2 * Copyright (c) 2009, 2010, 2011 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-provider.h"
26 #include "byte-order.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
36 #include "mac-learning.h"
37 #include "multipath.h"
44 #include "ofp-print.h"
45 #include "ofproto-dpif-sflow.h"
46 #include "poll-loop.h"
48 #include "unaligned.h"
50 #include "vlan-bitmap.h"
53 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
55 COVERAGE_DEFINE(ofproto_dpif_ctlr_action);
56 COVERAGE_DEFINE(ofproto_dpif_expired);
57 COVERAGE_DEFINE(ofproto_dpif_no_packet_in);
58 COVERAGE_DEFINE(ofproto_dpif_xlate);
59 COVERAGE_DEFINE(facet_changed_rule);
60 COVERAGE_DEFINE(facet_invalidated);
61 COVERAGE_DEFINE(facet_revalidate);
62 COVERAGE_DEFINE(facet_unexpected);
64 /* Maximum depth of flow table recursion (due to resubmit actions) in a
65 * flow translation. */
66 #define MAX_RESUBMIT_RECURSION 32
68 /* Number of implemented OpenFlow tables. */
69 enum { N_TABLES = 255 };
70 BUILD_ASSERT_DECL(N_TABLES >= 1 && N_TABLES <= 255);
78 long long int used; /* Time last used; time created if not used. */
82 * - Do include packets and bytes from facets that have been deleted or
83 * whose own statistics have been folded into the rule.
85 * - Do include packets and bytes sent "by hand" that were accounted to
86 * the rule without any facet being involved (this is a rare corner
87 * case in rule_execute()).
89 * - Do not include packet or bytes that can be obtained from any facet's
90 * packet_count or byte_count member or that can be obtained from the
91 * datapath by, e.g., dpif_flow_get() for any subfacet.
93 uint64_t packet_count; /* Number of packets received. */
94 uint64_t byte_count; /* Number of bytes received. */
96 tag_type tag; /* Caches rule_calculate_tag() result. */
98 struct list facets; /* List of "struct facet"s. */
101 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
103 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
106 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
107 const struct flow *, uint8_t table);
109 static void flow_push_stats(const struct rule_dpif *, const struct flow *,
110 uint64_t packets, uint64_t bytes,
113 static uint32_t rule_calculate_tag(const struct flow *,
114 const struct flow_wildcards *,
116 static void rule_invalidate(const struct rule_dpif *);
118 #define MAX_MIRRORS 32
119 typedef uint32_t mirror_mask_t;
120 #define MIRROR_MASK_C(X) UINT32_C(X)
121 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
123 struct ofproto_dpif *ofproto; /* Owning ofproto. */
124 size_t idx; /* In ofproto's "mirrors" array. */
125 void *aux; /* Key supplied by ofproto's client. */
126 char *name; /* Identifier for log messages. */
128 /* Selection criteria. */
129 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
130 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
131 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
133 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
134 struct ofbundle *out; /* Output port or NULL. */
135 int out_vlan; /* Output VLAN or -1. */
136 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
139 int64_t packet_count; /* Number of packets sent. */
140 int64_t byte_count; /* Number of bytes sent. */
143 static void mirror_destroy(struct ofmirror *);
144 static void update_mirror_stats(struct ofproto_dpif *ofproto,
145 mirror_mask_t mirrors,
146 uint64_t packets, uint64_t bytes);
149 struct ofproto_dpif *ofproto; /* Owning ofproto. */
150 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
151 void *aux; /* Key supplied by ofproto's client. */
152 char *name; /* Identifier for log messages. */
155 struct list ports; /* Contains "struct ofport"s. */
156 enum port_vlan_mode vlan_mode; /* VLAN mode */
157 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
158 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
159 * NULL if all VLANs are trunked. */
160 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
161 struct bond *bond; /* Nonnull iff more than one port. */
162 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
165 bool floodable; /* True if no port has OFPPC_NO_FLOOD set. */
167 /* Port mirroring info. */
168 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
169 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
170 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
173 static void bundle_remove(struct ofport *);
174 static void bundle_update(struct ofbundle *);
175 static void bundle_destroy(struct ofbundle *);
176 static void bundle_del_port(struct ofport_dpif *);
177 static void bundle_run(struct ofbundle *);
178 static void bundle_wait(struct ofbundle *);
179 static struct 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 original_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, bool clone);
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. */
525 long long int stp_last_tick;
527 /* VLAN splinters. */
528 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
529 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
532 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
533 * for debugging the asynchronous flow_mod implementation.) */
536 /* All existing ofproto_dpif instances, indexed by ->up.name. */
537 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
539 static void ofproto_dpif_unixctl_init(void);
541 static struct ofproto_dpif *
542 ofproto_dpif_cast(const struct ofproto *ofproto)
544 assert(ofproto->ofproto_class == &ofproto_dpif_class);
545 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
548 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
550 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
553 /* Packet processing. */
554 static void update_learning_table(struct ofproto_dpif *,
555 const struct flow *, int vlan,
558 #define FLOW_MISS_MAX_BATCH 50
559 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
561 /* Flow expiration. */
562 static int expire(struct ofproto_dpif *);
565 static void send_netflow_active_timeouts(struct ofproto_dpif *);
568 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
570 compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions,
571 const struct flow *, uint32_t odp_port);
572 static void add_mirror_actions(struct action_xlate_ctx *ctx,
573 const struct flow *flow);
574 /* Global variables. */
575 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
577 /* Factory functions. */
580 enumerate_types(struct sset *types)
582 dp_enumerate_types(types);
586 enumerate_names(const char *type, struct sset *names)
588 return dp_enumerate_names(type, names);
592 del(const char *type, const char *name)
597 error = dpif_open(name, type, &dpif);
599 error = dpif_delete(dpif);
605 /* Basic life-cycle. */
607 static struct ofproto *
610 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
615 dealloc(struct ofproto *ofproto_)
617 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
622 construct(struct ofproto *ofproto_, int *n_tablesp)
624 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
625 const char *name = ofproto->up.name;
629 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
631 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
635 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
636 ofproto->n_matches = 0;
638 dpif_flow_flush(ofproto->dpif);
639 dpif_recv_purge(ofproto->dpif);
641 error = dpif_recv_set_mask(ofproto->dpif,
642 ((1u << DPIF_UC_MISS) |
643 (1u << DPIF_UC_ACTION)));
645 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
646 dpif_close(ofproto->dpif);
650 ofproto->netflow = NULL;
651 ofproto->sflow = NULL;
653 hmap_init(&ofproto->bundles);
654 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
655 for (i = 0; i < MAX_MIRRORS; i++) {
656 ofproto->mirrors[i] = NULL;
658 ofproto->has_bonded_bundles = false;
660 timer_set_duration(&ofproto->next_expiration, 1000);
662 hmap_init(&ofproto->facets);
663 hmap_init(&ofproto->subfacets);
665 for (i = 0; i < N_TABLES; i++) {
666 struct table_dpif *table = &ofproto->tables[i];
668 table->catchall_table = NULL;
669 table->other_table = NULL;
670 table->basis = random_uint32();
672 ofproto->need_revalidate = false;
673 tag_set_init(&ofproto->revalidate_set);
675 list_init(&ofproto->completions);
677 ofproto_dpif_unixctl_init();
679 ofproto->has_bundle_action = false;
681 hmap_init(&ofproto->vlandev_map);
682 hmap_init(&ofproto->realdev_vid_map);
684 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
685 hash_string(ofproto->up.name, 0));
687 *n_tablesp = N_TABLES;
692 complete_operations(struct ofproto_dpif *ofproto)
694 struct dpif_completion *c, *next;
696 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
697 ofoperation_complete(c->op, 0);
698 list_remove(&c->list_node);
704 destruct(struct ofproto *ofproto_)
706 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
707 struct rule_dpif *rule, *next_rule;
708 struct classifier *table;
711 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
712 complete_operations(ofproto);
714 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
715 struct cls_cursor cursor;
717 cls_cursor_init(&cursor, table, NULL);
718 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
719 ofproto_rule_destroy(&rule->up);
723 for (i = 0; i < MAX_MIRRORS; i++) {
724 mirror_destroy(ofproto->mirrors[i]);
727 netflow_destroy(ofproto->netflow);
728 dpif_sflow_destroy(ofproto->sflow);
729 hmap_destroy(&ofproto->bundles);
730 mac_learning_destroy(ofproto->ml);
732 hmap_destroy(&ofproto->facets);
733 hmap_destroy(&ofproto->subfacets);
735 hmap_destroy(&ofproto->vlandev_map);
736 hmap_destroy(&ofproto->realdev_vid_map);
738 dpif_close(ofproto->dpif);
742 run_fast(struct ofproto *ofproto_)
744 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
747 /* Handle one or more batches of upcalls, until there's nothing left to do
748 * or until we do a fixed total amount of work.
750 * We do work in batches because it can be much cheaper to set up a number
751 * of flows and fire off their patches all at once. We do multiple batches
752 * because in some cases handling a packet can cause another packet to be
753 * queued almost immediately as part of the return flow. Both
754 * optimizations can make major improvements on some benchmarks and
755 * presumably for real traffic as well. */
757 while (work < FLOW_MISS_MAX_BATCH) {
758 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
768 run(struct ofproto *ofproto_)
770 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
771 struct ofport_dpif *ofport;
772 struct ofbundle *bundle;
776 complete_operations(ofproto);
778 dpif_run(ofproto->dpif);
780 error = run_fast(ofproto_);
785 if (timer_expired(&ofproto->next_expiration)) {
786 int delay = expire(ofproto);
787 timer_set_duration(&ofproto->next_expiration, delay);
790 if (ofproto->netflow) {
791 if (netflow_run(ofproto->netflow)) {
792 send_netflow_active_timeouts(ofproto);
795 if (ofproto->sflow) {
796 dpif_sflow_run(ofproto->sflow);
799 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
802 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
807 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
809 /* Now revalidate if there's anything to do. */
810 if (ofproto->need_revalidate
811 || !tag_set_is_empty(&ofproto->revalidate_set)) {
812 struct tag_set revalidate_set = ofproto->revalidate_set;
813 bool revalidate_all = ofproto->need_revalidate;
814 struct facet *facet, *next;
816 /* Clear the revalidation flags. */
817 tag_set_init(&ofproto->revalidate_set);
818 ofproto->need_revalidate = false;
820 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
822 || tag_set_intersects(&revalidate_set, facet->tags)) {
823 facet_revalidate(ofproto, facet);
832 wait(struct ofproto *ofproto_)
834 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
835 struct ofport_dpif *ofport;
836 struct ofbundle *bundle;
838 if (!clogged && !list_is_empty(&ofproto->completions)) {
839 poll_immediate_wake();
842 dpif_wait(ofproto->dpif);
843 dpif_recv_wait(ofproto->dpif);
844 if (ofproto->sflow) {
845 dpif_sflow_wait(ofproto->sflow);
847 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
848 poll_immediate_wake();
850 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
853 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
856 if (ofproto->netflow) {
857 netflow_wait(ofproto->netflow);
859 mac_learning_wait(ofproto->ml);
861 if (ofproto->need_revalidate) {
862 /* Shouldn't happen, but if it does just go around again. */
863 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
864 poll_immediate_wake();
866 timer_wait(&ofproto->next_expiration);
871 flush(struct ofproto *ofproto_)
873 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
874 struct facet *facet, *next_facet;
876 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
877 /* Mark the facet as not installed so that facet_remove() doesn't
878 * bother trying to uninstall it. There is no point in uninstalling it
879 * individually since we are about to blow away all the facets with
880 * dpif_flow_flush(). */
881 struct subfacet *subfacet;
883 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
884 subfacet->installed = false;
885 subfacet->dp_packet_count = 0;
886 subfacet->dp_byte_count = 0;
888 facet_remove(ofproto, facet);
890 dpif_flow_flush(ofproto->dpif);
894 get_features(struct ofproto *ofproto_ OVS_UNUSED,
895 bool *arp_match_ip, uint32_t *actions)
897 *arp_match_ip = true;
898 *actions = ((1u << OFPAT_OUTPUT) |
899 (1u << OFPAT_SET_VLAN_VID) |
900 (1u << OFPAT_SET_VLAN_PCP) |
901 (1u << OFPAT_STRIP_VLAN) |
902 (1u << OFPAT_SET_DL_SRC) |
903 (1u << OFPAT_SET_DL_DST) |
904 (1u << OFPAT_SET_NW_SRC) |
905 (1u << OFPAT_SET_NW_DST) |
906 (1u << OFPAT_SET_NW_TOS) |
907 (1u << OFPAT_SET_TP_SRC) |
908 (1u << OFPAT_SET_TP_DST) |
909 (1u << OFPAT_ENQUEUE));
913 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
915 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
916 struct dpif_dp_stats s;
918 strcpy(ots->name, "classifier");
920 dpif_get_dp_stats(ofproto->dpif, &s);
921 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
922 put_32aligned_be64(&ots->matched_count,
923 htonll(s.n_hit + ofproto->n_matches));
926 static struct ofport *
929 struct ofport_dpif *port = xmalloc(sizeof *port);
934 port_dealloc(struct ofport *port_)
936 struct ofport_dpif *port = ofport_dpif_cast(port_);
941 port_construct(struct ofport *port_)
943 struct ofport_dpif *port = ofport_dpif_cast(port_);
944 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
946 ofproto->need_revalidate = true;
947 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
950 port->tag = tag_create_random();
951 port->may_enable = true;
952 port->stp_port = NULL;
953 port->stp_state = STP_DISABLED;
954 hmap_init(&port->priorities);
955 port->realdev_ofp_port = 0;
956 port->vlandev_vid = 0;
958 if (ofproto->sflow) {
959 dpif_sflow_add_port(ofproto->sflow, port->odp_port,
960 netdev_get_name(port->up.netdev));
967 port_destruct(struct ofport *port_)
969 struct ofport_dpif *port = ofport_dpif_cast(port_);
970 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
972 ofproto->need_revalidate = true;
973 bundle_remove(port_);
974 set_cfm(port_, NULL);
975 if (ofproto->sflow) {
976 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
979 ofport_clear_priorities(port);
980 hmap_destroy(&port->priorities);
984 port_modified(struct ofport *port_)
986 struct ofport_dpif *port = ofport_dpif_cast(port_);
988 if (port->bundle && port->bundle->bond) {
989 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
994 port_reconfigured(struct ofport *port_, ovs_be32 old_config)
996 struct ofport_dpif *port = ofport_dpif_cast(port_);
997 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
998 ovs_be32 changed = old_config ^ port->up.opp.config;
1000 if (changed & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
1001 OFPPC_NO_FWD | OFPPC_NO_FLOOD)) {
1002 ofproto->need_revalidate = true;
1004 if (changed & htonl(OFPPC_NO_FLOOD) && port->bundle) {
1005 bundle_update(port->bundle);
1011 set_sflow(struct ofproto *ofproto_,
1012 const struct ofproto_sflow_options *sflow_options)
1014 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1015 struct dpif_sflow *ds = ofproto->sflow;
1017 if (sflow_options) {
1019 struct ofport_dpif *ofport;
1021 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1022 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1023 dpif_sflow_add_port(ds, ofport->odp_port,
1024 netdev_get_name(ofport->up.netdev));
1026 ofproto->need_revalidate = true;
1028 dpif_sflow_set_options(ds, sflow_options);
1031 dpif_sflow_destroy(ds);
1032 ofproto->need_revalidate = true;
1033 ofproto->sflow = NULL;
1040 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1042 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1049 struct ofproto_dpif *ofproto;
1051 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1052 ofproto->need_revalidate = true;
1053 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1056 if (cfm_configure(ofport->cfm, s)) {
1062 cfm_destroy(ofport->cfm);
1068 get_cfm_fault(const struct ofport *ofport_)
1070 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1072 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1076 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1079 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1082 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1089 /* Spanning Tree. */
1092 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1094 struct ofproto_dpif *ofproto = ofproto_;
1095 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1096 struct ofport_dpif *ofport;
1098 ofport = stp_port_get_aux(sp);
1100 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1101 ofproto->up.name, port_num);
