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 /* If nonnull, called just before executing a resubmit action.
219 * This is normally null so the client has to set it manually after
220 * calling action_xlate_ctx_init(). */
221 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *);
223 /* xlate_actions() initializes and uses these members. The client might want
224 * to look at them after it returns. */
226 struct ofpbuf *odp_actions; /* Datapath actions. */
227 tag_type tags; /* Tags associated with actions. */
228 bool may_set_up_flow; /* True ordinarily; false if the actions must
229 * be reassessed for every packet. */
230 bool has_learn; /* Actions include NXAST_LEARN? */
231 bool has_normal; /* Actions output to OFPP_NORMAL? */
232 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
233 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
235 /* xlate_actions() initializes and uses these members, but the client has no
236 * reason to look at them. */
238 int recurse; /* Recursion level, via xlate_table_action. */
239 struct flow base_flow; /* Flow at the last commit. */
240 uint32_t orig_skb_priority; /* Priority when packet arrived. */
241 uint8_t table_id; /* OpenFlow table ID where flow was found. */
242 uint32_t sflow_n_outputs; /* Number of output ports. */
243 uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
244 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
245 bool exit; /* No further actions should be processed. */
248 static void action_xlate_ctx_init(struct action_xlate_ctx *,
249 struct ofproto_dpif *, const struct flow *,
250 ovs_be16 initial_tci, const struct ofpbuf *);
251 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
252 const union ofp_action *in, size_t n_in);
254 /* An exact-match instantiation of an OpenFlow flow.
256 * A facet associates a "struct flow", which represents the Open vSwitch
257 * userspace idea of an exact-match flow, with one or more subfacets. Each
258 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
259 * the facet. When the kernel module (or other dpif implementation) and Open
260 * vSwitch userspace agree on the definition of a flow key, there is exactly
261 * one subfacet per facet. If the dpif implementation supports more-specific
262 * flow matching than userspace, however, a facet can have more than one
263 * subfacet, each of which corresponds to some distinction in flow that
264 * userspace simply doesn't understand.
266 * Flow expiration works in terms of subfacets, so a facet must have at least
267 * one subfacet or it will never expire, leaking memory. */
270 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
271 struct list list_node; /* In owning rule's 'facets' list. */
272 struct rule_dpif *rule; /* Owning rule. */
275 struct list subfacets;
276 long long int used; /* Time last used; time created if not used. */
283 * - Do include packets and bytes sent "by hand", e.g. with
286 * - Do include packets and bytes that were obtained from the datapath
287 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
288 * DPIF_FP_ZERO_STATS).
290 * - Do not include packets or bytes that can be obtained from the
291 * datapath for any existing subfacet.
293 uint64_t packet_count; /* Number of packets received. */
294 uint64_t byte_count; /* Number of bytes received. */
296 /* Resubmit statistics. */
297 uint64_t prev_packet_count; /* Number of packets from last stats push. */
298 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
299 long long int prev_used; /* Used time from last stats push. */
302 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
303 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
305 /* Properties of datapath actions.
307 * Every subfacet has its own actions because actions can differ slightly
308 * between splintered and non-splintered subfacets due to the VLAN tag
309 * being initially different (present vs. absent). All of them have these
310 * properties in common so we just store one copy of them here. */
311 bool may_install; /* Reassess actions for every packet? */
312 bool has_learn; /* Actions include NXAST_LEARN? */
313 bool has_normal; /* Actions output to OFPP_NORMAL? */
314 tag_type tags; /* Tags that would require revalidation. */
315 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
318 static struct facet *facet_create(struct rule_dpif *, const struct flow *);
319 static void facet_remove(struct ofproto_dpif *, struct facet *);
320 static void facet_free(struct facet *);
322 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
323 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
324 const struct flow *);
325 static bool facet_revalidate(struct ofproto_dpif *, struct facet *);
327 static bool execute_controller_action(struct ofproto_dpif *,
329 const struct nlattr *odp_actions,
331 struct ofpbuf *packet);
333 static void facet_flush_stats(struct ofproto_dpif *, struct facet *);
335 static void facet_update_time(struct ofproto_dpif *, struct facet *,
337 static void facet_reset_counters(struct facet *);
338 static void facet_push_stats(struct facet *);
339 static void facet_account(struct ofproto_dpif *, struct facet *);
341 static bool facet_is_controller_flow(struct facet *);
343 /* A dpif flow and actions associated with a facet.
345 * See also the large comment on struct facet. */
348 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
349 struct list list_node; /* In struct facet's 'facets' list. */
350 struct facet *facet; /* Owning facet. */
354 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
355 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
356 * regenerate the ODP flow key from ->facet->flow. */
357 enum odp_key_fitness key_fitness;
361 long long int used; /* Time last used; time created if not used. */
363 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
364 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
368 * These should be essentially identical for every subfacet in a facet, but
369 * may differ in trivial ways due to VLAN splinters. */
370 size_t actions_len; /* Number of bytes in actions[]. */
371 struct nlattr *actions; /* Datapath actions. */
373 bool installed; /* Installed in datapath? */
375 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
376 * splinters can cause it to differ. This value should be removed when
377 * the VLAN splinters feature is no longer needed. */
378 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
381 static struct subfacet *subfacet_create(struct ofproto_dpif *, struct facet *,
382 enum odp_key_fitness,
383 const struct nlattr *key,
384 size_t key_len, ovs_be16 initial_tci);
385 static struct subfacet *subfacet_find(struct ofproto_dpif *,
386 const struct nlattr *key, size_t key_len);
387 static void subfacet_destroy(struct ofproto_dpif *, struct subfacet *);
388 static void subfacet_destroy__(struct ofproto_dpif *, struct subfacet *);
389 static void subfacet_reset_dp_stats(struct subfacet *,
390 struct dpif_flow_stats *);
391 static void subfacet_update_time(struct ofproto_dpif *, struct subfacet *,
393 static void subfacet_update_stats(struct ofproto_dpif *, struct subfacet *,
394 const struct dpif_flow_stats *);
395 static void subfacet_make_actions(struct ofproto_dpif *, struct subfacet *,
396 const struct ofpbuf *packet);
397 static int subfacet_install(struct ofproto_dpif *, struct subfacet *,
398 const struct nlattr *actions, size_t actions_len,
399 struct dpif_flow_stats *);
400 static void subfacet_uninstall(struct ofproto_dpif *, struct subfacet *);
406 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
407 struct list bundle_node; /* In struct ofbundle's "ports" list. */
408 struct cfm *cfm; /* Connectivity Fault Management, if any. */
409 tag_type tag; /* Tag associated with this port. */
410 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
411 bool may_enable; /* May be enabled in bonds. */
414 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
415 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
416 long long int stp_state_entered;
418 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
420 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
422 * This is deprecated. It is only for compatibility with broken device
423 * drivers in old versions of Linux that do not properly support VLANs when
424 * VLAN devices are not used. When broken device drivers are no longer in
425 * widespread use, we will delete these interfaces. */
426 uint16_t realdev_ofp_port;
430 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
431 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
432 * traffic egressing the 'ofport' with that priority should be marked with. */
433 struct priority_to_dscp {
434 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
435 uint32_t priority; /* Priority of this queue (see struct flow). */
437 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
440 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
442 * This is deprecated. It is only for compatibility with broken device drivers
443 * in old versions of Linux that do not properly support VLANs when VLAN
444 * devices are not used. When broken device drivers are no longer in
445 * widespread use, we will delete these interfaces. */
446 struct vlan_splinter {
447 struct hmap_node realdev_vid_node;
448 struct hmap_node vlandev_node;
449 uint16_t realdev_ofp_port;
450 uint16_t vlandev_ofp_port;
454 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
455 uint32_t realdev, ovs_be16 vlan_tci);
456 static uint16_t vsp_vlandev_to_realdev(const struct ofproto_dpif *,
457 uint16_t vlandev, int *vid);
458 static void vsp_remove(struct ofport_dpif *);
459 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
461 static struct ofport_dpif *
462 ofport_dpif_cast(const struct ofport *ofport)
464 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
465 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
468 static void port_run(struct ofport_dpif *);
469 static void port_wait(struct ofport_dpif *);
470 static int set_cfm(struct ofport *, const struct cfm_settings *);
471 static void ofport_clear_priorities(struct ofport_dpif *);
473 struct dpif_completion {
474 struct list list_node;
475 struct ofoperation *op;
478 /* Extra information about a classifier table.
479 * Currently used just for optimized flow revalidation. */
481 /* If either of these is nonnull, then this table has a form that allows
482 * flows to be tagged to avoid revalidating most flows for the most common
483 * kinds of flow table changes. */
484 struct cls_table *catchall_table; /* Table that wildcards all fields. */
485 struct cls_table *other_table; /* Table with any other wildcard set. */
486 uint32_t basis; /* Keeps each table's tags separate. */
489 struct ofproto_dpif {
490 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
499 struct netflow *netflow;
500 struct dpif_sflow *sflow;
501 struct hmap bundles; /* Contains "struct ofbundle"s. */
502 struct mac_learning *ml;
503 struct ofmirror *mirrors[MAX_MIRRORS];
504 bool has_bonded_bundles;
507 struct timer next_expiration;
511 struct hmap subfacets;
514 struct table_dpif tables[N_TABLES];
515 bool need_revalidate;
516 struct tag_set revalidate_set;
518 /* Support for debugging async flow mods. */
519 struct list completions;
521 bool has_bundle_action; /* True when the first bundle action appears. */
522 struct netdev_stats stats; /* To account packets generated and consumed in
527 long long int stp_last_tick;
529 /* VLAN splinters. */
530 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
531 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
534 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
535 * for debugging the asynchronous flow_mod implementation.) */
538 /* All existing ofproto_dpif instances, indexed by ->up.name. */
539 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
541 static void ofproto_dpif_unixctl_init(void);
543 static struct ofproto_dpif *
544 ofproto_dpif_cast(const struct ofproto *ofproto)
546 assert(ofproto->ofproto_class == &ofproto_dpif_class);
547 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
550 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
552 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
555 /* Packet processing. */
556 static void update_learning_table(struct ofproto_dpif *,
557 const struct flow *, int vlan,
560 #define FLOW_MISS_MAX_BATCH 50
561 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
563 /* Flow expiration. */
564 static int expire(struct ofproto_dpif *);
567 static void send_netflow_active_timeouts(struct ofproto_dpif *);
570 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
572 compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions,
573 const struct flow *, uint32_t odp_port);
574 static void add_mirror_actions(struct action_xlate_ctx *ctx,
575 const struct flow *flow);
576 /* Global variables. */
577 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
579 /* Factory functions. */
582 enumerate_types(struct sset *types)
584 dp_enumerate_types(types);
588 enumerate_names(const char *type, struct sset *names)
590 return dp_enumerate_names(type, names);
594 del(const char *type, const char *name)
599 error = dpif_open(name, type, &dpif);
601 error = dpif_delete(dpif);
607 /* Basic life-cycle. */
609 static struct ofproto *
612 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
617 dealloc(struct ofproto *ofproto_)
619 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
624 construct(struct ofproto *ofproto_, int *n_tablesp)
626 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
627 const char *name = ofproto->up.name;
631 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
633 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
637 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
638 ofproto->n_matches = 0;
640 dpif_flow_flush(ofproto->dpif);
641 dpif_recv_purge(ofproto->dpif);
643 error = dpif_recv_set_mask(ofproto->dpif,
644 ((1u << DPIF_UC_MISS) |
645 (1u << DPIF_UC_ACTION)));
647 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
648 dpif_close(ofproto->dpif);
652 ofproto->netflow = NULL;
653 ofproto->sflow = NULL;
655 hmap_init(&ofproto->bundles);
656 ofproto->ml = mac_learning_create();
657 for (i = 0; i < MAX_MIRRORS; i++) {
658 ofproto->mirrors[i] = NULL;
660 ofproto->has_bonded_bundles = false;
662 timer_set_duration(&ofproto->next_expiration, 1000);
664 hmap_init(&ofproto->facets);
665 hmap_init(&ofproto->subfacets);
667 for (i = 0; i < N_TABLES; i++) {
668 struct table_dpif *table = &ofproto->tables[i];
670 table->catchall_table = NULL;
671 table->other_table = NULL;
672 table->basis = random_uint32();
674 ofproto->need_revalidate = false;
675 tag_set_init(&ofproto->revalidate_set);
677 list_init(&ofproto->completions);
679 ofproto_dpif_unixctl_init();
681 ofproto->has_bundle_action = false;
683 hmap_init(&ofproto->vlandev_map);
684 hmap_init(&ofproto->realdev_vid_map);
686 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
687 hash_string(ofproto->up.name, 0));
689 *n_tablesp = N_TABLES;
690 memset(&ofproto->stats, 0, sizeof ofproto->stats);
695 complete_operations(struct ofproto_dpif *ofproto)
697 struct dpif_completion *c, *next;
699 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
700 ofoperation_complete(c->op, 0);
701 list_remove(&c->list_node);
707 destruct(struct ofproto *ofproto_)
709 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
710 struct rule_dpif *rule, *next_rule;
711 struct classifier *table;
714 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
715 complete_operations(ofproto);
717 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
718 struct cls_cursor cursor;
720 cls_cursor_init(&cursor, table, NULL);
721 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
722 ofproto_rule_destroy(&rule->up);
726 for (i = 0; i < MAX_MIRRORS; i++) {
727 mirror_destroy(ofproto->mirrors[i]);
730 netflow_destroy(ofproto->netflow);
731 dpif_sflow_destroy(ofproto->sflow);
732 hmap_destroy(&ofproto->bundles);
733 mac_learning_destroy(ofproto->ml);
735 hmap_destroy(&ofproto->facets);
736 hmap_destroy(&ofproto->subfacets);
738 hmap_destroy(&ofproto->vlandev_map);
739 hmap_destroy(&ofproto->realdev_vid_map);
741 dpif_close(ofproto->dpif);
745 run_fast(struct ofproto *ofproto_)
747 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
750 /* Handle one or more batches of upcalls, until there's nothing left to do
751 * or until we do a fixed total amount of work.
