2 * Copyright (c) 2009, 2010, 2011, 2012 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-provider.h"
26 #include "byte-order.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
36 #include "mac-learning.h"
37 #include "meta-flow.h"
38 #include "multipath.h"
45 #include "ofp-print.h"
46 #include "ofproto-dpif-sflow.h"
47 #include "poll-loop.h"
49 #include "unaligned.h"
51 #include "vlan-bitmap.h"
54 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
56 COVERAGE_DEFINE(ofproto_dpif_ctlr_action);
57 COVERAGE_DEFINE(ofproto_dpif_expired);
58 COVERAGE_DEFINE(ofproto_dpif_no_packet_in);
59 COVERAGE_DEFINE(ofproto_dpif_xlate);
60 COVERAGE_DEFINE(facet_changed_rule);
61 COVERAGE_DEFINE(facet_invalidated);
62 COVERAGE_DEFINE(facet_revalidate);
63 COVERAGE_DEFINE(facet_unexpected);
65 /* Maximum depth of flow table recursion (due to resubmit actions) in a
66 * flow translation. */
67 #define MAX_RESUBMIT_RECURSION 32
69 /* Number of implemented OpenFlow tables. */
70 enum { N_TABLES = 255 };
71 BUILD_ASSERT_DECL(N_TABLES >= 1 && N_TABLES <= 255);
81 * - Do include packets and bytes from facets that have been deleted or
82 * whose own statistics have been folded into the rule.
84 * - Do include packets and bytes sent "by hand" that were accounted to
85 * the rule without any facet being involved (this is a rare corner
86 * case in rule_execute()).
88 * - Do not include packet or bytes that can be obtained from any facet's
89 * packet_count or byte_count member or that can be obtained from the
90 * datapath by, e.g., dpif_flow_get() for any subfacet.
92 uint64_t packet_count; /* Number of packets received. */
93 uint64_t byte_count; /* Number of bytes received. */
95 tag_type tag; /* Caches rule_calculate_tag() result. */
97 struct list facets; /* List of "struct facet"s. */
100 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
102 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
105 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
106 const struct flow *, uint8_t table);
108 static void rule_credit_stats(struct rule_dpif *,
109 const struct dpif_flow_stats *);
110 static void flow_push_stats(struct rule_dpif *, const struct flow *,
111 const struct dpif_flow_stats *);
112 static tag_type rule_calculate_tag(const struct flow *,
113 const struct flow_wildcards *,
115 static void rule_invalidate(const struct rule_dpif *);
117 #define MAX_MIRRORS 32
118 typedef uint32_t mirror_mask_t;
119 #define MIRROR_MASK_C(X) UINT32_C(X)
120 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
122 struct ofproto_dpif *ofproto; /* Owning ofproto. */
123 size_t idx; /* In ofproto's "mirrors" array. */
124 void *aux; /* Key supplied by ofproto's client. */
125 char *name; /* Identifier for log messages. */
127 /* Selection criteria. */
128 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
129 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
130 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
132 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
133 struct ofbundle *out; /* Output port or NULL. */
134 int out_vlan; /* Output VLAN or -1. */
135 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
138 int64_t packet_count; /* Number of packets sent. */
139 int64_t byte_count; /* Number of bytes sent. */
142 static void mirror_destroy(struct ofmirror *);
143 static void update_mirror_stats(struct ofproto_dpif *ofproto,
144 mirror_mask_t mirrors,
145 uint64_t packets, uint64_t bytes);
148 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
149 struct ofproto_dpif *ofproto; /* Owning ofproto. */
150 void *aux; /* Key supplied by ofproto's client. */
151 char *name; /* Identifier for log messages. */
154 struct list ports; /* Contains "struct ofport"s. */
155 enum port_vlan_mode vlan_mode; /* VLAN mode */
156 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
157 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
158 * NULL if all VLANs are trunked. */
159 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
160 struct bond *bond; /* Nonnull iff more than one port. */
161 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
164 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
166 /* Port mirroring info. */
167 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
168 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
169 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
172 static void bundle_remove(struct ofport *);
173 static void bundle_update(struct ofbundle *);
174 static void bundle_destroy(struct ofbundle *);
175 static void bundle_del_port(struct ofport_dpif *);
176 static void bundle_run(struct ofbundle *);
177 static void bundle_wait(struct ofbundle *);
178 static struct ofbundle *lookup_input_bundle(struct ofproto_dpif *,
179 uint16_t in_port, bool warn);
181 /* A controller may use OFPP_NONE as the ingress port to indicate that
182 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
183 * when an input bundle is needed for validation (e.g., mirroring or
184 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
185 * any 'port' structs, so care must be taken when dealing with it. */
186 static struct ofbundle ofpp_none_bundle = {
188 .vlan_mode = PORT_VLAN_TRUNK
191 static void stp_run(struct ofproto_dpif *ofproto);
192 static void stp_wait(struct ofproto_dpif *ofproto);
193 static int set_stp_port(struct ofport *,
194 const struct ofproto_port_stp_settings *);
196 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
198 struct action_xlate_ctx {
199 /* action_xlate_ctx_init() initializes these members. */
202 struct ofproto_dpif *ofproto;
204 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
205 * this flow when actions change header fields. */
208 /* The packet corresponding to 'flow', or a null pointer if we are
209 * revalidating without a packet to refer to. */
210 const struct ofpbuf *packet;
212 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
213 * actions update the flow table?
215 * We want to update these tables if we are actually processing a packet,
216 * or if we are accounting for packets that the datapath has processed, but
217 * not if we are just revalidating. */
220 /* The rule that we are currently translating, or NULL. */
221 struct rule_dpif *rule;
223 /* Union of the set of TCP flags seen so far in this flow. (Used only by
224 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
228 /* If nonnull, flow translation calls this function just before executing a
229 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
230 * when the recursion depth is exceeded.
232 * 'rule' is the rule being submitted into. It will be null if the
233 * resubmit or OFPP_TABLE action didn't find a matching rule.
235 * This is normally null so the client has to set it manually after
236 * calling action_xlate_ctx_init(). */
237 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule);
239 /* If nonnull, flow translation credits the specified statistics to each
240 * rule reached through a resubmit or OFPP_TABLE action.
242 * This is normally null so the client has to set it manually after
243 * calling action_xlate_ctx_init(). */
244 const struct dpif_flow_stats *resubmit_stats;
246 /* xlate_actions() initializes and uses these members. The client might want
247 * to look at them after it returns. */
249 struct ofpbuf *odp_actions; /* Datapath actions. */
250 tag_type tags; /* Tags associated with actions. */
251 bool may_set_up_flow; /* True ordinarily; false if the actions must
252 * be reassessed for every packet. */
253 bool has_learn; /* Actions include NXAST_LEARN? */
254 bool has_normal; /* Actions output to OFPP_NORMAL? */
255 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
256 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
257 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
259 /* xlate_actions() initializes and uses these members, but the client has no
260 * reason to look at them. */
262 int recurse; /* Recursion level, via xlate_table_action. */
263 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
264 struct flow base_flow; /* Flow at the last commit. */
265 uint32_t orig_skb_priority; /* Priority when packet arrived. */
266 uint8_t table_id; /* OpenFlow table ID where flow was found. */
267 uint32_t sflow_n_outputs; /* Number of output ports. */
268 uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
269 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
270 bool exit; /* No further actions should be processed. */
273 static void action_xlate_ctx_init(struct action_xlate_ctx *,
274 struct ofproto_dpif *, const struct flow *,
275 ovs_be16 initial_tci, struct rule_dpif *,
276 uint8_t tcp_flags, const struct ofpbuf *);
277 static void xlate_actions(struct action_xlate_ctx *,
278 const union ofp_action *in, size_t n_in,
279 struct ofpbuf *odp_actions);
280 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
281 const union ofp_action *in,
284 /* A dpif flow and actions associated with a facet.
286 * See also the large comment on struct facet. */
289 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
290 struct list list_node; /* In struct facet's 'facets' list. */
291 struct facet *facet; /* Owning facet. */
295 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
296 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
297 * regenerate the ODP flow key from ->facet->flow. */
298 enum odp_key_fitness key_fitness;
302 long long int used; /* Time last used; time created if not used. */
304 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
305 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
309 * These should be essentially identical for every subfacet in a facet, but
310 * may differ in trivial ways due to VLAN splinters. */
311 size_t actions_len; /* Number of bytes in actions[]. */
312 struct nlattr *actions; /* Datapath actions. */
314 bool installed; /* Installed in datapath? */
316 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
317 * splinters can cause it to differ. This value should be removed when
318 * the VLAN splinters feature is no longer needed. */
319 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
322 static struct subfacet *subfacet_create(struct facet *, enum odp_key_fitness,
323 const struct nlattr *key,
324 size_t key_len, ovs_be16 initial_tci);
325 static struct subfacet *subfacet_find(struct ofproto_dpif *,
326 const struct nlattr *key, size_t key_len);
327 static void subfacet_destroy(struct subfacet *);
328 static void subfacet_destroy__(struct subfacet *);
329 static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *,
331 static void subfacet_reset_dp_stats(struct subfacet *,
332 struct dpif_flow_stats *);
333 static void subfacet_update_time(struct subfacet *, long long int used);
334 static void subfacet_update_stats(struct subfacet *,
335 const struct dpif_flow_stats *);
336 static void subfacet_make_actions(struct subfacet *,
337 const struct ofpbuf *packet,
338 struct ofpbuf *odp_actions);
339 static int subfacet_install(struct subfacet *,
340 const struct nlattr *actions, size_t actions_len,
341 struct dpif_flow_stats *);
342 static void subfacet_uninstall(struct subfacet *);
344 /* An exact-match instantiation of an OpenFlow flow.
346 * A facet associates a "struct flow", which represents the Open vSwitch
347 * userspace idea of an exact-match flow, with one or more subfacets. Each
348 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
349 * the facet. When the kernel module (or other dpif implementation) and Open
350 * vSwitch userspace agree on the definition of a flow key, there is exactly
351 * one subfacet per facet. If the dpif implementation supports more-specific
352 * flow matching than userspace, however, a facet can have more than one
353 * subfacet, each of which corresponds to some distinction in flow that
354 * userspace simply doesn't understand.
356 * Flow expiration works in terms of subfacets, so a facet must have at least
357 * one subfacet or it will never expire, leaking memory. */
360 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
361 struct list list_node; /* In owning rule's 'facets' list. */
362 struct rule_dpif *rule; /* Owning rule. */
365 struct list subfacets;
366 long long int used; /* Time last used; time created if not used. */
373 * - Do include packets and bytes sent "by hand", e.g. with
376 * - Do include packets and bytes that were obtained from the datapath
377 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
378 * DPIF_FP_ZERO_STATS).
380 * - Do not include packets or bytes that can be obtained from the
381 * datapath for any existing subfacet.
383 uint64_t packet_count; /* Number of packets received. */
384 uint64_t byte_count; /* Number of bytes received. */
386 /* Resubmit statistics. */
387 uint64_t prev_packet_count; /* Number of packets from last stats push. */
388 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
389 long long int prev_used; /* Used time from last stats push. */
392 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
393 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
394 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
396 /* Properties of datapath actions.