1103 struct eth_header *eth = pkt->l2;
1105 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1106 if (eth_addr_is_zero(eth->eth_src)) {
1107 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1108 "with unknown MAC", ofproto->up.name, port_num);
1110 send_packet(ofport, pkt);
1116 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1118 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1120 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1122 /* Only revalidate flows if the configuration changed. */
1123 if (!s != !ofproto->stp) {
1124 ofproto->need_revalidate = true;
1128 if (!ofproto->stp) {
1129 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1130 send_bpdu_cb, ofproto);
1131 ofproto->stp_last_tick = time_msec();
1134 stp_set_bridge_id(ofproto->stp, s->system_id);
1135 stp_set_bridge_priority(ofproto->stp, s->priority);
1136 stp_set_hello_time(ofproto->stp, s->hello_time);
1137 stp_set_max_age(ofproto->stp, s->max_age);
1138 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1140 stp_destroy(ofproto->stp);
1141 ofproto->stp = NULL;
1148 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1150 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1154 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1155 s->designated_root = stp_get_designated_root(ofproto->stp);
1156 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1165 update_stp_port_state(struct ofport_dpif *ofport)
1167 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1168 enum stp_state state;
1170 /* Figure out new state. */
1171 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1175 if (ofport->stp_state != state) {
1179 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1180 netdev_get_name(ofport->up.netdev),
1181 stp_state_name(ofport->stp_state),
1182 stp_state_name(state));
1183 if (stp_learn_in_state(ofport->stp_state)
1184 != stp_learn_in_state(state)) {
1185 /* xxx Learning action flows should also be flushed. */
1186 mac_learning_flush(ofproto->ml);
1188 fwd_change = stp_forward_in_state(ofport->stp_state)
1189 != stp_forward_in_state(state);
1191 ofproto->need_revalidate = true;
1192 ofport->stp_state = state;
1193 ofport->stp_state_entered = time_msec();
1195 if (fwd_change && ofport->bundle) {
1196 bundle_update(ofport->bundle);
1199 /* Update the STP state bits in the OpenFlow port description. */
1200 of_state = (ofport->up.opp.state & htonl(~OFPPS_STP_MASK))
1201 | htonl(state == STP_LISTENING ? OFPPS_STP_LISTEN
1202 : state == STP_LEARNING ? OFPPS_STP_LEARN
1203 : state == STP_FORWARDING ? OFPPS_STP_FORWARD
1204 : state == STP_BLOCKING ? OFPPS_STP_BLOCK
1206 ofproto_port_set_state(&ofport->up, of_state);
1210 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1211 * caller is responsible for assigning STP port numbers and ensuring
1212 * there are no duplicates. */
1214 set_stp_port(struct ofport *ofport_,
1215 const struct ofproto_port_stp_settings *s)
1217 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1218 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1219 struct stp_port *sp = ofport->stp_port;
1221 if (!s || !s->enable) {
1223 ofport->stp_port = NULL;
1224 stp_port_disable(sp);
1225 update_stp_port_state(ofport);
1228 } else if (sp && stp_port_no(sp) != s->port_num
1229 && ofport == stp_port_get_aux(sp)) {
1230 /* The port-id changed, so disable the old one if it's not
1231 * already in use by another port. */
1232 stp_port_disable(sp);
1235 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1236 stp_port_enable(sp);
1238 stp_port_set_aux(sp, ofport);
1239 stp_port_set_priority(sp, s->priority);
1240 stp_port_set_path_cost(sp, s->path_cost);
1242 update_stp_port_state(ofport);
1248 get_stp_port_status(struct ofport *ofport_,
1249 struct ofproto_port_stp_status *s)
1251 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1252 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1253 struct stp_port *sp = ofport->stp_port;
1255 if (!ofproto->stp || !sp) {
1261 s->port_id = stp_port_get_id(sp);
1262 s->state = stp_port_get_state(sp);
1263 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1264 s->role = stp_port_get_role(sp);
1265 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1271 stp_run(struct ofproto_dpif *ofproto)
1274 long long int now = time_msec();
1275 long long int elapsed = now - ofproto->stp_last_tick;
1276 struct stp_port *sp;
1279 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1280 ofproto->stp_last_tick = now;
1282 while (stp_get_changed_port(ofproto->stp, &sp)) {
1283 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1286 update_stp_port_state(ofport);
1293 stp_wait(struct ofproto_dpif *ofproto)
1296 poll_timer_wait(1000);
1300 /* Returns true if STP should process 'flow'. */
1302 stp_should_process_flow(const struct flow *flow)
1304 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1308 stp_process_packet(const struct ofport_dpif *ofport,
1309 const struct ofpbuf *packet)
1311 struct ofpbuf payload = *packet;
1312 struct eth_header *eth = payload.data;
1313 struct stp_port *sp = ofport->stp_port;
1315 /* Sink packets on ports that have STP disabled when the bridge has
1317 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1321 /* Trim off padding on payload. */
1322 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1323 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1326 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1327 stp_received_bpdu(sp, payload.data, payload.size);
1331 static struct priority_to_dscp *
1332 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1334 struct priority_to_dscp *pdscp;
1337 hash = hash_int(priority, 0);
1338 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1339 if (pdscp->priority == priority) {
1347 ofport_clear_priorities(struct ofport_dpif *ofport)
1349 struct priority_to_dscp *pdscp, *next;
1351 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1352 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1358 set_queues(struct ofport *ofport_,
1359 const struct ofproto_port_queue *qdscp_list,
1362 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1363 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1364 struct hmap new = HMAP_INITIALIZER(&new);
1367 for (i = 0; i < n_qdscp; i++) {
1368 struct priority_to_dscp *pdscp;
1372 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1373 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1378 pdscp = get_priority(ofport, priority);
1380 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1382 pdscp = xmalloc(sizeof *pdscp);
1383 pdscp->priority = priority;
1385 ofproto->need_revalidate = true;
1388 if (pdscp->dscp != dscp) {
1390 ofproto->need_revalidate = true;
1393 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1396 if (!hmap_is_empty(&ofport->priorities)) {
1397 ofport_clear_priorities(ofport);
1398 ofproto->need_revalidate = true;
1401 hmap_swap(&new, &ofport->priorities);
1409 /* Expires all MAC learning entries associated with 'bundle' and forces its
1410 * ofproto to revalidate every flow.
1412 * Normally MAC learning entries are removed only from the ofproto associated
1413 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1414 * are removed from every ofproto. When patch ports and SLB bonds are in use
1415 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1416 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1417 * with the host from which it migrated. */
1419 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1421 struct ofproto_dpif *ofproto = bundle->ofproto;
1422 struct mac_learning *ml = ofproto->ml;
1423 struct mac_entry *mac, *next_mac;
1425 ofproto->need_revalidate = true;
1426 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1427 if (mac->port.p == bundle) {
1429 struct ofproto_dpif *o;
1431 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1433 struct mac_entry *e;
1435 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1438 tag_set_add(&o->revalidate_set, e->tag);
1439 mac_learning_expire(o->ml, e);
1445 mac_learning_expire(ml, mac);
1450 static struct ofbundle *
1451 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1453 struct ofbundle *bundle;
1455 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1456 &ofproto->bundles) {
1457 if (bundle->aux == aux) {
1464 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1465 * ones that are found to 'bundles'. */
1467 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1468 void **auxes, size_t n_auxes,
1469 struct hmapx *bundles)
1473 hmapx_init(bundles);
1474 for (i = 0; i < n_auxes; i++) {
1475 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1477 hmapx_add(bundles, bundle);
1483 bundle_update(struct ofbundle *bundle)
1485 struct ofport_dpif *port;
1487 bundle->floodable = true;
1488 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1489 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
1490 bundle->floodable = false;
1497 bundle_del_port(struct ofport_dpif *port)
1499 struct ofbundle *bundle = port->bundle;
1501 bundle->ofproto->need_revalidate = true;
1503 list_remove(&port->bundle_node);
1504 port->bundle = NULL;
1507 lacp_slave_unregister(bundle->lacp, port);
1510 bond_slave_unregister(bundle->bond, port);
1513 bundle_update(bundle);
1517 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1518 struct lacp_slave_settings *lacp,
1519 uint32_t bond_stable_id)
1521 struct ofport_dpif *port;
1523 port = get_ofp_port(bundle->ofproto, ofp_port);
1528 if (port->bundle != bundle) {
1529 bundle->ofproto->need_revalidate = true;
1531 bundle_del_port(port);
1534 port->bundle = bundle;
1535 list_push_back(&bundle->ports, &port->bundle_node);
1536 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
1537 bundle->floodable = false;
1541 port->bundle->ofproto->need_revalidate = true;
1542 lacp_slave_register(bundle->lacp, port, lacp);
1545 port->bond_stable_id = bond_stable_id;
1551 bundle_destroy(struct ofbundle *bundle)
1553 struct ofproto_dpif *ofproto;
1554 struct ofport_dpif *port, *next_port;
1561 ofproto = bundle->ofproto;
1562 for (i = 0; i < MAX_MIRRORS; i++) {
1563 struct ofmirror *m = ofproto->mirrors[i];
1565 if (m->out == bundle) {
1567 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1568 || hmapx_find_and_delete(&m->dsts, bundle)) {
1569 ofproto->need_revalidate = true;
1574 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1575 bundle_del_port(port);
1578 bundle_flush_macs(bundle, true);
1579 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1581 free(bundle->trunks);
1582 lacp_destroy(bundle->lacp);
1583 bond_destroy(bundle->bond);
1588 bundle_set(struct ofproto *ofproto_, void *aux,
1589 const struct ofproto_bundle_settings *s)
1591 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1592 bool need_flush = false;
1593 struct ofport_dpif *port;
1594 struct ofbundle *bundle;
1595 unsigned long *trunks;
1601 bundle_destroy(bundle_lookup(ofproto, aux));
1605 assert(s->n_slaves == 1 || s->bond != NULL);
1606 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1608 bundle = bundle_lookup(ofproto, aux);
1610 bundle = xmalloc(sizeof *bundle);
1612 bundle->ofproto = ofproto;
1613 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1614 hash_pointer(aux, 0));
1616 bundle->name = NULL;
1618 list_init(&bundle->ports);
1619 bundle->vlan_mode = PORT_VLAN_TRUNK;
1621 bundle->trunks = NULL;
1622 bundle->use_priority_tags = s->use_priority_tags;
1623 bundle->lacp = NULL;
1624 bundle->bond = NULL;
1626 bundle->floodable = true;
1628 bundle->src_mirrors = 0;
1629 bundle->dst_mirrors = 0;
1630 bundle->mirror_out = 0;
1633 if (!bundle->name || strcmp(s->name, bundle->name)) {
1635 bundle->name = xstrdup(s->name);
1640 if (!bundle->lacp) {
1641 ofproto->need_revalidate = true;
1642 bundle->lacp = lacp_create();
1644 lacp_configure(bundle->lacp, s->lacp);
1646 lacp_destroy(bundle->lacp);
1647 bundle->lacp = NULL;
1650 /* Update set of ports. */
1652 for (i = 0; i < s->n_slaves; i++) {
1653 if (!bundle_add_port(bundle, s->slaves[i],
1654 s->lacp ? &s->lacp_slaves[i] : NULL,
1655 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1659 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1660 struct ofport_dpif *next_port;
1662 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1663 for (i = 0; i < s->n_slaves; i++) {
1664 if (s->slaves[i] == port->up.ofp_port) {
1669 bundle_del_port(port);
1673 assert(list_size(&bundle->ports) <= s->n_slaves);
1675 if (list_is_empty(&bundle->ports)) {
1676 bundle_destroy(bundle);
1680 /* Set VLAN tagging mode */
1681 if (s->vlan_mode != bundle->vlan_mode
1682 || s->use_priority_tags != bundle->use_priority_tags) {
1683 bundle->vlan_mode = s->vlan_mode;
1684 bundle->use_priority_tags = s->use_priority_tags;
1689 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1690 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1692 if (vlan != bundle->vlan) {
1693 bundle->vlan = vlan;
1697 /* Get trunked VLANs. */
1698 switch (s->vlan_mode) {
1699 case PORT_VLAN_ACCESS:
1703 case PORT_VLAN_TRUNK:
1704 trunks = (unsigned long *) s->trunks;
1707 case PORT_VLAN_NATIVE_UNTAGGED:
1708 case PORT_VLAN_NATIVE_TAGGED:
1709 if (vlan != 0 && (!s->trunks
1710 || !bitmap_is_set(s->trunks, vlan)
1711 || bitmap_is_set(s->trunks, 0))) {
1712 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1714 trunks = bitmap_clone(s->trunks, 4096);
1716 trunks = bitmap_allocate1(4096);
1718 bitmap_set1(trunks, vlan);
1719 bitmap_set0(trunks, 0);
1721 trunks = (unsigned long *) s->trunks;
1728 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1729 free(bundle->trunks);
1730 if (trunks == s->trunks) {
1731 bundle->trunks = vlan_bitmap_clone(trunks);
1733 bundle->trunks = trunks;
1738 if (trunks != s->trunks) {
1743 if (!list_is_short(&bundle->ports)) {
1744 bundle->ofproto->has_bonded_bundles = true;
1746 if (bond_reconfigure(bundle->bond, s->bond)) {
1747 ofproto->need_revalidate = true;
1750 bundle->bond = bond_create(s->bond);
1751 ofproto->need_revalidate = true;
1754 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1755 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1759 bond_destroy(bundle->bond);
1760 bundle->bond = NULL;
1763 /* If we changed something that would affect MAC learning, un-learn
1764 * everything on this port and force flow revalidation. */
1766 bundle_flush_macs(bundle, false);
1773 bundle_remove(struct ofport *port_)
1775 struct ofport_dpif *port = ofport_dpif_cast(port_);
1776 struct ofbundle *bundle = port->bundle;
1779 bundle_del_port(port);
1780 if (list_is_empty(&bundle->ports)) {
1781 bundle_destroy(bundle);
1782 } else if (list_is_short(&bundle->ports)) {
1783 bond_destroy(bundle->bond);
1784 bundle->bond = NULL;
1790 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
1792 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1793 struct ofport_dpif *port = port_;
1794 uint8_t ea[ETH_ADDR_LEN];
1797 error = netdev_get_etheraddr(port->up.netdev, ea);
1799 struct ofpbuf packet;
1802 ofpbuf_init(&packet, 0);
1803 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1805 memcpy(packet_pdu, pdu, pdu_size);
1807 send_packet(port, &packet);
1808 ofpbuf_uninit(&packet);
1810 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1811 "%s (%s)", port->bundle->name,
1812 netdev_get_name(port->up.netdev), strerror(error));
1817 bundle_send_learning_packets(struct ofbundle *bundle)
1819 struct ofproto_dpif *ofproto = bundle->ofproto;
1820 int error, n_packets, n_errors;
1821 struct mac_entry *e;
1823 error = n_packets = n_errors = 0;
1824 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1825 if (e->port.p != bundle) {
1826 struct ofpbuf *learning_packet;
1827 struct ofport_dpif *port;
1830 learning_packet = bond_compose_learning_packet(bundle->bond, e->mac,
1833 ret = send_packet(port, learning_packet);
1834 ofpbuf_delete(learning_packet);
1844 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1845 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1846 "packets, last error was: %s",
1847 bundle->name, n_errors, n_packets, strerror(error));
1849 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1850 bundle->name, n_packets);
1855 bundle_run(struct ofbundle *bundle)
1858 lacp_run(bundle->lacp, send_pdu_cb);
1861 struct ofport_dpif *port;
1863 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1864 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1867 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1868 lacp_negotiated(bundle->lacp));
1869 if (bond_should_send_learning_packets(bundle->bond)) {
1870 bundle_send_learning_packets(bundle);
1876 bundle_wait(struct ofbundle *bundle)
1879 lacp_wait(bundle->lacp);
1882 bond_wait(bundle->bond);
1889 mirror_scan(struct ofproto_dpif *ofproto)
1893 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1894 if (!ofproto->mirrors[idx]) {
1901 static struct ofmirror *
1902 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1906 for (i = 0; i < MAX_MIRRORS; i++) {
1907 struct ofmirror *mirror = ofproto->mirrors[i];
1908 if (mirror && mirror->aux == aux) {
1916 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
1918 mirror_update_dups(struct ofproto_dpif *ofproto)
1922 for (i = 0; i < MAX_MIRRORS; i++) {
1923 struct ofmirror *m = ofproto->mirrors[i];
1926 m->dup_mirrors = MIRROR_MASK_C(1) << i;
1930 for (i = 0; i < MAX_MIRRORS; i++) {
1931 struct ofmirror *m1 = ofproto->mirrors[i];
1938 for (j = i + 1; j < MAX_MIRRORS; j++) {
1939 struct ofmirror *m2 = ofproto->mirrors[j];
1941 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
1942 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
1943 m2->dup_mirrors |= m1->dup_mirrors;
1950 mirror_set(struct ofproto *ofproto_, void *aux,