753 * We do work in batches because it can be much cheaper to set up a number
754 * of flows and fire off their patches all at once. We do multiple batches
755 * because in some cases handling a packet can cause another packet to be
756 * queued almost immediately as part of the return flow. Both
757 * optimizations can make major improvements on some benchmarks and
758 * presumably for real traffic as well. */
760 while (work < FLOW_MISS_MAX_BATCH) {
761 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
771 run(struct ofproto *ofproto_)
773 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
774 struct ofport_dpif *ofport;
775 struct ofbundle *bundle;
779 complete_operations(ofproto);
781 dpif_run(ofproto->dpif);
783 error = run_fast(ofproto_);
788 if (timer_expired(&ofproto->next_expiration)) {
789 int delay = expire(ofproto);
790 timer_set_duration(&ofproto->next_expiration, delay);
793 if (ofproto->netflow) {
794 if (netflow_run(ofproto->netflow)) {
795 send_netflow_active_timeouts(ofproto);
798 if (ofproto->sflow) {
799 dpif_sflow_run(ofproto->sflow);
802 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
805 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
810 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
812 /* Now revalidate if there's anything to do. */
813 if (ofproto->need_revalidate
814 || !tag_set_is_empty(&ofproto->revalidate_set)) {
815 struct tag_set revalidate_set = ofproto->revalidate_set;
816 bool revalidate_all = ofproto->need_revalidate;
817 struct facet *facet, *next;
819 /* Clear the revalidation flags. */
820 tag_set_init(&ofproto->revalidate_set);
821 ofproto->need_revalidate = false;
823 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
825 || tag_set_intersects(&revalidate_set, facet->tags)) {
826 facet_revalidate(ofproto, facet);
835 wait(struct ofproto *ofproto_)
837 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
838 struct ofport_dpif *ofport;
839 struct ofbundle *bundle;
841 if (!clogged && !list_is_empty(&ofproto->completions)) {
842 poll_immediate_wake();
845 dpif_wait(ofproto->dpif);
846 dpif_recv_wait(ofproto->dpif);
847 if (ofproto->sflow) {
848 dpif_sflow_wait(ofproto->sflow);
850 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
851 poll_immediate_wake();
853 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
856 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
859 if (ofproto->netflow) {
860 netflow_wait(ofproto->netflow);
862 mac_learning_wait(ofproto->ml);
864 if (ofproto->need_revalidate) {
865 /* Shouldn't happen, but if it does just go around again. */
866 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
867 poll_immediate_wake();
869 timer_wait(&ofproto->next_expiration);
874 flush(struct ofproto *ofproto_)
876 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
877 struct facet *facet, *next_facet;
879 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
880 /* Mark the facet as not installed so that facet_remove() doesn't
881 * bother trying to uninstall it. There is no point in uninstalling it
882 * individually since we are about to blow away all the facets with
883 * dpif_flow_flush(). */
884 struct subfacet *subfacet;
886 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
887 subfacet->installed = false;
888 subfacet->dp_packet_count = 0;
889 subfacet->dp_byte_count = 0;
891 facet_remove(ofproto, facet);
893 dpif_flow_flush(ofproto->dpif);
897 get_features(struct ofproto *ofproto_ OVS_UNUSED,
898 bool *arp_match_ip, uint32_t *actions)
900 *arp_match_ip = true;
901 *actions = ((1u << OFPAT_OUTPUT) |
902 (1u << OFPAT_SET_VLAN_VID) |
903 (1u << OFPAT_SET_VLAN_PCP) |
904 (1u << OFPAT_STRIP_VLAN) |
905 (1u << OFPAT_SET_DL_SRC) |
906 (1u << OFPAT_SET_DL_DST) |
907 (1u << OFPAT_SET_NW_SRC) |
908 (1u << OFPAT_SET_NW_DST) |
909 (1u << OFPAT_SET_NW_TOS) |
910 (1u << OFPAT_SET_TP_SRC) |
911 (1u << OFPAT_SET_TP_DST) |
912 (1u << OFPAT_ENQUEUE));
916 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
918 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
919 struct dpif_dp_stats s;
921 strcpy(ots->name, "classifier");
923 dpif_get_dp_stats(ofproto->dpif, &s);
924 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
925 put_32aligned_be64(&ots->matched_count,
926 htonll(s.n_hit + ofproto->n_matches));
929 static struct ofport *
932 struct ofport_dpif *port = xmalloc(sizeof *port);
937 port_dealloc(struct ofport *port_)
939 struct ofport_dpif *port = ofport_dpif_cast(port_);
944 port_construct(struct ofport *port_)
946 struct ofport_dpif *port = ofport_dpif_cast(port_);
947 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
949 ofproto->need_revalidate = true;
950 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
953 port->tag = tag_create_random();
954 port->may_enable = true;
955 port->stp_port = NULL;
956 port->stp_state = STP_DISABLED;
957 hmap_init(&port->priorities);
958 port->realdev_ofp_port = 0;
959 port->vlandev_vid = 0;
961 if (ofproto->sflow) {
962 dpif_sflow_add_port(ofproto->sflow, port_);
969 port_destruct(struct ofport *port_)
971 struct ofport_dpif *port = ofport_dpif_cast(port_);
972 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
974 ofproto->need_revalidate = true;
975 bundle_remove(port_);
976 set_cfm(port_, NULL);
977 if (ofproto->sflow) {
978 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
981 ofport_clear_priorities(port);
982 hmap_destroy(&port->priorities);
986 port_modified(struct ofport *port_)
988 struct ofport_dpif *port = ofport_dpif_cast(port_);
990 if (port->bundle && port->bundle->bond) {
991 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
996 port_reconfigured(struct ofport *port_, ovs_be32 old_config)
998 struct ofport_dpif *port = ofport_dpif_cast(port_);
999 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1000 ovs_be32 changed = old_config ^ port->up.opp.config;
1002 if (changed & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
1003 OFPPC_NO_FWD | OFPPC_NO_FLOOD)) {
1004 ofproto->need_revalidate = true;
1006 if (changed & htonl(OFPPC_NO_FLOOD) && port->bundle) {
1007 bundle_update(port->bundle);
1013 set_sflow(struct ofproto *ofproto_,
1014 const struct ofproto_sflow_options *sflow_options)
1016 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1017 struct dpif_sflow *ds = ofproto->sflow;
1019 if (sflow_options) {
1021 struct ofport_dpif *ofport;
1023 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1024 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1025 dpif_sflow_add_port(ds, &ofport->up);
1027 ofproto->need_revalidate = true;
1029 dpif_sflow_set_options(ds, sflow_options);
1032 dpif_sflow_destroy(ds);
1033 ofproto->need_revalidate = true;
1034 ofproto->sflow = NULL;
1041 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1043 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1050 struct ofproto_dpif *ofproto;
1052 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1053 ofproto->need_revalidate = true;
1054 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1057 if (cfm_configure(ofport->cfm, s)) {
1063 cfm_destroy(ofport->cfm);
1069 get_cfm_fault(const struct ofport *ofport_)
1071 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1073 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1077 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1080 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1083 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1090 /* Spanning Tree. */
1093 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1095 struct ofproto_dpif *ofproto = ofproto_;
1096 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1097 struct ofport_dpif *ofport;
1099 ofport = stp_port_get_aux(sp);
1101 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1102 ofproto->up.name, port_num);
1104 struct eth_header *eth = pkt->l2;
1106 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1107 if (eth_addr_is_zero(eth->eth_src)) {
1108 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1109 "with unknown MAC", ofproto->up.name, port_num);
1111 send_packet(ofport, pkt);
1117 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1119 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1121 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1123 /* Only revalidate flows if the configuration changed. */
1124 if (!s != !ofproto->stp) {
1125 ofproto->need_revalidate = true;
1129 if (!ofproto->stp) {
1130 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1131 send_bpdu_cb, ofproto);
1132 ofproto->stp_last_tick = time_msec();
1135 stp_set_bridge_id(ofproto->stp, s->system_id);
1136 stp_set_bridge_priority(ofproto->stp, s->priority);
1137 stp_set_hello_time(ofproto->stp, s->hello_time);
1138 stp_set_max_age(ofproto->stp, s->max_age);
1139 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1141 stp_destroy(ofproto->stp);
1142 ofproto->stp = NULL;
1149 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1151 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1155 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1156 s->designated_root = stp_get_designated_root(ofproto->stp);
1157 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1166 update_stp_port_state(struct ofport_dpif *ofport)
1168 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1169 enum stp_state state;
1171 /* Figure out new state. */
1172 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1176 if (ofport->stp_state != state) {
1180 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1181 netdev_get_name(ofport->up.netdev),
1182 stp_state_name(ofport->stp_state),
1183 stp_state_name(state));
1184 if (stp_learn_in_state(ofport->stp_state)
1185 != stp_learn_in_state(state)) {
1186 /* xxx Learning action flows should also be flushed. */
1187 mac_learning_flush(ofproto->ml);
1189 fwd_change = stp_forward_in_state(ofport->stp_state)
1190 != stp_forward_in_state(state);
1192 ofproto->need_revalidate = true;
1193 ofport->stp_state = state;
1194 ofport->stp_state_entered = time_msec();
1196 if (fwd_change && ofport->bundle) {
1197 bundle_update(ofport->bundle);
1200 /* Update the STP state bits in the OpenFlow port description. */
1201 of_state = (ofport->up.opp.state & htonl(~OFPPS_STP_MASK))
1202 | htonl(state == STP_LISTENING ? OFPPS_STP_LISTEN
1203 : state == STP_LEARNING ? OFPPS_STP_LEARN
1204 : state == STP_FORWARDING ? OFPPS_STP_FORWARD
1205 : state == STP_BLOCKING ? OFPPS_STP_BLOCK
1207 ofproto_port_set_state(&ofport->up, of_state);
1211 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1212 * caller is responsible for assigning STP port numbers and ensuring
1213 * there are no duplicates. */
1215 set_stp_port(struct ofport *ofport_,
1216 const struct ofproto_port_stp_settings *s)
1218 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1219 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1220 struct stp_port *sp = ofport->stp_port;
1222 if (!s || !s->enable) {
1224 ofport->stp_port = NULL;
1225 stp_port_disable(sp);
1226 update_stp_port_state(ofport);
1229 } else if (sp && stp_port_no(sp) != s->port_num
1230 && ofport == stp_port_get_aux(sp)) {
1231 /* The port-id changed, so disable the old one if it's not
1232 * already in use by another port. */
1233 stp_port_disable(sp);
1236 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1237 stp_port_enable(sp);
1239 stp_port_set_aux(sp, ofport);
1240 stp_port_set_priority(sp, s->priority);
1241 stp_port_set_path_cost(sp, s->path_cost);
1243 update_stp_port_state(ofport);
1249 get_stp_port_status(struct ofport *ofport_,
1250 struct ofproto_port_stp_status *s)
1252 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1253 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1254 struct stp_port *sp = ofport->stp_port;
1256 if (!ofproto->stp || !sp) {
1262 s->port_id = stp_port_get_id(sp);
1263 s->state = stp_port_get_state(sp);
1264 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1265 s->role = stp_port_get_role(sp);
1266 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1272 stp_run(struct ofproto_dpif *ofproto)
1275 long long int now = time_msec();
1276 long long int elapsed = now - ofproto->stp_last_tick;
1277 struct stp_port *sp;
1280 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1281 ofproto->stp_last_tick = now;
1283 while (stp_get_changed_port(ofproto->stp, &sp)) {
1284 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1287 update_stp_port_state(ofport);
1294 stp_wait(struct ofproto_dpif *ofproto)
1297 poll_timer_wait(1000);
1301 /* Returns true if STP should process 'flow'. */
1303 stp_should_process_flow(const struct flow *flow)
1305 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1309 stp_process_packet(const struct ofport_dpif *ofport,
1310 const struct ofpbuf *packet)
1312 struct ofpbuf payload = *packet;
1313 struct eth_header *eth = payload.data;
1314 struct stp_port *sp = ofport->stp_port;
1316 /* Sink packets on ports that have STP disabled when the bridge has
1318 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1322 /* Trim off padding on payload. */
1323 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1324 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1327 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1328 stp_received_bpdu(sp, payload.data, payload.size);
1332 static struct priority_to_dscp *
1333 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1335 struct priority_to_dscp *pdscp;
1338 hash = hash_int(priority, 0);
1339 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1340 if (pdscp->priority == priority) {
1348 ofport_clear_priorities(struct ofport_dpif *ofport)
1350 struct priority_to_dscp *pdscp, *next;
1352 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1353 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1359 set_queues(struct ofport *ofport_,
1360 const struct ofproto_port_queue *qdscp_list,
1363 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1364 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1365 struct hmap new = HMAP_INITIALIZER(&new);
1368 for (i = 0; i < n_qdscp; i++) {
1369 struct priority_to_dscp *pdscp;
1373 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1374 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1379 pdscp = get_priority(ofport, priority);
1381 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1383 pdscp = xmalloc(sizeof *pdscp);
1384 pdscp->priority = priority;
1386 ofproto->need_revalidate = true;
1389 if (pdscp->dscp != dscp) {
1391 ofproto->need_revalidate = true;
1394 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1397 if (!hmap_is_empty(&ofport->priorities)) {
1398 ofport_clear_priorities(ofport);
1399 ofproto->need_revalidate = true;
1402 hmap_swap(&new, &ofport->priorities);
1410 /* Expires all MAC learning entries associated with 'bundle' and forces its
1411 * ofproto to revalidate every flow.
1413 * Normally MAC learning entries are removed only from the ofproto associated
1414 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1415 * are removed from every ofproto. When patch ports and SLB bonds are in use
1416 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1417 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1418 * with the host from which it migrated. */
1420 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1422 struct ofproto_dpif *ofproto = bundle->ofproto;
1423 struct mac_learning *ml = ofproto->ml;
1424 struct mac_entry *mac, *next_mac;
1426 ofproto->need_revalidate = true;
1427 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1428 if (mac->port.p == bundle) {
1430 struct ofproto_dpif *o;
1432 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1434 struct mac_entry *e;
1436 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1439 tag_set_add(&o->revalidate_set, e->tag);
1440 mac_learning_expire(o->ml, e);
1446 mac_learning_expire(ml, mac);
1451 static struct ofbundle *
1452 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1454 struct ofbundle *bundle;
1456 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1457 &ofproto->bundles) {
1458 if (bundle->aux == aux) {
1465 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1466 * ones that are found to 'bundles'. */
1468 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1469 void **auxes, size_t n_auxes,
1470 struct hmapx *bundles)
1474 hmapx_init(bundles);
1475 for (i = 0; i < n_auxes; i++) {
1476 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1478 hmapx_add(bundles, bundle);
1484 bundle_update(struct ofbundle *bundle)
1486 struct ofport_dpif *port;
1488 bundle->floodable = true;
1489 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1490 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
1491 bundle->floodable = false;
1498 bundle_del_port(struct ofport_dpif *port)
1500 struct ofbundle *bundle = port->bundle;
1502 bundle->ofproto->need_revalidate = true;
1504 list_remove(&port->bundle_node);
1505 port->bundle = NULL;
1508 lacp_slave_unregister(bundle->lacp, port);
1511 bond_slave_unregister(bundle->bond, port);
1514 bundle_update(bundle);
1518 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1519 struct lacp_slave_settings *lacp,
1520 uint32_t bond_stable_id)
1522 struct ofport_dpif *port;
1524 port = get_ofp_port(bundle->ofproto, ofp_port);
1529 if (port->bundle != bundle) {
1530 bundle->ofproto->need_revalidate = true;
1532 bundle_del_port(port);
1535 port->bundle = bundle;
1536 list_push_back(&bundle->ports, &port->bundle_node);
1537 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
1538 bundle->floodable = false;
1542 port->bundle->ofproto->need_revalidate = true;
1543 lacp_slave_register(bundle->lacp, port, lacp);
1546 port->bond_stable_id = bond_stable_id;
1552 bundle_destroy(struct ofbundle *bundle)
1554 struct ofproto_dpif *ofproto;
1555 struct ofport_dpif *port, *next_port;
1562 ofproto = bundle->ofproto;
1563 for (i = 0; i < MAX_MIRRORS; i++) {
1564 struct ofmirror *m = ofproto->mirrors[i];
1566 if (m->out == bundle) {
1568 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1569 || hmapx_find_and_delete(&m->dsts, bundle)) {
1570 ofproto->need_revalidate = true;
1575 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1576 bundle_del_port(port);
1579 bundle_flush_macs(bundle, true);
1580 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1582 free(bundle->trunks);
1583 lacp_destroy(bundle->lacp);
1584 bond_destroy(bundle->bond);
1589 bundle_set(struct ofproto *ofproto_, void *aux,
1590 const struct ofproto_bundle_settings *s)
1592 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1593 bool need_flush = false;
1594 struct ofport_dpif *port;
1595 struct ofbundle *bundle;
1596 unsigned long *trunks;
1602 bundle_destroy(bundle_lookup(ofproto, aux));
1606 assert(s->n_slaves == 1 || s->bond != NULL);
1607 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1609 bundle = bundle_lookup(ofproto, aux);
1611 bundle = xmalloc(sizeof *bundle);
1613 bundle->ofproto = ofproto;
1614 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1615 hash_pointer(aux, 0));
1617 bundle->name = NULL;
1619 list_init(&bundle->ports);
1620 bundle->vlan_mode = PORT_VLAN_TRUNK;
1622 bundle->trunks = NULL;
1623 bundle->use_priority_tags = s->use_priority_tags;
1624 bundle->lacp = NULL;
1625 bundle->bond = NULL;
1627 bundle->floodable = true;
1629 bundle->src_mirrors = 0;
1630 bundle->dst_mirrors = 0;
1631 bundle->mirror_out = 0;
1634 if (!bundle->name || strcmp(s->name, bundle->name)) {
1636 bundle->name = xstrdup(s->name);
1641 if (!bundle->lacp) {
1642 ofproto->need_revalidate = true;
1643 bundle->lacp = lacp_create();
1645 lacp_configure(bundle->lacp, s->lacp);
1647 lacp_destroy(bundle->lacp);
1648 bundle->lacp = NULL;
1651 /* Update set of ports. */
1653 for (i = 0; i < s->n_slaves; i++) {
1654 if (!bundle_add_port(bundle, s->slaves[i],
1655 s->lacp ? &s->lacp_slaves[i] : NULL,
1656 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1660 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1661 struct ofport_dpif *next_port;