398 * Every subfacet has its own actions because actions can differ slightly
399 * between splintered and non-splintered subfacets due to the VLAN tag
400 * being initially different (present vs. absent). All of them have these
401 * properties in common so we just store one copy of them here. */
402 bool may_install; /* Reassess actions for every packet? */
403 bool has_learn; /* Actions include NXAST_LEARN? */
404 bool has_normal; /* Actions output to OFPP_NORMAL? */
405 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
406 tag_type tags; /* Tags that would require revalidation. */
407 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
409 /* Storage for a single subfacet, to reduce malloc() time and space
410 * overhead. (A facet always has at least one subfacet and in the common
411 * case has exactly one subfacet.) */
412 struct subfacet one_subfacet;
415 static struct facet *facet_create(struct rule_dpif *,
416 const struct flow *, uint32_t hash);
417 static void facet_remove(struct facet *);
418 static void facet_free(struct facet *);
420 static struct facet *facet_find(struct ofproto_dpif *,
421 const struct flow *, uint32_t hash);
422 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
423 const struct flow *, uint32_t hash);
424 static bool facet_revalidate(struct facet *);
425 static bool facet_check_consistency(struct facet *);
427 static void facet_flush_stats(struct facet *);
429 static void facet_update_time(struct facet *, long long int used);
430 static void facet_reset_counters(struct facet *);
431 static void facet_push_stats(struct facet *);
432 static void facet_learn(struct facet *);
433 static void facet_account(struct facet *);
435 static bool facet_is_controller_flow(struct facet *);
441 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
442 struct list bundle_node; /* In struct ofbundle's "ports" list. */
443 struct cfm *cfm; /* Connectivity Fault Management, if any. */
444 tag_type tag; /* Tag associated with this port. */
445 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
446 bool may_enable; /* May be enabled in bonds. */
447 long long int carrier_seq; /* Carrier status changes. */
450 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
451 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
452 long long int stp_state_entered;
454 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
456 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
458 * This is deprecated. It is only for compatibility with broken device
459 * drivers in old versions of Linux that do not properly support VLANs when
460 * VLAN devices are not used. When broken device drivers are no longer in
461 * widespread use, we will delete these interfaces. */
462 uint16_t realdev_ofp_port;
466 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
467 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
468 * traffic egressing the 'ofport' with that priority should be marked with. */
469 struct priority_to_dscp {
470 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
471 uint32_t priority; /* Priority of this queue (see struct flow). */
473 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
476 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
478 * This is deprecated. It is only for compatibility with broken device drivers
479 * in old versions of Linux that do not properly support VLANs when VLAN
480 * devices are not used. When broken device drivers are no longer in
481 * widespread use, we will delete these interfaces. */
482 struct vlan_splinter {
483 struct hmap_node realdev_vid_node;
484 struct hmap_node vlandev_node;
485 uint16_t realdev_ofp_port;
486 uint16_t vlandev_ofp_port;
490 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
491 uint32_t realdev, ovs_be16 vlan_tci);
492 static uint16_t vsp_vlandev_to_realdev(const struct ofproto_dpif *,
493 uint16_t vlandev, int *vid);
494 static void vsp_remove(struct ofport_dpif *);
495 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
497 static struct ofport_dpif *
498 ofport_dpif_cast(const struct ofport *ofport)
500 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
501 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
504 static void port_run(struct ofport_dpif *);
505 static void port_wait(struct ofport_dpif *);
506 static int set_cfm(struct ofport *, const struct cfm_settings *);
507 static void ofport_clear_priorities(struct ofport_dpif *);
509 struct dpif_completion {
510 struct list list_node;
511 struct ofoperation *op;
514 /* Extra information about a classifier table.
515 * Currently used just for optimized flow revalidation. */
517 /* If either of these is nonnull, then this table has a form that allows
518 * flows to be tagged to avoid revalidating most flows for the most common
519 * kinds of flow table changes. */
520 struct cls_table *catchall_table; /* Table that wildcards all fields. */
521 struct cls_table *other_table; /* Table with any other wildcard set. */
522 uint32_t basis; /* Keeps each table's tags separate. */
525 struct ofproto_dpif {
526 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
535 struct netflow *netflow;
536 struct dpif_sflow *sflow;
537 struct hmap bundles; /* Contains "struct ofbundle"s. */
538 struct mac_learning *ml;
539 struct ofmirror *mirrors[MAX_MIRRORS];
540 bool has_bonded_bundles;
543 struct timer next_expiration;
547 struct hmap subfacets;
550 struct table_dpif tables[N_TABLES];
551 bool need_revalidate;
552 struct tag_set revalidate_set;
554 /* Support for debugging async flow mods. */
555 struct list completions;
557 bool has_bundle_action; /* True when the first bundle action appears. */
558 struct netdev_stats stats; /* To account packets generated and consumed in
563 long long int stp_last_tick;
565 /* VLAN splinters. */
566 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
567 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
570 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
571 * for debugging the asynchronous flow_mod implementation.) */
574 /* All existing ofproto_dpif instances, indexed by ->up.name. */
575 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
577 static void ofproto_dpif_unixctl_init(void);
579 static struct ofproto_dpif *
580 ofproto_dpif_cast(const struct ofproto *ofproto)
582 assert(ofproto->ofproto_class == &ofproto_dpif_class);
583 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
586 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
588 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
590 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
591 const struct ofpbuf *, ovs_be16 initial_tci,
594 /* Packet processing. */
595 static void update_learning_table(struct ofproto_dpif *,
596 const struct flow *, int vlan,
599 #define FLOW_MISS_MAX_BATCH 50
600 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
602 /* Flow expiration. */
603 static int expire(struct ofproto_dpif *);
606 static void send_netflow_active_timeouts(struct ofproto_dpif *);
609 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
611 compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions,
612 const struct flow *, uint32_t odp_port);
613 static void add_mirror_actions(struct action_xlate_ctx *ctx,
614 const struct flow *flow);
615 /* Global variables. */
616 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
618 /* Factory functions. */
621 enumerate_types(struct sset *types)
623 dp_enumerate_types(types);
627 enumerate_names(const char *type, struct sset *names)
629 return dp_enumerate_names(type, names);
633 del(const char *type, const char *name)
638 error = dpif_open(name, type, &dpif);
640 error = dpif_delete(dpif);
646 /* Basic life-cycle. */
648 static struct ofproto *
651 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
656 dealloc(struct ofproto *ofproto_)
658 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
663 construct(struct ofproto *ofproto_)
665 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
666 const char *name = ofproto->up.name;
670 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
672 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
676 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
677 ofproto->n_matches = 0;
679 dpif_flow_flush(ofproto->dpif);
680 dpif_recv_purge(ofproto->dpif);
682 error = dpif_recv_set(ofproto->dpif, true);
684 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
685 dpif_close(ofproto->dpif);
689 ofproto->netflow = NULL;
690 ofproto->sflow = NULL;
692 hmap_init(&ofproto->bundles);
693 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
694 for (i = 0; i < MAX_MIRRORS; i++) {
695 ofproto->mirrors[i] = NULL;
697 ofproto->has_bonded_bundles = false;
699 timer_set_duration(&ofproto->next_expiration, 1000);
701 hmap_init(&ofproto->facets);
702 hmap_init(&ofproto->subfacets);
704 for (i = 0; i < N_TABLES; i++) {
705 struct table_dpif *table = &ofproto->tables[i];
707 table->catchall_table = NULL;
708 table->other_table = NULL;
709 table->basis = random_uint32();
711 ofproto->need_revalidate = false;
712 tag_set_init(&ofproto->revalidate_set);
714 list_init(&ofproto->completions);
716 ofproto_dpif_unixctl_init();
718 ofproto->has_bundle_action = false;
720 hmap_init(&ofproto->vlandev_map);
721 hmap_init(&ofproto->realdev_vid_map);
723 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
724 hash_string(ofproto->up.name, 0));
725 memset(&ofproto->stats, 0, sizeof ofproto->stats);
727 ofproto_init_tables(ofproto_, N_TABLES);
733 complete_operations(struct ofproto_dpif *ofproto)
735 struct dpif_completion *c, *next;
737 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
738 ofoperation_complete(c->op, 0);
739 list_remove(&c->list_node);
745 destruct(struct ofproto *ofproto_)
747 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
748 struct rule_dpif *rule, *next_rule;
749 struct oftable *table;
752 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
753 complete_operations(ofproto);
755 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
756 struct cls_cursor cursor;
758 cls_cursor_init(&cursor, &table->cls, NULL);
759 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
760 ofproto_rule_destroy(&rule->up);
764 for (i = 0; i < MAX_MIRRORS; i++) {
765 mirror_destroy(ofproto->mirrors[i]);
768 netflow_destroy(ofproto->netflow);
769 dpif_sflow_destroy(ofproto->sflow);
770 hmap_destroy(&ofproto->bundles);
771 mac_learning_destroy(ofproto->ml);
773 hmap_destroy(&ofproto->facets);
774 hmap_destroy(&ofproto->subfacets);
776 hmap_destroy(&ofproto->vlandev_map);
777 hmap_destroy(&ofproto->realdev_vid_map);
779 dpif_close(ofproto->dpif);
783 run_fast(struct ofproto *ofproto_)
785 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
788 /* Handle one or more batches of upcalls, until there's nothing left to do
789 * or until we do a fixed total amount of work.
791 * We do work in batches because it can be much cheaper to set up a number
792 * of flows and fire off their patches all at once. We do multiple batches
793 * because in some cases handling a packet can cause another packet to be
794 * queued almost immediately as part of the return flow. Both
795 * optimizations can make major improvements on some benchmarks and
796 * presumably for real traffic as well. */
798 while (work < FLOW_MISS_MAX_BATCH) {
799 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
809 run(struct ofproto *ofproto_)
811 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
812 struct ofport_dpif *ofport;
813 struct ofbundle *bundle;
817 complete_operations(ofproto);
819 dpif_run(ofproto->dpif);
821 error = run_fast(ofproto_);
826 if (timer_expired(&ofproto->next_expiration)) {
827 int delay = expire(ofproto);
828 timer_set_duration(&ofproto->next_expiration, delay);
831 if (ofproto->netflow) {
832 if (netflow_run(ofproto->netflow)) {
833 send_netflow_active_timeouts(ofproto);
836 if (ofproto->sflow) {
837 dpif_sflow_run(ofproto->sflow);
840 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
843 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
848 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
850 /* Now revalidate if there's anything to do. */
851 if (ofproto->need_revalidate
852 || !tag_set_is_empty(&ofproto->revalidate_set)) {
853 struct tag_set revalidate_set = ofproto->revalidate_set;
854 bool revalidate_all = ofproto->need_revalidate;
855 struct facet *facet, *next;
857 /* Clear the revalidation flags. */
858 tag_set_init(&ofproto->revalidate_set);
859 ofproto->need_revalidate = false;
861 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
863 || tag_set_intersects(&revalidate_set, facet->tags)) {
864 facet_revalidate(facet);
869 /* Check the consistency of a random facet, to aid debugging. */
870 if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) {
873 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
874 struct facet, hmap_node);
875 if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) {
876 if (!facet_check_consistency(facet)) {
877 ofproto->need_revalidate = true;
886 wait(struct ofproto *ofproto_)
888 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
889 struct ofport_dpif *ofport;
890 struct ofbundle *bundle;
892 if (!clogged && !list_is_empty(&ofproto->completions)) {
893 poll_immediate_wake();
896 dpif_wait(ofproto->dpif);
897 dpif_recv_wait(ofproto->dpif);
898 if (ofproto->sflow) {
899 dpif_sflow_wait(ofproto->sflow);
901 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
902 poll_immediate_wake();
904 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
907 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
910 if (ofproto->netflow) {
911 netflow_wait(ofproto->netflow);
913 mac_learning_wait(ofproto->ml);
915 if (ofproto->need_revalidate) {
916 /* Shouldn't happen, but if it does just go around again. */
917 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
918 poll_immediate_wake();
920 timer_wait(&ofproto->next_expiration);
925 flush(struct ofproto *ofproto_)
927 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
928 struct facet *facet, *next_facet;
930 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
931 /* Mark the facet as not installed so that facet_remove() doesn't
932 * bother trying to uninstall it. There is no point in uninstalling it
933 * individually since we are about to blow away all the facets with
934 * dpif_flow_flush(). */
935 struct subfacet *subfacet;
937 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
938 subfacet->installed = false;
939 subfacet->dp_packet_count = 0;
940 subfacet->dp_byte_count = 0;
944 dpif_flow_flush(ofproto->dpif);
948 get_features(struct ofproto *ofproto_ OVS_UNUSED,
949 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
951 *arp_match_ip = true;
952 *actions = (OFPUTIL_A_OUTPUT |
953 OFPUTIL_A_SET_VLAN_VID |
954 OFPUTIL_A_SET_VLAN_PCP |
955 OFPUTIL_A_STRIP_VLAN |
956 OFPUTIL_A_SET_DL_SRC |
957 OFPUTIL_A_SET_DL_DST |
958 OFPUTIL_A_SET_NW_SRC |
959 OFPUTIL_A_SET_NW_DST |
960 OFPUTIL_A_SET_NW_TOS |
961 OFPUTIL_A_SET_TP_SRC |
962 OFPUTIL_A_SET_TP_DST |