1951 const struct ofproto_mirror_settings *s)
1953 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1954 mirror_mask_t mirror_bit;
1955 struct ofbundle *bundle;
1956 struct ofmirror *mirror;
1957 struct ofbundle *out;
1958 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
1959 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
1962 mirror = mirror_lookup(ofproto, aux);
1964 mirror_destroy(mirror);
1970 idx = mirror_scan(ofproto);
1972 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
1974 ofproto->up.name, MAX_MIRRORS, s->name);
1978 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
1979 mirror->ofproto = ofproto;
1982 mirror->out_vlan = -1;
1983 mirror->name = NULL;
1986 if (!mirror->name || strcmp(s->name, mirror->name)) {
1988 mirror->name = xstrdup(s->name);
1991 /* Get the new configuration. */
1992 if (s->out_bundle) {
1993 out = bundle_lookup(ofproto, s->out_bundle);
1995 mirror_destroy(mirror);
2001 out_vlan = s->out_vlan;
2003 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2004 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2006 /* If the configuration has not changed, do nothing. */
2007 if (hmapx_equals(&srcs, &mirror->srcs)
2008 && hmapx_equals(&dsts, &mirror->dsts)
2009 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2010 && mirror->out == out
2011 && mirror->out_vlan == out_vlan)
2013 hmapx_destroy(&srcs);
2014 hmapx_destroy(&dsts);
2018 hmapx_swap(&srcs, &mirror->srcs);
2019 hmapx_destroy(&srcs);
2021 hmapx_swap(&dsts, &mirror->dsts);
2022 hmapx_destroy(&dsts);
2024 free(mirror->vlans);
2025 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2028 mirror->out_vlan = out_vlan;
2030 /* Update bundles. */
2031 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2032 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2033 if (hmapx_contains(&mirror->srcs, bundle)) {
2034 bundle->src_mirrors |= mirror_bit;
2036 bundle->src_mirrors &= ~mirror_bit;
2039 if (hmapx_contains(&mirror->dsts, bundle)) {
2040 bundle->dst_mirrors |= mirror_bit;
2042 bundle->dst_mirrors &= ~mirror_bit;
2045 if (mirror->out == bundle) {
2046 bundle->mirror_out |= mirror_bit;
2048 bundle->mirror_out &= ~mirror_bit;
2052 ofproto->need_revalidate = true;
2053 mac_learning_flush(ofproto->ml);
2054 mirror_update_dups(ofproto);
2060 mirror_destroy(struct ofmirror *mirror)
2062 struct ofproto_dpif *ofproto;
2063 mirror_mask_t mirror_bit;
2064 struct ofbundle *bundle;
2070 ofproto = mirror->ofproto;
2071 ofproto->need_revalidate = true;
2072 mac_learning_flush(ofproto->ml);
2074 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2075 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2076 bundle->src_mirrors &= ~mirror_bit;
2077 bundle->dst_mirrors &= ~mirror_bit;
2078 bundle->mirror_out &= ~mirror_bit;
2081 hmapx_destroy(&mirror->srcs);
2082 hmapx_destroy(&mirror->dsts);
2083 free(mirror->vlans);
2085 ofproto->mirrors[mirror->idx] = NULL;
2089 mirror_update_dups(ofproto);
2093 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2094 uint64_t *packets, uint64_t *bytes)
2096 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2097 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2100 *packets = *bytes = UINT64_MAX;
2104 *packets = mirror->packet_count;
2105 *bytes = mirror->byte_count;
2111 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2113 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2114 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2115 ofproto->need_revalidate = true;
2116 mac_learning_flush(ofproto->ml);
2122 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2124 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2125 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2126 return bundle && bundle->mirror_out != 0;
2130 forward_bpdu_changed(struct ofproto *ofproto_)
2132 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2133 /* Revalidate cached flows whenever forward_bpdu option changes. */
2134 ofproto->need_revalidate = true;
2138 set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time)
2140 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2141 mac_learning_set_idle_time(ofproto->ml, idle_time);
2146 static struct ofport_dpif *
2147 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
2149 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2150 return ofport ? ofport_dpif_cast(ofport) : NULL;
2153 static struct ofport_dpif *
2154 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
2156 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2160 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2161 struct dpif_port *dpif_port)
2163 ofproto_port->name = dpif_port->name;
2164 ofproto_port->type = dpif_port->type;
2165 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2169 port_run(struct ofport_dpif *ofport)
2171 bool enable = netdev_get_carrier(ofport->up.netdev);
2174 cfm_run(ofport->cfm);
2176 if (cfm_should_send_ccm(ofport->cfm)) {
2177 struct ofpbuf packet;
2179 ofpbuf_init(&packet, 0);
2180 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr);
2181 send_packet(ofport, &packet);
2182 ofpbuf_uninit(&packet);
2185 enable = enable && !cfm_get_fault(ofport->cfm)
2186 && cfm_get_opup(ofport->cfm);
2189 if (ofport->bundle) {
2190 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2193 if (ofport->may_enable != enable) {
2194 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2196 if (ofproto->has_bundle_action) {
2197 ofproto->need_revalidate = true;
2201 ofport->may_enable = enable;
2205 port_wait(struct ofport_dpif *ofport)
2208 cfm_wait(ofport->cfm);
2213 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2214 struct ofproto_port *ofproto_port)
2216 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2217 struct dpif_port dpif_port;
2220 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2222 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2228 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2230 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2234 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2236 *ofp_portp = odp_port_to_ofp_port(odp_port);
2242 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2244 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2247 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2249 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2251 /* The caller is going to close ofport->up.netdev. If this is a
2252 * bonded port, then the bond is using that netdev, so remove it
2253 * from the bond. The client will need to reconfigure everything
2254 * after deleting ports, so then the slave will get re-added. */
2255 bundle_remove(&ofport->up);
2261 struct port_dump_state {
2262 struct dpif_port_dump dump;
2267 port_dump_start(const struct ofproto *ofproto_, void **statep)
2269 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2270 struct port_dump_state *state;
2272 *statep = state = xmalloc(sizeof *state);
2273 dpif_port_dump_start(&state->dump, ofproto->dpif);
2274 state->done = false;
2279 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2280 struct ofproto_port *port)
2282 struct port_dump_state *state = state_;
2283 struct dpif_port dpif_port;
2285 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2286 ofproto_port_from_dpif_port(port, &dpif_port);
2289 int error = dpif_port_dump_done(&state->dump);
2291 return error ? error : EOF;
2296 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2298 struct port_dump_state *state = state_;
2301 dpif_port_dump_done(&state->dump);
2308 port_poll(const struct ofproto *ofproto_, char **devnamep)
2310 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2311 return dpif_port_poll(ofproto->dpif, devnamep);
2315 port_poll_wait(const struct ofproto *ofproto_)
2317 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2318 dpif_port_poll_wait(ofproto->dpif);
2322 port_is_lacp_current(const struct ofport *ofport_)
2324 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2325 return (ofport->bundle && ofport->bundle->lacp
2326 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2330 /* Upcall handling. */
2332 /* Flow miss batching.
2334 * Some dpifs implement operations faster when you hand them off in a batch.
2335 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2336 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2337 * more packets, plus possibly installing the flow in the dpif.
2339 * So far we only batch the operations that affect flow setup time the most.
2340 * It's possible to batch more than that, but the benefit might be minimal. */
2342 struct hmap_node hmap_node;
2344 enum odp_key_fitness key_fitness;
2345 const struct nlattr *key;
2347 ovs_be16 initial_tci;
2348 struct list packets;
2351 struct flow_miss_op {
2352 union dpif_op dpif_op;
2353 struct subfacet *subfacet;
2356 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2357 * OpenFlow controller as necessary according to their individual
2360 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
2361 * ownership is transferred to this function. */
2363 send_packet_in_miss(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
2364 const struct flow *flow, bool clone)
2366 struct ofputil_packet_in pin;
2368 pin.packet = packet;
2369 pin.in_port = flow->in_port;
2370 pin.reason = OFPR_NO_MATCH;
2371 pin.buffer_id = 0; /* not yet known */
2372 pin.send_len = 0; /* not used for flow table misses */
2373 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
2374 clone ? NULL : packet);
2377 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_ACTION to each
2378 * OpenFlow controller as necessary according to their individual
2381 * 'send_len' should be the number of bytes of 'packet' to send to the
2382 * controller, as specified in the action that caused the packet to be sent.
2384 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
2385 * Otherwise, ownership is transferred to this function. */
2387 send_packet_in_action(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
2388 uint64_t userdata, const struct flow *flow, bool clone)
2390 struct ofputil_packet_in pin;
2391 struct user_action_cookie cookie;
2393 memcpy(&cookie, &userdata, sizeof(cookie));
2395 pin.packet = packet;
2396 pin.in_port = flow->in_port;
2397 pin.reason = OFPR_ACTION;
2398 pin.buffer_id = 0; /* not yet known */
2399 pin.send_len = cookie.data;
2400 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
2401 clone ? NULL : packet);
2405 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2406 const struct ofpbuf *packet)
2408 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2414 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2416 cfm_process_heartbeat(ofport->cfm, packet);
2419 } else if (ofport->bundle && ofport->bundle->lacp
2420 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2422 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2425 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2427 stp_process_packet(ofport, packet);
2434 static struct flow_miss *
2435 flow_miss_create(struct hmap *todo, const struct flow *flow,
2436 enum odp_key_fitness key_fitness,
2437 const struct nlattr *key, size_t key_len,
2438 ovs_be16 initial_tci)
2440 uint32_t hash = flow_hash(flow, 0);
2441 struct flow_miss *miss;
2443 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2444 if (flow_equal(&miss->flow, flow)) {
2449 miss = xmalloc(sizeof *miss);
2450 hmap_insert(todo, &miss->hmap_node, hash);
2452 miss->key_fitness = key_fitness;
2454 miss->key_len = key_len;
2455 miss->initial_tci = initial_tci;
2456 list_init(&miss->packets);
2461 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2462 struct flow_miss_op *ops, size_t *n_ops)
2464 const struct flow *flow = &miss->flow;
2465 struct ofpbuf *packet, *next_packet;
2466 struct subfacet *subfacet;
2467 struct facet *facet;
2469 facet = facet_lookup_valid(ofproto, flow);
2471 struct rule_dpif *rule;
2473 rule = rule_dpif_lookup(ofproto, flow, 0);
2475 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
2476 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
2478 if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) {
2479 COVERAGE_INC(ofproto_dpif_no_packet_in);
2480 /* XXX install 'drop' flow entry */
2484 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
2488 LIST_FOR_EACH_SAFE (packet, next_packet, list_node,
2490 list_remove(&packet->list_node);
2491 send_packet_in_miss(ofproto, packet, flow, false);
2497 facet = facet_create(rule, flow);
2500 subfacet = subfacet_create(ofproto, facet,
2501 miss->key_fitness, miss->key, miss->key_len,
2504 LIST_FOR_EACH_SAFE (packet, next_packet, list_node, &miss->packets) {
2505 struct dpif_flow_stats stats;
2507 list_remove(&packet->list_node);
2508 ofproto->n_matches++;
2510 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2512 * Extra-special case for fail-open mode.
2514 * We are in fail-open mode and the packet matched the fail-open
2515 * rule, but we are connected to a controller too. We should send
2516 * the packet up to the controller in the hope that it will try to
2517 * set up a flow and thereby allow us to exit fail-open.
2519 * See the top-level comment in fail-open.c for more information.
2521 send_packet_in_miss(ofproto, packet, flow, true);
2524 if (!facet->may_install || !subfacet->actions) {
2525 subfacet_make_actions(ofproto, subfacet, packet);
2528 /* Credit statistics to subfacet for this packet. We must do this now
2529 * because execute_controller_action() below may destroy 'packet'. */
2530 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2531 subfacet_update_stats(ofproto, subfacet, &stats);
2533 if (!execute_controller_action(ofproto, &facet->flow,
2535 subfacet->actions_len, packet, true)) {
2536 struct flow_miss_op *op = &ops[(*n_ops)++];
2537 struct dpif_execute *execute = &op->dpif_op.execute;
2539 if (flow->vlan_tci != subfacet->initial_tci) {
2540 /* This packet was received on a VLAN splinter port. We added
2541 * a VLAN to the packet to make the packet resemble the flow,
2542 * but the actions were composed assuming that the packet
2543 * contained no VLAN. So, we must remove the VLAN header from
2544 * the packet before trying to execute the actions. */
2545 eth_pop_vlan(packet);
2548 op->subfacet = subfacet;
2549 execute->type = DPIF_OP_EXECUTE;
2550 execute->key = miss->key;
2551 execute->key_len = miss->key_len;
2553 = (facet->may_install
2555 : xmemdup(subfacet->actions, subfacet->actions_len));
2556 execute->actions_len = subfacet->actions_len;
2557 execute->packet = packet;
2561 if (facet->may_install && subfacet->key_fitness != ODP_FIT_TOO_LITTLE) {
2562 struct flow_miss_op *op = &ops[(*n_ops)++];
2563 struct dpif_flow_put *put = &op->dpif_op.flow_put;
2565 op->subfacet = subfacet;
2566 put->type = DPIF_OP_FLOW_PUT;
2567 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2568 put->key = miss->key;
2569 put->key_len = miss->key_len;
2570 put->actions = subfacet->actions;
2571 put->actions_len = subfacet->actions_len;
2576 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
2577 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
2578 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
2579 * what a flow key should contain.
2581 * This function also includes some logic to help make VLAN splinters
2582 * transparent to the rest of the upcall processing logic. In particular, if
2583 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
2584 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
2585 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
2587 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
2588 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
2589 * (This differs from the value returned in flow->vlan_tci only for packets
2590 * received on VLAN splinters.)
2592 static enum odp_key_fitness
2593 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
2594 const struct nlattr *key, size_t key_len,
2595 struct flow *flow, ovs_be16 *initial_tci,
2596 struct ofpbuf *packet)
2598 enum odp_key_fitness fitness;
2602 fitness = odp_flow_key_to_flow(key, key_len, flow);
2603 if (fitness == ODP_FIT_ERROR) {
2606 *initial_tci = flow->vlan_tci;
2608 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
2610 /* Cause the flow to be processed as if it came in on the real device
2611 * with the VLAN device's VLAN ID. */
2612 flow->in_port = realdev;
2613 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
2615 /* Make the packet resemble the flow, so that it gets sent to an
2616 * OpenFlow controller properly, so that it looks correct for
2617 * sFlow, and so that flow_extract() will get the correct vlan_tci
2618 * if it is called on 'packet'.
2620 * The allocated space inside 'packet' probably also contains
2621 * 'key', that is, both 'packet' and 'key' are probably part of a
2622 * struct dpif_upcall (see the large comment on that structure
2623 * definition), so pushing data on 'packet' is in general not a
2624 * good idea since it could overwrite 'key' or free it as a side
2625 * effect. However, it's OK in this special case because we know
2626 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
2627 * will just overwrite the 4-byte "struct nlattr", which is fine
2628 * since we don't need that header anymore. */
2629 eth_push_vlan(packet, flow->vlan_tci);
2632 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
2633 if (fitness == ODP_FIT_PERFECT) {
2634 fitness = ODP_FIT_TOO_MUCH;
2642 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
2645 struct dpif_upcall *upcall;
2646 struct flow_miss *miss, *next_miss;
2647 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
2648 union dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
2657 /* Construct the to-do list.