1663 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1664 for (i = 0; i < s->n_slaves; i++) {
1665 if (s->slaves[i] == port->up.ofp_port) {
1670 bundle_del_port(port);
1674 assert(list_size(&bundle->ports) <= s->n_slaves);
1676 if (list_is_empty(&bundle->ports)) {
1677 bundle_destroy(bundle);
1681 /* Set VLAN tagging mode */
1682 if (s->vlan_mode != bundle->vlan_mode
1683 || s->use_priority_tags != bundle->use_priority_tags) {
1684 bundle->vlan_mode = s->vlan_mode;
1685 bundle->use_priority_tags = s->use_priority_tags;
1690 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1691 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1693 if (vlan != bundle->vlan) {
1694 bundle->vlan = vlan;
1698 /* Get trunked VLANs. */
1699 switch (s->vlan_mode) {
1700 case PORT_VLAN_ACCESS:
1704 case PORT_VLAN_TRUNK:
1705 trunks = (unsigned long *) s->trunks;
1708 case PORT_VLAN_NATIVE_UNTAGGED:
1709 case PORT_VLAN_NATIVE_TAGGED:
1710 if (vlan != 0 && (!s->trunks
1711 || !bitmap_is_set(s->trunks, vlan)
1712 || bitmap_is_set(s->trunks, 0))) {
1713 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1715 trunks = bitmap_clone(s->trunks, 4096);
1717 trunks = bitmap_allocate1(4096);
1719 bitmap_set1(trunks, vlan);
1720 bitmap_set0(trunks, 0);
1722 trunks = (unsigned long *) s->trunks;
1729 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1730 free(bundle->trunks);
1731 if (trunks == s->trunks) {
1732 bundle->trunks = vlan_bitmap_clone(trunks);
1734 bundle->trunks = trunks;
1739 if (trunks != s->trunks) {
1744 if (!list_is_short(&bundle->ports)) {
1745 bundle->ofproto->has_bonded_bundles = true;
1747 if (bond_reconfigure(bundle->bond, s->bond)) {
1748 ofproto->need_revalidate = true;
1751 bundle->bond = bond_create(s->bond);
1752 ofproto->need_revalidate = true;
1755 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1756 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1760 bond_destroy(bundle->bond);
1761 bundle->bond = NULL;
1764 /* If we changed something that would affect MAC learning, un-learn
1765 * everything on this port and force flow revalidation. */
1767 bundle_flush_macs(bundle, false);
1774 bundle_remove(struct ofport *port_)
1776 struct ofport_dpif *port = ofport_dpif_cast(port_);
1777 struct ofbundle *bundle = port->bundle;
1780 bundle_del_port(port);
1781 if (list_is_empty(&bundle->ports)) {
1782 bundle_destroy(bundle);
1783 } else if (list_is_short(&bundle->ports)) {
1784 bond_destroy(bundle->bond);
1785 bundle->bond = NULL;
1791 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
1793 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1794 struct ofport_dpif *port = port_;
1795 uint8_t ea[ETH_ADDR_LEN];
1798 error = netdev_get_etheraddr(port->up.netdev, ea);
1800 struct ofpbuf packet;
1803 ofpbuf_init(&packet, 0);
1804 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1806 memcpy(packet_pdu, pdu, pdu_size);
1808 send_packet(port, &packet);
1809 ofpbuf_uninit(&packet);
1811 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1812 "%s (%s)", port->bundle->name,
1813 netdev_get_name(port->up.netdev), strerror(error));
1818 bundle_send_learning_packets(struct ofbundle *bundle)
1820 struct ofproto_dpif *ofproto = bundle->ofproto;
1821 int error, n_packets, n_errors;
1822 struct mac_entry *e;
1824 error = n_packets = n_errors = 0;
1825 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1826 if (e->port.p != bundle) {
1827 struct ofpbuf *learning_packet;
1828 struct ofport_dpif *port;
1832 /* The assignment to "port" is unnecessary but makes "grep"ing for
1833 * struct ofport_dpif more effective. */
1834 learning_packet = bond_compose_learning_packet(bundle->bond,
1838 ret = send_packet(port, learning_packet);
1839 ofpbuf_delete(learning_packet);
1849 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1850 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1851 "packets, last error was: %s",
1852 bundle->name, n_errors, n_packets, strerror(error));
1854 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1855 bundle->name, n_packets);
1860 bundle_run(struct ofbundle *bundle)
1863 lacp_run(bundle->lacp, send_pdu_cb);
1866 struct ofport_dpif *port;
1868 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1869 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1872 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1873 lacp_negotiated(bundle->lacp));
1874 if (bond_should_send_learning_packets(bundle->bond)) {
1875 bundle_send_learning_packets(bundle);
1881 bundle_wait(struct ofbundle *bundle)
1884 lacp_wait(bundle->lacp);
1887 bond_wait(bundle->bond);
1894 mirror_scan(struct ofproto_dpif *ofproto)
1898 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1899 if (!ofproto->mirrors[idx]) {
1906 static struct ofmirror *
1907 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1911 for (i = 0; i < MAX_MIRRORS; i++) {
1912 struct ofmirror *mirror = ofproto->mirrors[i];
1913 if (mirror && mirror->aux == aux) {
1921 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
1923 mirror_update_dups(struct ofproto_dpif *ofproto)
1927 for (i = 0; i < MAX_MIRRORS; i++) {
1928 struct ofmirror *m = ofproto->mirrors[i];
1931 m->dup_mirrors = MIRROR_MASK_C(1) << i;
1935 for (i = 0; i < MAX_MIRRORS; i++) {
1936 struct ofmirror *m1 = ofproto->mirrors[i];
1943 for (j = i + 1; j < MAX_MIRRORS; j++) {
1944 struct ofmirror *m2 = ofproto->mirrors[j];
1946 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
1947 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
1948 m2->dup_mirrors |= m1->dup_mirrors;
1955 mirror_set(struct ofproto *ofproto_, void *aux,
1956 const struct ofproto_mirror_settings *s)
1958 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1959 mirror_mask_t mirror_bit;
1960 struct ofbundle *bundle;
1961 struct ofmirror *mirror;
1962 struct ofbundle *out;
1963 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
1964 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
1967 mirror = mirror_lookup(ofproto, aux);
1969 mirror_destroy(mirror);
1975 idx = mirror_scan(ofproto);
1977 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
1979 ofproto->up.name, MAX_MIRRORS, s->name);
1983 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
1984 mirror->ofproto = ofproto;
1987 mirror->out_vlan = -1;
1988 mirror->name = NULL;
1991 if (!mirror->name || strcmp(s->name, mirror->name)) {
1993 mirror->name = xstrdup(s->name);
1996 /* Get the new configuration. */
1997 if (s->out_bundle) {
1998 out = bundle_lookup(ofproto, s->out_bundle);
2000 mirror_destroy(mirror);
2006 out_vlan = s->out_vlan;
2008 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2009 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2011 /* If the configuration has not changed, do nothing. */
2012 if (hmapx_equals(&srcs, &mirror->srcs)
2013 && hmapx_equals(&dsts, &mirror->dsts)
2014 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2015 && mirror->out == out
2016 && mirror->out_vlan == out_vlan)
2018 hmapx_destroy(&srcs);
2019 hmapx_destroy(&dsts);
2023 hmapx_swap(&srcs, &mirror->srcs);
2024 hmapx_destroy(&srcs);
2026 hmapx_swap(&dsts, &mirror->dsts);
2027 hmapx_destroy(&dsts);
2029 free(mirror->vlans);
2030 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2033 mirror->out_vlan = out_vlan;
2035 /* Update bundles. */
2036 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2037 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2038 if (hmapx_contains(&mirror->srcs, bundle)) {
2039 bundle->src_mirrors |= mirror_bit;
2041 bundle->src_mirrors &= ~mirror_bit;
2044 if (hmapx_contains(&mirror->dsts, bundle)) {
2045 bundle->dst_mirrors |= mirror_bit;
2047 bundle->dst_mirrors &= ~mirror_bit;
2050 if (mirror->out == bundle) {
2051 bundle->mirror_out |= mirror_bit;
2053 bundle->mirror_out &= ~mirror_bit;
2057 ofproto->need_revalidate = true;
2058 mac_learning_flush(ofproto->ml);
2059 mirror_update_dups(ofproto);
2065 mirror_destroy(struct ofmirror *mirror)
2067 struct ofproto_dpif *ofproto;
2068 mirror_mask_t mirror_bit;
2069 struct ofbundle *bundle;
2075 ofproto = mirror->ofproto;
2076 ofproto->need_revalidate = true;
2077 mac_learning_flush(ofproto->ml);
2079 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2080 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2081 bundle->src_mirrors &= ~mirror_bit;
2082 bundle->dst_mirrors &= ~mirror_bit;
2083 bundle->mirror_out &= ~mirror_bit;
2086 hmapx_destroy(&mirror->srcs);
2087 hmapx_destroy(&mirror->dsts);
2088 free(mirror->vlans);
2090 ofproto->mirrors[mirror->idx] = NULL;
2094 mirror_update_dups(ofproto);
2098 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2099 uint64_t *packets, uint64_t *bytes)
2101 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2102 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2105 *packets = *bytes = UINT64_MAX;
2109 *packets = mirror->packet_count;
2110 *bytes = mirror->byte_count;
2116 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2118 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2119 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2120 ofproto->need_revalidate = true;
2121 mac_learning_flush(ofproto->ml);
2127 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2129 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2130 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2131 return bundle && bundle->mirror_out != 0;
2135 forward_bpdu_changed(struct ofproto *ofproto_)
2137 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2138 /* Revalidate cached flows whenever forward_bpdu option changes. */
2139 ofproto->need_revalidate = true;
2144 static struct ofport_dpif *
2145 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
2147 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2148 return ofport ? ofport_dpif_cast(ofport) : NULL;
2151 static struct ofport_dpif *
2152 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
2154 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2158 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2159 struct dpif_port *dpif_port)
2161 ofproto_port->name = dpif_port->name;
2162 ofproto_port->type = dpif_port->type;
2163 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2167 port_run(struct ofport_dpif *ofport)
2169 bool enable = netdev_get_carrier(ofport->up.netdev);
2172 cfm_run(ofport->cfm);
2174 if (cfm_should_send_ccm(ofport->cfm)) {
2175 struct ofpbuf packet;
2177 ofpbuf_init(&packet, 0);
2178 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr);
2179 send_packet(ofport, &packet);
2180 ofpbuf_uninit(&packet);
2183 enable = enable && !cfm_get_fault(ofport->cfm)
2184 && cfm_get_opup(ofport->cfm);
2187 if (ofport->bundle) {
2188 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2191 if (ofport->may_enable != enable) {
2192 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2194 if (ofproto->has_bundle_action) {
2195 ofproto->need_revalidate = true;
2199 ofport->may_enable = enable;
2203 port_wait(struct ofport_dpif *ofport)
2206 cfm_wait(ofport->cfm);
2211 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2212 struct ofproto_port *ofproto_port)
2214 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2215 struct dpif_port dpif_port;
2218 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2220 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2226 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2228 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2232 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2234 *ofp_portp = odp_port_to_ofp_port(odp_port);
2240 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2242 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2245 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2247 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2249 /* The caller is going to close ofport->up.netdev. If this is a
2250 * bonded port, then the bond is using that netdev, so remove it
2251 * from the bond. The client will need to reconfigure everything
2252 * after deleting ports, so then the slave will get re-added. */
2253 bundle_remove(&ofport->up);
2260 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2262 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2265 error = netdev_get_stats(ofport->up.netdev, stats);
2267 if (!error && ofport->odp_port == OVSP_LOCAL) {
2268 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2270 /* ofproto->stats.tx_packets represents packets that we created
2271 * internally and sent to some port (e.g. packets sent with
2272 * send_packet()). Account for them as if they had come from
2273 * OFPP_LOCAL and got forwarded. */
2275 if (stats->rx_packets != UINT64_MAX) {
2276 stats->rx_packets += ofproto->stats.tx_packets;
2279 if (stats->rx_bytes != UINT64_MAX) {
2280 stats->rx_bytes += ofproto->stats.tx_bytes;
2283 /* ofproto->stats.rx_packets represents packets that were received on
2284 * some port and we processed internally and dropped (e.g. STP).
2285 * Account fro them as if they had been forwarded to OFPP_LOCAL. */
2287 if (stats->tx_packets != UINT64_MAX) {
2288 stats->tx_packets += ofproto->stats.rx_packets;
2291 if (stats->tx_bytes != UINT64_MAX) {
2292 stats->tx_bytes += ofproto->stats.rx_bytes;
2299 /* Account packets for LOCAL port. */
2301 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2302 size_t tx_size, size_t rx_size)
2304 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2307 ofproto->stats.rx_packets++;
2308 ofproto->stats.rx_bytes += rx_size;
2311 ofproto->stats.tx_packets++;
2312 ofproto->stats.tx_bytes += tx_size;
2316 struct port_dump_state {
2317 struct dpif_port_dump dump;
2322 port_dump_start(const struct ofproto *ofproto_, void **statep)
2324 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2325 struct port_dump_state *state;
2327 *statep = state = xmalloc(sizeof *state);
2328 dpif_port_dump_start(&state->dump, ofproto->dpif);
2329 state->done = false;
2334 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2335 struct ofproto_port *port)
2337 struct port_dump_state *state = state_;
2338 struct dpif_port dpif_port;
2340 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2341 ofproto_port_from_dpif_port(port, &dpif_port);
2344 int error = dpif_port_dump_done(&state->dump);
2346 return error ? error : EOF;
2351 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2353 struct port_dump_state *state = state_;
2356 dpif_port_dump_done(&state->dump);
2363 port_poll(const struct ofproto *ofproto_, char **devnamep)
2365 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2366 return dpif_port_poll(ofproto->dpif, devnamep);
2370 port_poll_wait(const struct ofproto *ofproto_)
2372 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2373 dpif_port_poll_wait(ofproto->dpif);
2377 port_is_lacp_current(const struct ofport *ofport_)
2379 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2380 return (ofport->bundle && ofport->bundle->lacp
2381 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2385 /* Upcall handling. */
2387 /* Flow miss batching.
2389 * Some dpifs implement operations faster when you hand them off in a batch.
2390 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2391 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2392 * more packets, plus possibly installing the flow in the dpif.
2394 * So far we only batch the operations that affect flow setup time the most.
2395 * It's possible to batch more than that, but the benefit might be minimal. */
2397 struct hmap_node hmap_node;
2399 enum odp_key_fitness key_fitness;
2400 const struct nlattr *key;
2402 ovs_be16 initial_tci;
2403 struct list packets;
2406 struct flow_miss_op {
2407 union dpif_op dpif_op;
2408 struct subfacet *subfacet;
2411 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2412 * OpenFlow controller as necessary according to their individual
2413 * configurations. */
2415 send_packet_in_miss(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
2416 const struct flow *flow)
2418 struct ofputil_packet_in pin;
2420 pin.packet = packet->data;
2421 pin.packet_len = packet->size;
2422 pin.total_len = packet->size;
2423 pin.in_port = flow->in_port;
2424 pin.reason = OFPR_NO_MATCH;
2425 pin.buffer_id = 0; /* not yet known */
2426 pin.send_len = 0; /* not used for flow table misses */
2427 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow);
2430 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_ACTION to each
2431 * OpenFlow controller as necessary according to their individual
2434 * 'send_len' should be the number of bytes of 'packet' to send to the
2435 * controller, as specified in the action that caused the packet to be sent. */
2437 send_packet_in_action(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
2438 uint64_t userdata, const struct flow *flow)
2440 struct ofputil_packet_in pin;
2441 struct user_action_cookie cookie;
2443 memcpy(&cookie, &userdata, sizeof(cookie));
2445 pin.packet = packet->data;
2446 pin.packet_len = packet->size;
2447 pin.total_len = packet->size;
2448 pin.in_port = flow->in_port;
2449 pin.reason = OFPR_ACTION;
2450 pin.buffer_id = 0; /* not yet known */
2451 pin.send_len = cookie.data;
2452 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow);
2456 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2457 const struct ofpbuf *packet)
2459 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2465 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2467 cfm_process_heartbeat(ofport->cfm, packet);
2470 } else if (ofport->bundle && ofport->bundle->lacp
2471 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2473 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2476 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2478 stp_process_packet(ofport, packet);
2485 static struct flow_miss *
2486 flow_miss_create(struct hmap *todo, const struct flow *flow,
2487 enum odp_key_fitness key_fitness,
2488 const struct nlattr *key, size_t key_len,
2489 ovs_be16 initial_tci)
2491 uint32_t hash = flow_hash(flow, 0);
2492 struct flow_miss *miss;
2494 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2495 if (flow_equal(&miss->flow, flow)) {
2500 miss = xmalloc(sizeof *miss);
2501 hmap_insert(todo, &miss->hmap_node, hash);
2503 miss->key_fitness = key_fitness;
2505 miss->key_len = key_len;
2506 miss->initial_tci = initial_tci;
2507 list_init(&miss->packets);
2512 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2513 struct flow_miss_op *ops, size_t *n_ops)
2515 const struct flow *flow = &miss->flow;
2516 struct ofpbuf *packet, *next_packet;
2517 struct subfacet *subfacet;
2518 struct facet *facet;
2520 facet = facet_lookup_valid(ofproto, flow);
2522 struct rule_dpif *rule;
2524 rule = rule_dpif_lookup(ofproto, flow, 0);
2526 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
2527 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
2529 if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) {
2530 COVERAGE_INC(ofproto_dpif_no_packet_in);
2531 /* XXX install 'drop' flow entry */
2535 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
2539 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2540 send_packet_in_miss(ofproto, packet, flow);
2546 facet = facet_create(rule, flow);
2549 subfacet = subfacet_create(ofproto, facet,
2550 miss->key_fitness, miss->key, miss->key_len,
2553 LIST_FOR_EACH_SAFE (packet, next_packet, list_node, &miss->packets) {
2554 struct dpif_flow_stats stats;
2556 list_remove(&packet->list_node);
2557 ofproto->n_matches++;
2559 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2561 * Extra-special case for fail-open mode.
2563 * We are in fail-open mode and the packet matched the fail-open
2564 * rule, but we are connected to a controller too. We should send
2565 * the packet up to the controller in the hope that it will try to
2566 * set up a flow and thereby allow us to exit fail-open.
2568 * See the top-level comment in fail-open.c for more information.
