_mirrors;
5127 output_normal(ctx, m->out, vlan);
5128 } else if (eth_dst_may_rspan(orig_flow->dl_dst)
5129 && vlan != m->out_vlan) {
5130 struct ofbundle *bundle;
5132 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5133 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5134 && !bundle->mirror_out) {
5135 output_normal(ctx, bundle, m->out_vlan);
5143 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5144 uint64_t packets, uint64_t bytes)
5150 for (; mirrors; mirrors &= mirrors - 1) {
5153 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5156 /* In normal circumstances 'm' will not be NULL. However,
5157 * if mirrors are reconfigured, we can temporarily get out
5158 * of sync in facet_revalidate(). We could "correct" the
5159 * mirror list before reaching here, but doing that would
5160 * not properly account the traffic stats we've currently
5161 * accumulated for previous mirror configuration. */
5165 m->packet_count += packets;
5166 m->byte_count += bytes;
5170 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
5171 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
5172 * indicate this; newer upstream kernels use gratuitous ARP requests. */
5174 is_gratuitous_arp(const struct flow *flow)
5176 return (flow->dl_type == htons(ETH_TYPE_ARP)
5177 && eth_addr_is_broadcast(flow->dl_dst)
5178 && (flow->nw_proto == ARP_OP_REPLY
5179 || (flow->nw_proto == ARP_OP_REQUEST
5180 && flow->nw_src == flow->nw_dst)));
5184 update_learning_table(struct ofproto_dpif *ofproto,
5185 const struct flow *flow, int vlan,
5186 struct ofbundle *in_bundle)
5188 struct mac_entry *mac;
5190 /* Don't learn the OFPP_NONE port. */
5191 if (in_bundle == &ofpp_none_bundle) {
5195 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
5199 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
5200 if (is_gratuitous_arp(flow)) {
5201 /* We don't want to learn from gratuitous ARP packets that are
5202 * reflected back over bond slaves so we lock the learning table. */
5203 if (!in_bundle->bond) {
5204 mac_entry_set_grat_arp_lock(mac);
5205 } else if (mac_entry_is_grat_arp_locked(mac)) {
5210 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
5211 /* The log messages here could actually be useful in debugging,
5212 * so keep the rate limit relatively high. */
5213 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5214 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
5215 "on port %s in VLAN %d",
5216 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
5217 in_bundle->name, vlan);
5219 mac->port.p = in_bundle;
5220 tag_set_add(&ofproto->revalidate_set,
5221 mac_learning_changed(ofproto->ml, mac));
5225 static struct ofbundle *
5226 lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn)
5228 struct ofport_dpif *ofport;
5230 /* Special-case OFPP_NONE, which a controller may use as the ingress
5231 * port for traffic that it is sourcing. */
5232 if (in_port == OFPP_NONE) {
5233 return &ofpp_none_bundle;
5236 /* Find the port and bundle for the received packet. */
5237 ofport = get_ofp_port(ofproto, in_port);
5238 if (ofport && ofport->bundle) {
5239 return ofport->bundle;
5242 /* Odd. A few possible reasons here:
5244 * - We deleted a port but there are still a few packets queued up
5247 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
5248 * we don't know about.
5250 * - The ofproto client didn't configure the port as part of a bundle.
5253 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5255 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
5256 "port %"PRIu16, ofproto->up.name, in_port);
5261 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
5262 * dropped. Returns true if they may be forwarded, false if they should be
5265 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
5266 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
5268 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
5269 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
5270 * checked by input_vid_is_valid().
5272 * May also add tags to '*tags', although the current implementation only does
5273 * so in one special case.