2659 * This just amounts to extracting the flow from each packet and sticking
2660 * the packets that have the same flow in the same "flow_miss" structure so
2661 * that we can process them together. */
2663 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
2664 enum odp_key_fitness fitness;
2665 struct flow_miss *miss;
2666 ovs_be16 initial_tci;
2669 /* Obtain metadata and check userspace/kernel agreement on flow match,
2670 * then set 'flow''s header pointers. */
2671 fitness = ofproto_dpif_extract_flow_key(ofproto,
2672 upcall->key, upcall->key_len,
2673 &flow, &initial_tci,
2675 if (fitness == ODP_FIT_ERROR) {
2676 ofpbuf_delete(upcall->packet);
2679 flow_extract(upcall->packet, flow.priority, flow.tun_id,
2680 flow.in_port, &flow);
2682 /* Handle 802.1ag, LACP, and STP specially. */
2683 if (process_special(ofproto, &flow, upcall->packet)) {
2684 ofpbuf_delete(upcall->packet);
2685 ofproto->n_matches++;
2689 /* Add other packets to a to-do list. */
2690 miss = flow_miss_create(&todo, &flow, fitness,
2691 upcall->key, upcall->key_len, initial_tci);
2692 list_push_back(&miss->packets, &upcall->packet->list_node);
2695 /* Process each element in the to-do list, constructing the set of
2696 * operations to batch. */
2698 HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) {
2699 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
2700 ofpbuf_list_delete(&miss->packets);
2701 hmap_remove(&todo, &miss->hmap_node);
2704 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
2705 hmap_destroy(&todo);
2707 /* Execute batch. */
2708 for (i = 0; i < n_ops; i++) {
2709 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
2711 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
2713 /* Free memory and update facets. */
2714 for (i = 0; i < n_ops; i++) {
2715 struct flow_miss_op *op = &flow_miss_ops[i];
2716 struct dpif_execute *execute;
2717 struct dpif_flow_put *put;
2719 switch (op->dpif_op.type) {
2720 case DPIF_OP_EXECUTE:
2721 execute = &op->dpif_op.execute;
2722 if (op->subfacet->actions != execute->actions) {
2723 free((struct nlattr *) execute->actions);
2725 ofpbuf_delete((struct ofpbuf *) execute->packet);
2728 case DPIF_OP_FLOW_PUT:
2729 put = &op->dpif_op.flow_put;
2731 op->subfacet->installed = true;
2739 handle_userspace_upcall(struct ofproto_dpif *ofproto,
2740 struct dpif_upcall *upcall)
2742 struct user_action_cookie cookie;
2743 enum odp_key_fitness fitness;
2744 ovs_be16 initial_tci;
2747 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
2749 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
2750 upcall->key_len, &flow,
2751 &initial_tci, upcall->packet);
2752 if (fitness == ODP_FIT_ERROR) {
2753 ofpbuf_delete(upcall->packet);
2757 if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
2758 if (ofproto->sflow) {
2759 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
2762 ofpbuf_delete(upcall->packet);
2763 } else if (cookie.type == USER_ACTION_COOKIE_CONTROLLER) {
2764 COVERAGE_INC(ofproto_dpif_ctlr_action);
2765 send_packet_in_action(ofproto, upcall->packet, upcall->userdata,
2768 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
2769 ofpbuf_delete(upcall->packet);
2774 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
2776 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
2780 assert (max_batch <= FLOW_MISS_MAX_BATCH);
2783 for (i = 0; i < max_batch; i++) {
2784 struct dpif_upcall *upcall = &misses[n_misses];
2787 error = dpif_recv(ofproto->dpif, upcall);
2792 switch (upcall->type) {
2793 case DPIF_UC_ACTION:
2794 handle_userspace_upcall(ofproto, upcall);
2798 /* Handle it later. */
2802 case DPIF_N_UC_TYPES:
2804 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32,
2810 handle_miss_upcalls(ofproto, misses, n_misses);
2815 /* Flow expiration. */
2817 static int subfacet_max_idle(const struct ofproto_dpif *);
2818 static void update_stats(struct ofproto_dpif *);
2819 static void rule_expire(struct rule_dpif *);
2820 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
2822 /* This function is called periodically by run(). Its job is to collect
2823 * updates for the flows that have been installed into the datapath, most
2824 * importantly when they last were used, and then use that information to
2825 * expire flows that have not been used recently.
2827 * Returns the number of milliseconds after which it should be called again. */
2829 expire(struct ofproto_dpif *ofproto)
2831 struct rule_dpif *rule, *next_rule;
2832 struct classifier *table;
2835 /* Update stats for each flow in the datapath. */
2836 update_stats(ofproto);
2838 /* Expire subfacets that have been idle too long. */
2839 dp_max_idle = subfacet_max_idle(ofproto);
2840 expire_subfacets(ofproto, dp_max_idle);
2842 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
2843 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
2844 struct cls_cursor cursor;
2846 cls_cursor_init(&cursor, table, NULL);
2847 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
2852 /* All outstanding data in existing flows has been accounted, so it's a
2853 * good time to do bond rebalancing. */
2854 if (ofproto->has_bonded_bundles) {
2855 struct ofbundle *bundle;
2857 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2859 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
2864 return MIN(dp_max_idle, 1000);
2867 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
2869 * This function also pushes statistics updates to rules which each facet
2870 * resubmits into. Generally these statistics will be accurate. However, if a
2871 * facet changes the rule it resubmits into at some time in between
2872 * update_stats() runs, it is possible that statistics accrued to the
2873 * old rule will be incorrectly attributed to the new rule. This could be
2874 * avoided by calling update_stats() whenever rules are created or
2875 * deleted. However, the performance impact of making so many calls to the
2876 * datapath do not justify the benefit of having perfectly accurate statistics.
2879 update_stats(struct ofproto_dpif *p)
2881 const struct dpif_flow_stats *stats;
2882 struct dpif_flow_dump dump;
2883 const struct nlattr *key;
2886 dpif_flow_dump_start(&dump, p->dpif);
2887 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
2888 struct subfacet *subfacet;
2890 subfacet = subfacet_find(p, key, key_len);
2891 if (subfacet && subfacet->installed) {
2892 struct facet *facet = subfacet->facet;
2894 if (stats->n_packets >= subfacet->dp_packet_count) {
2895 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
2896 facet->packet_count += extra;
2898 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
2901 if (stats->n_bytes >= subfacet->dp_byte_count) {
2902 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
2904 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
2907 subfacet->dp_packet_count = stats->n_packets;
2908 subfacet->dp_byte_count = stats->n_bytes;
2910 subfacet_update_time(p, subfacet, stats->used);
2911 facet_account(p, facet);
2912 facet_push_stats(facet);
2914 if (!VLOG_DROP_WARN(&rl)) {
2918 odp_flow_key_format(key, key_len, &s);
2919 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
2923 COVERAGE_INC(facet_unexpected);
2924 /* There's a flow in the datapath that we know nothing about, or a
2925 * flow that shouldn't be installed but was anyway. Delete it. */
2926 dpif_flow_del(p->dpif, key, key_len, NULL);
2929 dpif_flow_dump_done(&dump);
2932 /* Calculates and returns the number of milliseconds of idle time after which
2933 * subfacets should expire from the datapath. When a subfacet expires, we fold
2934 * its statistics into its facet, and when a facet's last subfacet expires, we
2935 * fold its statistic into its rule. */
2937 subfacet_max_idle(const struct ofproto_dpif *ofproto)
2940 * Idle time histogram.
2942 * Most of the time a switch has a relatively small number of subfacets.
2943 * When this is the case we might as well keep statistics for all of them
2944 * in userspace and to cache them in the kernel datapath for performance as
2947 * As the number of subfacets increases, the memory required to maintain
2948 * statistics about them in userspace and in the kernel becomes
2949 * significant. However, with a large number of subfacets it is likely
2950 * that only a few of them are "heavy hitters" that consume a large amount
2951 * of bandwidth. At this point, only heavy hitters are worth caching in
2952 * the kernel and maintaining in userspaces; other subfacets we can
2955 * The technique used to compute the idle time is to build a histogram with
2956 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
2957 * that is installed in the kernel gets dropped in the appropriate bucket.
2958 * After the histogram has been built, we compute the cutoff so that only
2959 * the most-recently-used 1% of subfacets (but at least
2960 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
2961 * the most-recently-used bucket of subfacets is kept, so actually an
2962 * arbitrary number of subfacets can be kept in any given expiration run
2963 * (though the next run will delete most of those unless they receive
2966 * This requires a second pass through the subfacets, in addition to the
2967 * pass made by update_stats(), because the former function never looks at
2968 * uninstallable subfacets.
2970 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
2971 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
2972 int buckets[N_BUCKETS] = { 0 };
2973 int total, subtotal, bucket;
2974 struct subfacet *subfacet;
2978 total = hmap_count(&ofproto->subfacets);
2979 if (total <= ofproto->up.flow_eviction_threshold) {
2980 return N_BUCKETS * BUCKET_WIDTH;
2983 /* Build histogram. */
2985 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
2986 long long int idle = now - subfacet->used;
2987 int bucket = (idle <= 0 ? 0
2988 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
2989 : (unsigned int) idle / BUCKET_WIDTH);
2993 /* Find the first bucket whose flows should be expired. */
2994 subtotal = bucket = 0;
2996 subtotal += buckets[bucket++];
2997 } while (bucket < N_BUCKETS &&
2998 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3000 if (VLOG_IS_DBG_ENABLED()) {
3004 ds_put_cstr(&s, "keep");
3005 for (i = 0; i < N_BUCKETS; i++) {
3007 ds_put_cstr(&s, ", drop");
3010 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3013 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3017 return bucket * BUCKET_WIDTH;
3021 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3023 long long int cutoff = time_msec() - dp_max_idle;
3024 struct subfacet *subfacet, *next_subfacet;
3026 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3027 &ofproto->subfacets) {
3028 if (subfacet->used < cutoff) {
3029 subfacet_destroy(ofproto, subfacet);
3034 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3035 * then delete it entirely. */
3037 rule_expire(struct rule_dpif *rule)
3039 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3040 struct facet *facet, *next_facet;
3044 /* Has 'rule' expired? */
3046 if (rule->up.hard_timeout
3047 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3048 reason = OFPRR_HARD_TIMEOUT;
3049 } else if (rule->up.idle_timeout && list_is_empty(&rule->facets)
3050 && now > rule->used + rule->up.idle_timeout * 1000) {
3051 reason = OFPRR_IDLE_TIMEOUT;
3056 COVERAGE_INC(ofproto_dpif_expired);
3058 /* Update stats. (This is a no-op if the rule expired due to an idle
3059 * timeout, because that only happens when the rule has no facets left.) */
3060 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3061 facet_remove(ofproto, facet);
3064 /* Get rid of the rule. */
3065 ofproto_rule_expire(&rule->up, reason);
3070 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3072 * The caller must already have determined that no facet with an identical
3073 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3074 * the ofproto's classifier table.
3076 * The facet will initially have no subfacets. The caller should create (at
3077 * least) one subfacet with subfacet_create(). */
3078 static struct facet *
3079 facet_create(struct rule_dpif *rule, const struct flow *flow)
3081 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3082 struct facet *facet;
3084 facet = xzalloc(sizeof *facet);
3085 facet->used = time_msec();
3086 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
3087 list_push_back(&rule->facets, &facet->list_node);
3089 facet->flow = *flow;
3090 list_init(&facet->subfacets);
3091 netflow_flow_init(&facet->nf_flow);
3092 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3098 facet_free(struct facet *facet)
3103 /* If the 'actions_len' bytes of actions in 'odp_actions' are just a single
3104 * OVS_ACTION_ATTR_USERSPACE action, executes it internally and returns true.
3105 * Otherwise, returns false without doing anything.
3107 * If 'clone' is true, the caller always retains ownership of 'packet'.
3108 * Otherwise, ownership is transferred to this function if it returns true. */
3110 execute_controller_action(struct ofproto_dpif *ofproto,
3111 const struct flow *flow,
3112 const struct nlattr *odp_actions, size_t actions_len,
3113 struct ofpbuf *packet, bool clone)
3116 && odp_actions->nla_type == OVS_ACTION_ATTR_USERSPACE
3117 && NLA_ALIGN(odp_actions->nla_len) == actions_len) {
3118 /* As an optimization, avoid a round-trip from userspace to kernel to
3119 * userspace. This also avoids possibly filling up kernel packet
3120 * buffers along the way.
3122 * This optimization will not accidentally catch sFlow
3123 * OVS_ACTION_ATTR_USERSPACE actions, since those are encapsulated
3124 * inside OVS_ACTION_ATTR_SAMPLE. */
3125 const struct nlattr *nla;
3127 nla = nl_attr_find_nested(odp_actions, OVS_USERSPACE_ATTR_USERDATA);
3128 send_packet_in_action(ofproto, packet, nl_attr_get_u64(nla), flow,
3136 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3137 * 'packet', which arrived on 'in_port'.
3139 * Takes ownership of 'packet'. */
3141 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3142 const struct nlattr *odp_actions, size_t actions_len,
3143 struct ofpbuf *packet)
3145 struct odputil_keybuf keybuf;
3149 if (execute_controller_action(ofproto, flow, odp_actions, actions_len,
3154 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3155 odp_flow_key_from_flow(&key, flow);
3157 error = dpif_execute(ofproto->dpif, key.data, key.size,
3158 odp_actions, actions_len, packet);
3160 ofpbuf_delete(packet);
3164 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3166 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3167 * rule's statistics, via subfacet_uninstall().
3169 * - Removes 'facet' from its rule and from ofproto->facets.
3172 facet_remove(struct ofproto_dpif *ofproto, struct facet *facet)
3174 struct subfacet *subfacet, *next_subfacet;
3176 assert(!list_is_empty(&facet->subfacets));
3178 /* First uninstall all of the subfacets to get final statistics. */
3179 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3180 subfacet_uninstall(ofproto, subfacet);
3183 /* Flush the final stats to the rule.
3185 * This might require us to have at least one subfacet around so that we
3186 * can use its actions for accounting in facet_account(), which is why we
3187 * have uninstalled but not yet destroyed the subfacets. */
3188 facet_flush_stats(ofproto, facet);
3190 /* Now we're really all done so destroy everything. */
3191 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3192 &facet->subfacets) {
3193 subfacet_destroy__(ofproto, subfacet);
3195 hmap_remove(&ofproto->facets, &facet->hmap_node);
3196 list_remove(&facet->list_node);
3201 facet_account(struct ofproto_dpif *ofproto, struct facet *facet)
3204 struct subfacet *subfacet;
3205 const struct nlattr *a;
3209 if (facet->byte_count <= facet->accounted_bytes) {
3212 n_bytes = facet->byte_count - facet->accounted_bytes;
3213 facet->accounted_bytes = facet->byte_count;
3215 /* Feed information from the active flows back into the learning table to
3216 * ensure that table is always in sync with what is actually flowing
3217 * through the datapath. */
3218 if (facet->has_learn || facet->has_normal) {
3219 struct action_xlate_ctx ctx;
3221 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3222 facet->flow.vlan_tci, NULL);
3223 ctx.may_learn = true;
3224 ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions,
3225 facet->rule->up.n_actions));
3228 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3232 /* This loop feeds byte counters to bond_account() for rebalancing to use
3233 * as a basis. We also need to track the actual VLAN on which the packet
3234 * is going to be sent to ensure that it matches the one passed to
3235 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3238 * We use the actions from an arbitrary subfacet because they should all
3239 * be equally valid for our purpose. */
3240 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3241 struct subfacet, list_node);
3242 vlan_tci = facet->flow.vlan_tci;
3243 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3244 subfacet->actions, subfacet->actions_len) {
3245 const struct ovs_action_push_vlan *vlan;
3246 struct ofport_dpif *port;
3248 switch (nl_attr_type(a)) {
3249 case OVS_ACTION_ATTR_OUTPUT:
3250 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3251 if (port && port->bundle && port->bundle->bond) {
3252 bond_account(port->bundle->bond, &facet->flow,
3253 vlan_tci_to_vid(vlan_tci), n_bytes);
3257 case OVS_ACTION_ATTR_POP_VLAN:
3258 vlan_tci = htons(0);
3261 case OVS_ACTION_ATTR_PUSH_VLAN:
3262 vlan = nl_attr_get(a);
3263 vlan_tci = vlan->vlan_tci;
3269 /* Returns true if the only action for 'facet' is to send to the controller.
3270 * (We don't report NetFlow expiration messages for such facets because they
3271 * are just part of the control logic for the network, not real traffic). */
3273 facet_is_controller_flow(struct facet *facet)
3276 && facet->rule->up.n_actions == 1
3277 && action_outputs_to_port(&facet->rule->up.actions[0],
3278 htons(OFPP_CONTROLLER)));
3281 /* Folds all of 'facet''s statistics into its rule. Also updates the
3282 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3283 * 'facet''s statistics in the datapath should have been zeroed and folded into
3284 * its packet and byte counts before this function is called. */
3286 facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet)
3288 struct subfacet *subfacet;
3290 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3291 assert(!subfacet->dp_byte_count);
3292 assert(!subfacet->dp_packet_count);
3295 facet_push_stats(facet);
3296 facet_account(ofproto, facet);
3298 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3299 struct ofexpired expired;
3300 expired.flow = facet->flow;
3301 expired.packet_count = facet->packet_count;
3302 expired.byte_count = facet->byte_count;
3303 expired.used = facet->used;
3304 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3307 facet->rule->packet_count += facet->packet_count;
3308 facet->rule->byte_count += facet->byte_count;
3310 /* Reset counters to prevent double counting if 'facet' ever gets
3312 facet_reset_counters(facet);
3314 netflow_flow_clear(&facet->nf_flow);
3317 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3318 * Returns it if found, otherwise a null pointer.