2570 send_packet_in_miss(ofproto, packet, flow);
2573 if (!facet->may_install || !subfacet->actions) {
2574 subfacet_make_actions(ofproto, subfacet, packet);
2577 /* Credit statistics to subfacet for this packet. We must do this now
2578 * because execute_controller_action() below may destroy 'packet'. */
2579 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2580 subfacet_update_stats(ofproto, subfacet, &stats);
2582 if (!execute_controller_action(ofproto, &facet->flow,
2584 subfacet->actions_len, packet)
2585 && subfacet->actions_len > 0) {
2586 struct flow_miss_op *op = &ops[(*n_ops)++];
2587 struct dpif_execute *execute = &op->dpif_op.execute;
2589 if (flow->vlan_tci != subfacet->initial_tci) {
2590 /* This packet was received on a VLAN splinter port. We added
2591 * a VLAN to the packet to make the packet resemble the flow,
2592 * but the actions were composed assuming that the packet
2593 * contained no VLAN. So, we must remove the VLAN header from
2594 * the packet before trying to execute the actions. */
2595 eth_pop_vlan(packet);
2598 op->subfacet = subfacet;
2599 execute->type = DPIF_OP_EXECUTE;
2600 execute->key = miss->key;
2601 execute->key_len = miss->key_len;
2603 = (facet->may_install
2605 : xmemdup(subfacet->actions, subfacet->actions_len));
2606 execute->actions_len = subfacet->actions_len;
2607 execute->packet = packet;
2611 if (facet->may_install && subfacet->key_fitness != ODP_FIT_TOO_LITTLE) {
2612 struct flow_miss_op *op = &ops[(*n_ops)++];
2613 struct dpif_flow_put *put = &op->dpif_op.flow_put;
2615 op->subfacet = subfacet;
2616 put->type = DPIF_OP_FLOW_PUT;
2617 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2618 put->key = miss->key;
2619 put->key_len = miss->key_len;
2620 put->actions = subfacet->actions;
2621 put->actions_len = subfacet->actions_len;
2626 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
2627 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
2628 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
2629 * what a flow key should contain.
2631 * This function also includes some logic to help make VLAN splinters
2632 * transparent to the rest of the upcall processing logic. In particular, if
2633 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
2634 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
2635 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
2637 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
2638 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
2639 * (This differs from the value returned in flow->vlan_tci only for packets
2640 * received on VLAN splinters.)
2642 static enum odp_key_fitness
2643 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
2644 const struct nlattr *key, size_t key_len,
2645 struct flow *flow, ovs_be16 *initial_tci,
2646 struct ofpbuf *packet)
2648 enum odp_key_fitness fitness;
2652 fitness = odp_flow_key_to_flow(key, key_len, flow);
2653 if (fitness == ODP_FIT_ERROR) {
2656 *initial_tci = flow->vlan_tci;
2658 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
2660 /* Cause the flow to be processed as if it came in on the real device
2661 * with the VLAN device's VLAN ID. */
2662 flow->in_port = realdev;
2663 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
2665 /* Make the packet resemble the flow, so that it gets sent to an
2666 * OpenFlow controller properly, so that it looks correct for
2667 * sFlow, and so that flow_extract() will get the correct vlan_tci
2668 * if it is called on 'packet'.
2670 * The allocated space inside 'packet' probably also contains
2671 * 'key', that is, both 'packet' and 'key' are probably part of a
2672 * struct dpif_upcall (see the large comment on that structure
2673 * definition), so pushing data on 'packet' is in general not a
2674 * good idea since it could overwrite 'key' or free it as a side
2675 * effect. However, it's OK in this special case because we know
2676 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
2677 * will just overwrite the 4-byte "struct nlattr", which is fine
2678 * since we don't need that header anymore. */
2679 eth_push_vlan(packet, flow->vlan_tci);
2682 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
2683 if (fitness == ODP_FIT_PERFECT) {
2684 fitness = ODP_FIT_TOO_MUCH;
2692 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
2695 struct dpif_upcall *upcall;
2696 struct flow_miss *miss, *next_miss;
2697 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
2698 union dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
2707 /* Construct the to-do list.
2709 * This just amounts to extracting the flow from each packet and sticking
2710 * the packets that have the same flow in the same "flow_miss" structure so
2711 * that we can process them together. */
2713 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
2714 enum odp_key_fitness fitness;
2715 struct flow_miss *miss;
2716 ovs_be16 initial_tci;
2719 /* Obtain metadata and check userspace/kernel agreement on flow match,
2720 * then set 'flow''s header pointers. */
2721 fitness = ofproto_dpif_extract_flow_key(ofproto,
2722 upcall->key, upcall->key_len,
2723 &flow, &initial_tci,
2725 if (fitness == ODP_FIT_ERROR) {
2726 ofpbuf_delete(upcall->packet);
2729 flow_extract(upcall->packet, flow.skb_priority, flow.tun_id,
2730 flow.in_port, &flow);
2732 /* Handle 802.1ag, LACP, and STP specially. */
2733 if (process_special(ofproto, &flow, upcall->packet)) {
2734 ofproto_update_local_port_stats(&ofproto->up,
2735 0, upcall->packet->size);
2736 ofpbuf_delete(upcall->packet);
2737 ofproto->n_matches++;
2741 /* Add other packets to a to-do list. */
2742 miss = flow_miss_create(&todo, &flow, fitness,
2743 upcall->key, upcall->key_len, initial_tci);
2744 list_push_back(&miss->packets, &upcall->packet->list_node);
2747 /* Process each element in the to-do list, constructing the set of
2748 * operations to batch. */
2750 HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) {
2751 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
2752 ofpbuf_list_delete(&miss->packets);
2753 hmap_remove(&todo, &miss->hmap_node);
2756 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
2757 hmap_destroy(&todo);
2759 /* Execute batch. */
2760 for (i = 0; i < n_ops; i++) {
2761 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
2763 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
2765 /* Free memory and update facets. */
2766 for (i = 0; i < n_ops; i++) {
2767 struct flow_miss_op *op = &flow_miss_ops[i];
2768 struct dpif_execute *execute;
2769 struct dpif_flow_put *put;
2771 switch (op->dpif_op.type) {
2772 case DPIF_OP_EXECUTE:
2773 execute = &op->dpif_op.execute;
2774 if (op->subfacet->actions != execute->actions) {
2775 free((struct nlattr *) execute->actions);
2777 ofpbuf_delete((struct ofpbuf *) execute->packet);
2780 case DPIF_OP_FLOW_PUT:
2781 put = &op->dpif_op.flow_put;
2783 op->subfacet->installed = true;
2791 handle_userspace_upcall(struct ofproto_dpif *ofproto,
2792 struct dpif_upcall *upcall)
2794 struct user_action_cookie cookie;
2795 enum odp_key_fitness fitness;
2796 ovs_be16 initial_tci;
2799 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
2801 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
2802 upcall->key_len, &flow,
2803 &initial_tci, upcall->packet);
2804 if (fitness == ODP_FIT_ERROR) {
2805 ofpbuf_delete(upcall->packet);
2809 if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
2810 if (ofproto->sflow) {
2811 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
2814 } else if (cookie.type == USER_ACTION_COOKIE_CONTROLLER) {
2815 COVERAGE_INC(ofproto_dpif_ctlr_action);
2816 send_packet_in_action(ofproto, upcall->packet, upcall->userdata,
2819 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
2821 ofpbuf_delete(upcall->packet);
2825 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
2827 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
2831 assert (max_batch <= FLOW_MISS_MAX_BATCH);
2834 for (i = 0; i < max_batch; i++) {
2835 struct dpif_upcall *upcall = &misses[n_misses];
2838 error = dpif_recv(ofproto->dpif, upcall);
2843 switch (upcall->type) {
2844 case DPIF_UC_ACTION:
2845 handle_userspace_upcall(ofproto, upcall);
2849 /* Handle it later. */
2853 case DPIF_N_UC_TYPES:
2855 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32,
2861 handle_miss_upcalls(ofproto, misses, n_misses);
2866 /* Flow expiration. */
2868 static int subfacet_max_idle(const struct ofproto_dpif *);
2869 static void update_stats(struct ofproto_dpif *);
2870 static void rule_expire(struct rule_dpif *);
2871 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
2873 /* This function is called periodically by run(). Its job is to collect
2874 * updates for the flows that have been installed into the datapath, most
2875 * importantly when they last were used, and then use that information to
2876 * expire flows that have not been used recently.
2878 * Returns the number of milliseconds after which it should be called again. */
2880 expire(struct ofproto_dpif *ofproto)
2882 struct rule_dpif *rule, *next_rule;
2883 struct classifier *table;
2886 /* Update stats for each flow in the datapath. */
2887 update_stats(ofproto);
2889 /* Expire subfacets that have been idle too long. */
2890 dp_max_idle = subfacet_max_idle(ofproto);
2891 expire_subfacets(ofproto, dp_max_idle);
2893 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
2894 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
2895 struct cls_cursor cursor;
2897 cls_cursor_init(&cursor, table, NULL);
2898 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
2903 /* All outstanding data in existing flows has been accounted, so it's a
2904 * good time to do bond rebalancing. */
2905 if (ofproto->has_bonded_bundles) {
2906 struct ofbundle *bundle;
2908 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2910 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
2915 return MIN(dp_max_idle, 1000);
2918 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
2920 * This function also pushes statistics updates to rules which each facet
2921 * resubmits into. Generally these statistics will be accurate. However, if a
2922 * facet changes the rule it resubmits into at some time in between
2923 * update_stats() runs, it is possible that statistics accrued to the
2924 * old rule will be incorrectly attributed to the new rule. This could be
2925 * avoided by calling update_stats() whenever rules are created or
2926 * deleted. However, the performance impact of making so many calls to the
2927 * datapath do not justify the benefit of having perfectly accurate statistics.
2930 update_stats(struct ofproto_dpif *p)
2932 const struct dpif_flow_stats *stats;
2933 struct dpif_flow_dump dump;
2934 const struct nlattr *key;
2937 dpif_flow_dump_start(&dump, p->dpif);
2938 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
2939 struct subfacet *subfacet;
2941 subfacet = subfacet_find(p, key, key_len);
2942 if (subfacet && subfacet->installed) {
2943 struct facet *facet = subfacet->facet;
2945 if (stats->n_packets >= subfacet->dp_packet_count) {
2946 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
2947 facet->packet_count += extra;
2949 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
2952 if (stats->n_bytes >= subfacet->dp_byte_count) {
2953 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
2955 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
2958 subfacet->dp_packet_count = stats->n_packets;
2959 subfacet->dp_byte_count = stats->n_bytes;
2961 subfacet_update_time(p, subfacet, stats->used);
2962 facet_account(p, facet);
2963 facet_push_stats(facet);
2965 if (!VLOG_DROP_WARN(&rl)) {
2969 odp_flow_key_format(key, key_len, &s);
2970 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
2974 COVERAGE_INC(facet_unexpected);
2975 /* There's a flow in the datapath that we know nothing about, or a
2976 * flow that shouldn't be installed but was anyway. Delete it. */
2977 dpif_flow_del(p->dpif, key, key_len, NULL);
2980 dpif_flow_dump_done(&dump);
2983 /* Calculates and returns the number of milliseconds of idle time after which
2984 * subfacets should expire from the datapath. When a subfacet expires, we fold
2985 * its statistics into its facet, and when a facet's last subfacet expires, we
2986 * fold its statistic into its rule. */
2988 subfacet_max_idle(const struct ofproto_dpif *ofproto)
2991 * Idle time histogram.
2993 * Most of the time a switch has a relatively small number of subfacets.
2994 * When this is the case we might as well keep statistics for all of them
2995 * in userspace and to cache them in the kernel datapath for performance as
2998 * As the number of subfacets increases, the memory required to maintain
2999 * statistics about them in userspace and in the kernel becomes
3000 * significant. However, with a large number of subfacets it is likely
3001 * that only a few of them are "heavy hitters" that consume a large amount
3002 * of bandwidth. At this point, only heavy hitters are worth caching in
3003 * the kernel and maintaining in userspaces; other subfacets we can
3006 * The technique used to compute the idle time is to build a histogram with
3007 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3008 * that is installed in the kernel gets dropped in the appropriate bucket.
3009 * After the histogram has been built, we compute the cutoff so that only
3010 * the most-recently-used 1% of subfacets (but at least
3011 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3012 * the most-recently-used bucket of subfacets is kept, so actually an
3013 * arbitrary number of subfacets can be kept in any given expiration run
3014 * (though the next run will delete most of those unless they receive
3017 * This requires a second pass through the subfacets, in addition to the
3018 * pass made by update_stats(), because the former function never looks at
3019 * uninstallable subfacets.
3021 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3022 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3023 int buckets[N_BUCKETS] = { 0 };
3024 int total, subtotal, bucket;
3025 struct subfacet *subfacet;
3029 total = hmap_count(&ofproto->subfacets);
3030 if (total <= ofproto->up.flow_eviction_threshold) {
3031 return N_BUCKETS * BUCKET_WIDTH;
3034 /* Build histogram. */
3036 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3037 long long int idle = now - subfacet->used;
3038 int bucket = (idle <= 0 ? 0
3039 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3040 : (unsigned int) idle / BUCKET_WIDTH);
3044 /* Find the first bucket whose flows should be expired. */
3045 subtotal = bucket = 0;
3047 subtotal += buckets[bucket++];
3048 } while (bucket < N_BUCKETS &&
3049 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3051 if (VLOG_IS_DBG_ENABLED()) {
3055 ds_put_cstr(&s, "keep");
3056 for (i = 0; i < N_BUCKETS; i++) {
3058 ds_put_cstr(&s, ", drop");
3061 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3064 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3068 return bucket * BUCKET_WIDTH;
3072 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3074 long long int cutoff = time_msec() - dp_max_idle;
3075 struct subfacet *subfacet, *next_subfacet;
3077 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3078 &ofproto->subfacets) {
3079 if (subfacet->used < cutoff) {
3080 subfacet_destroy(ofproto, subfacet);
3085 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3086 * then delete it entirely. */
3088 rule_expire(struct rule_dpif *rule)
3090 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3091 struct facet *facet, *next_facet;
3095 /* Has 'rule' expired? */
3097 if (rule->up.hard_timeout
3098 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3099 reason = OFPRR_HARD_TIMEOUT;
3100 } else if (rule->up.idle_timeout && list_is_empty(&rule->facets)
3101 && now > rule->used + rule->up.idle_timeout * 1000) {
3102 reason = OFPRR_IDLE_TIMEOUT;
3107 COVERAGE_INC(ofproto_dpif_expired);
3109 /* Update stats. (This is a no-op if the rule expired due to an idle
3110 * timeout, because that only happens when the rule has no facets left.) */
3111 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3112 facet_remove(ofproto, facet);
3115 /* Get rid of the rule. */
3116 ofproto_rule_expire(&rule->up, reason);
3121 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3123 * The caller must already have determined that no facet with an identical
3124 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3125 * the ofproto's classifier table.
3127 * The facet will initially have no subfacets. The caller should create (at
3128 * least) one subfacet with subfacet_create(). */
3129 static struct facet *
3130 facet_create(struct rule_dpif *rule, const struct flow *flow)
3132 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3133 struct facet *facet;
3135 facet = xzalloc(sizeof *facet);
3136 facet->used = time_msec();
3137 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
3138 list_push_back(&rule->facets, &facet->list_node);
3140 facet->flow = *flow;
3141 list_init(&facet->subfacets);
3142 netflow_flow_init(&facet->nf_flow);
3143 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3149 facet_free(struct facet *facet)
3154 /* If the 'actions_len' bytes of actions in 'odp_actions' are just a single
3155 * OVS_ACTION_ATTR_USERSPACE action, executes it internally and returns true.
3156 * Otherwise, returns false without doing anything. */
3158 execute_controller_action(struct ofproto_dpif *ofproto,
3159 const struct flow *flow,
3160 const struct nlattr *odp_actions, size_t actions_len,
3161 struct ofpbuf *packet)
3164 && odp_actions->nla_type == OVS_ACTION_ATTR_USERSPACE
3165 && NLA_ALIGN(odp_actions->nla_len) == actions_len) {
3166 /* As an optimization, avoid a round-trip from userspace to kernel to
3167 * userspace. This also avoids possibly filling up kernel packet
3168 * buffers along the way.
3170 * This optimization will not accidentally catch sFlow
3171 * OVS_ACTION_ATTR_USERSPACE actions, since those are encapsulated
3172 * inside OVS_ACTION_ATTR_SAMPLE. */
3173 const struct nlattr *nla;
3175 nla = nl_attr_find_nested(odp_actions, OVS_USERSPACE_ATTR_USERDATA);
3176 send_packet_in_action(ofproto, packet, nl_attr_get_u64(nla), flow);
3183 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3184 * 'packet', which arrived on 'in_port'.
3186 * Takes ownership of 'packet'. */
3188 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3189 const struct nlattr *odp_actions, size_t actions_len,
3190 struct ofpbuf *packet)
3192 struct odputil_keybuf keybuf;
3196 if (execute_controller_action(ofproto, flow, odp_actions, actions_len,
3198 ofpbuf_delete(packet);
3202 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3203 odp_flow_key_from_flow(&key, flow);
3205 error = dpif_execute(ofproto->dpif, key.data, key.size,
3206 odp_actions, actions_len, packet);
3208 ofpbuf_delete(packet);
3212 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3214 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3215 * rule's statistics, via subfacet_uninstall().