967 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
969 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
970 struct dpif_dp_stats s;
972 strcpy(ots->name, "classifier");
974 dpif_get_dp_stats(ofproto->dpif, &s);
975 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
976 put_32aligned_be64(&ots->matched_count,
977 htonll(s.n_hit + ofproto->n_matches));
980 static struct ofport *
983 struct ofport_dpif *port = xmalloc(sizeof *port);
988 port_dealloc(struct ofport *port_)
990 struct ofport_dpif *port = ofport_dpif_cast(port_);
995 port_construct(struct ofport *port_)
997 struct ofport_dpif *port = ofport_dpif_cast(port_);
998 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1000 ofproto->need_revalidate = true;
1001 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
1002 port->bundle = NULL;
1004 port->tag = tag_create_random();
1005 port->may_enable = true;
1006 port->stp_port = NULL;
1007 port->stp_state = STP_DISABLED;
1008 hmap_init(&port->priorities);
1009 port->realdev_ofp_port = 0;
1010 port->vlandev_vid = 0;
1011 port->carrier_seq = netdev_get_carrier_resets(port->up.netdev);
1013 if (ofproto->sflow) {
1014 dpif_sflow_add_port(ofproto->sflow, port_);
1021 port_destruct(struct ofport *port_)
1023 struct ofport_dpif *port = ofport_dpif_cast(port_);
1024 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1026 ofproto->need_revalidate = true;
1027 bundle_remove(port_);
1028 set_cfm(port_, NULL);
1029 if (ofproto->sflow) {
1030 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1033 ofport_clear_priorities(port);
1034 hmap_destroy(&port->priorities);
1038 port_modified(struct ofport *port_)
1040 struct ofport_dpif *port = ofport_dpif_cast(port_);
1042 if (port->bundle && port->bundle->bond) {
1043 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1048 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1050 struct ofport_dpif *port = ofport_dpif_cast(port_);
1051 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1052 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1054 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1055 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD)) {
1056 ofproto->need_revalidate = true;
1058 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1059 bundle_update(port->bundle);
1065 set_sflow(struct ofproto *ofproto_,
1066 const struct ofproto_sflow_options *sflow_options)
1068 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1069 struct dpif_sflow *ds = ofproto->sflow;
1071 if (sflow_options) {
1073 struct ofport_dpif *ofport;
1075 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1076 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1077 dpif_sflow_add_port(ds, &ofport->up);
1079 ofproto->need_revalidate = true;
1081 dpif_sflow_set_options(ds, sflow_options);
1084 dpif_sflow_destroy(ds);
1085 ofproto->need_revalidate = true;
1086 ofproto->sflow = NULL;
1093 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1095 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1102 struct ofproto_dpif *ofproto;
1104 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1105 ofproto->need_revalidate = true;
1106 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1109 if (cfm_configure(ofport->cfm, s)) {
1115 cfm_destroy(ofport->cfm);
1121 get_cfm_fault(const struct ofport *ofport_)
1123 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1125 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1129 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1132 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1135 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1143 get_cfm_health(const struct ofport *ofport_)
1145 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1147 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1150 /* Spanning Tree. */
1153 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1155 struct ofproto_dpif *ofproto = ofproto_;
1156 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1157 struct ofport_dpif *ofport;
1159 ofport = stp_port_get_aux(sp);
1161 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1162 ofproto->up.name, port_num);
1164 struct eth_header *eth = pkt->l2;
1166 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1167 if (eth_addr_is_zero(eth->eth_src)) {
1168 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1169 "with unknown MAC", ofproto->up.name, port_num);
1171 send_packet(ofport, pkt);
1177 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1179 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1181 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1183 /* Only revalidate flows if the configuration changed. */
1184 if (!s != !ofproto->stp) {
1185 ofproto->need_revalidate = true;
1189 if (!ofproto->stp) {
1190 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1191 send_bpdu_cb, ofproto);
1192 ofproto->stp_last_tick = time_msec();
1195 stp_set_bridge_id(ofproto->stp, s->system_id);
1196 stp_set_bridge_priority(ofproto->stp, s->priority);
1197 stp_set_hello_time(ofproto->stp, s->hello_time);
1198 stp_set_max_age(ofproto->stp, s->max_age);
1199 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1201 struct ofport *ofport;
1203 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1204 set_stp_port(ofport, NULL);
1207 stp_destroy(ofproto->stp);
1208 ofproto->stp = NULL;
1215 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1217 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1221 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1222 s->designated_root = stp_get_designated_root(ofproto->stp);
1223 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1232 update_stp_port_state(struct ofport_dpif *ofport)
1234 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1235 enum stp_state state;
1237 /* Figure out new state. */
1238 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1242 if (ofport->stp_state != state) {
1243 enum ofputil_port_state of_state;
1246 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1247 netdev_get_name(ofport->up.netdev),
1248 stp_state_name(ofport->stp_state),
1249 stp_state_name(state));
1250 if (stp_learn_in_state(ofport->stp_state)
1251 != stp_learn_in_state(state)) {
1252 /* xxx Learning action flows should also be flushed. */
1253 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1255 fwd_change = stp_forward_in_state(ofport->stp_state)
1256 != stp_forward_in_state(state);
1258 ofproto->need_revalidate = true;
1259 ofport->stp_state = state;
1260 ofport->stp_state_entered = time_msec();
1262 if (fwd_change && ofport->bundle) {
1263 bundle_update(ofport->bundle);
1266 /* Update the STP state bits in the OpenFlow port description. */
1267 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1268 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1269 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1270 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1271 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1273 ofproto_port_set_state(&ofport->up, of_state);
1277 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1278 * caller is responsible for assigning STP port numbers and ensuring
1279 * there are no duplicates. */
1281 set_stp_port(struct ofport *ofport_,
1282 const struct ofproto_port_stp_settings *s)
1284 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1285 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1286 struct stp_port *sp = ofport->stp_port;
1288 if (!s || !s->enable) {
1290 ofport->stp_port = NULL;
1291 stp_port_disable(sp);
1292 update_stp_port_state(ofport);
1295 } else if (sp && stp_port_no(sp) != s->port_num
1296 && ofport == stp_port_get_aux(sp)) {
1297 /* The port-id changed, so disable the old one if it's not
1298 * already in use by another port. */
1299 stp_port_disable(sp);
1302 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1303 stp_port_enable(sp);
1305 stp_port_set_aux(sp, ofport);
1306 stp_port_set_priority(sp, s->priority);
1307 stp_port_set_path_cost(sp, s->path_cost);
1309 update_stp_port_state(ofport);
1315 get_stp_port_status(struct ofport *ofport_,
1316 struct ofproto_port_stp_status *s)
1318 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1319 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1320 struct stp_port *sp = ofport->stp_port;
1322 if (!ofproto->stp || !sp) {
1328 s->port_id = stp_port_get_id(sp);
1329 s->state = stp_port_get_state(sp);
1330 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1331 s->role = stp_port_get_role(sp);
1332 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1338 stp_run(struct ofproto_dpif *ofproto)
1341 long long int now = time_msec();
1342 long long int elapsed = now - ofproto->stp_last_tick;
1343 struct stp_port *sp;
1346 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1347 ofproto->stp_last_tick = now;
1349 while (stp_get_changed_port(ofproto->stp, &sp)) {
1350 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1353 update_stp_port_state(ofport);
1357 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1358 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1364 stp_wait(struct ofproto_dpif *ofproto)
1367 poll_timer_wait(1000);
1371 /* Returns true if STP should process 'flow'. */
1373 stp_should_process_flow(const struct flow *flow)
1375 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1379 stp_process_packet(const struct ofport_dpif *ofport,
1380 const struct ofpbuf *packet)
1382 struct ofpbuf payload = *packet;
1383 struct eth_header *eth = payload.data;
1384 struct stp_port *sp = ofport->stp_port;
1386 /* Sink packets on ports that have STP disabled when the bridge has
1388 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1392 /* Trim off padding on payload. */
1393 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1394 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1397 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1398 stp_received_bpdu(sp, payload.data, payload.size);
1402 static struct priority_to_dscp *
1403 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1405 struct priority_to_dscp *pdscp;
1408 hash = hash_int(priority, 0);
1409 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1410 if (pdscp->priority == priority) {
1418 ofport_clear_priorities(struct ofport_dpif *ofport)
1420 struct priority_to_dscp *pdscp, *next;
1422 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1423 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1429 set_queues(struct ofport *ofport_,
1430 const struct ofproto_port_queue *qdscp_list,
1433 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1434 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1435 struct hmap new = HMAP_INITIALIZER(&new);
1438 for (i = 0; i < n_qdscp; i++) {
1439 struct priority_to_dscp *pdscp;
1443 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1444 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1449 pdscp = get_priority(ofport, priority);
1451 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1453 pdscp = xmalloc(sizeof *pdscp);
1454 pdscp->priority = priority;
1456 ofproto->need_revalidate = true;
1459 if (pdscp->dscp != dscp) {
1461 ofproto->need_revalidate = true;
1464 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1467 if (!hmap_is_empty(&ofport->priorities)) {
1468 ofport_clear_priorities(ofport);
1469 ofproto->need_revalidate = true;
1472 hmap_swap(&new, &ofport->priorities);
1480 /* Expires all MAC learning entries associated with 'bundle' and forces its
1481 * ofproto to revalidate every flow.
1483 * Normally MAC learning entries are removed only from the ofproto associated
1484 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1485 * are removed from every ofproto. When patch ports and SLB bonds are in use
1486 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1487 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1488 * with the host from which it migrated. */
1490 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1492 struct ofproto_dpif *ofproto = bundle->ofproto;
1493 struct mac_learning *ml = ofproto->ml;
1494 struct mac_entry *mac, *next_mac;
1496 ofproto->need_revalidate = true;
1497 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1498 if (mac->port.p == bundle) {
1500 struct ofproto_dpif *o;
1502 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1504 struct mac_entry *e;
1506 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1509 tag_set_add(&o->revalidate_set, e->tag);
1510 mac_learning_expire(o->ml, e);
1516 mac_learning_expire(ml, mac);
1521 static struct ofbundle *
1522 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1524 struct ofbundle *bundle;
1526 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1527 &ofproto->bundles) {
1528 if (bundle->aux == aux) {
1535 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1536 * ones that are found to 'bundles'. */
1538 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1539 void **auxes, size_t n_auxes,
1540 struct hmapx *bundles)
1544 hmapx_init(bundles);
1545 for (i = 0; i < n_auxes; i++) {
1546 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1548 hmapx_add(bundles, bundle);
1554 bundle_update(struct ofbundle *bundle)
1556 struct ofport_dpif *port;
1558 bundle->floodable = true;
1559 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1560 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1561 || !stp_forward_in_state(port->stp_state)) {
1562 bundle->floodable = false;
1569 bundle_del_port(struct ofport_dpif *port)
1571 struct ofbundle *bundle = port->bundle;
1573 bundle->ofproto->need_revalidate = true;
1575 list_remove(&port->bundle_node);
1576 port->bundle = NULL;
1579 lacp_slave_unregister(bundle->lacp, port);
1582 bond_slave_unregister(bundle->bond, port);
1585 bundle_update(bundle);
1589 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1590 struct lacp_slave_settings *lacp,
1591 uint32_t bond_stable_id)
1593 struct ofport_dpif *port;
1595 port = get_ofp_port(bundle->ofproto, ofp_port);
1600 if (port->bundle != bundle) {
1601 bundle->ofproto->need_revalidate = true;
1603 bundle_del_port(port);
1606 port->bundle = bundle;
1607 list_push_back(&bundle->ports, &port->bundle_node);
1608 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1609 || !stp_forward_in_state(port->stp_state)) {
1610 bundle->floodable = false;
1614 port->bundle->ofproto->need_revalidate = true;
1615 lacp_slave_register(bundle->lacp, port, lacp);
1618 port->bond_stable_id = bond_stable_id;
1624 bundle_destroy(struct ofbundle *bundle)
1626 struct ofproto_dpif *ofproto;
1627 struct ofport_dpif *port, *next_port;
1634 ofproto = bundle->ofproto;
1635 for (i = 0; i < MAX_MIRRORS; i++) {
1636 struct ofmirror *m = ofproto->mirrors[i];
1638 if (m->out == bundle) {
1640 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1641 || hmapx_find_and_delete(&m->dsts, bundle)) {
1642 ofproto->need_revalidate = true;
1647 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1648 bundle_del_port(port);
1651 bundle_flush_macs(bundle, true);
1652 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1654 free(bundle->trunks);