5276 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
5277 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
5279 struct ofbundle *in_bundle = in_port->bundle;
5281 /* Drop frames for reserved multicast addresses
5282 * only if forward_bpdu option is absent. */
5283 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
5287 if (in_bundle->bond) {
5288 struct mac_entry *mac;
5290 switch (bond_check_admissibility(in_bundle->bond, in_port,
5291 flow->dl_dst, tags)) {
5298 case BV_DROP_IF_MOVED:
5299 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
5300 if (mac && mac->port.p != in_bundle &&
5301 (!is_gratuitous_arp(flow)
5302 || mac_entry_is_grat_arp_locked(mac))) {
5313 xlate_normal(struct action_xlate_ctx *ctx)
5315 struct ofport_dpif *in_port;
5316 struct ofbundle *in_bundle;
5317 struct mac_entry *mac;
5321 ctx->has_normal = true;
5323 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
5324 ctx->packet != NULL);
5329 /* We know 'in_port' exists unless it is "ofpp_none_bundle",
5330 * since lookup_input_bundle() succeeded. */
5331 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5333 /* Drop malformed frames. */
5334 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
5335 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
5336 if (ctx->packet != NULL) {
5337 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5338 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
5339 "VLAN tag received on port %s",
5340 ctx->ofproto->up.name, in_bundle->name);
5345 /* Drop frames on bundles reserved for mirroring. */
5346 if (in_bundle->mirror_out) {
5347 if (ctx->packet != NULL) {
5348 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5349 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5350 "%s, which is reserved exclusively for mirroring",
5351 ctx->ofproto->up.name, in_bundle->name);
5357 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
5358 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5361 vlan = input_vid_to_vlan(in_bundle, vid);
5363 /* Check other admissibility requirements. */
5365 !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
5369 /* Learn source MAC. */
5370 if (ctx->may_learn) {
5371 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
5374 /* Determine output bundle. */
5375 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
5378 if (mac->port.p != in_bundle) {
5379 output_normal(ctx, mac->port.p, vlan);
5382 struct ofbundle *bundle;
5384 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
5385 if (bundle != in_bundle
5386 && ofbundle_includes_vlan(bundle, vlan)
5387 && bundle->floodable
5388 && !bundle->mirror_out) {
5389 output_normal(ctx, bundle, vlan);
5392 ctx->nf_output_iface = NF_OUT_FLOOD;
5396 /* Optimized flow revalidation.
5398 * It's a difficult problem, in general, to tell which facets need to have
5399 * their actions recalculated whenever the OpenFlow flow table changes. We
5400 * don't try to solve that general problem: for most kinds of OpenFlow flow
5401 * table changes, we recalculate the actions for every facet. This is
5402 * relatively expensive, but it's good enough if the OpenFlow flow table
5403 * doesn't change very often.
5405 * However, we can expect one particular kind of OpenFlow flow table change to
5406 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5407 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5408 * table, we add a special case that applies to flow tables in which every rule
5409 * has the same form (that is, the same wildcards), except that the table is
5410 * also allowed to have a single "catch-all" flow that matches all packets. We
5411 * optimize this case by tagging all of the facets that resubmit into the table
5412 * and invalidating the same tag whenever a flow changes in that table. The
5413 * end result is that we revalidate just the facets that need it (and sometimes
5414 * a few more, but not all of the facets or even all of the facets that
5415 * resubmit to the table modified by MAC learning). */
5417 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
5418 * into an OpenFlow table with the given 'basis'. */
5420 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
5423 if (flow_wildcards_is_catchall(wc)) {
5426 struct flow tag_flow = *flow;
5427 flow_zero_wildcards(&tag_flow, wc);
5428 return tag_create_deterministic(flow_hash(&tag_flow, secret));
5432 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5433 * taggability of that table.
5435 * This function must be called after *each* change to a flow table. If you
5436 * skip calling it on some changes then the pointer comparisons at the end can
5437 * be invalid if you get unlucky. For example, if a flow removal causes a
5438 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5439 * different wildcards to be created with the same address, then this function
5440 * will incorrectly skip revalidation. */
5442 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5444 struct table_dpif *table = &ofproto->tables[table_id];
5445 const struct classifier *cls = &ofproto->up.tables[table_id];
5446 struct cls_table *catchall, *other;
5447 struct cls_table *t;
5449 catchall = other = NULL;
5451 switch (hmap_count(&cls->tables)) {
5453 /* We could tag this OpenFlow table but it would make the logic a
5454 * little harder and it's a corner case that doesn't seem worth it
5460 HMAP_FOR_EACH (t, hmap_node, &cls->tables) {
5461 if (cls_table_is_catchall(t)) {
5463 } else if (!other) {
5466 /* Indicate that we can't tag this by setting both tables to
5467 * NULL. (We know that 'catchall' is already NULL.) */
5474 /* Can't tag this table. */
5478 if (table->catchall_table != catchall || table->other_table != other) {
5479 table->catchall_table = catchall;
5480 table->other_table = other;
5481 ofproto->need_revalidate = true;
5485 /* Given 'rule' that has changed in some way (either it is a rule being
5486 * inserted, a rule being deleted, or a rule whose actions are being
5487 * modified), marks facets for revalidation to ensure that packets will be
5488 * forwarded correctly according to the new state of the flow table.