3217 * - Removes 'facet' from its rule and from ofproto->facets.
3220 facet_remove(struct ofproto_dpif *ofproto, struct facet *facet)
3222 struct subfacet *subfacet, *next_subfacet;
3224 assert(!list_is_empty(&facet->subfacets));
3226 /* First uninstall all of the subfacets to get final statistics. */
3227 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3228 subfacet_uninstall(ofproto, subfacet);
3231 /* Flush the final stats to the rule.
3233 * This might require us to have at least one subfacet around so that we
3234 * can use its actions for accounting in facet_account(), which is why we
3235 * have uninstalled but not yet destroyed the subfacets. */
3236 facet_flush_stats(ofproto, facet);
3238 /* Now we're really all done so destroy everything. */
3239 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3240 &facet->subfacets) {
3241 subfacet_destroy__(ofproto, subfacet);
3243 hmap_remove(&ofproto->facets, &facet->hmap_node);
3244 list_remove(&facet->list_node);
3249 facet_account(struct ofproto_dpif *ofproto, struct facet *facet)
3252 struct subfacet *subfacet;
3253 const struct nlattr *a;
3257 if (facet->byte_count <= facet->accounted_bytes) {
3260 n_bytes = facet->byte_count - facet->accounted_bytes;
3261 facet->accounted_bytes = facet->byte_count;
3263 /* Feed information from the active flows back into the learning table to
3264 * ensure that table is always in sync with what is actually flowing
3265 * through the datapath. */
3266 if (facet->has_learn || facet->has_normal) {
3267 struct action_xlate_ctx ctx;
3269 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3270 facet->flow.vlan_tci, NULL);
3271 ctx.may_learn = true;
3272 ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions,
3273 facet->rule->up.n_actions));
3276 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3280 /* This loop feeds byte counters to bond_account() for rebalancing to use
3281 * as a basis. We also need to track the actual VLAN on which the packet
3282 * is going to be sent to ensure that it matches the one passed to
3283 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3286 * We use the actions from an arbitrary subfacet because they should all
3287 * be equally valid for our purpose. */
3288 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3289 struct subfacet, list_node);
3290 vlan_tci = facet->flow.vlan_tci;
3291 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3292 subfacet->actions, subfacet->actions_len) {
3293 const struct ovs_action_push_vlan *vlan;
3294 struct ofport_dpif *port;
3296 switch (nl_attr_type(a)) {
3297 case OVS_ACTION_ATTR_OUTPUT:
3298 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3299 if (port && port->bundle && port->bundle->bond) {
3300 bond_account(port->bundle->bond, &facet->flow,
3301 vlan_tci_to_vid(vlan_tci), n_bytes);
3305 case OVS_ACTION_ATTR_POP_VLAN:
3306 vlan_tci = htons(0);
3309 case OVS_ACTION_ATTR_PUSH_VLAN:
3310 vlan = nl_attr_get(a);
3311 vlan_tci = vlan->vlan_tci;
3317 /* Returns true if the only action for 'facet' is to send to the controller.
3318 * (We don't report NetFlow expiration messages for such facets because they
3319 * are just part of the control logic for the network, not real traffic). */
3321 facet_is_controller_flow(struct facet *facet)
3324 && facet->rule->up.n_actions == 1
3325 && action_outputs_to_port(&facet->rule->up.actions[0],
3326 htons(OFPP_CONTROLLER)));
3329 /* Folds all of 'facet''s statistics into its rule. Also updates the
3330 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3331 * 'facet''s statistics in the datapath should have been zeroed and folded into
3332 * its packet and byte counts before this function is called. */
3334 facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet)
3336 struct subfacet *subfacet;
3338 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3339 assert(!subfacet->dp_byte_count);
3340 assert(!subfacet->dp_packet_count);
3343 facet_push_stats(facet);
3344 facet_account(ofproto, facet);
3346 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3347 struct ofexpired expired;
3348 expired.flow = facet->flow;
3349 expired.packet_count = facet->packet_count;
3350 expired.byte_count = facet->byte_count;
3351 expired.used = facet->used;
3352 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3355 facet->rule->packet_count += facet->packet_count;
3356 facet->rule->byte_count += facet->byte_count;
3358 /* Reset counters to prevent double counting if 'facet' ever gets
3360 facet_reset_counters(facet);
3362 netflow_flow_clear(&facet->nf_flow);
3365 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3366 * Returns it if found, otherwise a null pointer.
3368 * The returned facet might need revalidation; use facet_lookup_valid()
3369 * instead if that is important. */
3370 static struct facet *
3371 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
3373 struct facet *facet;
3375 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
3377 if (flow_equal(flow, &facet->flow)) {
3385 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3386 * Returns it if found, otherwise a null pointer.
3388 * The returned facet is guaranteed to be valid. */
3389 static struct facet *
3390 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
3392 struct facet *facet = facet_find(ofproto, flow);
3394 /* The facet we found might not be valid, since we could be in need of
3395 * revalidation. If it is not valid, don't return it. */
3397 && (ofproto->need_revalidate
3398 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))
3399 && !facet_revalidate(ofproto, facet)) {
3400 COVERAGE_INC(facet_invalidated);
3407 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
3409 * - If the rule found is different from 'facet''s current rule, moves
3410 * 'facet' to the new rule and recompiles its actions.
3412 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3413 * where it is and recompiles its actions anyway.
3415 * - If there is none, destroys 'facet'.
3417 * Returns true if 'facet' still exists, false if it has been destroyed. */
3419 facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet)
3422 struct nlattr *odp_actions;
3425 struct actions *new_actions;
3427 struct action_xlate_ctx ctx;
3428 struct rule_dpif *new_rule;
3429 struct subfacet *subfacet;
3430 bool actions_changed;
3433 COVERAGE_INC(facet_revalidate);
3435 /* Determine the new rule. */
3436 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3438 /* No new rule, so delete the facet. */
3439 facet_remove(ofproto, facet);
3443 /* Calculate new datapath actions.
3445 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3446 * emit a NetFlow expiration and, if so, we need to have the old state
3447 * around to properly compose it. */
3449 /* If the datapath actions changed or the installability changed,
3450 * then we need to talk to the datapath. */
3453 memset(&ctx, 0, sizeof ctx);
3454 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3455 struct ofpbuf *odp_actions;
3456 bool should_install;
3458 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3459 subfacet->initial_tci, NULL);
3460 odp_actions = xlate_actions(&ctx, new_rule->up.actions,
3461 new_rule->up.n_actions);
3462 actions_changed = (subfacet->actions_len != odp_actions->size
3463 || memcmp(subfacet->actions, odp_actions->data,
3464 subfacet->actions_len));
3466 should_install = (ctx.may_set_up_flow
3467 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3468 if (actions_changed || should_install != subfacet->installed) {
3469 if (should_install) {
3470 struct dpif_flow_stats stats;
3472 subfacet_install(ofproto, subfacet,
3473 odp_actions->data, odp_actions->size, &stats);
3474 subfacet_update_stats(ofproto, subfacet, &stats);
3476 subfacet_uninstall(ofproto, subfacet);
3480 new_actions = xcalloc(list_size(&facet->subfacets),
3481 sizeof *new_actions);
3483 new_actions[i].odp_actions = xmemdup(odp_actions->data,
3485 new_actions[i].actions_len = odp_actions->size;
3488 ofpbuf_delete(odp_actions);
3492 facet_flush_stats(ofproto, facet);
3495 /* Update 'facet' now that we've taken care of all the old state. */
3496 facet->tags = ctx.tags;
3497 facet->nf_flow.output_iface = ctx.nf_output_iface;
3498 facet->may_install = ctx.may_set_up_flow;
3499 facet->has_learn = ctx.has_learn;
3500 facet->has_normal = ctx.has_normal;
3501 facet->mirrors = ctx.mirrors;
3504 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3505 if (new_actions[i].odp_actions) {
3506 free(subfacet->actions);
3507 subfacet->actions = new_actions[i].odp_actions;
3508 subfacet->actions_len = new_actions[i].actions_len;
3514 if (facet->rule != new_rule) {
3515 COVERAGE_INC(facet_changed_rule);
3516 list_remove(&facet->list_node);
3517 list_push_back(&new_rule->facets, &facet->list_node);
3518 facet->rule = new_rule;
3519 facet->used = new_rule->up.created;
3520 facet->prev_used = facet->used;
3526 /* Updates 'facet''s used time. Caller is responsible for calling
3527 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3529 facet_update_time(struct ofproto_dpif *ofproto, struct facet *facet,
3532 if (used > facet->used) {
3534 if (used > facet->rule->used) {
3535 facet->rule->used = used;
3537 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3542 facet_reset_counters(struct facet *facet)
3544 facet->packet_count = 0;
3545 facet->byte_count = 0;
3546 facet->prev_packet_count = 0;
3547 facet->prev_byte_count = 0;
3548 facet->accounted_bytes = 0;
3552 facet_push_stats(struct facet *facet)
3554 uint64_t new_packets, new_bytes;
3556 assert(facet->packet_count >= facet->prev_packet_count);
3557 assert(facet->byte_count >= facet->prev_byte_count);
3558 assert(facet->used >= facet->prev_used);
3560 new_packets = facet->packet_count - facet->prev_packet_count;
3561 new_bytes = facet->byte_count - facet->prev_byte_count;
3563 if (new_packets || new_bytes || facet->used > facet->prev_used) {
3564 facet->prev_packet_count = facet->packet_count;
3565 facet->prev_byte_count = facet->byte_count;
3566 facet->prev_used = facet->used;
3568 flow_push_stats(facet->rule, &facet->flow,
3569 new_packets, new_bytes, facet->used);
3571 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
3572 facet->mirrors, new_packets, new_bytes);
3576 struct ofproto_push {
3577 struct action_xlate_ctx ctx;
3584 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
3586 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3589 rule->packet_count += push->packets;
3590 rule->byte_count += push->bytes;
3591 rule->used = MAX(push->used, rule->used);
3595 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3596 * 'rule''s actions and mirrors. */
3598 flow_push_stats(const struct rule_dpif *rule,
3599 const struct flow *flow, uint64_t packets, uint64_t bytes,
3602 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3603 struct ofproto_push push;
3605 push.packets = packets;
3609 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, NULL);
3610 push.ctx.resubmit_hook = push_resubmit;
3611 ofpbuf_delete(xlate_actions(&push.ctx,
3612 rule->up.actions, rule->up.n_actions));
3617 static struct subfacet *
3618 subfacet_find__(struct ofproto_dpif *ofproto,
3619 const struct nlattr *key, size_t key_len, uint32_t key_hash,
3620 const struct flow *flow)
3622 struct subfacet *subfacet;
3624 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
3625 &ofproto->subfacets) {
3627 ? (subfacet->key_len == key_len
3628 && !memcmp(key, subfacet->key, key_len))
3629 : flow_equal(flow, &subfacet->facet->flow)) {
3637 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
3638 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
3639 * there is one, otherwise creates and returns a new subfacet.
3641 * If the returned subfacet is new, then subfacet->actions will be NULL, in
3642 * which case the caller must populate the actions with
3643 * subfacet_make_actions(). */
3644 static struct subfacet *
3645 subfacet_create(struct ofproto_dpif *ofproto, struct facet *facet,
3646 enum odp_key_fitness key_fitness,
3647 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
3649 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3650 struct subfacet *subfacet;
3652 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
3654 if (subfacet->facet == facet) {
3658 /* This shouldn't happen. */
3659 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
3660 subfacet_destroy(ofproto, subfacet);
3663 subfacet = xzalloc(sizeof *subfacet);
3664 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
3665 list_push_back(&facet->subfacets, &subfacet->list_node);
3666 subfacet->facet = facet;
3667 subfacet->used = time_msec();
3668 subfacet->key_fitness = key_fitness;
3669 if (key_fitness != ODP_FIT_PERFECT) {
3670 subfacet->key = xmemdup(key, key_len);
3671 subfacet->key_len = key_len;
3673 subfacet->installed = false;
3674 subfacet->initial_tci = initial_tci;
3679 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
3680 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
3681 static struct subfacet *
3682 subfacet_find(struct ofproto_dpif *ofproto,
3683 const struct nlattr *key, size_t key_len)
3685 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3686 enum odp_key_fitness fitness;
3689 fitness = odp_flow_key_to_flow(key, key_len, &flow);
3690 if (fitness == ODP_FIT_ERROR) {
3694 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
3697 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
3698 * its facet within 'ofproto', and frees it. */
3700 subfacet_destroy__(struct ofproto_dpif *ofproto, struct subfacet *subfacet)
3702 subfacet_uninstall(ofproto, subfacet);
3703 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
3704 list_remove(&subfacet->list_node);
3705 free(subfacet->key);
3706 free(subfacet->actions);
3710 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
3711 * last remaining subfacet in its facet destroys the facet too. */
3713 subfacet_destroy(struct ofproto_dpif *ofproto, struct subfacet *subfacet)
3715 struct facet *facet = subfacet->facet;
3717 if (list_is_singleton(&facet->subfacets)) {
3718 /* facet_remove() needs at least one subfacet (it will remove it). */
3719 facet_remove(ofproto, facet);
3721 subfacet_destroy__(ofproto, subfacet);
3725 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
3726 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
3727 * for use as temporary storage. */
3729 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
3732 if (!subfacet->key) {
3733 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
3734 odp_flow_key_from_flow(key, &subfacet->facet->flow);
3736 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
3740 /* Composes the datapath actions for 'subfacet' based on its rule's actions. */
3742 subfacet_make_actions(struct ofproto_dpif *p, struct subfacet *subfacet,
3743 const struct ofpbuf *packet)
3745 struct facet *facet = subfacet->facet;
3746 const struct rule_dpif *rule = facet->rule;
3747 struct ofpbuf *odp_actions;
3748 struct action_xlate_ctx ctx;
3750 action_xlate_ctx_init(&ctx, p, &facet->flow, subfacet->initial_tci,
3752 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3753 facet->tags = ctx.tags;
3754 facet->may_install = ctx.may_set_up_flow;
3755 facet->has_learn = ctx.has_learn;
3756 facet->has_normal = ctx.has_normal;
3757 facet->nf_flow.output_iface = ctx.nf_output_iface;
3758 facet->mirrors = ctx.mirrors;
3760 if (subfacet->actions_len != odp_actions->size
3761 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
3762 free(subfacet->actions);
3763 subfacet->actions_len = odp_actions->size;
3764 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
3767 ofpbuf_delete(odp_actions);
3770 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
3771 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
3772 * in the datapath will be zeroed and 'stats' will be updated with traffic new
3773 * since 'subfacet' was last updated.