1655 lacp_destroy(bundle->lacp);
1656 bond_destroy(bundle->bond);
1661 bundle_set(struct ofproto *ofproto_, void *aux,
1662 const struct ofproto_bundle_settings *s)
1664 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1665 bool need_flush = false;
1666 struct ofport_dpif *port;
1667 struct ofbundle *bundle;
1668 unsigned long *trunks;
1674 bundle_destroy(bundle_lookup(ofproto, aux));
1678 assert(s->n_slaves == 1 || s->bond != NULL);
1679 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1681 bundle = bundle_lookup(ofproto, aux);
1683 bundle = xmalloc(sizeof *bundle);
1685 bundle->ofproto = ofproto;
1686 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1687 hash_pointer(aux, 0));
1689 bundle->name = NULL;
1691 list_init(&bundle->ports);
1692 bundle->vlan_mode = PORT_VLAN_TRUNK;
1694 bundle->trunks = NULL;
1695 bundle->use_priority_tags = s->use_priority_tags;
1696 bundle->lacp = NULL;
1697 bundle->bond = NULL;
1699 bundle->floodable = true;
1701 bundle->src_mirrors = 0;
1702 bundle->dst_mirrors = 0;
1703 bundle->mirror_out = 0;
1706 if (!bundle->name || strcmp(s->name, bundle->name)) {
1708 bundle->name = xstrdup(s->name);
1713 if (!bundle->lacp) {
1714 ofproto->need_revalidate = true;
1715 bundle->lacp = lacp_create();
1717 lacp_configure(bundle->lacp, s->lacp);
1719 lacp_destroy(bundle->lacp);
1720 bundle->lacp = NULL;
1723 /* Update set of ports. */
1725 for (i = 0; i < s->n_slaves; i++) {
1726 if (!bundle_add_port(bundle, s->slaves[i],
1727 s->lacp ? &s->lacp_slaves[i] : NULL,
1728 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1732 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1733 struct ofport_dpif *next_port;
1735 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1736 for (i = 0; i < s->n_slaves; i++) {
1737 if (s->slaves[i] == port->up.ofp_port) {
1742 bundle_del_port(port);
1746 assert(list_size(&bundle->ports) <= s->n_slaves);
1748 if (list_is_empty(&bundle->ports)) {
1749 bundle_destroy(bundle);
1753 /* Set VLAN tagging mode */
1754 if (s->vlan_mode != bundle->vlan_mode
1755 || s->use_priority_tags != bundle->use_priority_tags) {
1756 bundle->vlan_mode = s->vlan_mode;
1757 bundle->use_priority_tags = s->use_priority_tags;
1762 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1763 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1765 if (vlan != bundle->vlan) {
1766 bundle->vlan = vlan;
1770 /* Get trunked VLANs. */
1771 switch (s->vlan_mode) {
1772 case PORT_VLAN_ACCESS:
1776 case PORT_VLAN_TRUNK:
1777 trunks = (unsigned long *) s->trunks;
1780 case PORT_VLAN_NATIVE_UNTAGGED:
1781 case PORT_VLAN_NATIVE_TAGGED:
1782 if (vlan != 0 && (!s->trunks
1783 || !bitmap_is_set(s->trunks, vlan)
1784 || bitmap_is_set(s->trunks, 0))) {
1785 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1787 trunks = bitmap_clone(s->trunks, 4096);
1789 trunks = bitmap_allocate1(4096);
1791 bitmap_set1(trunks, vlan);
1792 bitmap_set0(trunks, 0);
1794 trunks = (unsigned long *) s->trunks;
1801 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1802 free(bundle->trunks);
1803 if (trunks == s->trunks) {
1804 bundle->trunks = vlan_bitmap_clone(trunks);
1806 bundle->trunks = trunks;
1811 if (trunks != s->trunks) {
1816 if (!list_is_short(&bundle->ports)) {
1817 bundle->ofproto->has_bonded_bundles = true;
1819 if (bond_reconfigure(bundle->bond, s->bond)) {
1820 ofproto->need_revalidate = true;
1823 bundle->bond = bond_create(s->bond);
1824 ofproto->need_revalidate = true;
1827 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1828 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1832 bond_destroy(bundle->bond);
1833 bundle->bond = NULL;
1836 /* If we changed something that would affect MAC learning, un-learn
1837 * everything on this port and force flow revalidation. */
1839 bundle_flush_macs(bundle, false);
1846 bundle_remove(struct ofport *port_)
1848 struct ofport_dpif *port = ofport_dpif_cast(port_);
1849 struct ofbundle *bundle = port->bundle;
1852 bundle_del_port(port);
1853 if (list_is_empty(&bundle->ports)) {
1854 bundle_destroy(bundle);
1855 } else if (list_is_short(&bundle->ports)) {
1856 bond_destroy(bundle->bond);
1857 bundle->bond = NULL;
1863 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
1865 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1866 struct ofport_dpif *port = port_;
1867 uint8_t ea[ETH_ADDR_LEN];
1870 error = netdev_get_etheraddr(port->up.netdev, ea);
1872 struct ofpbuf packet;
1875 ofpbuf_init(&packet, 0);
1876 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1878 memcpy(packet_pdu, pdu, pdu_size);
1880 send_packet(port, &packet);
1881 ofpbuf_uninit(&packet);
1883 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1884 "%s (%s)", port->bundle->name,
1885 netdev_get_name(port->up.netdev), strerror(error));
1890 bundle_send_learning_packets(struct ofbundle *bundle)
1892 struct ofproto_dpif *ofproto = bundle->ofproto;
1893 int error, n_packets, n_errors;
1894 struct mac_entry *e;
1896 error = n_packets = n_errors = 0;
1897 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1898 if (e->port.p != bundle) {
1899 struct ofpbuf *learning_packet;
1900 struct ofport_dpif *port;
1904 /* The assignment to "port" is unnecessary but makes "grep"ing for
1905 * struct ofport_dpif more effective. */
1906 learning_packet = bond_compose_learning_packet(bundle->bond,
1910 ret = send_packet(port, learning_packet);
1911 ofpbuf_delete(learning_packet);
1921 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1922 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1923 "packets, last error was: %s",
1924 bundle->name, n_errors, n_packets, strerror(error));
1926 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1927 bundle->name, n_packets);
1932 bundle_run(struct ofbundle *bundle)
1935 lacp_run(bundle->lacp, send_pdu_cb);
1938 struct ofport_dpif *port;
1940 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1941 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1944 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1945 lacp_status(bundle->lacp));
1946 if (bond_should_send_learning_packets(bundle->bond)) {
1947 bundle_send_learning_packets(bundle);
1953 bundle_wait(struct ofbundle *bundle)
1956 lacp_wait(bundle->lacp);
1959 bond_wait(bundle->bond);
1966 mirror_scan(struct ofproto_dpif *ofproto)
1970 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1971 if (!ofproto->mirrors[idx]) {
1978 static struct ofmirror *
1979 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1983 for (i = 0; i < MAX_MIRRORS; i++) {
1984 struct ofmirror *mirror = ofproto->mirrors[i];
1985 if (mirror && mirror->aux == aux) {
1993 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
1995 mirror_update_dups(struct ofproto_dpif *ofproto)
1999 for (i = 0; i < MAX_MIRRORS; i++) {
2000 struct ofmirror *m = ofproto->mirrors[i];
2003 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2007 for (i = 0; i < MAX_MIRRORS; i++) {
2008 struct ofmirror *m1 = ofproto->mirrors[i];
2015 for (j = i + 1; j < MAX_MIRRORS; j++) {
2016 struct ofmirror *m2 = ofproto->mirrors[j];
2018 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2019 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2020 m2->dup_mirrors |= m1->dup_mirrors;
2027 mirror_set(struct ofproto *ofproto_, void *aux,
2028 const struct ofproto_mirror_settings *s)
2030 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2031 mirror_mask_t mirror_bit;
2032 struct ofbundle *bundle;
2033 struct ofmirror *mirror;
2034 struct ofbundle *out;
2035 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2036 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2039 mirror = mirror_lookup(ofproto, aux);
2041 mirror_destroy(mirror);
2047 idx = mirror_scan(ofproto);
2049 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2051 ofproto->up.name, MAX_MIRRORS, s->name);
2055 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2056 mirror->ofproto = ofproto;
2059 mirror->out_vlan = -1;
2060 mirror->name = NULL;
2063 if (!mirror->name || strcmp(s->name, mirror->name)) {
2065 mirror->name = xstrdup(s->name);
2068 /* Get the new configuration. */
2069 if (s->out_bundle) {
2070 out = bundle_lookup(ofproto, s->out_bundle);
2072 mirror_destroy(mirror);
2078 out_vlan = s->out_vlan;
2080 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2081 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2083 /* If the configuration has not changed, do nothing. */
2084 if (hmapx_equals(&srcs, &mirror->srcs)
2085 && hmapx_equals(&dsts, &mirror->dsts)
2086 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2087 && mirror->out == out
2088 && mirror->out_vlan == out_vlan)
2090 hmapx_destroy(&srcs);
2091 hmapx_destroy(&dsts);
2095 hmapx_swap(&srcs, &mirror->srcs);
2096 hmapx_destroy(&srcs);
2098 hmapx_swap(&dsts, &mirror->dsts);
2099 hmapx_destroy(&dsts);
2101 free(mirror->vlans);
2102 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2105 mirror->out_vlan = out_vlan;
2107 /* Update bundles. */
2108 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2109 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2110 if (hmapx_contains(&mirror->srcs, bundle)) {
2111 bundle->src_mirrors |= mirror_bit;
2113 bundle->src_mirrors &= ~mirror_bit;
2116 if (hmapx_contains(&mirror->dsts, bundle)) {
2117 bundle->dst_mirrors |= mirror_bit;
2119 bundle->dst_mirrors &= ~mirror_bit;
2122 if (mirror->out == bundle) {
2123 bundle->mirror_out |= mirror_bit;
2125 bundle->mirror_out &= ~mirror_bit;
2129 ofproto->need_revalidate = true;
2130 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2131 mirror_update_dups(ofproto);
2137 mirror_destroy(struct ofmirror *mirror)
2139 struct ofproto_dpif *ofproto;
2140 mirror_mask_t mirror_bit;
2141 struct ofbundle *bundle;
2147 ofproto = mirror->ofproto;
2148 ofproto->need_revalidate = true;
2149 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2151 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2152 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2153 bundle->src_mirrors &= ~mirror_bit;
2154 bundle->dst_mirrors &= ~mirror_bit;
2155 bundle->mirror_out &= ~mirror_bit;
2158 hmapx_destroy(&mirror->srcs);
2159 hmapx_destroy(&mirror->dsts);
2160 free(mirror->vlans);
2162 ofproto->mirrors[mirror->idx] = NULL;
2166 mirror_update_dups(ofproto);
2170 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2171 uint64_t *packets, uint64_t *bytes)
2173 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2174 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2177 *packets = *bytes = UINT64_MAX;
2181 *packets = mirror->packet_count;
2182 *bytes = mirror->byte_count;
2188 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2190 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2191 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2192 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2198 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2200 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2201 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2202 return bundle && bundle->mirror_out != 0;
2206 forward_bpdu_changed(struct ofproto *ofproto_)
2208 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2209 /* Revalidate cached flows whenever forward_bpdu option changes. */
2210 ofproto->need_revalidate = true;
2214 set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time)
2216 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2217 mac_learning_set_idle_time(ofproto->ml, idle_time);
2222 static struct ofport_dpif *
2223 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
2225 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2226 return ofport ? ofport_dpif_cast(ofport) : NULL;
2229 static struct ofport_dpif *
2230 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
2232 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2236 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2237 struct dpif_port *dpif_port)
2239 ofproto_port->name = dpif_port->name;
2240 ofproto_port->type = dpif_port->type;
2241 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2245 port_run(struct ofport_dpif *ofport)
2247 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2248 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2249 bool enable = netdev_get_carrier(ofport->up.netdev);
2251 ofport->carrier_seq = carrier_seq;
2254 cfm_run(ofport->cfm);
2256 if (cfm_should_send_ccm(ofport->cfm)) {
2257 struct ofpbuf packet;
2259 ofpbuf_init(&packet, 0);
2260 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2261 send_packet(ofport, &packet);
2262 ofpbuf_uninit(&packet);
2265 enable = enable && !cfm_get_fault(ofport->cfm)
2266 && cfm_get_opup(ofport->cfm);
2269 if (ofport->bundle) {
2270 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2271 if (carrier_changed) {
2272 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2276 if (ofport->may_enable != enable) {
2277 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2279 if (ofproto->has_bundle_action) {
2280 ofproto->need_revalidate = true;
2284 ofport->may_enable = enable;
2288 port_wait(struct ofport_dpif *ofport)
2291 cfm_wait(ofport->cfm);
2296 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2297 struct ofproto_port *ofproto_port)
2299 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2300 struct dpif_port dpif_port;
2303 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2305 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2311 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2313 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2317 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2319 *ofp_portp = odp_port_to_ofp_port(odp_port);
2325 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2327 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2330 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2332 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2334 /* The caller is going to close ofport->up.netdev. If this is a
2335 * bonded port, then the bond is using that netdev, so remove it
2336 * from the bond. The client will need to reconfigure everything
2337 * after deleting ports, so then the slave will get re-added. */
2338 bundle_remove(&ofport->up);
2345 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2347 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2350 error = netdev_get_stats(ofport->up.netdev, stats);
2352 if (!error && ofport->odp_port == OVSP_LOCAL) {
2353 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2355 /* ofproto->stats.tx_packets represents packets that we created
2356 * internally and sent to some port (e.g. packets sent with
2357 * send_packet()). Account for them as if they had come from
2358 * OFPP_LOCAL and got forwarded. */
2360 if (stats->rx_packets != UINT64_MAX) {
2361 stats->rx_packets += ofproto->stats.tx_packets;
2364 if (stats->rx_bytes != UINT64_MAX) {
2365 stats->rx_bytes += ofproto->stats.tx_bytes;
2368 /* ofproto->stats.rx_packets represents packets that were received on
2369 * some port and we processed internally and dropped (e.g. STP).