5490 * This function must be called after *each* change to a flow table. See
5491 * the comment on table_update_taggable() for more information. */
5493 rule_invalidate(const struct rule_dpif *rule)
5495 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5497 table_update_taggable(ofproto, rule->up.table_id);
5499 if (!ofproto->need_revalidate) {
5500 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5502 if (table->other_table && rule->tag) {
5503 tag_set_add(&ofproto->revalidate_set, rule->tag);
5505 ofproto->need_revalidate = true;
5511 set_frag_handling(struct ofproto *ofproto_,
5512 enum ofp_config_flags frag_handling)
5514 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5516 if (frag_handling != OFPC_FRAG_REASM) {
5517 ofproto->need_revalidate = true;
5525 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5526 const struct flow *flow,
5527 const union ofp_action *ofp_actions, size_t n_ofp_actions)
5529 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5532 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
5533 return ofp_mkerr_nicira(OFPET_BAD_REQUEST, NXBRC_BAD_IN_PORT);
5536 error = validate_actions(ofp_actions, n_ofp_actions, flow,
5537 ofproto->max_ports);
5539 struct odputil_keybuf keybuf;
5540 struct ofpbuf *odp_actions;
5541 struct ofproto_push push;
5544 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5545 odp_flow_key_from_flow(&key, flow);
5547 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, 0,
5550 /* Ensure that resubmits in 'ofp_actions' get accounted to their
5551 * matching rules. */
5553 push.bytes = packet->size;
5554 push.used = time_msec();
5555 push.ctx.resubmit_hook = push_resubmit;
5557 odp_actions = xlate_actions(&push.ctx, ofp_actions, n_ofp_actions);
5558 dpif_execute(ofproto->dpif, key.data, key.size,
5559 odp_actions->data, odp_actions->size, packet);
5560 ofpbuf_delete(odp_actions);
5568 set_netflow(struct ofproto *ofproto_,
5569 const struct netflow_options *netflow_options)
5571 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5573 if (netflow_options) {
5574 if (!ofproto->netflow) {
5575 ofproto->netflow = netflow_create();
5577 return netflow_set_options(ofproto->netflow, netflow_options);
5579 netflow_destroy(ofproto->netflow);
5580 ofproto->netflow = NULL;
5586 get_netflow_ids(const struct ofproto *ofproto_,
5587 uint8_t *engine_type, uint8_t *engine_id)
5589 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5591 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
5595 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5597 if (!facet_is_controller_flow(facet) &&
5598 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5599 struct subfacet *subfacet;
5600 struct ofexpired expired;
5602 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5603 if (subfacet->installed) {
5604 struct dpif_flow_stats stats;
5606 subfacet_install(ofproto, subfacet, subfacet->actions,
5607 subfacet->actions_len, &stats);
5608 subfacet_update_stats(ofproto, subfacet, &stats);
5612 expired.flow = facet->flow;
5613 expired.packet_count = facet->packet_count;
5614 expired.byte_count = facet->byte_count;
5615 expired.used = facet->used;
5616 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5621 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5623 struct facet *facet;
5625 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5626 send_active_timeout(ofproto, facet);
5630 static struct ofproto_dpif *
5631 ofproto_dpif_lookup(const char *name)
5633 struct ofproto_dpif *ofproto;
5635 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5636 hash_string(name, 0), &all_ofproto_dpifs) {
5637 if (!strcmp(ofproto->up.name, name)) {
5645 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc OVS_UNUSED,
5646 const char *argv[], void *aux OVS_UNUSED)
5648 const struct ofproto_dpif *ofproto;
5650 ofproto = ofproto_dpif_lookup(argv[1]);
5652 unixctl_command_reply(conn, 501, "no such bridge");
5655 mac_learning_flush(ofproto->ml);
5657 unixctl_command_reply(conn, 200, "table successfully flushed");
5661 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5662 const char *argv[], void *aux OVS_UNUSED)
5664 struct ds ds = DS_EMPTY_INITIALIZER;
5665 const struct ofproto_dpif *ofproto;
5666 const struct mac_entry *e;
5668 ofproto = ofproto_dpif_lookup(argv[1]);
5670 unixctl_command_reply(conn, 501, "no such bridge");
5674 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5675 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5676 struct ofbundle *bundle = e->port.p;
5677 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
5678 ofbundle_get_a_port(bundle)->odp_port,
5679 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
5681 unixctl_command_reply(conn, 200, ds_cstr(&ds));