3775 * Returns 0 if successful, otherwise a positive errno value. */
3777 subfacet_install(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3778 const struct nlattr *actions, size_t actions_len,
3779 struct dpif_flow_stats *stats)
3781 struct odputil_keybuf keybuf;
3782 enum dpif_flow_put_flags flags;
3786 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3788 flags |= DPIF_FP_ZERO_STATS;
3791 subfacet_get_key(subfacet, &keybuf, &key);
3792 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
3793 actions, actions_len, stats);
3796 subfacet_reset_dp_stats(subfacet, stats);
3802 /* If 'subfacet' is installed in the datapath, uninstalls it. */
3804 subfacet_uninstall(struct ofproto_dpif *p, struct subfacet *subfacet)
3806 if (subfacet->installed) {
3807 struct odputil_keybuf keybuf;
3808 struct dpif_flow_stats stats;
3812 subfacet_get_key(subfacet, &keybuf, &key);
3813 error = dpif_flow_del(p->dpif, key.data, key.size, &stats);
3814 subfacet_reset_dp_stats(subfacet, &stats);
3816 subfacet_update_stats(p, subfacet, &stats);
3818 subfacet->installed = false;
3820 assert(subfacet->dp_packet_count == 0);
3821 assert(subfacet->dp_byte_count == 0);
3825 /* Resets 'subfacet''s datapath statistics counters. This should be called
3826 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
3827 * non-null, it should contain the statistics returned by dpif when 'subfacet'
3828 * was reset in the datapath. 'stats' will be modified to include only
3829 * statistics new since 'subfacet' was last updated. */
3831 subfacet_reset_dp_stats(struct subfacet *subfacet,
3832 struct dpif_flow_stats *stats)
3835 && subfacet->dp_packet_count <= stats->n_packets
3836 && subfacet->dp_byte_count <= stats->n_bytes) {
3837 stats->n_packets -= subfacet->dp_packet_count;
3838 stats->n_bytes -= subfacet->dp_byte_count;
3841 subfacet->dp_packet_count = 0;
3842 subfacet->dp_byte_count = 0;
3845 /* Updates 'subfacet''s used time. The caller is responsible for calling
3846 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
3848 subfacet_update_time(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3851 if (used > subfacet->used) {
3852 subfacet->used = used;
3853 facet_update_time(ofproto, subfacet->facet, used);
3857 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
3859 * Because of the meaning of a subfacet's counters, it only makes sense to do
3860 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
3861 * represents a packet that was sent by hand or if it represents statistics
3862 * that have been cleared out of the datapath. */
3864 subfacet_update_stats(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3865 const struct dpif_flow_stats *stats)
3867 if (stats->n_packets || stats->used > subfacet->used) {
3868 struct facet *facet = subfacet->facet;
3870 subfacet_update_time(ofproto, subfacet, stats->used);
3871 facet->packet_count += stats->n_packets;
3872 facet->byte_count += stats->n_bytes;
3873 facet_push_stats(facet);
3874 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3880 static struct rule_dpif *
3881 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
3884 struct cls_rule *cls_rule;
3885 struct classifier *cls;
3887 if (table_id >= N_TABLES) {
3891 cls = &ofproto->up.tables[table_id];
3892 if (flow->nw_frag & FLOW_NW_FRAG_ANY
3893 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3894 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
3895 * are unavailable. */
3896 struct flow ofpc_normal_flow = *flow;
3897 ofpc_normal_flow.tp_src = htons(0);
3898 ofpc_normal_flow.tp_dst = htons(0);
3899 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
3901 cls_rule = classifier_lookup(cls, flow);
3903 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
3907 complete_operation(struct rule_dpif *rule)
3909 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3911 rule_invalidate(rule);
3913 struct dpif_completion *c = xmalloc(sizeof *c);
3914 c->op = rule->up.pending;
3915 list_push_back(&ofproto->completions, &c->list_node);
3917 ofoperation_complete(rule->up.pending, 0);
3921 static struct rule *
3924 struct rule_dpif *rule = xmalloc(sizeof *rule);
3929 rule_dealloc(struct rule *rule_)
3931 struct rule_dpif *rule = rule_dpif_cast(rule_);
3936 rule_construct(struct rule *rule_)
3938 struct rule_dpif *rule = rule_dpif_cast(rule_);
3939 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3940 struct rule_dpif *victim;
3944 error = validate_actions(rule->up.actions, rule->up.n_actions,
3945 &rule->up.cr.flow, ofproto->max_ports);
3950 rule->used = rule->up.created;
3951 rule->packet_count = 0;
3952 rule->byte_count = 0;
3954 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
3955 if (victim && !list_is_empty(&victim->facets)) {
3956 struct facet *facet;
3958 rule->facets = victim->facets;
3959 list_moved(&rule->facets);
3960 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3961 /* XXX: We're only clearing our local counters here. It's possible
3962 * that quite a few packets are unaccounted for in the datapath
3963 * statistics. These will be accounted to the new rule instead of
3964 * cleared as required. This could be fixed by clearing out the
3965 * datapath statistics for this facet, but currently it doesn't
3967 facet_reset_counters(facet);
3971 /* Must avoid list_moved() in this case. */
3972 list_init(&rule->facets);
3975 table_id = rule->up.table_id;
3976 rule->tag = (victim ? victim->tag
3978 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
3979 ofproto->tables[table_id].basis));
3981 complete_operation(rule);
3986 rule_destruct(struct rule *rule_)
3988 struct rule_dpif *rule = rule_dpif_cast(rule_);
3989 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3990 struct facet *facet, *next_facet;
3992 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3993 facet_revalidate(ofproto, facet);
3996 complete_operation(rule);
4000 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4002 struct rule_dpif *rule = rule_dpif_cast(rule_);
4003 struct facet *facet;
4005 /* Start from historical data for 'rule' itself that are no longer tracked
4006 * in facets. This counts, for example, facets that have expired. */
4007 *packets = rule->packet_count;
4008 *bytes = rule->byte_count;
4010 /* Add any statistics that are tracked by facets. This includes
4011 * statistical data recently updated by ofproto_update_stats() as well as
4012 * stats for packets that were executed "by hand" via dpif_execute(). */
4013 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4014 *packets += facet->packet_count;
4015 *bytes += facet->byte_count;
4020 rule_execute(struct rule *rule_, const struct flow *flow,
4021 struct ofpbuf *packet)
4023 struct rule_dpif *rule = rule_dpif_cast(rule_);
4024 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4025 struct action_xlate_ctx ctx;
4026 struct ofpbuf *odp_actions;
4029 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, packet);
4030 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
4031 size = packet->size;
4032 if (execute_odp_actions(ofproto, flow, odp_actions->data,
4033 odp_actions->size, packet)) {
4034 rule->used = time_msec();
4035 rule->packet_count++;
4036 rule->byte_count += size;
4037 flow_push_stats(rule, flow, 1, size, rule->used);
4039 ofpbuf_delete(odp_actions);
4045 rule_modify_actions(struct rule *rule_)
4047 struct rule_dpif *rule = rule_dpif_cast(rule_);
4048 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4051 error = validate_actions(rule->up.actions, rule->up.n_actions,
4052 &rule->up.cr.flow, ofproto->max_ports);
4054 ofoperation_complete(rule->up.pending, error);
4058 complete_operation(rule);
4061 /* Sends 'packet' out 'ofport'.
4062 * May modify 'packet'.
4063 * Returns 0 if successful, otherwise a positive errno value. */
4065 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4067 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4068 struct ofpbuf key, odp_actions;
4069 struct odputil_keybuf keybuf;
4074 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
4075 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4077 if (odp_port != ofport->odp_port) {
4078 eth_pop_vlan(packet);
4079 flow.vlan_tci = htons(0);
4082 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4083 odp_flow_key_from_flow(&key, &flow);
4085 ofpbuf_init(&odp_actions, 32);
4086 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4088 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4089 error = dpif_execute(ofproto->dpif,
4091 odp_actions.data, odp_actions.size,
4093 ofpbuf_uninit(&odp_actions);
4096 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4097 ofproto->up.name, odp_port, strerror(error));
4099 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
4103 /* OpenFlow to datapath action translation. */
4105 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
4106 struct action_xlate_ctx *ctx);
4107 static void xlate_normal(struct action_xlate_ctx *);
4110 put_userspace_action(const struct ofproto_dpif *ofproto,
4111 struct ofpbuf *odp_actions,
4112 const struct flow *flow,
4113 const struct user_action_cookie *cookie)
4117 pid = dpif_port_get_pid(ofproto->dpif,
4118 ofp_port_to_odp_port(flow->in_port));
4120 return odp_put_userspace_action(pid, cookie, odp_actions);
4123 /* Compose SAMPLE action for sFlow. */
4125 compose_sflow_action(const struct ofproto_dpif *ofproto,
4126 struct ofpbuf *odp_actions,
4127 const struct flow *flow,
4130 uint32_t port_ifindex;
4131 uint32_t probability;
4132 struct user_action_cookie cookie;
4133 size_t sample_offset, actions_offset;
4134 int cookie_offset, n_output;
4136 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4140 if (odp_port == OVSP_NONE) {
4144 port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4148 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4150 /* Number of packets out of UINT_MAX to sample. */
4151 probability = dpif_sflow_get_probability(ofproto->sflow);
4152 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4154 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4156 cookie.type = USER_ACTION_COOKIE_SFLOW;
4157 cookie.data = port_ifindex;
4158 cookie.n_output = n_output;
4159 cookie.vlan_tci = 0;
4160 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4162 nl_msg_end_nested(odp_actions, actions_offset);
4163 nl_msg_end_nested(odp_actions, sample_offset);
4164 return cookie_offset;
4167 /* SAMPLE action must be first action in any given list of actions.
4168 * At this point we do not have all information required to build it. So try to
4169 * build sample action as complete as possible. */
4171 add_sflow_action(struct action_xlate_ctx *ctx)
4173 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4175 &ctx->flow, OVSP_NONE);
4176 ctx->sflow_odp_port = 0;
4177 ctx->sflow_n_outputs = 0;
4180 /* Fix SAMPLE action according to data collected while composing ODP actions.
4181 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4182 * USERSPACE action's user-cookie which is required for sflow. */
4184 fix_sflow_action(struct action_xlate_ctx *ctx)
4186 const struct flow *base = &ctx->base_flow;
4187 struct user_action_cookie *cookie;
4189 if (!ctx->user_cookie_offset) {
4193 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4195 assert(cookie != NULL);
4196 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4198 if (ctx->sflow_n_outputs) {
4199 cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow,
4200 ctx->sflow_odp_port);
4202 if (ctx->sflow_n_outputs >= 255) {
4203 cookie->n_output = 255;
4205 cookie->n_output = ctx->sflow_n_outputs;
4207 cookie->vlan_tci = base->vlan_tci;
4211 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4214 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4215 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4216 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4217 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4221 struct priority_to_dscp *pdscp;
4223 if (ofport->up.opp.config & htonl(OFPPC_NO_FWD)
4224 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4228 pdscp = get_priority(ofport, ctx->flow.skb_priority);
4230 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4231 ctx->flow.nw_tos |= pdscp->dscp;
4234 /* We may not have an ofport record for this port, but it doesn't hurt
4235 * to allow forwarding to it anyhow. Maybe such a port will appear
4236 * later and we're pre-populating the flow table. */
4239 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4240 ctx->flow.vlan_tci);
4241 if (out_port != odp_port) {
4242 ctx->flow.vlan_tci = htons(0);
4244 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4245 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4247 ctx->sflow_odp_port = odp_port;
4248 ctx->sflow_n_outputs++;
4249 ctx->nf_output_iface = ofp_port;
4250 ctx->flow.vlan_tci = flow_vlan_tci;
4251 ctx->flow.nw_tos = flow_nw_tos;
4255 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4257 compose_output_action__(ctx, ofp_port, true);
4261 xlate_table_action(struct action_xlate_ctx *ctx,
4262 uint16_t in_port, uint8_t table_id)
4264 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4265 struct ofproto_dpif *ofproto = ctx->ofproto;
4266 struct rule_dpif *rule;
4267 uint16_t old_in_port;
4268 uint8_t old_table_id;
4270 old_table_id = ctx->table_id;
4271 ctx->table_id = table_id;
4273 /* Look up a flow with 'in_port' as the input port. */
4274 old_in_port = ctx->flow.in_port;
4275 ctx->flow.in_port = in_port;
4276 rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
4279 if (table_id > 0 && table_id < N_TABLES) {
4280 struct table_dpif *table = &ofproto->tables[table_id];
4281 if (table->other_table) {
4284 : rule_calculate_tag(&ctx->flow,
4285 &table->other_table->wc,
4290 /* Restore the original input port. Otherwise OFPP_NORMAL and
4291 * OFPP_IN_PORT will have surprising behavior. */
4292 ctx->flow.in_port = old_in_port;
4294 if (ctx->resubmit_hook) {
4295 ctx->resubmit_hook(ctx, rule);
4300 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
4304 ctx->table_id = old_table_id;
4306 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4308 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
4309 MAX_RESUBMIT_RECURSION);
4314 xlate_resubmit_table(struct action_xlate_ctx *ctx,
4315 const struct nx_action_resubmit *nar)
4320 in_port = (nar->in_port == htons(OFPP_IN_PORT)
4322 : ntohs(nar->in_port));
4323 table_id = nar->table == 255 ? ctx->table_id : nar->table;
4325 xlate_table_action(ctx, in_port, table_id);
4329 flood_packets(struct action_xlate_ctx *ctx, bool all)
4331 struct ofport_dpif *ofport;
4333 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
4334 uint16_t ofp_port = ofport->up.ofp_port;
4336 if (ofp_port == ctx->flow.in_port) {
4341 compose_output_action__(ctx, ofp_port, false);
4342 } else if (!(ofport->up.opp.config & htonl(OFPPC_NO_FLOOD))) {
4343 compose_output_action(ctx, ofp_port);
4347 ctx->nf_output_iface = NF_OUT_FLOOD;
4351 compose_controller_action(struct action_xlate_ctx *ctx, int len)
4353 struct user_action_cookie cookie;
4355 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4356 cookie.type = USER_ACTION_COOKIE_CONTROLLER;
4358 cookie.n_output = 0;
4359 cookie.vlan_tci = 0;
4360 put_userspace_action(ctx->ofproto, ctx->odp_actions, &ctx->flow, &cookie);
4364 xlate_output_action__(struct action_xlate_ctx *ctx,
4365 uint16_t port, uint16_t max_len)
4367 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
4369 ctx->nf_output_iface = NF_OUT_DROP;
4373 compose_output_action(ctx, ctx->flow.in_port);
4376 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
4382 flood_packets(ctx, false);
4385 flood_packets(ctx, true);
4387 case OFPP_CONTROLLER:
4388 compose_controller_action(ctx, max_len);
4391 compose_output_action(ctx, OFPP_LOCAL);
4396 if (port != ctx->flow.in_port) {
4397 compose_output_action(ctx, port);
4402 if (prev_nf_output_iface == NF_OUT_FLOOD) {
4403 ctx->nf_output_iface = NF_OUT_FLOOD;
4404 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
4405 ctx->nf_output_iface = prev_nf_output_iface;
4406 } else if (prev_nf_output_iface != NF_OUT_DROP &&
4407 ctx->nf_output_iface != NF_OUT_FLOOD) {
4408 ctx->nf_output_iface = NF_OUT_MULTI;
4413 xlate_output_reg_action(struct action_xlate_ctx *ctx,
4414 const struct nx_action_output_reg *naor)
4418 ofp_port = nxm_read_field_bits(naor->src, naor->ofs_nbits, &ctx->flow);
4420 if (ofp_port <= UINT16_MAX) {
4421 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
4426 xlate_output_action(struct action_xlate_ctx *ctx,
4427 const struct ofp_action_output *oao)
4429 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
4433 xlate_enqueue_action(struct action_xlate_ctx *ctx,
4434 const struct ofp_action_enqueue *oae)
4437 uint32_t flow_priority, priority;
4440 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
4443 /* Fall back to ordinary output action. */
4444 xlate_output_action__(ctx, ntohs(oae->port), 0);
4448 /* Figure out datapath output port. */
4449 ofp_port = ntohs(oae->port);
4450 if (ofp_port == OFPP_IN_PORT) {
4451 ofp_port = ctx->flow.in_port;
4452 } else if (ofp_port == ctx->flow.in_port) {
4456 /* Add datapath actions. */
4457 flow_priority = ctx->flow.skb_priority;
4458 ctx->flow.skb_priority = priority;
4459 compose_output_action(ctx, ofp_port);
4460 ctx->flow.skb_priority = flow_priority;
4462 /* Update NetFlow output port. */
4463 if (ctx->nf_output_iface == NF_OUT_DROP) {
4464 ctx->nf_output_iface = ofp_port;
4465 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
4466 ctx->nf_output_iface = NF_OUT_MULTI;
4471 xlate_set_queue_action(struct action_xlate_ctx *ctx,
4472 const struct nx_action_set_queue *nasq)
4477 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
4480 /* Couldn't translate queue to a priority, so ignore. A warning
4481 * has already been logged. */
4485 ctx->flow.skb_priority = priority;
4488 struct xlate_reg_state {
4494 xlate_autopath(struct action_xlate_ctx *ctx,
4495 const struct nx_action_autopath *naa)
4497 uint16_t ofp_port = ntohl(naa->id);
4498 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
4500 if (!port || !port->bundle) {
4501 ofp_port = OFPP_NONE;
4502 } else if (port->bundle->bond) {
4503 /* Autopath does not support VLAN hashing. */
4504 struct ofport_dpif *slave = bond_choose_output_slave(
4505 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