2370 * Account fro them as if they had been forwarded to OFPP_LOCAL. */
2372 if (stats->tx_packets != UINT64_MAX) {
2373 stats->tx_packets += ofproto->stats.rx_packets;
2376 if (stats->tx_bytes != UINT64_MAX) {
2377 stats->tx_bytes += ofproto->stats.rx_bytes;
2384 /* Account packets for LOCAL port. */
2386 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2387 size_t tx_size, size_t rx_size)
2389 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2392 ofproto->stats.rx_packets++;
2393 ofproto->stats.rx_bytes += rx_size;
2396 ofproto->stats.tx_packets++;
2397 ofproto->stats.tx_bytes += tx_size;
2401 struct port_dump_state {
2402 struct dpif_port_dump dump;
2407 port_dump_start(const struct ofproto *ofproto_, void **statep)
2409 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2410 struct port_dump_state *state;
2412 *statep = state = xmalloc(sizeof *state);
2413 dpif_port_dump_start(&state->dump, ofproto->dpif);
2414 state->done = false;
2419 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2420 struct ofproto_port *port)
2422 struct port_dump_state *state = state_;
2423 struct dpif_port dpif_port;
2425 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2426 ofproto_port_from_dpif_port(port, &dpif_port);
2429 int error = dpif_port_dump_done(&state->dump);
2431 return error ? error : EOF;
2436 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2438 struct port_dump_state *state = state_;
2441 dpif_port_dump_done(&state->dump);
2448 port_poll(const struct ofproto *ofproto_, char **devnamep)
2450 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2451 return dpif_port_poll(ofproto->dpif, devnamep);
2455 port_poll_wait(const struct ofproto *ofproto_)
2457 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2458 dpif_port_poll_wait(ofproto->dpif);
2462 port_is_lacp_current(const struct ofport *ofport_)
2464 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2465 return (ofport->bundle && ofport->bundle->lacp
2466 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2470 /* Upcall handling. */
2472 /* Flow miss batching.
2474 * Some dpifs implement operations faster when you hand them off in a batch.
2475 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2476 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2477 * more packets, plus possibly installing the flow in the dpif.
2479 * So far we only batch the operations that affect flow setup time the most.
2480 * It's possible to batch more than that, but the benefit might be minimal. */
2482 struct hmap_node hmap_node;
2484 enum odp_key_fitness key_fitness;
2485 const struct nlattr *key;
2487 ovs_be16 initial_tci;
2488 struct list packets;
2491 struct flow_miss_op {
2492 struct dpif_op dpif_op;
2493 struct subfacet *subfacet; /* Subfacet */
2494 void *garbage; /* Pointer to pass to free(), NULL if none. */
2495 uint64_t stub[1024 / 8]; /* Temporary buffer. */
2498 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2499 * OpenFlow controller as necessary according to their individual
2500 * configurations. */
2502 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
2503 const struct flow *flow)
2505 struct ofputil_packet_in pin;
2507 pin.packet = packet->data;
2508 pin.packet_len = packet->size;
2509 pin.reason = OFPR_NO_MATCH;
2510 pin.controller_id = 0;
2515 pin.send_len = 0; /* not used for flow table misses */
2517 flow_get_metadata(flow, &pin.fmd);
2519 /* Registers aren't meaningful on a miss. */
2520 memset(pin.fmd.reg_masks, 0, sizeof pin.fmd.reg_masks);
2522 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
2526 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2527 const struct ofpbuf *packet)
2529 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2535 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2537 cfm_process_heartbeat(ofport->cfm, packet);
2540 } else if (ofport->bundle && ofport->bundle->lacp
2541 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2543 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2546 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2548 stp_process_packet(ofport, packet);
2555 static struct flow_miss *
2556 flow_miss_find(struct hmap *todo, const struct flow *flow, uint32_t hash)
2558 struct flow_miss *miss;
2560 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2561 if (flow_equal(&miss->flow, flow)) {
2570 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2571 struct flow_miss_op *ops, size_t *n_ops)
2573 const struct flow *flow = &miss->flow;
2574 struct subfacet *subfacet;
2575 struct ofpbuf *packet;
2576 struct facet *facet;
2579 /* The caller must ensure that miss->hmap_node.hash contains
2580 * flow_hash(miss->flow, 0). */
2581 hash = miss->hmap_node.hash;
2583 facet = facet_lookup_valid(ofproto, flow, hash);
2585 struct rule_dpif *rule;
2587 rule = rule_dpif_lookup(ofproto, flow, 0);
2589 /* Don't send a packet-in if OFPUTIL_PC_NO_PACKET_IN asserted. */
2590 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
2592 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
2593 COVERAGE_INC(ofproto_dpif_no_packet_in);
2594 /* XXX install 'drop' flow entry */
2598 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
2602 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2603 send_packet_in_miss(ofproto, packet, flow);
2609 facet = facet_create(rule, flow, hash);
2612 subfacet = subfacet_create(facet,
2613 miss->key_fitness, miss->key, miss->key_len,
2616 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2617 struct flow_miss_op *op = &ops[*n_ops];
2618 struct dpif_execute *execute = &op->dpif_op.u.execute;
2619 struct dpif_flow_stats stats;
2621 struct ofpbuf odp_actions;
2623 ofproto->n_matches++;
2625 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2627 * Extra-special case for fail-open mode.
2629 * We are in fail-open mode and the packet matched the fail-open
2630 * rule, but we are connected to a controller too. We should send
2631 * the packet up to the controller in the hope that it will try to
2632 * set up a flow and thereby allow us to exit fail-open.
2634 * See the top-level comment in fail-open.c for more information.
2636 send_packet_in_miss(ofproto, packet, flow);
2639 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
2640 if (!facet->may_install || !subfacet->actions) {
2641 subfacet_make_actions(subfacet, packet, &odp_actions);
2644 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2645 subfacet_update_stats(subfacet, &stats);
2647 if (!subfacet->actions_len) {
2648 /* No actions to execute, so skip talking to the dpif. */
2649 ofpbuf_uninit(&odp_actions);
2653 if (flow->vlan_tci != subfacet->initial_tci) {
2654 /* This packet was received on a VLAN splinter port. We added
2655 * a VLAN to the packet to make the packet resemble the flow,
2656 * but the actions were composed assuming that the packet
2657 * contained no VLAN. So, we must remove the VLAN header from
2658 * the packet before trying to execute the actions. */
2659 eth_pop_vlan(packet);
2662 /* Set up operation. */
2663 op->dpif_op.type = DPIF_OP_EXECUTE;
2664 execute->key = miss->key;
2665 execute->key_len = miss->key_len;
2666 if (facet->may_install) {
2667 execute->actions = subfacet->actions;
2668 execute->actions_len = subfacet->actions_len;
2669 ofpbuf_uninit(&odp_actions);
2672 execute->actions = odp_actions.data;
2673 execute->actions_len = odp_actions.size;
2674 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
2676 execute->packet = packet;
2681 if (facet->may_install && subfacet->key_fitness != ODP_FIT_TOO_LITTLE) {
2682 struct flow_miss_op *op = &ops[(*n_ops)++];
2683 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
2685 op->subfacet = subfacet;
2687 op->dpif_op.type = DPIF_OP_FLOW_PUT;
2688 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2689 put->key = miss->key;
2690 put->key_len = miss->key_len;
2691 put->actions = subfacet->actions;
2692 put->actions_len = subfacet->actions_len;
2697 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
2698 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
2699 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
2700 * what a flow key should contain.
2702 * This function also includes some logic to help make VLAN splinters
2703 * transparent to the rest of the upcall processing logic. In particular, if
2704 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
2705 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
2706 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
2708 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
2709 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
2710 * (This differs from the value returned in flow->vlan_tci only for packets
2711 * received on VLAN splinters.)
2713 static enum odp_key_fitness
2714 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
2715 const struct nlattr *key, size_t key_len,
2716 struct flow *flow, ovs_be16 *initial_tci,
2717 struct ofpbuf *packet)
2719 enum odp_key_fitness fitness;
2723 fitness = odp_flow_key_to_flow(key, key_len, flow);
2724 if (fitness == ODP_FIT_ERROR) {
2727 *initial_tci = flow->vlan_tci;
2729 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
2731 /* Cause the flow to be processed as if it came in on the real device
2732 * with the VLAN device's VLAN ID. */
2733 flow->in_port = realdev;
2734 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
2736 /* Make the packet resemble the flow, so that it gets sent to an
2737 * OpenFlow controller properly, so that it looks correct for
2738 * sFlow, and so that flow_extract() will get the correct vlan_tci
2739 * if it is called on 'packet'.
2741 * The allocated space inside 'packet' probably also contains
2742 * 'key', that is, both 'packet' and 'key' are probably part of a
2743 * struct dpif_upcall (see the large comment on that structure
2744 * definition), so pushing data on 'packet' is in general not a
2745 * good idea since it could overwrite 'key' or free it as a side
2746 * effect. However, it's OK in this special case because we know
2747 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
2748 * will just overwrite the 4-byte "struct nlattr", which is fine
2749 * since we don't need that header anymore. */
2750 eth_push_vlan(packet, flow->vlan_tci);
2753 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
2754 if (fitness == ODP_FIT_PERFECT) {
2755 fitness = ODP_FIT_TOO_MUCH;
2763 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
2766 struct dpif_upcall *upcall;
2767 struct flow_miss *miss;
2768 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
2769 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
2770 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
2780 /* Construct the to-do list.
2782 * This just amounts to extracting the flow from each packet and sticking
2783 * the packets that have the same flow in the same "flow_miss" structure so
2784 * that we can process them together. */
2787 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
2788 struct flow_miss *miss = &misses[n_misses];
2789 struct flow_miss *existing_miss;
2792 /* Obtain metadata and check userspace/kernel agreement on flow match,
2793 * then set 'flow''s header pointers. */
2794 miss->key_fitness = ofproto_dpif_extract_flow_key(
2795 ofproto, upcall->key, upcall->key_len,
2796 &miss->flow, &miss->initial_tci, upcall->packet);
2797 if (miss->key_fitness == ODP_FIT_ERROR) {
2800 flow_extract(upcall->packet, miss->flow.skb_priority,
2801 miss->flow.tun_id, miss->flow.in_port, &miss->flow);
2803 /* Handle 802.1ag, LACP, and STP specially. */
2804 if (process_special(ofproto, &miss->flow, upcall->packet)) {
2805 ofproto_update_local_port_stats(&ofproto->up,
2806 0, upcall->packet->size);
2807 ofproto->n_matches++;
2811 /* Add other packets to a to-do list. */
2812 hash = flow_hash(&miss->flow, 0);
2813 existing_miss = flow_miss_find(&todo, &miss->flow, hash);
2814 if (!existing_miss) {
2815 hmap_insert(&todo, &miss->hmap_node, hash);
2816 miss->key = upcall->key;
2817 miss->key_len = upcall->key_len;
2818 list_init(&miss->packets);
2822 miss = existing_miss;
2824 list_push_back(&miss->packets, &upcall->packet->list_node);
2827 /* Process each element in the to-do list, constructing the set of
2828 * operations to batch. */
2830 HMAP_FOR_EACH (miss, hmap_node, &todo) {
2831 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
2833 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
2835 /* Execute batch. */
2836 for (i = 0; i < n_ops; i++) {
2837 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
2839 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
2841 /* Free memory and update facets. */
2842 for (i = 0; i < n_ops; i++) {
2843 struct flow_miss_op *op = &flow_miss_ops[i];
2845 switch (op->dpif_op.type) {
2846 case DPIF_OP_EXECUTE:
2849 case DPIF_OP_FLOW_PUT:
2850 if (!op->dpif_op.error) {
2851 op->subfacet->installed = true;
2855 case DPIF_OP_FLOW_DEL:
2861 hmap_destroy(&todo);
2865 handle_userspace_upcall(struct ofproto_dpif *ofproto,
2866 struct dpif_upcall *upcall)
2868 struct user_action_cookie cookie;
2869 enum odp_key_fitness fitness;
2870 ovs_be16 initial_tci;
2873 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
2875 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
2876 upcall->key_len, &flow,
2877 &initial_tci, upcall->packet);
2878 if (fitness == ODP_FIT_ERROR) {
2882 if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
2883 if (ofproto->sflow) {
2884 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
2888 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
2893 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
2895 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
2896 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
2897 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
2902 assert(max_batch <= FLOW_MISS_MAX_BATCH);
2906 for (n_processed = 0; n_processed < max_batch; n_processed++) {
2907 struct dpif_upcall *upcall = &misses[n_misses];
2908 struct ofpbuf *buf = &miss_bufs[n_misses];
2911 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
2912 sizeof miss_buf_stubs[n_misses]);
2913 error = dpif_recv(ofproto->dpif, upcall, buf);
2919 switch (upcall->type) {
2920 case DPIF_UC_ACTION:
2921 handle_userspace_upcall(ofproto, upcall);
2926 /* Handle it later. */
2930 case DPIF_N_UC_TYPES:
2932 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32,
2938 handle_miss_upcalls(ofproto, misses, n_misses);
2939 for (i = 0; i < n_misses; i++) {
2940 ofpbuf_uninit(&miss_bufs[i]);
2946 /* Flow expiration. */
2948 static int subfacet_max_idle(const struct ofproto_dpif *);
2949 static void update_stats(struct ofproto_dpif *);
2950 static void rule_expire(struct rule_dpif *);
2951 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
2953 /* This function is called periodically by run(). Its job is to collect
2954 * updates for the flows that have been installed into the datapath, most
2955 * importantly when they last were used, and then use that information to
2956 * expire flows that have not been used recently.