3320 * The returned facet might need revalidation; use facet_lookup_valid()
3321 * instead if that is important. */
3322 static struct facet *
3323 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
3325 struct facet *facet;
3327 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
3329 if (flow_equal(flow, &facet->flow)) {
3337 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3338 * Returns it if found, otherwise a null pointer.
3340 * The returned facet is guaranteed to be valid. */
3341 static struct facet *
3342 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
3344 struct facet *facet = facet_find(ofproto, flow);
3346 /* The facet we found might not be valid, since we could be in need of
3347 * revalidation. If it is not valid, don't return it. */
3349 && (ofproto->need_revalidate
3350 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))
3351 && !facet_revalidate(ofproto, facet)) {
3352 COVERAGE_INC(facet_invalidated);
3359 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
3361 * - If the rule found is different from 'facet''s current rule, moves
3362 * 'facet' to the new rule and recompiles its actions.
3364 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3365 * where it is and recompiles its actions anyway.
3367 * - If there is none, destroys 'facet'.
3369 * Returns true if 'facet' still exists, false if it has been destroyed. */
3371 facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet)
3374 struct nlattr *odp_actions;
3377 struct actions *new_actions;
3379 struct action_xlate_ctx ctx;
3380 struct rule_dpif *new_rule;
3381 struct subfacet *subfacet;
3382 bool actions_changed;
3385 COVERAGE_INC(facet_revalidate);
3387 /* Determine the new rule. */
3388 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3390 /* No new rule, so delete the facet. */
3391 facet_remove(ofproto, facet);
3395 /* Calculate new datapath actions.
3397 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3398 * emit a NetFlow expiration and, if so, we need to have the old state
3399 * around to properly compose it. */
3401 /* If the datapath actions changed or the installability changed,
3402 * then we need to talk to the datapath. */
3405 memset(&ctx, 0, sizeof ctx);
3406 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3407 struct ofpbuf *odp_actions;
3408 bool should_install;
3410 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3411 subfacet->initial_tci, NULL);
3412 odp_actions = xlate_actions(&ctx, new_rule->up.actions,
3413 new_rule->up.n_actions);
3414 actions_changed = (subfacet->actions_len != odp_actions->size
3415 || memcmp(subfacet->actions, odp_actions->data,
3416 subfacet->actions_len));
3418 should_install = (ctx.may_set_up_flow
3419 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3420 if (actions_changed || should_install != subfacet->installed) {
3421 if (should_install) {
3422 struct dpif_flow_stats stats;
3424 subfacet_install(ofproto, subfacet,
3425 odp_actions->data, odp_actions->size, &stats);
3426 subfacet_update_stats(ofproto, subfacet, &stats);
3428 subfacet_uninstall(ofproto, subfacet);
3432 new_actions = xcalloc(list_size(&facet->subfacets),
3433 sizeof *new_actions);
3435 new_actions[i].odp_actions = xmemdup(odp_actions->data,
3437 new_actions[i].actions_len = odp_actions->size;
3440 ofpbuf_delete(odp_actions);
3444 facet_flush_stats(ofproto, facet);
3447 /* Update 'facet' now that we've taken care of all the old state. */
3448 facet->tags = ctx.tags;
3449 facet->nf_flow.output_iface = ctx.nf_output_iface;
3450 facet->may_install = ctx.may_set_up_flow;
3451 facet->has_learn = ctx.has_learn;
3452 facet->has_normal = ctx.has_normal;
3453 facet->mirrors = ctx.mirrors;
3456 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3457 if (new_actions[i].odp_actions) {
3458 free(subfacet->actions);
3459 subfacet->actions = new_actions[i].odp_actions;
3460 subfacet->actions_len = new_actions[i].actions_len;
3466 if (facet->rule != new_rule) {
3467 COVERAGE_INC(facet_changed_rule);
3468 list_remove(&facet->list_node);
3469 list_push_back(&new_rule->facets, &facet->list_node);
3470 facet->rule = new_rule;
3471 facet->used = new_rule->up.created;
3472 facet->prev_used = facet->used;
3478 /* Updates 'facet''s used time. Caller is responsible for calling
3479 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3481 facet_update_time(struct ofproto_dpif *ofproto, struct facet *facet,
3484 if (used > facet->used) {
3486 if (used > facet->rule->used) {
3487 facet->rule->used = used;
3489 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3494 facet_reset_counters(struct facet *facet)
3496 facet->packet_count = 0;
3497 facet->byte_count = 0;
3498 facet->prev_packet_count = 0;
3499 facet->prev_byte_count = 0;
3500 facet->accounted_bytes = 0;
3504 facet_push_stats(struct facet *facet)
3506 uint64_t new_packets, new_bytes;
3508 assert(facet->packet_count >= facet->prev_packet_count);
3509 assert(facet->byte_count >= facet->prev_byte_count);
3510 assert(facet->used >= facet->prev_used);
3512 new_packets = facet->packet_count - facet->prev_packet_count;
3513 new_bytes = facet->byte_count - facet->prev_byte_count;
3515 if (new_packets || new_bytes || facet->used > facet->prev_used) {
3516 facet->prev_packet_count = facet->packet_count;
3517 facet->prev_byte_count = facet->byte_count;
3518 facet->prev_used = facet->used;
3520 flow_push_stats(facet->rule, &facet->flow,
3521 new_packets, new_bytes, facet->used);
3523 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
3524 facet->mirrors, new_packets, new_bytes);
3528 struct ofproto_push {
3529 struct action_xlate_ctx ctx;
3536 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
3538 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3541 rule->packet_count += push->packets;
3542 rule->byte_count += push->bytes;
3543 rule->used = MAX(push->used, rule->used);
3547 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3548 * 'rule''s actions and mirrors. */
3550 flow_push_stats(const struct rule_dpif *rule,
3551 const struct flow *flow, uint64_t packets, uint64_t bytes,
3554 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3555 struct ofproto_push push;
3557 push.packets = packets;
3561 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, NULL);
3562 push.ctx.resubmit_hook = push_resubmit;
3563 ofpbuf_delete(xlate_actions(&push.ctx,
3564 rule->up.actions, rule->up.n_actions));
3569 static struct subfacet *
3570 subfacet_find__(struct ofproto_dpif *ofproto,
3571 const struct nlattr *key, size_t key_len, uint32_t key_hash,
3572 const struct flow *flow)
3574 struct subfacet *subfacet;
3576 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
3577 &ofproto->subfacets) {
3579 ? (subfacet->key_len == key_len
3580 && !memcmp(key, subfacet->key, key_len))
3581 : flow_equal(flow, &subfacet->facet->flow)) {
3589 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
3590 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
3591 * there is one, otherwise creates and returns a new subfacet.
3593 * If the returned subfacet is new, then subfacet->actions will be NULL, in
3594 * which case the caller must populate the actions with
3595 * subfacet_make_actions(). */
3596 static struct subfacet *
3597 subfacet_create(struct ofproto_dpif *ofproto, struct facet *facet,
3598 enum odp_key_fitness key_fitness,
3599 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
3601 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3602 struct subfacet *subfacet;
3604 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
3606 if (subfacet->facet == facet) {
3610 /* This shouldn't happen. */
3611 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
3612 subfacet_destroy(ofproto, subfacet);
3615 subfacet = xzalloc(sizeof *subfacet);
3616 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
3617 list_push_back(&facet->subfacets, &subfacet->list_node);
3618 subfacet->facet = facet;
3619 subfacet->used = time_msec();
3620 subfacet->key_fitness = key_fitness;
3621 if (key_fitness != ODP_FIT_PERFECT) {
3622 subfacet->key = xmemdup(key, key_len);
3623 subfacet->key_len = key_len;
3625 subfacet->installed = false;
3626 subfacet->initial_tci = initial_tci;
3631 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
3632 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
3633 static struct subfacet *
3634 subfacet_find(struct ofproto_dpif *ofproto,
3635 const struct nlattr *key, size_t key_len)
3637 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3638 enum odp_key_fitness fitness;
3641 fitness = odp_flow_key_to_flow(key, key_len, &flow);
3642 if (fitness == ODP_FIT_ERROR) {
3646 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
3649 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
3650 * its facet within 'ofproto', and frees it. */
3652 subfacet_destroy__(struct ofproto_dpif *ofproto, struct subfacet *subfacet)
3654 subfacet_uninstall(ofproto, subfacet);
3655 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
3656 list_remove(&subfacet->list_node);
3657 free(subfacet->key);
3658 free(subfacet->actions);
3662 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
3663 * last remaining subfacet in its facet destroys the facet too. */
3665 subfacet_destroy(struct ofproto_dpif *ofproto, struct subfacet *subfacet)
3667 struct facet *facet = subfacet->facet;
3669 if (list_is_singleton(&facet->subfacets)) {
3670 /* facet_remove() needs at least one subfacet (it will remove it). */
3671 facet_remove(ofproto, facet);
3673 subfacet_destroy__(ofproto, subfacet);
3677 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
3678 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
3679 * for use as temporary storage. */
3681 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
3684 if (!subfacet->key) {
3685 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
3686 odp_flow_key_from_flow(key, &subfacet->facet->flow);
3688 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
3692 /* Composes the datapath actions for 'subfacet' based on its rule's actions. */
3694 subfacet_make_actions(struct ofproto_dpif *p, struct subfacet *subfacet,
3695 const struct ofpbuf *packet)
3697 struct facet *facet = subfacet->facet;
3698 const struct rule_dpif *rule = facet->rule;
3699 struct ofpbuf *odp_actions;
3700 struct action_xlate_ctx ctx;
3702 action_xlate_ctx_init(&ctx, p, &facet->flow, subfacet->initial_tci,
3704 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3705 facet->tags = ctx.tags;
3706 facet->may_install = ctx.may_set_up_flow;
3707 facet->has_learn = ctx.has_learn;
3708 facet->has_normal = ctx.has_normal;
3709 facet->nf_flow.output_iface = ctx.nf_output_iface;
3710 facet->mirrors = ctx.mirrors;
3712 if (subfacet->actions_len != odp_actions->size
3713 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
3714 free(subfacet->actions);
3715 subfacet->actions_len = odp_actions->size;
3716 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
3719 ofpbuf_delete(odp_actions);
3722 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
3723 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
3724 * in the datapath will be zeroed and 'stats' will be updated with traffic new
3725 * since 'subfacet' was last updated.
3727 * Returns 0 if successful, otherwise a positive errno value. */
3729 subfacet_install(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3730 const struct nlattr *actions, size_t actions_len,
3731 struct dpif_flow_stats *stats)
3733 struct odputil_keybuf keybuf;
3734 enum dpif_flow_put_flags flags;
3738 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3740 flags |= DPIF_FP_ZERO_STATS;
3743 subfacet_get_key(subfacet, &keybuf, &key);
3744 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
3745 actions, actions_len, stats);
3748 subfacet_reset_dp_stats(subfacet, stats);
3754 /* If 'subfacet' is installed in the datapath, uninstalls it. */
3756 subfacet_uninstall(struct ofproto_dpif *p, struct subfacet *subfacet)
3758 if (subfacet->installed) {
3759 struct odputil_keybuf keybuf;
3760 struct dpif_flow_stats stats;
3764 subfacet_get_key(subfacet, &keybuf, &key);
3765 error = dpif_flow_del(p->dpif, key.data, key.size, &stats);
3766 subfacet_reset_dp_stats(subfacet, &stats);
3768 subfacet_update_stats(p, subfacet, &stats);
3770 subfacet->installed = false;
3772 assert(subfacet->dp_packet_count == 0);
3773 assert(subfacet->dp_byte_count == 0);
3777 /* Resets 'subfacet''s datapath statistics counters. This should be called
3778 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
3779 * non-null, it should contain the statistics returned by dpif when 'subfacet'
3780 * was reset in the datapath. 'stats' will be modified to include only
3781 * statistics new since 'subfacet' was last updated. */
3783 subfacet_reset_dp_stats(struct subfacet *subfacet,
3784 struct dpif_flow_stats *stats)
3787 && subfacet->dp_packet_count <= stats->n_packets
3788 && subfacet->dp_byte_count <= stats->n_bytes) {
3789 stats->n_packets -= subfacet->dp_packet_count;
3790 stats->n_bytes -= subfacet->dp_byte_count;
3793 subfacet->dp_packet_count = 0;
3794 subfacet->dp_byte_count = 0;
3797 /* Updates 'subfacet''s used time. The caller is responsible for calling
3798 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
3800 subfacet_update_time(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3803 if (used > subfacet->used) {
3804 subfacet->used = used;
3805 facet_update_time(ofproto, subfacet->facet, used);
3809 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
3811 * Because of the meaning of a subfacet's counters, it only makes sense to do
3812 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
3813 * represents a packet that was sent by hand or if it represents statistics
3814 * that have been cleared out of the datapath. */
3816 subfacet_update_stats(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3817 const struct dpif_flow_stats *stats)
3819 if (stats->n_packets || stats->used > subfacet->used) {
3820 struct facet *facet = subfacet->facet;
3822 subfacet_update_time(ofproto, subfacet, stats->used);
3823 facet->packet_count += stats->n_packets;
3824 facet->byte_count += stats->n_bytes;
3825 facet_push_stats(facet);
3826 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3832 static struct rule_dpif *
3833 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
3836 struct cls_rule *cls_rule;
3837 struct classifier *cls;
3839 if (table_id >= N_TABLES) {
3843 cls = &ofproto->up.tables[table_id];
3844 if (flow->nw_frag & FLOW_NW_FRAG_ANY
3845 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3846 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
3847 * are unavailable. */
3848 struct flow ofpc_normal_flow = *flow;
3849 ofpc_normal_flow.tp_src = htons(0);
3850 ofpc_normal_flow.tp_dst = htons(0);
3851 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
3853 cls_rule = classifier_lookup(cls, flow);
3855 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
3859 complete_operation(struct rule_dpif *rule)
3861 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3863 rule_invalidate(rule);
3865 struct dpif_completion *c = xmalloc(sizeof *c);
3866 c->op = rule->up.pending;
3867 list_push_back(&ofproto->completions, &c->list_node);
3869 ofoperation_complete(rule->up.pending, 0);
3873 static struct rule *
3876 struct rule_dpif *rule = xmalloc(sizeof *rule);
3881 rule_dealloc(struct rule *rule_)
3883 struct rule_dpif *rule = rule_dpif_cast(rule_);
3888 rule_construct(struct rule *rule_)
3890 struct rule_dpif *rule = rule_dpif_cast(rule_);
3891 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3892 struct rule_dpif *victim;
3896 error = validate_actions(rule->up.actions, rule->up.n_actions,
3897 &rule->up.cr.flow, ofproto->max_ports);
3902 rule->used = rule->up.created;
3903 rule->packet_count = 0;
3904 rule->byte_count = 0;
3906 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
3907 if (victim && !list_is_empty(&victim->facets)) {
3908 struct facet *facet;
3910 rule->facets = victim->facets;
3911 list_moved(&rule->facets);
3912 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3913 /* XXX: We're only clearing our local counters here. It's possible
3914 * that quite a few packets are unaccounted for in the datapath
3915 * statistics. These will be accounted to the new rule instead of
3916 * cleared as required. This could be fixed by clearing out the
3917 * datapath statistics for this facet, but currently it doesn't
3919 facet_reset_counters(facet);
3923 /* Must avoid list_moved() in this case. */
3924 list_init(&rule->facets);
3927 table_id = rule->up.table_id;
3928 rule->tag = (victim ? victim->tag
3930 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
3931 ofproto->tables[table_id].basis));
3933 complete_operation(rule);
3938 rule_destruct(struct rule *rule_)
3940 struct rule_dpif *rule = rule_dpif_cast(rule_);
3941 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3942 struct facet *facet, *next_facet;
3944 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3945 facet_revalidate(ofproto, facet);
3948 complete_operation(rule);
3952 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
3954 struct rule_dpif *rule = rule_dpif_cast(rule_);
3955 struct facet *facet;
3957 /* Start from historical data for 'rule' itself that are no longer tracked
3958 * in facets. This counts, for example, facets that have expired. */
3959 *packets = rule->packet_count;
3960 *bytes = rule->byte_count;
3962 /* Add any statistics that are tracked by facets. This includes
3963 * statistical data recently updated by ofproto_update_stats() as well as
3964 * stats for packets that were executed "by hand" via dpif_execute(). */
3965 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3966 *packets += facet->packet_count;
3967 *bytes += facet->byte_count;
3972 rule_execute(struct rule *rule_, const struct flow *flow,
3973 struct ofpbuf *packet)
3975 struct rule_dpif *rule = rule_dpif_cast(rule_);
3976 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3977 struct action_xlate_ctx ctx;
3978 struct ofpbuf *odp_actions;
3981 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, packet);
3982 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3983 size = packet->size;
3984 if (execute_odp_actions(ofproto, flow, odp_actions->data,
3985 odp_actions->size, packet)) {
3986 rule->used = time_msec();
3987 rule->packet_count++;
3988 rule->byte_count += size;
3989 flow_push_stats(rule, flow, 1, size, rule->used);
3991 ofpbuf_delete(odp_actions);
3997 rule_modify_actions(struct rule *rule_)
3999 struct rule_dpif *rule = rule_dpif_cast(rule_);
4000 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4003 error = validate_actions(rule->up.actions, rule->up.n_actions,
4004 &rule->up.cr.flow, ofproto->max_ports);
4006 ofoperation_complete(rule->up.pending, error);
4010 complete_operation(rule);
4013 /* Sends 'packet' out 'ofport'.