4507 ofp_port = slave->up.ofp_port;
4510 autopath_execute(naa, &ctx->flow, ofp_port);
4514 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
4516 struct ofproto_dpif *ofproto = ofproto_;
4517 struct ofport_dpif *port;
4527 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
4530 port = get_ofp_port(ofproto, ofp_port);
4531 return port ? port->may_enable : false;
4536 xlate_learn_action(struct action_xlate_ctx *ctx,
4537 const struct nx_action_learn *learn)
4539 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
4540 struct ofputil_flow_mod fm;
4543 learn_execute(learn, &ctx->flow, &fm);
4545 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
4546 if (error && !VLOG_DROP_WARN(&rl)) {
4547 char *msg = ofputil_error_to_string(error);
4548 VLOG_WARN("learning action failed to modify flow table (%s)", msg);
4556 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
4558 if (port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
4559 ? htonl(OFPPC_NO_RECV_STP)
4560 : htonl(OFPPC_NO_RECV))) {
4564 /* Only drop packets here if both forwarding and learning are
4565 * disabled. If just learning is enabled, we need to have
4566 * OFPP_NORMAL and the learning action have a look at the packet
4567 * before we can drop it. */
4568 if (!stp_forward_in_state(port->stp_state)
4569 && !stp_learn_in_state(port->stp_state)) {
4577 do_xlate_actions(const union ofp_action *in, size_t n_in,
4578 struct action_xlate_ctx *ctx)
4580 const struct ofport_dpif *port;
4581 const union ofp_action *ia;
4584 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
4585 if (port && !may_receive(port, ctx)) {
4586 /* Drop this flow. */
4590 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
4591 const struct ofp_action_dl_addr *oada;
4592 const struct nx_action_resubmit *nar;
4593 const struct nx_action_set_tunnel *nast;
4594 const struct nx_action_set_queue *nasq;
4595 const struct nx_action_multipath *nam;
4596 const struct nx_action_autopath *naa;
4597 const struct nx_action_bundle *nab;
4598 const struct nx_action_output_reg *naor;
4599 enum ofputil_action_code code;
4606 code = ofputil_decode_action_unsafe(ia);
4608 case OFPUTIL_OFPAT_OUTPUT:
4609 xlate_output_action(ctx, &ia->output);
4612 case OFPUTIL_OFPAT_SET_VLAN_VID:
4613 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
4614 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
4617 case OFPUTIL_OFPAT_SET_VLAN_PCP:
4618 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
4619 ctx->flow.vlan_tci |= htons(
4620 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
4623 case OFPUTIL_OFPAT_STRIP_VLAN:
4624 ctx->flow.vlan_tci = htons(0);
4627 case OFPUTIL_OFPAT_SET_DL_SRC:
4628 oada = ((struct ofp_action_dl_addr *) ia);
4629 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
4632 case OFPUTIL_OFPAT_SET_DL_DST:
4633 oada = ((struct ofp_action_dl_addr *) ia);
4634 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
4637 case OFPUTIL_OFPAT_SET_NW_SRC:
4638 ctx->flow.nw_src = ia->nw_addr.nw_addr;
4641 case OFPUTIL_OFPAT_SET_NW_DST:
4642 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
4645 case OFPUTIL_OFPAT_SET_NW_TOS:
4646 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4647 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
4650 case OFPUTIL_OFPAT_SET_TP_SRC:
4651 ctx->flow.tp_src = ia->tp_port.tp_port;
4654 case OFPUTIL_OFPAT_SET_TP_DST:
4655 ctx->flow.tp_dst = ia->tp_port.tp_port;
4658 case OFPUTIL_OFPAT_ENQUEUE:
4659 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
4662 case OFPUTIL_NXAST_RESUBMIT:
4663 nar = (const struct nx_action_resubmit *) ia;
4664 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
4667 case OFPUTIL_NXAST_RESUBMIT_TABLE:
4668 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
4671 case OFPUTIL_NXAST_SET_TUNNEL:
4672 nast = (const struct nx_action_set_tunnel *) ia;
4673 tun_id = htonll(ntohl(nast->tun_id));
4674 ctx->flow.tun_id = tun_id;
4677 case OFPUTIL_NXAST_SET_QUEUE:
4678 nasq = (const struct nx_action_set_queue *) ia;
4679 xlate_set_queue_action(ctx, nasq);
4682 case OFPUTIL_NXAST_POP_QUEUE:
4683 ctx->flow.skb_priority = ctx->orig_skb_priority;
4686 case OFPUTIL_NXAST_REG_MOVE:
4687 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
4691 case OFPUTIL_NXAST_REG_LOAD:
4692 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
4696 case OFPUTIL_NXAST_NOTE:
4697 /* Nothing to do. */
4700 case OFPUTIL_NXAST_SET_TUNNEL64:
4701 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
4702 ctx->flow.tun_id = tun_id;
4705 case OFPUTIL_NXAST_MULTIPATH:
4706 nam = (const struct nx_action_multipath *) ia;
4707 multipath_execute(nam, &ctx->flow);
4710 case OFPUTIL_NXAST_AUTOPATH:
4711 naa = (const struct nx_action_autopath *) ia;
4712 xlate_autopath(ctx, naa);
4715 case OFPUTIL_NXAST_BUNDLE:
4716 ctx->ofproto->has_bundle_action = true;
4717 nab = (const struct nx_action_bundle *) ia;
4718 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
4723 case OFPUTIL_NXAST_BUNDLE_LOAD:
4724 ctx->ofproto->has_bundle_action = true;
4725 nab = (const struct nx_action_bundle *) ia;
4726 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
4730 case OFPUTIL_NXAST_OUTPUT_REG:
4731 naor = (const struct nx_action_output_reg *) ia;
4732 xlate_output_reg_action(ctx, naor);
4735 case OFPUTIL_NXAST_LEARN:
4736 ctx->has_learn = true;
4737 if (ctx->may_learn) {
4738 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
4742 case OFPUTIL_NXAST_EXIT:
4748 /* We've let OFPP_NORMAL and the learning action look at the packet,
4749 * so drop it now if forwarding is disabled. */
4750 if (port && !stp_forward_in_state(port->stp_state)) {
4751 ofpbuf_clear(ctx->odp_actions);
4752 add_sflow_action(ctx);
4757 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
4758 struct ofproto_dpif *ofproto, const struct flow *flow,
4759 ovs_be16 initial_tci, const struct ofpbuf *packet)
4761 ctx->ofproto = ofproto;
4763 ctx->base_flow = ctx->flow;
4764 ctx->base_flow.tun_id = 0;
4765 ctx->base_flow.vlan_tci = initial_tci;
4766 ctx->packet = packet;
4767 ctx->may_learn = packet != NULL;
4768 ctx->resubmit_hook = NULL;
4771 static struct ofpbuf *
4772 xlate_actions(struct action_xlate_ctx *ctx,
4773 const union ofp_action *in, size_t n_in)
4775 struct flow orig_flow = ctx->flow;
4777 COVERAGE_INC(ofproto_dpif_xlate);
4779 ctx->odp_actions = ofpbuf_new(512);
4780 ofpbuf_reserve(ctx->odp_actions, NL_A_U32_SIZE);
4782 ctx->may_set_up_flow = true;
4783 ctx->has_learn = false;
4784 ctx->has_normal = false;
4785 ctx->nf_output_iface = NF_OUT_DROP;
4788 ctx->orig_skb_priority = ctx->flow.skb_priority;
4792 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
4793 switch (ctx->ofproto->up.frag_handling) {
4794 case OFPC_FRAG_NORMAL:
4795 /* We must pretend that transport ports are unavailable. */
4796 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
4797 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
4800 case OFPC_FRAG_DROP:
4801 return ctx->odp_actions;
4803 case OFPC_FRAG_REASM:
4806 case OFPC_FRAG_NX_MATCH:
4807 /* Nothing to do. */
4812 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
4813 ctx->may_set_up_flow = false;
4814 return ctx->odp_actions;
4816 add_sflow_action(ctx);
4817 do_xlate_actions(in, n_in, ctx);
4819 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
4820 ctx->odp_actions->data,
4821 ctx->odp_actions->size)) {
4822 ctx->may_set_up_flow = false;
4824 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
4826 compose_output_action(ctx, OFPP_LOCAL);
4829 add_mirror_actions(ctx, &orig_flow);
4830 fix_sflow_action(ctx);
4833 return ctx->odp_actions;
4836 /* OFPP_NORMAL implementation. */
4838 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
4840 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
4841 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
4842 * the bundle on which the packet was received, returns the VLAN to which the
4845 * Both 'vid' and the return value are in the range 0...4095. */
4847 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
4849 switch (in_bundle->vlan_mode) {
4850 case PORT_VLAN_ACCESS:
4851 return in_bundle->vlan;
4854 case PORT_VLAN_TRUNK:
4857 case PORT_VLAN_NATIVE_UNTAGGED:
4858 case PORT_VLAN_NATIVE_TAGGED:
4859 return vid ? vid : in_bundle->vlan;
4866 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
4867 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
4870 * 'vid' should be the VID obtained from the 802.1Q header that was received as
4871 * part of a packet (specify 0 if there was no 802.1Q header), in the range
4874 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
4876 /* Allow any VID on the OFPP_NONE port. */
4877 if (in_bundle == &ofpp_none_bundle) {
4881 switch (in_bundle->vlan_mode) {
4882 case PORT_VLAN_ACCESS:
4885 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4886 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
4887 "packet received on port %s configured as VLAN "
4888 "%"PRIu16" access port",
4889 in_bundle->ofproto->up.name, vid,
4890 in_bundle->name, in_bundle->vlan);
4896 case PORT_VLAN_NATIVE_UNTAGGED:
4897 case PORT_VLAN_NATIVE_TAGGED:
4899 /* Port must always carry its native VLAN. */
4903 case PORT_VLAN_TRUNK:
4904 if (!ofbundle_includes_vlan(in_bundle, vid)) {
4906 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4907 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
4908 "received on port %s not configured for trunking "
4910 in_bundle->ofproto->up.name, vid,
4911 in_bundle->name, vid);
4923 /* Given 'vlan', the VLAN that a packet belongs to, and
4924 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
4925 * that should be included in the 802.1Q header. (If the return value is 0,
4926 * then the 802.1Q header should only be included in the packet if there is a
4929 * Both 'vlan' and the return value are in the range 0...4095. */
4931 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
4933 switch (out_bundle->vlan_mode) {
4934 case PORT_VLAN_ACCESS:
4937 case PORT_VLAN_TRUNK:
4938 case PORT_VLAN_NATIVE_TAGGED:
4941 case PORT_VLAN_NATIVE_UNTAGGED:
4942 return vlan == out_bundle->vlan ? 0 : vlan;
4950 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
4953 struct ofport_dpif *port;
4955 ovs_be16 tci, old_tci;
4957 vid = output_vlan_to_vid(out_bundle, vlan);
4958 if (!out_bundle->bond) {
4959 port = ofbundle_get_a_port(out_bundle);
4961 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
4964 /* No slaves enabled, so drop packet. */
4969 old_tci = ctx->flow.vlan_tci;
4971 if (tci || out_bundle->use_priority_tags) {
4972 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
4974 tci |= htons(VLAN_CFI);
4977 ctx->flow.vlan_tci = tci;
4979 compose_output_action(ctx, port->up.ofp_port);
4980 ctx->flow.vlan_tci = old_tci;
4984 mirror_mask_ffs(mirror_mask_t mask)
4986 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
4991 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
4993 return (bundle->vlan_mode != PORT_VLAN_ACCESS
4994 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
4998 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
5000 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
5003 /* Returns an arbitrary interface within 'bundle'. */
5004 static struct ofport_dpif *
5005 ofbundle_get_a_port(const struct ofbundle *bundle)
5007 return CONTAINER_OF(list_front(&bundle->ports),
5008 struct ofport_dpif, bundle_node);
5012 vlan_is_mirrored(const struct ofmirror *m, int vlan)
5014 return !m->vlans || bitmap_is_set(m->vlans, vlan);
5017 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
5018 * to a VLAN. In general most packets may be mirrored but we want to drop
5019 * protocols that may confuse switches. */
5021 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
5023 /* If you change this function's behavior, please update corresponding
5024 * documentation in vswitch.xml at the same time. */
5025 if (dst[0] != 0x01) {
5026 /* All the currently banned MACs happen to start with 01 currently, so
5027 * this is a quick way to eliminate most of the good ones. */
5029 if (eth_addr_is_reserved(dst)) {
5030 /* Drop STP, IEEE pause frames, and other reserved protocols
5031 * (01-80-c2-00-00-0x). */
5035 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
5037 if ((dst[3] & 0xfe) == 0xcc &&
5038 (dst[4] & 0xfe) == 0xcc &&
5039 (dst[5] & 0xfe) == 0xcc) {
5040 /* Drop the following protocols plus others following the same
5043 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
5044 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
5045 STP Uplink Fast (01-00-0c-cd-cd-cd) */
5049 if (!(dst[3] | dst[4] | dst[5])) {
5050 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
5059 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5061 struct ofproto_dpif *ofproto = ctx->ofproto;
5062 mirror_mask_t mirrors;
5063 struct ofbundle *in_bundle;
5066 const struct nlattr *a;
5069 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5070 ctx->packet != NULL);
5074 mirrors = in_bundle->src_mirrors;
5076 /* Drop frames on bundles reserved for mirroring. */
5077 if (in_bundle->mirror_out) {
5078 if (ctx->packet != NULL) {
5079 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5080 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5081 "%s, which is reserved exclusively for mirroring",
5082 ctx->ofproto->up.name, in_bundle->name);
5088 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5089 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5092 vlan = input_vid_to_vlan(in_bundle, vid);
5094 /* Look at the output ports to check for destination selections. */
5096 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5097 ctx->odp_actions->size) {
5098 enum ovs_action_attr type = nl_attr_type(a);
5099 struct ofport_dpif *ofport;
5101 if (type != OVS_ACTION_ATTR_OUTPUT) {
5105 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5106 if (ofport && ofport->bundle) {
5107 mirrors |= ofport->bundle->dst_mirrors;
5115 /* Restore the original packet before adding the mirror actions. */
5116 ctx->flow = *orig_flow;
5121 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5123 if (!vlan_is_mirrored(m, vlan)) {
5124 mirrors &= mirrors - 1;
5128 mirrors &= ~m->dup_mirrors;
5129 ctx->mirrors |= m->dup_mirrors;
5131 output_normal(ctx, m->out, vlan);
5132 } else if (eth_dst_may_rspan(orig_flow->dl_dst)
5133 && vlan != m->out_vlan) {
5134 struct ofbundle *bundle;
5136 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5137 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5138 && !bundle->mirror_out) {
5139 output_normal(ctx, bundle, m->out_vlan);
5147 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5148 uint64_t packets, uint64_t bytes)
5154 for (; mirrors; mirrors &= mirrors - 1) {
5157 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5160 /* In normal circumstances 'm' will not be NULL. However,
5161 * if mirrors are reconfigured, we can temporarily get out
5162 * of sync in facet_revalidate(). We could "correct" the
5163 * mirror list before reaching here, but doing that would
5164 * not properly account the traffic stats we've currently
5165 * accumulated for previous mirror configuration. */
5169 m->packet_count += packets;
5170 m->byte_count += bytes;
5174 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
5175 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
5176 * indicate this; newer upstream kernels use gratuitous ARP requests. */
5178 is_gratuitous_arp(const struct flow *flow)
5180 return (flow->dl_type == htons(ETH_TYPE_ARP)
5181 && eth_addr_is_broadcast(flow->dl_dst)
5182 && (flow->nw_proto == ARP_OP_REPLY
5183 || (flow->nw_proto == ARP_OP_REQUEST
5184 && flow->nw_src == flow->nw_dst)));
5188 update_learning_table(struct ofproto_dpif *ofproto,
5189 const struct flow *flow, int vlan,
5190 struct ofbundle *in_bundle)
5192 struct mac_entry *mac;
5194 /* Don't learn the OFPP_NONE port. */
5195 if (in_bundle == &ofpp_none_bundle) {
5199 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
5203 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
5204 if (is_gratuitous_arp(flow)) {
5205 /* We don't want to learn from gratuitous ARP packets that are
5206 * reflected back over bond slaves so we lock the learning table. */
5207 if (!in_bundle->bond) {
5208 mac_entry_set_grat_arp_lock(mac);
5209 } else if (mac_entry_is_grat_arp_locked(mac)) {
5214 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
5215 /* The log messages here could actually be useful in debugging,
5216 * so keep the rate limit relatively high. */
5217 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5218 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
5219 "on port %s in VLAN %d",
5220 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
5221 in_bundle->name, vlan);
5223 mac->port.p = in_bundle;
5224 tag_set_add(&ofproto->revalidate_set,
5225 mac_learning_changed(ofproto->ml, mac));
5229 static struct ofbundle *
5230 lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn)
5232 struct ofport_dpif *ofport;
5234 /* Special-case OFPP_NONE, which a controller may use as the ingress
5235 * port for traffic that it is sourcing. */
5236 if (in_port == OFPP_NONE) {
5237 return &ofpp_none_bundle;
5240 /* Find the port and bundle for the received packet. */
5241 ofport = get_ofp_port(ofproto, in_port);
5242 if (ofport && ofport->bundle) {
5243 return ofport->bundle;
5246 /* Odd. A few possible reasons here:
5248 * - We deleted a port but there are still a few packets queued up
5251 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
5252 * we don't know about.
5254 * - The ofproto client didn't configure the port as part of a bundle.
5257 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5259 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
5260 "port %"PRIu16, ofproto->up.name, in_port);
5265 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
5266 * dropped. Returns true if they may be forwarded, false if they should be
5269 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
5270 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
5272 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
5273 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
5274 * checked by input_vid_is_valid().
5276 * May also add tags to '*tags', although the current implementation only does
5277 * so in one special case.