2958 * Returns the number of milliseconds after which it should be called again. */
2960 expire(struct ofproto_dpif *ofproto)
2962 struct rule_dpif *rule, *next_rule;
2963 struct oftable *table;
2966 /* Update stats for each flow in the datapath. */
2967 update_stats(ofproto);
2969 /* Expire subfacets that have been idle too long. */
2970 dp_max_idle = subfacet_max_idle(ofproto);
2971 expire_subfacets(ofproto, dp_max_idle);
2973 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
2974 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
2975 struct cls_cursor cursor;
2977 cls_cursor_init(&cursor, &table->cls, NULL);
2978 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
2983 /* All outstanding data in existing flows has been accounted, so it's a
2984 * good time to do bond rebalancing. */
2985 if (ofproto->has_bonded_bundles) {
2986 struct ofbundle *bundle;
2988 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2990 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
2995 return MIN(dp_max_idle, 1000);
2998 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3000 * This function also pushes statistics updates to rules which each facet
3001 * resubmits into. Generally these statistics will be accurate. However, if a
3002 * facet changes the rule it resubmits into at some time in between
3003 * update_stats() runs, it is possible that statistics accrued to the
3004 * old rule will be incorrectly attributed to the new rule. This could be
3005 * avoided by calling update_stats() whenever rules are created or
3006 * deleted. However, the performance impact of making so many calls to the
3007 * datapath do not justify the benefit of having perfectly accurate statistics.
3010 update_stats(struct ofproto_dpif *p)
3012 const struct dpif_flow_stats *stats;
3013 struct dpif_flow_dump dump;
3014 const struct nlattr *key;
3017 dpif_flow_dump_start(&dump, p->dpif);
3018 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
3019 struct subfacet *subfacet;
3021 subfacet = subfacet_find(p, key, key_len);
3022 if (subfacet && subfacet->installed) {
3023 struct facet *facet = subfacet->facet;
3025 if (stats->n_packets >= subfacet->dp_packet_count) {
3026 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
3027 facet->packet_count += extra;
3029 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3032 if (stats->n_bytes >= subfacet->dp_byte_count) {
3033 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
3035 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3038 subfacet->dp_packet_count = stats->n_packets;
3039 subfacet->dp_byte_count = stats->n_bytes;
3041 facet->tcp_flags |= stats->tcp_flags;
3043 subfacet_update_time(subfacet, stats->used);
3044 if (facet->accounted_bytes < facet->byte_count) {
3046 facet_account(facet);
3047 facet->accounted_bytes = facet->byte_count;
3049 facet_push_stats(facet);
3051 if (!VLOG_DROP_WARN(&rl)) {
3055 odp_flow_key_format(key, key_len, &s);
3056 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
3060 COVERAGE_INC(facet_unexpected);
3061 /* There's a flow in the datapath that we know nothing about, or a
3062 * flow that shouldn't be installed but was anyway. Delete it. */
3063 dpif_flow_del(p->dpif, key, key_len, NULL);
3066 dpif_flow_dump_done(&dump);
3069 /* Calculates and returns the number of milliseconds of idle time after which
3070 * subfacets should expire from the datapath. When a subfacet expires, we fold
3071 * its statistics into its facet, and when a facet's last subfacet expires, we
3072 * fold its statistic into its rule. */
3074 subfacet_max_idle(const struct ofproto_dpif *ofproto)
3077 * Idle time histogram.
3079 * Most of the time a switch has a relatively small number of subfacets.
3080 * When this is the case we might as well keep statistics for all of them
3081 * in userspace and to cache them in the kernel datapath for performance as
3084 * As the number of subfacets increases, the memory required to maintain
3085 * statistics about them in userspace and in the kernel becomes
3086 * significant. However, with a large number of subfacets it is likely
3087 * that only a few of them are "heavy hitters" that consume a large amount
3088 * of bandwidth. At this point, only heavy hitters are worth caching in
3089 * the kernel and maintaining in userspaces; other subfacets we can
3092 * The technique used to compute the idle time is to build a histogram with
3093 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3094 * that is installed in the kernel gets dropped in the appropriate bucket.
3095 * After the histogram has been built, we compute the cutoff so that only
3096 * the most-recently-used 1% of subfacets (but at least
3097 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3098 * the most-recently-used bucket of subfacets is kept, so actually an
3099 * arbitrary number of subfacets can be kept in any given expiration run
3100 * (though the next run will delete most of those unless they receive
3103 * This requires a second pass through the subfacets, in addition to the
3104 * pass made by update_stats(), because the former function never looks at
3105 * uninstallable subfacets.
3107 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3108 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3109 int buckets[N_BUCKETS] = { 0 };
3110 int total, subtotal, bucket;
3111 struct subfacet *subfacet;
3115 total = hmap_count(&ofproto->subfacets);
3116 if (total <= ofproto->up.flow_eviction_threshold) {
3117 return N_BUCKETS * BUCKET_WIDTH;
3120 /* Build histogram. */
3122 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3123 long long int idle = now - subfacet->used;
3124 int bucket = (idle <= 0 ? 0
3125 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3126 : (unsigned int) idle / BUCKET_WIDTH);
3130 /* Find the first bucket whose flows should be expired. */
3131 subtotal = bucket = 0;
3133 subtotal += buckets[bucket++];
3134 } while (bucket < N_BUCKETS &&
3135 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3137 if (VLOG_IS_DBG_ENABLED()) {
3141 ds_put_cstr(&s, "keep");
3142 for (i = 0; i < N_BUCKETS; i++) {
3144 ds_put_cstr(&s, ", drop");
3147 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3150 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3154 return bucket * BUCKET_WIDTH;
3157 enum { EXPIRE_MAX_BATCH = 50 };
3160 expire_batch(struct ofproto_dpif *ofproto, struct subfacet **subfacets, int n)
3162 struct odputil_keybuf keybufs[EXPIRE_MAX_BATCH];
3163 struct dpif_op ops[EXPIRE_MAX_BATCH];
3164 struct dpif_op *opsp[EXPIRE_MAX_BATCH];
3165 struct ofpbuf keys[EXPIRE_MAX_BATCH];
3166 struct dpif_flow_stats stats[EXPIRE_MAX_BATCH];
3169 for (i = 0; i < n; i++) {
3170 ops[i].type = DPIF_OP_FLOW_DEL;
3171 subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]);
3172 ops[i].u.flow_del.key = keys[i].data;
3173 ops[i].u.flow_del.key_len = keys[i].size;
3174 ops[i].u.flow_del.stats = &stats[i];
3178 dpif_operate(ofproto->dpif, opsp, n);
3179 for (i = 0; i < n; i++) {
3180 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
3181 subfacets[i]->installed = false;
3182 subfacet_destroy(subfacets[i]);
3187 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3189 long long int cutoff = time_msec() - dp_max_idle;
3191 struct subfacet *subfacet, *next_subfacet;
3192 struct subfacet *batch[EXPIRE_MAX_BATCH];
3196 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3197 &ofproto->subfacets) {
3198 if (subfacet->used < cutoff) {
3199 if (subfacet->installed) {
3200 batch[n_batch++] = subfacet;
3201 if (n_batch >= EXPIRE_MAX_BATCH) {
3202 expire_batch(ofproto, batch, n_batch);
3206 subfacet_destroy(subfacet);
3212 expire_batch(ofproto, batch, n_batch);
3216 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3217 * then delete it entirely. */
3219 rule_expire(struct rule_dpif *rule)
3221 struct facet *facet, *next_facet;
3225 /* Has 'rule' expired? */
3227 if (rule->up.hard_timeout
3228 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3229 reason = OFPRR_HARD_TIMEOUT;
3230 } else if (rule->up.idle_timeout
3231 && now > rule->up.used + rule->up.idle_timeout * 1000) {
3232 reason = OFPRR_IDLE_TIMEOUT;
3237 COVERAGE_INC(ofproto_dpif_expired);
3239 /* Update stats. (This is a no-op if the rule expired due to an idle
3240 * timeout, because that only happens when the rule has no facets left.) */
3241 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3242 facet_remove(facet);
3245 /* Get rid of the rule. */
3246 ofproto_rule_expire(&rule->up, reason);
3251 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3253 * The caller must already have determined that no facet with an identical
3254 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3255 * the ofproto's classifier table.
3257 * 'hash' must be the return value of flow_hash(flow, 0).
3259 * The facet will initially have no subfacets. The caller should create (at
3260 * least) one subfacet with subfacet_create(). */
3261 static struct facet *
3262 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
3264 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3265 struct facet *facet;
3267 facet = xzalloc(sizeof *facet);
3268 facet->used = time_msec();
3269 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
3270 list_push_back(&rule->facets, &facet->list_node);
3272 facet->flow = *flow;
3273 list_init(&facet->subfacets);
3274 netflow_flow_init(&facet->nf_flow);
3275 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3281 facet_free(struct facet *facet)
3286 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3287 * 'packet', which arrived on 'in_port'.
3289 * Takes ownership of 'packet'. */
3291 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3292 const struct nlattr *odp_actions, size_t actions_len,
3293 struct ofpbuf *packet)
3295 struct odputil_keybuf keybuf;
3299 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3300 odp_flow_key_from_flow(&key, flow);
3302 error = dpif_execute(ofproto->dpif, key.data, key.size,
3303 odp_actions, actions_len, packet);
3305 ofpbuf_delete(packet);
3309 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3311 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3312 * rule's statistics, via subfacet_uninstall().
3314 * - Removes 'facet' from its rule and from ofproto->facets.
3317 facet_remove(struct facet *facet)
3319 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3320 struct subfacet *subfacet, *next_subfacet;
3322 assert(!list_is_empty(&facet->subfacets));
3324 /* First uninstall all of the subfacets to get final statistics. */
3325 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3326 subfacet_uninstall(subfacet);
3329 /* Flush the final stats to the rule.
3331 * This might require us to have at least one subfacet around so that we
3332 * can use its actions for accounting in facet_account(), which is why we
3333 * have uninstalled but not yet destroyed the subfacets. */
3334 facet_flush_stats(facet);
3336 /* Now we're really all done so destroy everything. */
3337 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3338 &facet->subfacets) {
3339 subfacet_destroy__(subfacet);
3341 hmap_remove(&ofproto->facets, &facet->hmap_node);
3342 list_remove(&facet->list_node);
3346 /* Feed information from 'facet' back into the learning table to keep it in
3347 * sync with what is actually flowing through the datapath. */
3349 facet_learn(struct facet *facet)
3351 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3352 struct action_xlate_ctx ctx;
3354 if (!facet->has_learn
3355 && !facet->has_normal
3356 && (!facet->has_fin_timeout
3357 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
3361 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3362 facet->flow.vlan_tci,
3363 facet->rule, facet->tcp_flags, NULL);
3364 ctx.may_learn = true;
3365 xlate_actions_for_side_effects(&ctx, facet->rule->up.actions,
3366 facet->rule->up.n_actions);
3370 facet_account(struct facet *facet)
3372 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3373 struct subfacet *subfacet;
3374 const struct nlattr *a;
3379 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3382 n_bytes = facet->byte_count - facet->accounted_bytes;
3384 /* This loop feeds byte counters to bond_account() for rebalancing to use
3385 * as a basis. We also need to track the actual VLAN on which the packet
3386 * is going to be sent to ensure that it matches the one passed to
3387 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3390 * We use the actions from an arbitrary subfacet because they should all
3391 * be equally valid for our purpose. */
3392 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3393 struct subfacet, list_node);
3394 vlan_tci = facet->flow.vlan_tci;
3395 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3396 subfacet->actions, subfacet->actions_len) {
3397 const struct ovs_action_push_vlan *vlan;
3398 struct ofport_dpif *port;
3400 switch (nl_attr_type(a)) {
3401 case OVS_ACTION_ATTR_OUTPUT:
3402 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3403 if (port && port->bundle && port->bundle->bond) {
3404 bond_account(port->bundle->bond, &facet->flow,
3405 vlan_tci_to_vid(vlan_tci), n_bytes);
3409 case OVS_ACTION_ATTR_POP_VLAN:
3410 vlan_tci = htons(0);
3413 case OVS_ACTION_ATTR_PUSH_VLAN:
3414 vlan = nl_attr_get(a);
3415 vlan_tci = vlan->vlan_tci;
3421 /* Returns true if the only action for 'facet' is to send to the controller.