4014 * May modify 'packet'.
4015 * Returns 0 if successful, otherwise a positive errno value. */
4017 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4019 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4020 struct ofpbuf key, odp_actions;
4021 struct odputil_keybuf keybuf;
4026 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
4027 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4029 if (odp_port != ofport->odp_port) {
4030 eth_pop_vlan(packet);
4031 flow.vlan_tci = htons(0);
4034 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4035 odp_flow_key_from_flow(&key, &flow);
4037 ofpbuf_init(&odp_actions, 32);
4038 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4040 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4041 error = dpif_execute(ofproto->dpif,
4043 odp_actions.data, odp_actions.size,
4045 ofpbuf_uninit(&odp_actions);
4048 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4049 ofproto->up.name, odp_port, strerror(error));
4054 /* OpenFlow to datapath action translation. */
4056 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
4057 struct action_xlate_ctx *ctx);
4058 static void xlate_normal(struct action_xlate_ctx *);
4061 put_userspace_action(const struct ofproto_dpif *ofproto,
4062 struct ofpbuf *odp_actions,
4063 const struct flow *flow,
4064 const struct user_action_cookie *cookie)
4068 pid = dpif_port_get_pid(ofproto->dpif,
4069 ofp_port_to_odp_port(flow->in_port));
4071 return odp_put_userspace_action(pid, cookie, odp_actions);
4074 /* Compose SAMPLE action for sFlow. */
4076 compose_sflow_action(const struct ofproto_dpif *ofproto,
4077 struct ofpbuf *odp_actions,
4078 const struct flow *flow,
4081 uint32_t port_ifindex;
4082 uint32_t probability;
4083 struct user_action_cookie cookie;
4084 size_t sample_offset, actions_offset;
4085 int cookie_offset, n_output;
4087 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4091 if (odp_port == OVSP_NONE) {
4095 port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4099 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4101 /* Number of packets out of UINT_MAX to sample. */
4102 probability = dpif_sflow_get_probability(ofproto->sflow);
4103 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4105 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4107 cookie.type = USER_ACTION_COOKIE_SFLOW;
4108 cookie.data = port_ifindex;
4109 cookie.n_output = n_output;
4110 cookie.vlan_tci = 0;
4111 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4113 nl_msg_end_nested(odp_actions, actions_offset);
4114 nl_msg_end_nested(odp_actions, sample_offset);
4115 return cookie_offset;
4118 /* SAMPLE action must be first action in any given list of actions.
4119 * At this point we do not have all information required to build it. So try to
4120 * build sample action as complete as possible. */
4122 add_sflow_action(struct action_xlate_ctx *ctx)
4124 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4126 &ctx->flow, OVSP_NONE);
4127 ctx->sflow_odp_port = 0;
4128 ctx->sflow_n_outputs = 0;
4131 /* Fix SAMPLE action according to data collected while composing ODP actions.
4132 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4133 * USERSPACE action's user-cookie which is required for sflow. */
4135 fix_sflow_action(struct action_xlate_ctx *ctx)
4137 const struct flow *base = &ctx->base_flow;
4138 struct user_action_cookie *cookie;
4140 if (!ctx->user_cookie_offset) {
4144 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4146 assert(cookie != NULL);
4147 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4149 if (ctx->sflow_n_outputs) {
4150 cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow,
4151 ctx->sflow_odp_port);
4153 if (ctx->sflow_n_outputs >= 255) {
4154 cookie->n_output = 255;
4156 cookie->n_output = ctx->sflow_n_outputs;
4158 cookie->vlan_tci = base->vlan_tci;
4162 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4165 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4166 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4167 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4168 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4172 struct priority_to_dscp *pdscp;
4174 if (ofport->up.opp.config & htonl(OFPPC_NO_FWD)
4175 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4179 pdscp = get_priority(ofport, ctx->flow.priority);
4181 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4182 ctx->flow.nw_tos |= pdscp->dscp;
4185 /* We may not have an ofport record for this port, but it doesn't hurt
4186 * to allow forwarding to it anyhow. Maybe such a port will appear
4187 * later and we're pre-populating the flow table. */
4190 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4191 ctx->flow.vlan_tci);
4192 if (out_port != odp_port) {
4193 ctx->flow.vlan_tci = htons(0);
4195 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4196 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4198 ctx->sflow_odp_port = odp_port;
4199 ctx->sflow_n_outputs++;
4200 ctx->nf_output_iface = ofp_port;
4201 ctx->flow.vlan_tci = flow_vlan_tci;
4202 ctx->flow.nw_tos = flow_nw_tos;
4206 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4208 compose_output_action__(ctx, ofp_port, true);
4212 xlate_table_action(struct action_xlate_ctx *ctx,
4213 uint16_t in_port, uint8_t table_id)
4215 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4216 struct ofproto_dpif *ofproto = ctx->ofproto;
4217 struct rule_dpif *rule;
4218 uint16_t old_in_port;
4219 uint8_t old_table_id;
4221 old_table_id = ctx->table_id;
4222 ctx->table_id = table_id;
4224 /* Look up a flow with 'in_port' as the input port. */
4225 old_in_port = ctx->flow.in_port;
4226 ctx->flow.in_port = in_port;
4227 rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
4230 if (table_id > 0 && table_id < N_TABLES) {
4231 struct table_dpif *table = &ofproto->tables[table_id];
4232 if (table->other_table) {
4235 : rule_calculate_tag(&ctx->flow,
4236 &table->other_table->wc,
4241 /* Restore the original input port. Otherwise OFPP_NORMAL and
4242 * OFPP_IN_PORT will have surprising behavior. */
4243 ctx->flow.in_port = old_in_port;
4245 if (ctx->resubmit_hook) {
4246 ctx->resubmit_hook(ctx, rule);
4251 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
4255 ctx->table_id = old_table_id;
4257 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4259 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
4260 MAX_RESUBMIT_RECURSION);
4265 xlate_resubmit_table(struct action_xlate_ctx *ctx,
4266 const struct nx_action_resubmit *nar)
4271 in_port = (nar->in_port == htons(OFPP_IN_PORT)
4273 : ntohs(nar->in_port));
4274 table_id = nar->table == 255 ? ctx->table_id : nar->table;
4276 xlate_table_action(ctx, in_port, table_id);
4280 flood_packets(struct action_xlate_ctx *ctx, bool all)
4282 struct ofport_dpif *ofport;
4284 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
4285 uint16_t ofp_port = ofport->up.ofp_port;
4287 if (ofp_port == ctx->flow.in_port) {
4292 compose_output_action__(ctx, ofp_port, false);
4293 } else if (!(ofport->up.opp.config & htonl(OFPPC_NO_FLOOD))) {
4294 compose_output_action(ctx, ofp_port);
4298 ctx->nf_output_iface = NF_OUT_FLOOD;
4302 compose_controller_action(struct action_xlate_ctx *ctx, int len)
4304 struct user_action_cookie cookie;
4306 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4307 cookie.type = USER_ACTION_COOKIE_CONTROLLER;
4309 cookie.n_output = 0;
4310 cookie.vlan_tci = 0;
4311 put_userspace_action(ctx->ofproto, ctx->odp_actions, &ctx->flow, &cookie);
4315 xlate_output_action__(struct action_xlate_ctx *ctx,
4316 uint16_t port, uint16_t max_len)
4318 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
4320 ctx->nf_output_iface = NF_OUT_DROP;
4324 compose_output_action(ctx, ctx->flow.in_port);
4327 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
4333 flood_packets(ctx, false);
4336 flood_packets(ctx, true);
4338 case OFPP_CONTROLLER:
4339 compose_controller_action(ctx, max_len);
4345 if (port != ctx->flow.in_port) {
4346 compose_output_action(ctx, port);
4351 if (prev_nf_output_iface == NF_OUT_FLOOD) {
4352 ctx->nf_output_iface = NF_OUT_FLOOD;
4353 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
4354 ctx->nf_output_iface = prev_nf_output_iface;
4355 } else if (prev_nf_output_iface != NF_OUT_DROP &&
4356 ctx->nf_output_iface != NF_OUT_FLOOD) {
4357 ctx->nf_output_iface = NF_OUT_MULTI;
4362 xlate_output_reg_action(struct action_xlate_ctx *ctx,
4363 const struct nx_action_output_reg *naor)
4367 ofp_port = nxm_read_field_bits(naor->src, naor->ofs_nbits, &ctx->flow);
4369 if (ofp_port <= UINT16_MAX) {
4370 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
4375 xlate_output_action(struct action_xlate_ctx *ctx,
4376 const struct ofp_action_output *oao)
4378 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
4382 xlate_enqueue_action(struct action_xlate_ctx *ctx,
4383 const struct ofp_action_enqueue *oae)
4386 uint32_t flow_priority, priority;
4389 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
4392 /* Fall back to ordinary output action. */
4393 xlate_output_action__(ctx, ntohs(oae->port), 0);
4397 /* Figure out datapath output port. */
4398 ofp_port = ntohs(oae->port);
4399 if (ofp_port == OFPP_IN_PORT) {
4400 ofp_port = ctx->flow.in_port;
4401 } else if (ofp_port == ctx->flow.in_port) {
4405 /* Add datapath actions. */
4406 flow_priority = ctx->flow.priority;
4407 ctx->flow.priority = priority;
4408 compose_output_action(ctx, ofp_port);
4409 ctx->flow.priority = flow_priority;
4411 /* Update NetFlow output port. */
4412 if (ctx->nf_output_iface == NF_OUT_DROP) {
4413 ctx->nf_output_iface = ofp_port;
4414 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
4415 ctx->nf_output_iface = NF_OUT_MULTI;
4420 xlate_set_queue_action(struct action_xlate_ctx *ctx,
4421 const struct nx_action_set_queue *nasq)
4426 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
4429 /* Couldn't translate queue to a priority, so ignore. A warning
4430 * has already been logged. */
4434 ctx->flow.priority = priority;
4437 struct xlate_reg_state {
4443 xlate_autopath(struct action_xlate_ctx *ctx,
4444 const struct nx_action_autopath *naa)
4446 uint16_t ofp_port = ntohl(naa->id);
4447 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
4449 if (!port || !port->bundle) {
4450 ofp_port = OFPP_NONE;
4451 } else if (port->bundle->bond) {
4452 /* Autopath does not support VLAN hashing. */
4453 struct ofport_dpif *slave = bond_choose_output_slave(
4454 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
4456 ofp_port = slave->up.ofp_port;
4459 autopath_execute(naa, &ctx->flow, ofp_port);
4463 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
4465 struct ofproto_dpif *ofproto = ofproto_;
4466 struct ofport_dpif *port;
4476 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
4479 port = get_ofp_port(ofproto, ofp_port);
4480 return port ? port->may_enable : false;
4485 xlate_learn_action(struct action_xlate_ctx *ctx,
4486 const struct nx_action_learn *learn)
4488 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
4489 struct ofputil_flow_mod fm;
4492 learn_execute(learn, &ctx->flow, &fm);
4494 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
4495 if (error && !VLOG_DROP_WARN(&rl)) {
4496 char *msg = ofputil_error_to_string(error);
4497 VLOG_WARN("learning action failed to modify flow table (%s)", msg);
4505 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
4507 if (port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
4508 ? htonl(OFPPC_NO_RECV_STP)
4509 : htonl(OFPPC_NO_RECV))) {
4513 /* Only drop packets here if both forwarding and learning are
4514 * disabled. If just learning is enabled, we need to have
4515 * OFPP_NORMAL and the learning action have a look at the packet
4516 * before we can drop it. */
4517 if (!stp_forward_in_state(port->stp_state)
4518 && !stp_learn_in_state(port->stp_state)) {
4526 do_xlate_actions(const union ofp_action *in, size_t n_in,
4527 struct action_xlate_ctx *ctx)
4529 const struct ofport_dpif *port;
4530 const union ofp_action *ia;
4533 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
4534 if (port && !may_receive(port, ctx)) {
4535 /* Drop this flow. */
4539 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
4540 const struct ofp_action_dl_addr *oada;
4541 const struct nx_action_resubmit *nar;
4542 const struct nx_action_set_tunnel *nast;
4543 const struct nx_action_set_queue *nasq;
4544 const struct nx_action_multipath *nam;
4545 const struct nx_action_autopath *naa;
4546 const struct nx_action_bundle *nab;
4547 const struct nx_action_output_reg *naor;
4548 enum ofputil_action_code code;
4555 code = ofputil_decode_action_unsafe(ia);
4557 case OFPUTIL_OFPAT_OUTPUT:
4558 xlate_output_action(ctx, &ia->output);
4561 case OFPUTIL_OFPAT_SET_VLAN_VID:
4562 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
4563 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
4566 case OFPUTIL_OFPAT_SET_VLAN_PCP:
4567 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
4568 ctx->flow.vlan_tci |= htons(
4569 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
4572 case OFPUTIL_OFPAT_STRIP_VLAN:
4573 ctx->flow.vlan_tci = htons(0);
4576 case OFPUTIL_OFPAT_SET_DL_SRC:
4577 oada = ((struct ofp_action_dl_addr *) ia);
4578 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
4581 case OFPUTIL_OFPAT_SET_DL_DST:
4582 oada = ((struct ofp_action_dl_addr *) ia);
4583 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
4586 case OFPUTIL_OFPAT_SET_NW_SRC:
4587 ctx->flow.nw_src = ia->nw_addr.nw_addr;
4590 case OFPUTIL_OFPAT_SET_NW_DST:
4591 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
4594 case OFPUTIL_OFPAT_SET_NW_TOS:
4595 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4596 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
4599 case OFPUTIL_OFPAT_SET_TP_SRC:
4600 ctx->flow.tp_src = ia->tp_port.tp_port;
4603 case OFPUTIL_OFPAT_SET_TP_DST:
4604 ctx->flow.tp_dst = ia->tp_port.tp_port;
4607 case OFPUTIL_OFPAT_ENQUEUE:
4608 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
4611 case OFPUTIL_NXAST_RESUBMIT:
4612 nar = (const struct nx_action_resubmit *) ia;
4613 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
4616 case OFPUTIL_NXAST_RESUBMIT_TABLE:
4617 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
4620 case OFPUTIL_NXAST_SET_TUNNEL:
4621 nast = (const struct nx_action_set_tunnel *) ia;
4622 tun_id = htonll(ntohl(nast->tun_id));
4623 ctx->flow.tun_id = tun_id;
4626 case OFPUTIL_NXAST_SET_QUEUE:
4627 nasq = (const struct nx_action_set_queue *) ia;
4628 xlate_set_queue_action(ctx, nasq);
4631 case OFPUTIL_NXAST_POP_QUEUE:
4632 ctx->flow.priority = ctx->original_priority;
4635 case OFPUTIL_NXAST_REG_MOVE:
4636 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
4640 case OFPUTIL_NXAST_REG_LOAD:
4641 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
4645 case OFPUTIL_NXAST_NOTE:
4646 /* Nothing to do. */
4649 case OFPUTIL_NXAST_SET_TUNNEL64:
4650 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
4651 ctx->flow.tun_id = tun_id;
4654 case OFPUTIL_NXAST_MULTIPATH:
4655 nam = (const struct nx_action_multipath *) ia;
4656 multipath_execute(nam, &ctx->flow);
4659 case OFPUTIL_NXAST_AUTOPATH:
4660 naa = (const struct nx_action_autopath *) ia;
4661 xlate_autopath(ctx, naa);
4664 case OFPUTIL_NXAST_BUNDLE:
4665 ctx->ofproto->has_bundle_action = true;
4666 nab = (const struct nx_action_bundle *) ia;
4667 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
4672 case OFPUTIL_NXAST_BUNDLE_LOAD:
4673 ctx->ofproto->has_bundle_action = true;
4674 nab = (const struct nx_action_bundle *) ia;
4675 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
4679 case OFPUTIL_NXAST_OUTPUT_REG:
4680 naor = (const struct nx_action_output_reg *) ia;
4681 xlate_output_reg_action(ctx, naor);
4684 case OFPUTIL_NXAST_LEARN:
4685 ctx->has_learn = true;
4686 if (ctx->may_learn) {
4687 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
4691 case OFPUTIL_NXAST_EXIT:
4697 /* We've let OFPP_NORMAL and the learning action look at the packet,
4698 * so drop it now if forwarding is disabled. */
4699 if (port && !stp_forward_in_state(port->stp_state)) {
4700 ofpbuf_clear(ctx->odp_actions);
4701 add_sflow_action(ctx);