5685 struct ofproto_trace {
5686 struct action_xlate_ctx ctx;
5692 trace_format_rule(struct ds *result, uint8_t table_id, int level,
5693 const struct rule_dpif *rule)
5695 ds_put_char_multiple(result, '\t', level);
5697 ds_put_cstr(result, "No match\n");
5701 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5702 table_id, ntohll(rule->up.flow_cookie));
5703 cls_rule_format(&rule->up.cr, result);
5704 ds_put_char(result, '\n');
5706 ds_put_char_multiple(result, '\t', level);
5707 ds_put_cstr(result, "OpenFlow ");
5708 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
5709 ds_put_char(result, '\n');
5713 trace_format_flow(struct ds *result, int level, const char *title,
5714 struct ofproto_trace *trace)
5716 ds_put_char_multiple(result, '\t', level);
5717 ds_put_format(result, "%s: ", title);
5718 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
5719 ds_put_cstr(result, "unchanged");
5721 flow_format(result, &trace->ctx.flow);
5722 trace->flow = trace->ctx.flow;
5724 ds_put_char(result, '\n');
5728 trace_format_regs(struct ds *result, int level, const char *title,
5729 struct ofproto_trace *trace)
5733 ds_put_char_multiple(result, '\t', level);
5734 ds_put_format(result, "%s:", title);
5735 for (i = 0; i < FLOW_N_REGS; i++) {
5736 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5738 ds_put_char(result, '\n');
5742 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
5744 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
5745 struct ds *result = trace->result;
5747 ds_put_char(result, '\n');
5748 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
5749 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
5750 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
5754 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5755 void *aux OVS_UNUSED)
5757 const char *dpname = argv[1];
5758 struct ofproto_dpif *ofproto;
5759 struct ofpbuf odp_key;
5760 struct ofpbuf *packet;
5761 struct rule_dpif *rule;
5762 ovs_be16 initial_tci;
5768 ofpbuf_init(&odp_key, 0);
5771 ofproto = ofproto_dpif_lookup(dpname);
5773 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
5777 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
5778 /* ofproto/trace dpname flow [-generate] */
5779 const char *flow_s = argv[2];
5780 const char *generate_s = argv[3];
5783 /* Convert string to datapath key. */
5784 ofpbuf_init(&odp_key, 0);
5785 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
5787 unixctl_command_reply(conn, 501, "Bad flow syntax");
5791 /* Convert odp_key to flow. */
5792 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
5793 odp_key.size, &flow,
5794 &initial_tci, NULL);
5795 if (error == ODP_FIT_ERROR) {
5796 unixctl_command_reply(conn, 501, "Invalid flow");
5800 /* Generate a packet, if requested. */
5802 packet = ofpbuf_new(0);
5803 flow_compose(packet, &flow);
5805 } else if (argc == 6) {
5806 /* ofproto/trace dpname priority tun_id in_port packet */
5807 const char *priority_s = argv[2];
5808 const char *tun_id_s = argv[3];
5809 const char *in_port_s = argv[4];
5810 const char *packet_s = argv[5];
5811 uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s));
5812 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
5813 uint32_t priority = atoi(priority_s);
5816 msg = eth_from_hex(packet_s, &packet);
5818 unixctl_command_reply(conn, 501, msg);
5822 ds_put_cstr(&result, "Packet: ");
5823 s = ofp_packet_to_string(packet->data, packet->size);
5824 ds_put_cstr(&result, s);
5827 flow_extract(packet, priority, tun_id, in_port, &flow);
5828 initial_tci = flow.vlan_tci;
5830 unixctl_command_reply(conn, 501, "Bad command syntax");
5834 ds_put_cstr(&result, "Flow: ");
5835 flow_format(&result, &flow);
5836 ds_put_char(&result, '\n');
5838 rule = rule_dpif_lookup(ofproto, &flow, 0);
5839 trace_format_rule(&result, 0, 0, rule);
5841 struct ofproto_trace trace;
5842 struct ofpbuf *odp_actions;
5844 trace.result = &result;
5846 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, initial_tci,
5847 rule->up.flow_cookie, packet);
5848 trace.ctx.resubmit_hook = trace_resubmit;
5849 odp_actions = xlate_actions(&trace.ctx,
5850 rule->up.actions, rule->up.n_actions);
5852 ds_put_char(&result, '\n');
5853 trace_format_flow(&result, 0, "Final flow", &trace);
5854 ds_put_cstr(&result, "Datapath actions: ");
5855 format_odp_actions(&result, odp_actions->data, odp_actions->size);
5856 ofpbuf_delete(odp_actions);
5858 if (!trace.ctx.may_set_up_flow) {
5860 ds_put_cstr(&result, "\nThis flow is not cachable.");
5862 ds_put_cstr(&result, "\nThe datapath actions are incomplete--"
5863 "for complete actions, please supply a packet.");
5868 unixctl_command_reply(conn, 200, ds_cstr(&result));