3422 * (We don't report NetFlow expiration messages for such facets because they
3423 * are just part of the control logic for the network, not real traffic). */
3425 facet_is_controller_flow(struct facet *facet)
3428 && facet->rule->up.n_actions == 1
3429 && action_outputs_to_port(&facet->rule->up.actions[0],
3430 htons(OFPP_CONTROLLER)));
3433 /* Folds all of 'facet''s statistics into its rule. Also updates the
3434 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3435 * 'facet''s statistics in the datapath should have been zeroed and folded into
3436 * its packet and byte counts before this function is called. */
3438 facet_flush_stats(struct facet *facet)
3440 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3441 struct subfacet *subfacet;
3443 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3444 assert(!subfacet->dp_byte_count);
3445 assert(!subfacet->dp_packet_count);
3448 facet_push_stats(facet);
3449 if (facet->accounted_bytes < facet->byte_count) {
3450 facet_account(facet);
3451 facet->accounted_bytes = facet->byte_count;
3454 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3455 struct ofexpired expired;
3456 expired.flow = facet->flow;
3457 expired.packet_count = facet->packet_count;
3458 expired.byte_count = facet->byte_count;
3459 expired.used = facet->used;
3460 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3463 facet->rule->packet_count += facet->packet_count;
3464 facet->rule->byte_count += facet->byte_count;
3466 /* Reset counters to prevent double counting if 'facet' ever gets
3468 facet_reset_counters(facet);
3470 netflow_flow_clear(&facet->nf_flow);
3471 facet->tcp_flags = 0;
3474 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3475 * Returns it if found, otherwise a null pointer.
3477 * 'hash' must be the return value of flow_hash(flow, 0).
3479 * The returned facet might need revalidation; use facet_lookup_valid()
3480 * instead if that is important. */
3481 static struct facet *
3482 facet_find(struct ofproto_dpif *ofproto,
3483 const struct flow *flow, uint32_t hash)
3485 struct facet *facet;
3487 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
3488 if (flow_equal(flow, &facet->flow)) {
3496 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3497 * Returns it if found, otherwise a null pointer.
3499 * 'hash' must be the return value of flow_hash(flow, 0).
3501 * The returned facet is guaranteed to be valid. */
3502 static struct facet *
3503 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
3506 struct facet *facet = facet_find(ofproto, flow, hash);
3508 /* The facet we found might not be valid, since we could be in need of
3509 * revalidation. If it is not valid, don't return it. */
3511 && (ofproto->need_revalidate
3512 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))
3513 && !facet_revalidate(facet)) {
3514 COVERAGE_INC(facet_invalidated);
3522 facet_check_consistency(struct facet *facet)
3524 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3526 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3528 uint64_t odp_actions_stub[1024 / 8];
3529 struct ofpbuf odp_actions;
3531 struct rule_dpif *rule;
3532 struct subfacet *subfacet;
3533 bool may_log = false;
3536 /* Check the rule for consistency. */
3537 rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3539 if (!VLOG_DROP_WARN(&rl)) {
3540 char *s = flow_to_string(&facet->flow);
3541 VLOG_WARN("%s: facet should not exist", s);
3545 } else if (rule != facet->rule) {
3546 may_log = !VLOG_DROP_WARN(&rl);
3552 flow_format(&s, &facet->flow);
3553 ds_put_format(&s, ": facet associated with wrong rule (was "
3554 "table=%"PRIu8",", facet->rule->up.table_id);
3555 cls_rule_format(&facet->rule->up.cr, &s);
3556 ds_put_format(&s, ") (should have been table=%"PRIu8",",
3558 cls_rule_format(&rule->up.cr, &s);
3559 ds_put_char(&s, ')');
3561 VLOG_WARN("%s", ds_cstr(&s));
3568 /* Check the datapath actions for consistency. */
3569 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
3570 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3571 struct action_xlate_ctx ctx;
3572 bool actions_changed;
3573 bool should_install;
3575 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3576 subfacet->initial_tci, rule, 0, NULL);
3577 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions,
3580 should_install = (ctx.may_set_up_flow
3581 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3582 if (!should_install && !subfacet->installed) {
3583 /* The actions for uninstallable flows may vary from one packet to
3584 * the next, so don't compare the actions. */
3588 actions_changed = (subfacet->actions_len != odp_actions.size
3589 || memcmp(subfacet->actions, odp_actions.data,
3590 subfacet->actions_len));
3591 if (should_install != subfacet->installed || actions_changed) {
3593 may_log = !VLOG_DROP_WARN(&rl);
3598 struct odputil_keybuf keybuf;
3603 subfacet_get_key(subfacet, &keybuf, &key);
3604 odp_flow_key_format(key.data, key.size, &s);
3606 ds_put_cstr(&s, ": inconsistency in subfacet");
3607 if (should_install != subfacet->installed) {
3608 enum odp_key_fitness fitness = subfacet->key_fitness;
3610 ds_put_format(&s, " (should%s have been installed)",
3611 should_install ? "" : " not");
3612 ds_put_format(&s, " (may_set_up_flow=%s, fitness=%s)",
3613 ctx.may_set_up_flow ? "true" : "false",
3614 odp_key_fitness_to_string(fitness));
3616 if (actions_changed) {
3617 ds_put_cstr(&s, " (actions were: ");
3618 format_odp_actions(&s, subfacet->actions,
3619 subfacet->actions_len);
3620 ds_put_cstr(&s, ") (correct actions: ");
3621 format_odp_actions(&s, odp_actions.data, odp_actions.size);
3622 ds_put_char(&s, ')');
3624 ds_put_cstr(&s, " (actions: ");
3625 format_odp_actions(&s, subfacet->actions,
3626 subfacet->actions_len);
3627 ds_put_char(&s, ')');
3629 VLOG_WARN("%s", ds_cstr(&s));
3634 ofpbuf_uninit(&odp_actions);
3639 /* Re-searches the classifier for 'facet':
3641 * - If the rule found is different from 'facet''s current rule, moves
3642 * 'facet' to the new rule and recompiles its actions.
3644 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3645 * where it is and recompiles its actions anyway.
3647 * - If there is none, destroys 'facet'.
3649 * Returns true if 'facet' still exists, false if it has been destroyed. */
3651 facet_revalidate(struct facet *facet)
3653 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3655 struct nlattr *odp_actions;
3658 struct actions *new_actions;
3660 struct action_xlate_ctx ctx;
3661 uint64_t odp_actions_stub[1024 / 8];
3662 struct ofpbuf odp_actions;
3664 struct rule_dpif *new_rule;
3665 struct subfacet *subfacet;
3666 bool actions_changed;
3669 COVERAGE_INC(facet_revalidate);
3671 /* Determine the new rule. */
3672 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3674 /* No new rule, so delete the facet. */
3675 facet_remove(facet);
3679 /* Calculate new datapath actions.
3681 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3682 * emit a NetFlow expiration and, if so, we need to have the old state
3683 * around to properly compose it. */
3685 /* If the datapath actions changed or the installability changed,
3686 * then we need to talk to the datapath. */
3689 memset(&ctx, 0, sizeof ctx);
3690 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
3691 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3692 bool should_install;
3694 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3695 subfacet->initial_tci, new_rule, 0, NULL);
3696 xlate_actions(&ctx, new_rule->up.actions, new_rule->up.n_actions,
3698 actions_changed = (subfacet->actions_len != odp_actions.size
3699 || memcmp(subfacet->actions, odp_actions.data,
3700 subfacet->actions_len));
3702 should_install = (ctx.may_set_up_flow
3703 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3704 if (actions_changed || should_install != subfacet->installed) {
3705 if (should_install) {
3706 struct dpif_flow_stats stats;
3708 subfacet_install(subfacet,
3709 odp_actions.data, odp_actions.size, &stats);
3710 subfacet_update_stats(subfacet, &stats);
3712 subfacet_uninstall(subfacet);
3716 new_actions = xcalloc(list_size(&facet->subfacets),
3717 sizeof *new_actions);
3719 new_actions[i].odp_actions = xmemdup(odp_actions.data,
3721 new_actions[i].actions_len = odp_actions.size;
3726 ofpbuf_uninit(&odp_actions);
3729 facet_flush_stats(facet);
3732 /* Update 'facet' now that we've taken care of all the old state. */
3733 facet->tags = ctx.tags;
3734 facet->nf_flow.output_iface = ctx.nf_output_iface;
3735 facet->may_install = ctx.may_set_up_flow;
3736 facet->has_learn = ctx.has_learn;
3737 facet->has_normal = ctx.has_normal;
3738 facet->has_fin_timeout = ctx.has_fin_timeout;
3739 facet->mirrors = ctx.mirrors;
3742 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3743 if (new_actions[i].odp_actions) {
3744 free(subfacet->actions);
3745 subfacet->actions = new_actions[i].odp_actions;
3746 subfacet->actions_len = new_actions[i].actions_len;
3752 if (facet->rule != new_rule) {
3753 COVERAGE_INC(facet_changed_rule);
3754 list_remove(&facet->list_node);
3755 list_push_back(&new_rule->facets, &facet->list_node);
3756 facet->rule = new_rule;
3757 facet->used = new_rule->up.created;
3758 facet->prev_used = facet->used;
3764 /* Updates 'facet''s used time. Caller is responsible for calling
3765 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3767 facet_update_time(struct facet *facet, long long int used)
3769 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3770 if (used > facet->used) {
3772 ofproto_rule_update_used(&facet->rule->up, used);
3773 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3778 facet_reset_counters(struct facet *facet)
3780 facet->packet_count = 0;
3781 facet->byte_count = 0;
3782 facet->prev_packet_count = 0;
3783 facet->prev_byte_count = 0;
3784 facet->accounted_bytes = 0;
3788 facet_push_stats(struct facet *facet)
3790 struct dpif_flow_stats stats;
3792 assert(facet->packet_count >= facet->prev_packet_count);
3793 assert(facet->byte_count >= facet->prev_byte_count);
3794 assert(facet->used >= facet->prev_used);
3796 stats.n_packets = facet->packet_count - facet->prev_packet_count;
3797 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
3798 stats.used = facet->used;
3799 stats.tcp_flags = 0;
3801 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
3802 facet->prev_packet_count = facet->packet_count;
3803 facet->prev_byte_count = facet->byte_count;
3804 facet->prev_used = facet->used;
3806 flow_push_stats(facet->rule, &facet->flow, &stats);
3808 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
3809 facet->mirrors, stats.n_packets, stats.n_bytes);
3814 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
3816 rule->packet_count += stats->n_packets;
3817 rule->byte_count += stats->n_bytes;
3818 ofproto_rule_update_used(&rule->up, stats->used);
3821 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3822 * 'rule''s actions and mirrors. */
3824 flow_push_stats(struct rule_dpif *rule,
3825 const struct flow *flow, const struct dpif_flow_stats *stats)
3827 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3828 struct action_xlate_ctx ctx;
3830 ofproto_rule_update_used(&rule->up, stats->used);
3832 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule,
3834 ctx.resubmit_stats = stats;
3835 xlate_actions_for_side_effects(&ctx, rule->up.actions, rule->up.n_actions);
3840 static struct subfacet *
3841 subfacet_find__(struct ofproto_dpif *ofproto,
3842 const struct nlattr *key, size_t key_len, uint32_t key_hash,
3843 const struct flow *flow)
3845 struct subfacet *subfacet;
3847 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
3848 &ofproto->subfacets) {
3850 ? (subfacet->key_len == key_len
3851 && !memcmp(key, subfacet->key, key_len))
3852 : flow_equal(flow, &subfacet->facet->flow)) {
3860 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
3861 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
3862 * there is one, otherwise creates and returns a new subfacet.
3864 * If the returned subfacet is new, then subfacet->actions will be NULL, in
3865 * which case the caller must populate the actions with
3866 * subfacet_make_actions(). */
3867 static struct subfacet *
3868 subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness,
3869 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
3871 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3872 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3873 struct subfacet *subfacet;
3875 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
3877 if (subfacet->facet == facet) {
3881 /* This shouldn't happen. */
3882 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
3883 subfacet_destroy(subfacet);
3886 subfacet = (list_is_empty(&facet->subfacets)
3887 ? &facet->one_subfacet
3888 : xmalloc(sizeof *subfacet));
3889 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
3890 list_push_back(&facet->subfacets, &subfacet->list_node);
3891 subfacet->facet = facet;
3892 subfacet->key_fitness = key_fitness;
3893 if (key_fitness != ODP_FIT_PERFECT) {
3894 subfacet->key = xmemdup(key, key_len);
3895 subfacet->key_len = key_len;
3897 subfacet->key = NULL;
3898 subfacet->key_len = 0;
3900 subfacet->used = time_msec();
3901 subfacet->dp_packet_count = 0;
3902 subfacet->dp_byte_count = 0;
3903 subfacet->actions_len = 0;
3904 subfacet->actions = NULL;
3905 subfacet->installed = false;
3906 subfacet->initial_tci = initial_tci;
3911 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
3912 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
3913 static struct subfacet *
3914 subfacet_find(struct ofproto_dpif *ofproto,
3915 const struct nlattr *key, size_t key_len)
3917 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3918 enum odp_key_fitness fitness;
3921 fitness = odp_flow_key_to_flow(key, key_len, &flow);
3922 if (fitness == ODP_FIT_ERROR) {
3926 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
3929 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
3930 * its facet within 'ofproto', and frees it. */
3932 subfacet_destroy__(struct subfacet *subfacet)
3934 struct facet *facet = subfacet->facet;
3935 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3937 subfacet_uninstall(subfacet);
3938 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
3939 list_remove(&subfacet->list_node);
3940 free(subfacet->key);
3941 free(subfacet->actions);
3942 if (subfacet != &facet->one_subfacet) {
3947 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
3948 * last remaining subfacet in its facet destroys the facet too. */
3950 subfacet_destroy(struct subfacet *subfacet)
3952 struct facet *facet = subfacet->facet;
3954 if (list_is_singleton(&facet->subfacets)) {
3955 /* facet_remove() needs at least one subfacet (it will remove it). */
3956 facet_remove(facet);
3958 subfacet_destroy__(subfacet);
3962 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
3963 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
3964 * for use as temporary storage. */
3966 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
3969 if (!subfacet->key) {
3970 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
3971 odp_flow_key_from_flow(key, &subfacet->facet->flow);
3973 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
3977 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
3978 * Translates the actions into 'odp_actions', which the caller must have
3979 * initialized and is responsible for uninitializing. */
3981 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
3982 struct ofpbuf *odp_actions)
3984 struct facet *facet = subfacet->facet;
3985 struct rule_dpif *rule = facet->rule;
3986 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3988 struct action_xlate_ctx ctx;
3990 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
3992 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions, odp_actions);
3993 facet->tags = ctx.tags;
3994 facet->may_install = ctx.may_set_up_flow;
3995 facet->has_learn = ctx.has_learn;
3996 facet->has_normal = ctx.has_normal;
3997 facet->has_fin_timeout = ctx.has_fin_timeout;
3998 facet->nf_flow.output_iface = ctx.nf_output_iface;
3999 facet->mirrors = ctx.mirrors;
4001 if (subfacet->actions_len != odp_actions->size
4002 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
4003 free(subfacet->actions);
4004 subfacet->actions_len = odp_actions->size;
4005 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
4009 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4010 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4011 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4012 * since 'subfacet' was last updated.