4706 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
4707 struct ofproto_dpif *ofproto, const struct flow *flow,
4708 ovs_be16 initial_tci, const struct ofpbuf *packet)
4710 ctx->ofproto = ofproto;
4712 ctx->base_flow = ctx->flow;
4713 ctx->base_flow.tun_id = 0;
4714 ctx->base_flow.vlan_tci = initial_tci;
4715 ctx->packet = packet;
4716 ctx->may_learn = packet != NULL;
4717 ctx->resubmit_hook = NULL;
4720 static struct ofpbuf *
4721 xlate_actions(struct action_xlate_ctx *ctx,
4722 const union ofp_action *in, size_t n_in)
4724 struct flow orig_flow = ctx->flow;
4726 COVERAGE_INC(ofproto_dpif_xlate);
4728 ctx->odp_actions = ofpbuf_new(512);
4729 ofpbuf_reserve(ctx->odp_actions, NL_A_U32_SIZE);
4731 ctx->may_set_up_flow = true;
4732 ctx->has_learn = false;
4733 ctx->has_normal = false;
4734 ctx->nf_output_iface = NF_OUT_DROP;
4737 ctx->original_priority = ctx->flow.priority;
4741 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
4742 switch (ctx->ofproto->up.frag_handling) {
4743 case OFPC_FRAG_NORMAL:
4744 /* We must pretend that transport ports are unavailable. */
4745 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
4746 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
4749 case OFPC_FRAG_DROP:
4750 return ctx->odp_actions;
4752 case OFPC_FRAG_REASM:
4755 case OFPC_FRAG_NX_MATCH:
4756 /* Nothing to do. */
4761 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
4762 ctx->may_set_up_flow = false;
4763 return ctx->odp_actions;
4765 add_sflow_action(ctx);
4766 do_xlate_actions(in, n_in, ctx);
4768 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
4769 ctx->odp_actions->data,
4770 ctx->odp_actions->size)) {
4771 ctx->may_set_up_flow = false;
4773 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
4775 compose_output_action(ctx, OFPP_LOCAL);
4778 add_mirror_actions(ctx, &orig_flow);
4779 fix_sflow_action(ctx);
4782 return ctx->odp_actions;
4785 /* OFPP_NORMAL implementation. */
4787 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
4789 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
4790 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
4791 * the bundle on which the packet was received, returns the VLAN to which the
4794 * Both 'vid' and the return value are in the range 0...4095. */
4796 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
4798 switch (in_bundle->vlan_mode) {
4799 case PORT_VLAN_ACCESS:
4800 return in_bundle->vlan;
4803 case PORT_VLAN_TRUNK:
4806 case PORT_VLAN_NATIVE_UNTAGGED:
4807 case PORT_VLAN_NATIVE_TAGGED:
4808 return vid ? vid : in_bundle->vlan;
4815 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
4816 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
4819 * 'vid' should be the VID obtained from the 802.1Q header that was received as
4820 * part of a packet (specify 0 if there was no 802.1Q header), in the range
4823 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
4825 /* Allow any VID on the OFPP_NONE port. */
4826 if (in_bundle == &ofpp_none_bundle) {
4830 switch (in_bundle->vlan_mode) {
4831 case PORT_VLAN_ACCESS:
4834 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4835 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
4836 "packet received on port %s configured as VLAN "
4837 "%"PRIu16" access port",
4838 in_bundle->ofproto->up.name, vid,
4839 in_bundle->name, in_bundle->vlan);
4845 case PORT_VLAN_NATIVE_UNTAGGED:
4846 case PORT_VLAN_NATIVE_TAGGED:
4848 /* Port must always carry its native VLAN. */
4852 case PORT_VLAN_TRUNK:
4853 if (!ofbundle_includes_vlan(in_bundle, vid)) {
4855 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4856 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
4857 "received on port %s not configured for trunking "
4859 in_bundle->ofproto->up.name, vid,
4860 in_bundle->name, vid);
4872 /* Given 'vlan', the VLAN that a packet belongs to, and
4873 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
4874 * that should be included in the 802.1Q header. (If the return value is 0,
4875 * then the 802.1Q header should only be included in the packet if there is a
4878 * Both 'vlan' and the return value are in the range 0...4095. */
4880 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
4882 switch (out_bundle->vlan_mode) {
4883 case PORT_VLAN_ACCESS:
4886 case PORT_VLAN_TRUNK:
4887 case PORT_VLAN_NATIVE_TAGGED:
4890 case PORT_VLAN_NATIVE_UNTAGGED:
4891 return vlan == out_bundle->vlan ? 0 : vlan;
4899 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
4902 struct ofport_dpif *port;
4904 ovs_be16 tci, old_tci;
4906 vid = output_vlan_to_vid(out_bundle, vlan);
4907 if (!out_bundle->bond) {
4908 port = ofbundle_get_a_port(out_bundle);
4910 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
4913 /* No slaves enabled, so drop packet. */
4918 old_tci = ctx->flow.vlan_tci;
4920 if (tci || out_bundle->use_priority_tags) {
4921 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
4923 tci |= htons(VLAN_CFI);
4926 ctx->flow.vlan_tci = tci;
4928 compose_output_action(ctx, port->up.ofp_port);
4929 ctx->flow.vlan_tci = old_tci;
4933 mirror_mask_ffs(mirror_mask_t mask)
4935 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
4940 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
4942 return (bundle->vlan_mode != PORT_VLAN_ACCESS
4943 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
4947 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
4949 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
4952 /* Returns an arbitrary interface within 'bundle'. */
4953 static struct ofport_dpif *
4954 ofbundle_get_a_port(const struct ofbundle *bundle)
4956 return CONTAINER_OF(list_front(&bundle->ports),
4957 struct ofport_dpif, bundle_node);
4961 vlan_is_mirrored(const struct ofmirror *m, int vlan)
4963 return !m->vlans || bitmap_is_set(m->vlans, vlan);
4966 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
4967 * to a VLAN. In general most packets may be mirrored but we want to drop
4968 * protocols that may confuse switches. */
4970 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
4972 /* If you change this function's behavior, please update corresponding
4973 * documentation in vswitch.xml at the same time. */
4974 if (dst[0] != 0x01) {
4975 /* All the currently banned MACs happen to start with 01 currently, so
4976 * this is a quick way to eliminate most of the good ones. */
4978 if (eth_addr_is_reserved(dst)) {
4979 /* Drop STP, IEEE pause frames, and other reserved protocols
4980 * (01-80-c2-00-00-0x). */
4984 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
4986 if ((dst[3] & 0xfe) == 0xcc &&
4987 (dst[4] & 0xfe) == 0xcc &&
4988 (dst[5] & 0xfe) == 0xcc) {
4989 /* Drop the following protocols plus others following the same
4992 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
4993 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
4994 STP Uplink Fast (01-00-0c-cd-cd-cd) */
4998 if (!(dst[3] | dst[4] | dst[5])) {
4999 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
5008 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5010 struct ofproto_dpif *ofproto = ctx->ofproto;
5011 mirror_mask_t mirrors;
5012 struct ofbundle *in_bundle;
5015 const struct nlattr *a;
5018 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5019 ctx->packet != NULL);
5023 mirrors = in_bundle->src_mirrors;
5025 /* Drop frames on bundles reserved for mirroring. */
5026 if (in_bundle->mirror_out) {
5027 if (ctx->packet != NULL) {
5028 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5029 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5030 "%s, which is reserved exclusively for mirroring",
5031 ctx->ofproto->up.name, in_bundle->name);
5037 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5038 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5041 vlan = input_vid_to_vlan(in_bundle, vid);
5043 /* Look at the output ports to check for destination selections. */
5045 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5046 ctx->odp_actions->size) {
5047 enum ovs_action_attr type = nl_attr_type(a);
5048 struct ofport_dpif *ofport;
5050 if (type != OVS_ACTION_ATTR_OUTPUT) {
5054 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5055 if (ofport && ofport->bundle) {
5056 mirrors |= ofport->bundle->dst_mirrors;
5064 /* Restore the original packet before adding the mirror actions. */
5065 ctx->flow = *orig_flow;
5070 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5072 if (!vlan_is_mirrored(m, vlan)) {
5073 mirrors &= mirrors - 1;
5077 mirrors &= ~m->dup_mirrors;
5078 ctx->mirrors |= m->dup_mirrors;
5080 output_normal(ctx, m->out, vlan);
5081 } else if (eth_dst_may_rspan(orig_flow->dl_dst)
5082 && vlan != m->out_vlan) {
5083 struct ofbundle *bundle;
5085 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5086 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5087 && !bundle->mirror_out) {
5088 output_normal(ctx, bundle, m->out_vlan);
5096 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5097 uint64_t packets, uint64_t bytes)
5103 for (; mirrors; mirrors &= mirrors - 1) {
5106 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5109 /* In normal circumstances 'm' will not be NULL. However,
5110 * if mirrors are reconfigured, we can temporarily get out
5111 * of sync in facet_revalidate(). We could "correct" the
5112 * mirror list before reaching here, but doing that would
5113 * not properly account the traffic stats we've currently
5114 * accumulated for previous mirror configuration. */
5118 m->packet_count += packets;
5119 m->byte_count += bytes;
5123 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
5124 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
5125 * indicate this; newer upstream kernels use gratuitous ARP requests. */
5127 is_gratuitous_arp(const struct flow *flow)
5129 return (flow->dl_type == htons(ETH_TYPE_ARP)
5130 && eth_addr_is_broadcast(flow->dl_dst)
5131 && (flow->nw_proto == ARP_OP_REPLY
5132 || (flow->nw_proto == ARP_OP_REQUEST
5133 && flow->nw_src == flow->nw_dst)));
5137 update_learning_table(struct ofproto_dpif *ofproto,
5138 const struct flow *flow, int vlan,
5139 struct ofbundle *in_bundle)
5141 struct mac_entry *mac;
5143 /* Don't learn the OFPP_NONE port. */
5144 if (in_bundle == &ofpp_none_bundle) {
5148 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
5152 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
5153 if (is_gratuitous_arp(flow)) {
5154 /* We don't want to learn from gratuitous ARP packets that are
5155 * reflected back over bond slaves so we lock the learning table. */
5156 if (!in_bundle->bond) {
5157 mac_entry_set_grat_arp_lock(mac);
5158 } else if (mac_entry_is_grat_arp_locked(mac)) {
5163 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
5164 /* The log messages here could actually be useful in debugging,
5165 * so keep the rate limit relatively high. */
5166 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5167 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
5168 "on port %s in VLAN %d",
5169 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
5170 in_bundle->name, vlan);
5172 mac->port.p = in_bundle;
5173 tag_set_add(&ofproto->revalidate_set,
5174 mac_learning_changed(ofproto->ml, mac));
5178 static struct ofbundle *
5179 lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn)
5181 struct ofport_dpif *ofport;
5183 /* Special-case OFPP_NONE, which a controller may use as the ingress
5184 * port for traffic that it is sourcing. */
5185 if (in_port == OFPP_NONE) {
5186 return &ofpp_none_bundle;
5189 /* Find the port and bundle for the received packet. */
5190 ofport = get_ofp_port(ofproto, in_port);
5191 if (ofport && ofport->bundle) {
5192 return ofport->bundle;
5195 /* Odd. A few possible reasons here:
5197 * - We deleted a port but there are still a few packets queued up
5200 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
5201 * we don't know about.
5203 * - The ofproto client didn't configure the port as part of a bundle.
5206 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5208 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
5209 "port %"PRIu16, ofproto->up.name, in_port);
5214 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
5215 * dropped. Returns true if they may be forwarded, false if they should be
5218 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
5219 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
5221 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
5222 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
5223 * checked by input_vid_is_valid().
5225 * May also add tags to '*tags', although the current implementation only does
5226 * so in one special case.
5229 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
5230 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
5232 struct ofbundle *in_bundle = in_port->bundle;
5234 /* Drop frames for reserved multicast addresses
5235 * only if forward_bpdu option is absent. */
5236 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
5240 if (in_bundle->bond) {
5241 struct mac_entry *mac;
5243 switch (bond_check_admissibility(in_bundle->bond, in_port,
5244 flow->dl_dst, tags)) {
5251 case BV_DROP_IF_MOVED:
5252 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
5253 if (mac && mac->port.p != in_bundle &&
5254 (!is_gratuitous_arp(flow)
5255 || mac_entry_is_grat_arp_locked(mac))) {
5266 xlate_normal(struct action_xlate_ctx *ctx)
5268 struct ofport_dpif *in_port;
5269 struct ofbundle *in_bundle;
5270 struct mac_entry *mac;
5274 ctx->has_normal = true;
5276 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
5277 ctx->packet != NULL);
5282 /* We know 'in_port' exists unless it is "ofpp_none_bundle",
5283 * since lookup_input_bundle() succeeded. */
5284 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5286 /* Drop malformed frames. */
5287 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
5288 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
5289 if (ctx->packet != NULL) {
5290 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5291 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
5292 "VLAN tag received on port %s",
5293 ctx->ofproto->up.name, in_bundle->name);
5298 /* Drop frames on bundles reserved for mirroring. */
5299 if (in_bundle->mirror_out) {
5300 if (ctx->packet != NULL) {
5301 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5302 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5303 "%s, which is reserved exclusively for mirroring",
5304 ctx->ofproto->up.name, in_bundle->name);
5310 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
5311 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5314 vlan = input_vid_to_vlan(in_bundle, vid);
5316 /* Check other admissibility requirements. */
5318 !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
5322 /* Learn source MAC. */
5323 if (ctx->may_learn) {
5324 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
5327 /* Determine output bundle. */
5328 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
5331 if (mac->port.p != in_bundle) {
5332 output_normal(ctx, mac->port.p, vlan);
5334 } else if (!ctx->packet && !eth_addr_is_multicast(ctx->flow.dl_dst)) {
5335 /* If we are revalidating but don't have a learning entry then eject
5336 * the flow. Installing a flow that floods packets opens up a window
5337 * of time where we could learn from a packet reflected on a bond and
5338 * blackhole packets before the learning table is updated to reflect
5339 * the correct port. */
5340 ctx->may_set_up_flow = false;
5343 struct ofbundle *bundle;
5345 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
5346 if (bundle != in_bundle
5347 && ofbundle_includes_vlan(bundle, vlan)
5348 && bundle->floodable
5349 && !bundle->mirror_out) {
5350 output_normal(ctx, bundle, vlan);
5353 ctx->nf_output_iface = NF_OUT_FLOOD;
5357 /* Optimized flow revalidation.
5359 * It's a difficult problem, in general, to tell which facets need to have
5360 * their actions recalculated whenever the OpenFlow flow table changes. We
5361 * don't try to solve that general problem: for most kinds of OpenFlow flow
5362 * table changes, we recalculate the actions for every facet. This is
5363 * relatively expensive, but it's good enough if the OpenFlow flow table
5364 * doesn't change very often.