5871 ds_destroy(&result);
5872 ofpbuf_delete(packet);
5873 ofpbuf_uninit(&odp_key);
5877 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5878 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5881 unixctl_command_reply(conn, 200, NULL);
5885 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5886 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5889 unixctl_command_reply(conn, 200, NULL);
5893 ofproto_dpif_unixctl_init(void)
5895 static bool registered;
5901 unixctl_command_register(
5903 "bridge {tun_id in_port packet | odp_flow [-generate]}",
5904 2, 4, ofproto_unixctl_trace, NULL);
5905 unixctl_command_register("fdb/flush", "bridge", 1, 1,
5906 ofproto_unixctl_fdb_flush, NULL);
5907 unixctl_command_register("fdb/show", "bridge", 1, 1,
5908 ofproto_unixctl_fdb_show, NULL);
5909 unixctl_command_register("ofproto/clog", "", 0, 0,
5910 ofproto_dpif_clog, NULL);
5911 unixctl_command_register("ofproto/unclog", "", 0, 0,
5912 ofproto_dpif_unclog, NULL);
5915 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5917 * This is deprecated. It is only for compatibility with broken device drivers
5918 * in old versions of Linux that do not properly support VLANs when VLAN
5919 * devices are not used. When broken device drivers are no longer in
5920 * widespread use, we will delete these interfaces. */
5923 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
5925 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
5926 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
5928 if (realdev_ofp_port == ofport->realdev_ofp_port
5929 && vid == ofport->vlandev_vid) {
5933 ofproto->need_revalidate = true;
5935 if (ofport->realdev_ofp_port) {
5938 if (realdev_ofp_port && ofport->bundle) {
5939 /* vlandevs are enslaved to their realdevs, so they are not allowed to
5940 * themselves be part of a bundle. */
5941 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
5944 ofport->realdev_ofp_port = realdev_ofp_port;
5945 ofport->vlandev_vid = vid;
5947 if (realdev_ofp_port) {
5948 vsp_add(ofport, realdev_ofp_port, vid);
5955 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
5957 return hash_2words(realdev_ofp_port, vid);
5961 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
5962 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
5964 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
5965 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
5966 int vid = vlan_tci_to_vid(vlan_tci);
5967 const struct vlan_splinter *vsp;
5969 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
5970 hash_realdev_vid(realdev_ofp_port, vid),
5971 &ofproto->realdev_vid_map) {
5972 if (vsp->realdev_ofp_port == realdev_ofp_port
5973 && vsp->vid == vid) {
5974 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
5978 return realdev_odp_port;
5981 static struct vlan_splinter *
5982 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
5984 struct vlan_splinter *vsp;
5986 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
5987 &ofproto->vlandev_map) {
5988 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
5997 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
5998 uint16_t vlandev_ofp_port, int *vid)
6000 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6001 const struct vlan_splinter *vsp;
6003 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6008 return vsp->realdev_ofp_port;
6015 vsp_remove(struct ofport_dpif *port)
6017 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6018 struct vlan_splinter *vsp;
6020 vsp = vlandev_find(ofproto, port->up.ofp_port);
6022 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6023 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6026 port->realdev_ofp_port = 0;
6028 VLOG_ERR("missing vlan device record");
6033 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
6035 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6037 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6038 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6039 == realdev_ofp_port)) {
6040 struct vlan_splinter *vsp;
6042 vsp = xmalloc(sizeof *vsp);
6043 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6044 hash_int(port->up.ofp_port, 0));
6045 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6046 hash_realdev_vid(realdev_ofp_port, vid));
6047 vsp->realdev_ofp_port = realdev_ofp_port;
6048 vsp->vlandev_ofp_port = port->up.ofp_port;
6051 port->realdev_ofp_port = realdev_ofp_port;
6053 VLOG_ERR("duplicate vlan device record");
6057 const struct ofproto_class ofproto_dpif_class = {
6086 port_is_lacp_current,
6087 NULL, /* rule_choose_table */
6094 rule_modify_actions,
6102 get_cfm_remote_mpids,
6106 get_stp_port_status,
6113 is_mirror_output_bundle,
6114 forward_bpdu_changed,