5280 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
5281 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
5283 struct ofbundle *in_bundle = in_port->bundle;
5285 /* Drop frames for reserved multicast addresses
5286 * only if forward_bpdu option is absent. */
5287 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
5291 if (in_bundle->bond) {
5292 struct mac_entry *mac;
5294 switch (bond_check_admissibility(in_bundle->bond, in_port,
5295 flow->dl_dst, tags)) {
5302 case BV_DROP_IF_MOVED:
5303 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
5304 if (mac && mac->port.p != in_bundle &&
5305 (!is_gratuitous_arp(flow)
5306 || mac_entry_is_grat_arp_locked(mac))) {
5317 xlate_normal(struct action_xlate_ctx *ctx)
5319 struct ofport_dpif *in_port;
5320 struct ofbundle *in_bundle;
5321 struct mac_entry *mac;
5325 ctx->has_normal = true;
5327 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
5328 ctx->packet != NULL);
5333 /* We know 'in_port' exists unless it is "ofpp_none_bundle",
5334 * since lookup_input_bundle() succeeded. */
5335 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5337 /* Drop malformed frames. */
5338 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
5339 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
5340 if (ctx->packet != NULL) {
5341 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5342 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
5343 "VLAN tag received on port %s",
5344 ctx->ofproto->up.name, in_bundle->name);
5349 /* Drop frames on bundles reserved for mirroring. */
5350 if (in_bundle->mirror_out) {
5351 if (ctx->packet != NULL) {
5352 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5353 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5354 "%s, which is reserved exclusively for mirroring",
5355 ctx->ofproto->up.name, in_bundle->name);
5361 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
5362 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5365 vlan = input_vid_to_vlan(in_bundle, vid);
5367 /* Check other admissibility requirements. */
5369 !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
5373 /* Learn source MAC. */
5374 if (ctx->may_learn) {
5375 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
5378 /* Determine output bundle. */
5379 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
5382 if (mac->port.p != in_bundle) {
5383 output_normal(ctx, mac->port.p, vlan);
5386 struct ofbundle *bundle;
5388 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
5389 if (bundle != in_bundle
5390 && ofbundle_includes_vlan(bundle, vlan)
5391 && bundle->floodable
5392 && !bundle->mirror_out) {
5393 output_normal(ctx, bundle, vlan);
5396 ctx->nf_output_iface = NF_OUT_FLOOD;
5400 /* Optimized flow revalidation.
5402 * It's a difficult problem, in general, to tell which facets need to have
5403 * their actions recalculated whenever the OpenFlow flow table changes. We
5404 * don't try to solve that general problem: for most kinds of OpenFlow flow
5405 * table changes, we recalculate the actions for every facet. This is
5406 * relatively expensive, but it's good enough if the OpenFlow flow table
5407 * doesn't change very often.
5409 * However, we can expect one particular kind of OpenFlow flow table change to
5410 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5411 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5412 * table, we add a special case that applies to flow tables in which every rule
5413 * has the same form (that is, the same wildcards), except that the table is
5414 * also allowed to have a single "catch-all" flow that matches all packets. We
5415 * optimize this case by tagging all of the facets that resubmit into the table
5416 * and invalidating the same tag whenever a flow changes in that table. The
5417 * end result is that we revalidate just the facets that need it (and sometimes
5418 * a few more, but not all of the facets or even all of the facets that
5419 * resubmit to the table modified by MAC learning). */
5421 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
5422 * into an OpenFlow table with the given 'basis'. */
5424 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
5427 if (flow_wildcards_is_catchall(wc)) {
5430 struct flow tag_flow = *flow;
5431 flow_zero_wildcards(&tag_flow, wc);
5432 return tag_create_deterministic(flow_hash(&tag_flow, secret));
5436 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5437 * taggability of that table.
5439 * This function must be called after *each* change to a flow table. If you
5440 * skip calling it on some changes then the pointer comparisons at the end can
5441 * be invalid if you get unlucky. For example, if a flow removal causes a
5442 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5443 * different wildcards to be created with the same address, then this function
5444 * will incorrectly skip revalidation. */
5446 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5448 struct table_dpif *table = &ofproto->tables[table_id];
5449 const struct classifier *cls = &ofproto->up.tables[table_id];
5450 struct cls_table *catchall, *other;
5451 struct cls_table *t;
5453 catchall = other = NULL;
5455 switch (hmap_count(&cls->tables)) {
5457 /* We could tag this OpenFlow table but it would make the logic a
5458 * little harder and it's a corner case that doesn't seem worth it
5464 HMAP_FOR_EACH (t, hmap_node, &cls->tables) {
5465 if (cls_table_is_catchall(t)) {
5467 } else if (!other) {
5470 /* Indicate that we can't tag this by setting both tables to
5471 * NULL. (We know that 'catchall' is already NULL.) */
5478 /* Can't tag this table. */
5482 if (table->catchall_table != catchall || table->other_table != other) {
5483 table->catchall_table = catchall;
5484 table->other_table = other;
5485 ofproto->need_revalidate = true;
5489 /* Given 'rule' that has changed in some way (either it is a rule being
5490 * inserted, a rule being deleted, or a rule whose actions are being
5491 * modified), marks facets for revalidation to ensure that packets will be
5492 * forwarded correctly according to the new state of the flow table.
5494 * This function must be called after *each* change to a flow table. See
5495 * the comment on table_update_taggable() for more information. */
5497 rule_invalidate(const struct rule_dpif *rule)
5499 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5501 table_update_taggable(ofproto, rule->up.table_id);
5503 if (!ofproto->need_revalidate) {
5504 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5506 if (table->other_table && rule->tag) {
5507 tag_set_add(&ofproto->revalidate_set, rule->tag);
5509 ofproto->need_revalidate = true;
5515 set_frag_handling(struct ofproto *ofproto_,
5516 enum ofp_config_flags frag_handling)
5518 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5520 if (frag_handling != OFPC_FRAG_REASM) {
5521 ofproto->need_revalidate = true;
5529 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5530 const struct flow *flow,
5531 const union ofp_action *ofp_actions, size_t n_ofp_actions)
5533 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5536 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
5537 return ofp_mkerr_nicira(OFPET_BAD_REQUEST, NXBRC_BAD_IN_PORT);
5540 error = validate_actions(ofp_actions, n_ofp_actions, flow,
5541 ofproto->max_ports);
5543 struct odputil_keybuf keybuf;
5544 struct action_xlate_ctx ctx;
5545 struct ofpbuf *odp_actions;
5548 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5549 odp_flow_key_from_flow(&key, flow);
5551 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, packet);
5552 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
5553 dpif_execute(ofproto->dpif, key.data, key.size,
5554 odp_actions->data, odp_actions->size, packet);
5555 ofpbuf_delete(odp_actions);
5563 set_netflow(struct ofproto *ofproto_,
5564 const struct netflow_options *netflow_options)
5566 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5568 if (netflow_options) {
5569 if (!ofproto->netflow) {
5570 ofproto->netflow = netflow_create();
5572 return netflow_set_options(ofproto->netflow, netflow_options);
5574 netflow_destroy(ofproto->netflow);
5575 ofproto->netflow = NULL;
5581 get_netflow_ids(const struct ofproto *ofproto_,
5582 uint8_t *engine_type, uint8_t *engine_id)
5584 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5586 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
5590 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5592 if (!facet_is_controller_flow(facet) &&
5593 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5594 struct subfacet *subfacet;
5595 struct ofexpired expired;
5597 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5598 if (subfacet->installed) {
5599 struct dpif_flow_stats stats;
5601 subfacet_install(ofproto, subfacet, subfacet->actions,
5602 subfacet->actions_len, &stats);
5603 subfacet_update_stats(ofproto, subfacet, &stats);
5607 expired.flow = facet->flow;
5608 expired.packet_count = facet->packet_count;
5609 expired.byte_count = facet->byte_count;
5610 expired.used = facet->used;
5611 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5616 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5618 struct facet *facet;
5620 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5621 send_active_timeout(ofproto, facet);
5625 static struct ofproto_dpif *
5626 ofproto_dpif_lookup(const char *name)
5628 struct ofproto_dpif *ofproto;
5630 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5631 hash_string(name, 0), &all_ofproto_dpifs) {
5632 if (!strcmp(ofproto->up.name, name)) {
5640 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc OVS_UNUSED,
5641 const char *argv[], void *aux OVS_UNUSED)
5643 const struct ofproto_dpif *ofproto;
5645 ofproto = ofproto_dpif_lookup(argv[1]);
5647 unixctl_command_reply(conn, 501, "no such bridge");
5650 mac_learning_flush(ofproto->ml);
5652 unixctl_command_reply(conn, 200, "table successfully flushed");
5656 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5657 const char *argv[], void *aux OVS_UNUSED)
5659 struct ds ds = DS_EMPTY_INITIALIZER;
5660 const struct ofproto_dpif *ofproto;
5661 const struct mac_entry *e;
5663 ofproto = ofproto_dpif_lookup(argv[1]);
5665 unixctl_command_reply(conn, 501, "no such bridge");
5669 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5670 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5671 struct ofbundle *bundle = e->port.p;
5672 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
5673 ofbundle_get_a_port(bundle)->odp_port,
5674 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
5676 unixctl_command_reply(conn, 200, ds_cstr(&ds));
5680 struct ofproto_trace {
5681 struct action_xlate_ctx ctx;
5687 trace_format_rule(struct ds *result, uint8_t table_id, int level,
5688 const struct rule_dpif *rule)
5690 ds_put_char_multiple(result, '\t', level);
5692 ds_put_cstr(result, "No match\n");
5696 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5697 table_id, ntohll(rule->up.flow_cookie));
5698 cls_rule_format(&rule->up.cr, result);
5699 ds_put_char(result, '\n');
5701 ds_put_char_multiple(result, '\t', level);
5702 ds_put_cstr(result, "OpenFlow ");
5703 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
5704 ds_put_char(result, '\n');
5708 trace_format_flow(struct ds *result, int level, const char *title,
5709 struct ofproto_trace *trace)
5711 ds_put_char_multiple(result, '\t', level);
5712 ds_put_format(result, "%s: ", title);
5713 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
5714 ds_put_cstr(result, "unchanged");
5716 flow_format(result, &trace->ctx.flow);
5717 trace->flow = trace->ctx.flow;
5719 ds_put_char(result, '\n');
5723 trace_format_regs(struct ds *result, int level, const char *title,
5724 struct ofproto_trace *trace)
5728 ds_put_char_multiple(result, '\t', level);
5729 ds_put_format(result, "%s:", title);
5730 for (i = 0; i < FLOW_N_REGS; i++) {
5731 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5733 ds_put_char(result, '\n');
5737 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
5739 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
5740 struct ds *result = trace->result;
5742 ds_put_char(result, '\n');
5743 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
5744 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
5745 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
5749 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5750 void *aux OVS_UNUSED)
5752 const char *dpname = argv[1];
5753 struct ofproto_dpif *ofproto;
5754 struct ofpbuf odp_key;
5755 struct ofpbuf *packet;
5756 struct rule_dpif *rule;
5757 ovs_be16 initial_tci;
5763 ofpbuf_init(&odp_key, 0);
5766 ofproto = ofproto_dpif_lookup(dpname);
5768 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
5772 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
5773 /* ofproto/trace dpname flow [-generate] */
5774 const char *flow_s = argv[2];
5775 const char *generate_s = argv[3];
5778 /* Convert string to datapath key. */
5779 ofpbuf_init(&odp_key, 0);
5780 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
5782 unixctl_command_reply(conn, 501, "Bad flow syntax");
5786 /* Convert odp_key to flow. */
5787 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
5788 odp_key.size, &flow,
5789 &initial_tci, NULL);
5790 if (error == ODP_FIT_ERROR) {
5791 unixctl_command_reply(conn, 501, "Invalid flow");
5795 /* Generate a packet, if requested. */
5797 packet = ofpbuf_new(0);
5798 flow_compose(packet, &flow);
5800 } else if (argc == 6) {
5801 /* ofproto/trace dpname priority tun_id in_port packet */
5802 const char *priority_s = argv[2];
5803 const char *tun_id_s = argv[3];
5804 const char *in_port_s = argv[4];
5805 const char *packet_s = argv[5];
5806 uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s));
5807 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
5808 uint32_t priority = atoi(priority_s);
5811 msg = eth_from_hex(packet_s, &packet);
5813 unixctl_command_reply(conn, 501, msg);
5817 ds_put_cstr(&result, "Packet: ");
5818 s = ofp_packet_to_string(packet->data, packet->size);
5819 ds_put_cstr(&result, s);
5822 flow_extract(packet, priority, tun_id, in_port, &flow);
5823 initial_tci = flow.vlan_tci;
5825 unixctl_command_reply(conn, 501, "Bad command syntax");
5829 ds_put_cstr(&result, "Flow: ");
5830 flow_format(&result, &flow);
5831 ds_put_char(&result, '\n');
5833 rule = rule_dpif_lookup(ofproto, &flow, 0);
5834 trace_format_rule(&result, 0, 0, rule);
5836 struct ofproto_trace trace;
5837 struct ofpbuf *odp_actions;
5839 trace.result = &result;
5841 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, initial_tci, packet);
5842 trace.ctx.resubmit_hook = trace_resubmit;
5843 odp_actions = xlate_actions(&trace.ctx,
5844 rule->up.actions, rule->up.n_actions);
5846 ds_put_char(&result, '\n');
5847 trace_format_flow(&result, 0, "Final flow", &trace);
5848 ds_put_cstr(&result, "Datapath actions: ");
5849 format_odp_actions(&result, odp_actions->data, odp_actions->size);
5850 ofpbuf_delete(odp_actions);
5852 if (!trace.ctx.may_set_up_flow) {
5854 ds_put_cstr(&result, "\nThis flow is not cachable.");
5856 ds_put_cstr(&result, "\nThe datapath actions are incomplete--"
5857 "for complete actions, please supply a packet.");
5862 unixctl_command_reply(conn, 200, ds_cstr(&result));
5865 ds_destroy(&result);
5866 ofpbuf_delete(packet);
5867 ofpbuf_uninit(&odp_key);
5871 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5872 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5875 unixctl_command_reply(conn, 200, NULL);
5879 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5880 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5883 unixctl_command_reply(conn, 200, NULL);
5887 ofproto_dpif_unixctl_init(void)
5889 static bool registered;
5895 unixctl_command_register(
5897 "bridge {tun_id in_port packet | odp_flow [-generate]}",
5898 2, 4, ofproto_unixctl_trace, NULL);
5899 unixctl_command_register("fdb/flush", "bridge", 1, 1,
5900 ofproto_unixctl_fdb_flush, NULL);
5901 unixctl_command_register("fdb/show", "bridge", 1, 1,
5902 ofproto_unixctl_fdb_show, NULL);
5903 unixctl_command_register("ofproto/clog", "", 0, 0,
5904 ofproto_dpif_clog, NULL);
5905 unixctl_command_register("ofproto/unclog", "", 0, 0,
5906 ofproto_dpif_unclog, NULL);
5909 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5911 * This is deprecated. It is only for compatibility with broken device drivers
5912 * in old versions of Linux that do not properly support VLANs when VLAN
5913 * devices are not used. When broken device drivers are no longer in
5914 * widespread use, we will delete these interfaces. */
5917 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
5919 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
5920 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
5922 if (realdev_ofp_port == ofport->realdev_ofp_port
5923 && vid == ofport->vlandev_vid) {
5927 ofproto->need_revalidate = true;
5929 if (ofport->realdev_ofp_port) {
5932 if (realdev_ofp_port && ofport->bundle) {
5933 /* vlandevs are enslaved to their realdevs, so they are not allowed to
5934 * themselves be part of a bundle. */
5935 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
5938 ofport->realdev_ofp_port = realdev_ofp_port;
5939 ofport->vlandev_vid = vid;
5941 if (realdev_ofp_port) {
5942 vsp_add(ofport, realdev_ofp_port, vid);
5949 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
5951 return hash_2words(realdev_ofp_port, vid);
5955 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
5956 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
5958 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
5959 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
5960 int vid = vlan_tci_to_vid(vlan_tci);
5961 const struct vlan_splinter *vsp;
5963 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
5964 hash_realdev_vid(realdev_ofp_port, vid),
5965 &ofproto->realdev_vid_map) {
5966 if (vsp->realdev_ofp_port == realdev_ofp_port
5967 && vsp->vid == vid) {
5968 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
5972 return realdev_odp_port;
5975 static struct vlan_splinter *
5976 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
5978 struct vlan_splinter *vsp;
5980 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
5981 &ofproto->vlandev_map) {
5982 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
5991 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
5992 uint16_t vlandev_ofp_port, int *vid)
5994 if (!hmap_is_empty(&ofproto->vlandev_map)) {
5995 const struct vlan_splinter *vsp;
5997 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6002 return vsp->realdev_ofp_port;
6009 vsp_remove(struct ofport_dpif *port)
6011 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6012 struct vlan_splinter *vsp;
6014 vsp = vlandev_find(ofproto, port->up.ofp_port);
6016 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6017 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6020 port->realdev_ofp_port = 0;
6022 VLOG_ERR("missing vlan device record");
6027 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
6029 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6031 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6032 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6033 == realdev_ofp_port)) {
6034 struct vlan_splinter *vsp;
6036 vsp = xmalloc(sizeof *vsp);
6037 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6038 hash_int(port->up.ofp_port, 0));
6039 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6040 hash_realdev_vid(realdev_ofp_port, vid));
6041 vsp->realdev_ofp_port = realdev_ofp_port;
6042 vsp->vlandev_ofp_port = port->up.ofp_port;
6045 port->realdev_ofp_port = realdev_ofp_port;
6047 VLOG_ERR("duplicate vlan device record");
6051 const struct ofproto_class ofproto_dpif_class = {
6080 port_is_lacp_current,
6081 NULL, /* rule_choose_table */
6088 rule_modify_actions,
6096 get_cfm_remote_mpids,
6100 get_stp_port_status,
6107 is_mirror_output_bundle,
6108 forward_bpdu_changed,