5366 * However, we can expect one particular kind of OpenFlow flow table change to
5367 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5368 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5369 * table, we add a special case that applies to flow tables in which every rule
5370 * has the same form (that is, the same wildcards), except that the table is
5371 * also allowed to have a single "catch-all" flow that matches all packets. We
5372 * optimize this case by tagging all of the facets that resubmit into the table
5373 * and invalidating the same tag whenever a flow changes in that table. The
5374 * end result is that we revalidate just the facets that need it (and sometimes
5375 * a few more, but not all of the facets or even all of the facets that
5376 * resubmit to the table modified by MAC learning). */
5378 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
5379 * into an OpenFlow table with the given 'basis'. */
5381 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
5384 if (flow_wildcards_is_catchall(wc)) {
5387 struct flow tag_flow = *flow;
5388 flow_zero_wildcards(&tag_flow, wc);
5389 return tag_create_deterministic(flow_hash(&tag_flow, secret));
5393 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5394 * taggability of that table.
5396 * This function must be called after *each* change to a flow table. If you
5397 * skip calling it on some changes then the pointer comparisons at the end can
5398 * be invalid if you get unlucky. For example, if a flow removal causes a
5399 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5400 * different wildcards to be created with the same address, then this function
5401 * will incorrectly skip revalidation. */
5403 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5405 struct table_dpif *table = &ofproto->tables[table_id];
5406 const struct classifier *cls = &ofproto->up.tables[table_id];
5407 struct cls_table *catchall, *other;
5408 struct cls_table *t;
5410 catchall = other = NULL;
5412 switch (hmap_count(&cls->tables)) {
5414 /* We could tag this OpenFlow table but it would make the logic a
5415 * little harder and it's a corner case that doesn't seem worth it
5421 HMAP_FOR_EACH (t, hmap_node, &cls->tables) {
5422 if (cls_table_is_catchall(t)) {
5424 } else if (!other) {
5427 /* Indicate that we can't tag this by setting both tables to
5428 * NULL. (We know that 'catchall' is already NULL.) */
5435 /* Can't tag this table. */
5439 if (table->catchall_table != catchall || table->other_table != other) {
5440 table->catchall_table = catchall;
5441 table->other_table = other;
5442 ofproto->need_revalidate = true;
5446 /* Given 'rule' that has changed in some way (either it is a rule being
5447 * inserted, a rule being deleted, or a rule whose actions are being
5448 * modified), marks facets for revalidation to ensure that packets will be
5449 * forwarded correctly according to the new state of the flow table.
5451 * This function must be called after *each* change to a flow table. See
5452 * the comment on table_update_taggable() for more information. */
5454 rule_invalidate(const struct rule_dpif *rule)
5456 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5458 table_update_taggable(ofproto, rule->up.table_id);
5460 if (!ofproto->need_revalidate) {
5461 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5463 if (table->other_table && rule->tag) {
5464 tag_set_add(&ofproto->revalidate_set, rule->tag);
5466 ofproto->need_revalidate = true;
5472 set_frag_handling(struct ofproto *ofproto_,
5473 enum ofp_config_flags frag_handling)
5475 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5477 if (frag_handling != OFPC_FRAG_REASM) {
5478 ofproto->need_revalidate = true;
5486 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5487 const struct flow *flow,
5488 const union ofp_action *ofp_actions, size_t n_ofp_actions)
5490 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5493 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
5494 return ofp_mkerr_nicira(OFPET_BAD_REQUEST, NXBRC_BAD_IN_PORT);
5497 error = validate_actions(ofp_actions, n_ofp_actions, flow,
5498 ofproto->max_ports);
5500 struct odputil_keybuf keybuf;
5501 struct action_xlate_ctx ctx;
5502 struct ofpbuf *odp_actions;
5505 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5506 odp_flow_key_from_flow(&key, flow);
5508 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, packet);
5509 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
5510 dpif_execute(ofproto->dpif, key.data, key.size,
5511 odp_actions->data, odp_actions->size, packet);
5512 ofpbuf_delete(odp_actions);
5520 set_netflow(struct ofproto *ofproto_,
5521 const struct netflow_options *netflow_options)
5523 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5525 if (netflow_options) {
5526 if (!ofproto->netflow) {
5527 ofproto->netflow = netflow_create();
5529 return netflow_set_options(ofproto->netflow, netflow_options);
5531 netflow_destroy(ofproto->netflow);
5532 ofproto->netflow = NULL;
5538 get_netflow_ids(const struct ofproto *ofproto_,
5539 uint8_t *engine_type, uint8_t *engine_id)
5541 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5543 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
5547 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5549 if (!facet_is_controller_flow(facet) &&
5550 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5551 struct subfacet *subfacet;
5552 struct ofexpired expired;
5554 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5555 if (subfacet->installed) {
5556 struct dpif_flow_stats stats;
5558 subfacet_install(ofproto, subfacet, subfacet->actions,
5559 subfacet->actions_len, &stats);
5560 subfacet_update_stats(ofproto, subfacet, &stats);
5564 expired.flow = facet->flow;
5565 expired.packet_count = facet->packet_count;
5566 expired.byte_count = facet->byte_count;
5567 expired.used = facet->used;
5568 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5573 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5575 struct facet *facet;
5577 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5578 send_active_timeout(ofproto, facet);
5582 static struct ofproto_dpif *
5583 ofproto_dpif_lookup(const char *name)
5585 struct ofproto_dpif *ofproto;
5587 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5588 hash_string(name, 0), &all_ofproto_dpifs) {
5589 if (!strcmp(ofproto->up.name, name)) {
5597 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn,
5598 const char *args, void *aux OVS_UNUSED)
5600 struct ofproto_dpif *ofproto;
5602 if (args[0] != '\0') {
5603 ofproto = ofproto_dpif_lookup(args);
5605 unixctl_command_reply(conn, 501, "no such bridge");
5608 mac_learning_flush(ofproto->ml);
5609 ofproto->need_revalidate = true;
5611 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5612 mac_learning_flush(ofproto->ml);
5613 ofproto->need_revalidate = true;
5617 unixctl_command_reply(conn, 200, "table successfully flushed");
5621 ofproto_unixctl_fdb_show(struct unixctl_conn *conn,
5622 const char *args, void *aux OVS_UNUSED)
5624 struct ds ds = DS_EMPTY_INITIALIZER;
5625 const struct ofproto_dpif *ofproto;
5626 const struct mac_entry *e;
5628 ofproto = ofproto_dpif_lookup(args);
5630 unixctl_command_reply(conn, 501, "no such bridge");
5634 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5635 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5636 struct ofbundle *bundle = e->port.p;
5637 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
5638 ofbundle_get_a_port(bundle)->odp_port,
5639 e->vlan, ETH_ADDR_ARGS(e->mac),
5640 mac_entry_age(ofproto->ml, e));
5642 unixctl_command_reply(conn, 200, ds_cstr(&ds));
5646 struct ofproto_trace {
5647 struct action_xlate_ctx ctx;
5653 trace_format_rule(struct ds *result, uint8_t table_id, int level,
5654 const struct rule_dpif *rule)
5656 ds_put_char_multiple(result, '\t', level);
5658 ds_put_cstr(result, "No match\n");
5662 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5663 table_id, ntohll(rule->up.flow_cookie));
5664 cls_rule_format(&rule->up.cr, result);
5665 ds_put_char(result, '\n');
5667 ds_put_char_multiple(result, '\t', level);
5668 ds_put_cstr(result, "OpenFlow ");
5669 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
5670 ds_put_char(result, '\n');
5674 trace_format_flow(struct ds *result, int level, const char *title,
5675 struct ofproto_trace *trace)
5677 ds_put_char_multiple(result, '\t', level);
5678 ds_put_format(result, "%s: ", title);
5679 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
5680 ds_put_cstr(result, "unchanged");
5682 flow_format(result, &trace->ctx.flow);
5683 trace->flow = trace->ctx.flow;
5685 ds_put_char(result, '\n');
5689 trace_format_regs(struct ds *result, int level, const char *title,
5690 struct ofproto_trace *trace)
5694 ds_put_char_multiple(result, '\t', level);
5695 ds_put_format(result, "%s:", title);
5696 for (i = 0; i < FLOW_N_REGS; i++) {
5697 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5699 ds_put_char(result, '\n');
5703 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
5705 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
5706 struct ds *result = trace->result;
5708 ds_put_char(result, '\n');
5709 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
5710 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
5711 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
5715 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
5716 void *aux OVS_UNUSED)
5718 char *dpname, *arg1, *arg2, *arg3, *arg4;
5719 char *args = xstrdup(args_);
5720 char *save_ptr = NULL;
5721 struct ofproto_dpif *ofproto;
5722 struct ofpbuf odp_key;
5723 struct ofpbuf *packet;
5724 struct rule_dpif *rule;
5725 ovs_be16 initial_tci;
5731 ofpbuf_init(&odp_key, 0);
5734 dpname = strtok_r(args, " ", &save_ptr);
5736 unixctl_command_reply(conn, 501, "Bad command syntax");
5740 ofproto = ofproto_dpif_lookup(dpname);
5742 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
5746 arg1 = strtok_r(NULL, " ", &save_ptr);
5747 arg2 = strtok_r(NULL, " ", &save_ptr);
5748 arg3 = strtok_r(NULL, " ", &save_ptr);
5749 arg4 = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
5750 if (dpname && arg1 && (!arg2 || !strcmp(arg2, "-generate")) && !arg3) {
5751 /* ofproto/trace dpname flow [-generate] */
5754 /* Convert string to datapath key. */
5755 ofpbuf_init(&odp_key, 0);
5756 error = odp_flow_key_from_string(arg1, NULL, &odp_key);
5758 unixctl_command_reply(conn, 501, "Bad flow syntax");
5762 /* Convert odp_key to flow. */
5763 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
5764 odp_key.size, &flow,
5765 &initial_tci, NULL);
5766 if (error == ODP_FIT_ERROR) {
5767 unixctl_command_reply(conn, 501, "Invalid flow");
5771 /* Generate a packet, if requested. */
5773 packet = ofpbuf_new(0);
5774 flow_compose(packet, &flow);
5776 } else if (dpname && arg1 && arg2 && arg3 && arg4) {
5777 /* ofproto/trace dpname priority tun_id in_port packet */
5782 priority = atoi(arg1);
5783 tun_id = htonll(strtoull(arg2, NULL, 0));
5784 in_port = ofp_port_to_odp_port(atoi(arg3));
5786 packet = ofpbuf_new(strlen(args) / 2);
5787 arg4 = ofpbuf_put_hex(packet, arg4, NULL);
5788 arg4 += strspn(arg4, " ");
5789 if (*arg4 != '\0') {
5790 unixctl_command_reply(conn, 501, "Trailing garbage in command");
5793 if (packet->size < ETH_HEADER_LEN) {
5794 unixctl_command_reply(conn, 501,
5795 "Packet data too short for Ethernet");
5799 ds_put_cstr(&result, "Packet: ");
5800 s = ofp_packet_to_string(packet->data, packet->size, packet->size);
5801 ds_put_cstr(&result, s);
5804 flow_extract(packet, priority, tun_id, in_port, &flow);
5805 initial_tci = flow.vlan_tci;
5807 unixctl_command_reply(conn, 501, "Bad command syntax");
5811 ds_put_cstr(&result, "Flow: ");
5812 flow_format(&result, &flow);
5813 ds_put_char(&result, '\n');
5815 rule = rule_dpif_lookup(ofproto, &flow, 0);
5816 trace_format_rule(&result, 0, 0, rule);
5818 struct ofproto_trace trace;
5819 struct ofpbuf *odp_actions;
5821 trace.result = &result;
5823 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, initial_tci, packet);
5824 trace.ctx.resubmit_hook = trace_resubmit;
5825 odp_actions = xlate_actions(&trace.ctx,
5826 rule->up.actions, rule->up.n_actions);
5828 ds_put_char(&result, '\n');
5829 trace_format_flow(&result, 0, "Final flow", &trace);
5830 ds_put_cstr(&result, "Datapath actions: ");
5831 format_odp_actions(&result, odp_actions->data, odp_actions->size);
5832 ofpbuf_delete(odp_actions);
5834 if (!trace.ctx.may_set_up_flow) {
5836 ds_put_cstr(&result, "\nThis flow is not cachable.");
5838 ds_put_cstr(&result, "\nThe datapath actions are incomplete--"
5839 "for complete actions, please supply a packet.");
5844 unixctl_command_reply(conn, 200, ds_cstr(&result));
5847 ds_destroy(&result);
5848 ofpbuf_delete(packet);
5849 ofpbuf_uninit(&odp_key);
5854 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED,
5855 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
5858 unixctl_command_reply(conn, 200, NULL);
5862 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED,
5863 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
5866 unixctl_command_reply(conn, 200, NULL);
5870 ofproto_dpif_unixctl_init(void)
5872 static bool registered;
5878 unixctl_command_register("ofproto/trace",
5879 "bridge {tun_id in_port packet | odp_flow [-generate]}",
5880 ofproto_unixctl_trace, NULL);
5881 unixctl_command_register("fdb/flush", "[bridge]",
5882 ofproto_unixctl_fdb_flush, NULL);
5883 unixctl_command_register("fdb/show", "bridge", ofproto_unixctl_fdb_show,
5885 unixctl_command_register("ofproto/clog", "", ofproto_dpif_clog, NULL);
5886 unixctl_command_register("ofproto/unclog", "", ofproto_dpif_unclog, NULL);
5889 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5891 * This is deprecated. It is only for compatibility with broken device drivers
5892 * in old versions of Linux that do not properly support VLANs when VLAN
5893 * devices are not used. When broken device drivers are no longer in
5894 * widespread use, we will delete these interfaces. */
5897 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
5899 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
5900 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
5902 if (realdev_ofp_port == ofport->realdev_ofp_port
5903 && vid == ofport->vlandev_vid) {
5907 ofproto->need_revalidate = true;
5909 if (ofport->realdev_ofp_port) {
5912 if (realdev_ofp_port && ofport->bundle) {
5913 /* vlandevs are enslaved to their realdevs, so they are not allowed to
5914 * themselves be part of a bundle. */
5915 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
5918 ofport->realdev_ofp_port = realdev_ofp_port;
5919 ofport->vlandev_vid = vid;
5921 if (realdev_ofp_port) {
5922 vsp_add(ofport, realdev_ofp_port, vid);
5929 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
5931 return hash_2words(realdev_ofp_port, vid);
5935 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
5936 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
5938 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
5939 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
5940 int vid = vlan_tci_to_vid(vlan_tci);
5941 const struct vlan_splinter *vsp;
5943 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
5944 hash_realdev_vid(realdev_ofp_port, vid),
5945 &ofproto->realdev_vid_map) {
5946 if (vsp->realdev_ofp_port == realdev_ofp_port
5947 && vsp->vid == vid) {
5948 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
5952 return realdev_odp_port;
5955 static struct vlan_splinter *
5956 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
5958 struct vlan_splinter *vsp;
5960 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
5961 &ofproto->vlandev_map) {
5962 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
5971 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
5972 uint16_t vlandev_ofp_port, int *vid)
5974 if (!hmap_is_empty(&ofproto->vlandev_map)) {
5975 const struct vlan_splinter *vsp;
5977 vsp = vlandev_find(ofproto, vlandev_ofp_port);
5982 return vsp->realdev_ofp_port;
5989 vsp_remove(struct ofport_dpif *port)
5991 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
5992 struct vlan_splinter *vsp;
5994 vsp = vlandev_find(ofproto, port->up.ofp_port);
5996 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
5997 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6000 port->realdev_ofp_port = 0;
6002 VLOG_ERR("missing vlan device record");
6007 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
6009 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6011 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6012 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6013 == realdev_ofp_port)) {
6014 struct vlan_splinter *vsp;
6016 vsp = xmalloc(sizeof *vsp);
6017 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6018 hash_int(port->up.ofp_port, 0));
6019 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6020 hash_realdev_vid(realdev_ofp_port, vid));
6021 vsp->realdev_ofp_port = realdev_ofp_port;
6022 vsp->vlandev_ofp_port = port->up.ofp_port;
6025 port->realdev_ofp_port = realdev_ofp_port;
6027 VLOG_ERR("duplicate vlan device record");
6031 const struct ofproto_class ofproto_dpif_class = {
6059 port_is_lacp_current,
6060 NULL, /* rule_choose_table */
6067 rule_modify_actions,
6075 get_cfm_remote_mpids,
6079 get_stp_port_status,
6086 is_mirror_output_bundle,
6087 forward_bpdu_changed,