4014 * Returns 0 if successful, otherwise a positive errno value. */
4016 subfacet_install(struct subfacet *subfacet,
4017 const struct nlattr *actions, size_t actions_len,
4018 struct dpif_flow_stats *stats)
4020 struct facet *facet = subfacet->facet;
4021 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4022 struct odputil_keybuf keybuf;
4023 enum dpif_flow_put_flags flags;
4027 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
4029 flags |= DPIF_FP_ZERO_STATS;
4032 subfacet_get_key(subfacet, &keybuf, &key);
4033 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
4034 actions, actions_len, stats);
4037 subfacet_reset_dp_stats(subfacet, stats);
4043 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4045 subfacet_uninstall(struct subfacet *subfacet)
4047 if (subfacet->installed) {
4048 struct rule_dpif *rule = subfacet->facet->rule;
4049 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4050 struct odputil_keybuf keybuf;
4051 struct dpif_flow_stats stats;
4055 subfacet_get_key(subfacet, &keybuf, &key);
4056 error = dpif_flow_del(ofproto->dpif, key.data, key.size, &stats);
4057 subfacet_reset_dp_stats(subfacet, &stats);
4059 subfacet_update_stats(subfacet, &stats);
4061 subfacet->installed = false;
4063 assert(subfacet->dp_packet_count == 0);
4064 assert(subfacet->dp_byte_count == 0);
4068 /* Resets 'subfacet''s datapath statistics counters. This should be called
4069 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4070 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4071 * was reset in the datapath. 'stats' will be modified to include only
4072 * statistics new since 'subfacet' was last updated. */
4074 subfacet_reset_dp_stats(struct subfacet *subfacet,
4075 struct dpif_flow_stats *stats)
4078 && subfacet->dp_packet_count <= stats->n_packets
4079 && subfacet->dp_byte_count <= stats->n_bytes) {
4080 stats->n_packets -= subfacet->dp_packet_count;
4081 stats->n_bytes -= subfacet->dp_byte_count;
4084 subfacet->dp_packet_count = 0;
4085 subfacet->dp_byte_count = 0;
4088 /* Updates 'subfacet''s used time. The caller is responsible for calling
4089 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
4091 subfacet_update_time(struct subfacet *subfacet, long long int used)
4093 if (used > subfacet->used) {
4094 subfacet->used = used;
4095 facet_update_time(subfacet->facet, used);
4099 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4101 * Because of the meaning of a subfacet's counters, it only makes sense to do
4102 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4103 * represents a packet that was sent by hand or if it represents statistics
4104 * that have been cleared out of the datapath. */
4106 subfacet_update_stats(struct subfacet *subfacet,
4107 const struct dpif_flow_stats *stats)
4109 if (stats->n_packets || stats->used > subfacet->used) {
4110 struct facet *facet = subfacet->facet;
4112 subfacet_update_time(subfacet, stats->used);
4113 facet->packet_count += stats->n_packets;
4114 facet->byte_count += stats->n_bytes;
4115 facet->tcp_flags |= stats->tcp_flags;
4116 facet_push_stats(facet);
4117 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
4123 static struct rule_dpif *
4124 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4127 struct cls_rule *cls_rule;
4128 struct classifier *cls;
4130 if (table_id >= N_TABLES) {
4134 cls = &ofproto->up.tables[table_id].cls;
4135 if (flow->nw_frag & FLOW_NW_FRAG_ANY
4136 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4137 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
4138 * are unavailable. */
4139 struct flow ofpc_normal_flow = *flow;
4140 ofpc_normal_flow.tp_src = htons(0);
4141 ofpc_normal_flow.tp_dst = htons(0);
4142 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
4144 cls_rule = classifier_lookup(cls, flow);
4146 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
4150 complete_operation(struct rule_dpif *rule)
4152 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4154 rule_invalidate(rule);
4156 struct dpif_completion *c = xmalloc(sizeof *c);
4157 c->op = rule->up.pending;
4158 list_push_back(&ofproto->completions, &c->list_node);
4160 ofoperation_complete(rule->up.pending, 0);
4164 static struct rule *
4167 struct rule_dpif *rule = xmalloc(sizeof *rule);
4172 rule_dealloc(struct rule *rule_)
4174 struct rule_dpif *rule = rule_dpif_cast(rule_);
4179 rule_construct(struct rule *rule_)
4181 struct rule_dpif *rule = rule_dpif_cast(rule_);
4182 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4183 struct rule_dpif *victim;
4187 error = validate_actions(rule->up.actions, rule->up.n_actions,
4188 &rule->up.cr.flow, ofproto->max_ports);
4193 rule->packet_count = 0;
4194 rule->byte_count = 0;
4196 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
4197 if (victim && !list_is_empty(&victim->facets)) {
4198 struct facet *facet;
4200 rule->facets = victim->facets;
4201 list_moved(&rule->facets);
4202 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4203 /* XXX: We're only clearing our local counters here. It's possible
4204 * that quite a few packets are unaccounted for in the datapath
4205 * statistics. These will be accounted to the new rule instead of
4206 * cleared as required. This could be fixed by clearing out the
4207 * datapath statistics for this facet, but currently it doesn't
4209 facet_reset_counters(facet);
4213 /* Must avoid list_moved() in this case. */
4214 list_init(&rule->facets);
4217 table_id = rule->up.table_id;
4218 rule->tag = (victim ? victim->tag
4220 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
4221 ofproto->tables[table_id].basis));
4223 complete_operation(rule);
4228 rule_destruct(struct rule *rule_)
4230 struct rule_dpif *rule = rule_dpif_cast(rule_);
4231 struct facet *facet, *next_facet;
4233 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4234 facet_revalidate(facet);
4237 complete_operation(rule);
4241 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4243 struct rule_dpif *rule = rule_dpif_cast(rule_);
4244 struct facet *facet;
4246 /* Start from historical data for 'rule' itself that are no longer tracked
4247 * in facets. This counts, for example, facets that have expired. */
4248 *packets = rule->packet_count;
4249 *bytes = rule->byte_count;
4251 /* Add any statistics that are tracked by facets. This includes
4252 * statistical data recently updated by ofproto_update_stats() as well as
4253 * stats for packets that were executed "by hand" via dpif_execute(). */
4254 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4255 *packets += facet->packet_count;
4256 *bytes += facet->byte_count;
4261 rule_execute(struct rule *rule_, const struct flow *flow,
4262 struct ofpbuf *packet)
4264 struct rule_dpif *rule = rule_dpif_cast(rule_);
4265 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4267 struct dpif_flow_stats stats;
4269 struct action_xlate_ctx ctx;
4270 uint64_t odp_actions_stub[1024 / 8];
4271 struct ofpbuf odp_actions;
4273 dpif_flow_stats_extract(flow, packet, &stats);
4274 rule_credit_stats(rule, &stats);
4276 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4277 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
4278 rule, stats.tcp_flags, packet);
4279 ctx.resubmit_stats = &stats;
4280 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions, &odp_actions);
4282 execute_odp_actions(ofproto, flow, odp_actions.data,
4283 odp_actions.size, packet);
4285 ofpbuf_uninit(&odp_actions);
4291 rule_modify_actions(struct rule *rule_)
4293 struct rule_dpif *rule = rule_dpif_cast(rule_);
4294 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4297 error = validate_actions(rule->up.actions, rule->up.n_actions,
4298 &rule->up.cr.flow, ofproto->max_ports);
4300 ofoperation_complete(rule->up.pending, error);
4304 complete_operation(rule);
4307 /* Sends 'packet' out 'ofport'.
4308 * May modify 'packet'.
4309 * Returns 0 if successful, otherwise a positive errno value. */
4311 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4313 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4314 struct ofpbuf key, odp_actions;
4315 struct odputil_keybuf keybuf;
4320 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
4321 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4323 if (odp_port != ofport->odp_port) {
4324 eth_pop_vlan(packet);
4325 flow.vlan_tci = htons(0);
4328 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4329 odp_flow_key_from_flow(&key, &flow);
4331 ofpbuf_init(&odp_actions, 32);
4332 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4334 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4335 error = dpif_execute(ofproto->dpif,
4337 odp_actions.data, odp_actions.size,
4339 ofpbuf_uninit(&odp_actions);
4342 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4343 ofproto->up.name, odp_port, strerror(error));
4345 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
4349 /* OpenFlow to datapath action translation. */
4351 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
4352 struct action_xlate_ctx *ctx);
4353 static void xlate_normal(struct action_xlate_ctx *);
4356 put_userspace_action(const struct ofproto_dpif *ofproto,
4357 struct ofpbuf *odp_actions,
4358 const struct flow *flow,
4359 const struct user_action_cookie *cookie)
4363 pid = dpif_port_get_pid(ofproto->dpif,
4364 ofp_port_to_odp_port(flow->in_port));
4366 return odp_put_userspace_action(pid, cookie, odp_actions);
4369 /* Compose SAMPLE action for sFlow. */
4371 compose_sflow_action(const struct ofproto_dpif *ofproto,
4372 struct ofpbuf *odp_actions,
4373 const struct flow *flow,
4376 uint32_t port_ifindex;
4377 uint32_t probability;
4378 struct user_action_cookie cookie;
4379 size_t sample_offset, actions_offset;
4380 int cookie_offset, n_output;
4382 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4386 if (odp_port == OVSP_NONE) {
4390 port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4394 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4396 /* Number of packets out of UINT_MAX to sample. */
4397 probability = dpif_sflow_get_probability(ofproto->sflow);
4398 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4400 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4402 cookie.type = USER_ACTION_COOKIE_SFLOW;
4403 cookie.data = port_ifindex;
4404 cookie.n_output = n_output;
4405 cookie.vlan_tci = 0;
4406 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4408 nl_msg_end_nested(odp_actions, actions_offset);
4409 nl_msg_end_nested(odp_actions, sample_offset);
4410 return cookie_offset;
4413 /* SAMPLE action must be first action in any given list of actions.
4414 * At this point we do not have all information required to build it. So try to
4415 * build sample action as complete as possible. */
4417 add_sflow_action(struct action_xlate_ctx *ctx)
4419 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4421 &ctx->flow, OVSP_NONE);
4422 ctx->sflow_odp_port = 0;
4423 ctx->sflow_n_outputs = 0;
4426 /* Fix SAMPLE action according to data collected while composing ODP actions.
4427 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4428 * USERSPACE action's user-cookie which is required for sflow. */
4430 fix_sflow_action(struct action_xlate_ctx *ctx)
4432 const struct flow *base = &ctx->base_flow;
4433 struct user_action_cookie *cookie;
4435 if (!ctx->user_cookie_offset) {
4439 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4441 assert(cookie != NULL);
4442 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4444 if (ctx->sflow_n_outputs) {
4445 cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow,
4446 ctx->sflow_odp_port);
4448 if (ctx->sflow_n_outputs >= 255) {
4449 cookie->n_output = 255;
4451 cookie->n_output = ctx->sflow_n_outputs;
4453 cookie->vlan_tci = base->vlan_tci;
4457 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4460 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4461 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4462 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4463 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4467 struct priority_to_dscp *pdscp;
4469 if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD
4470 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4474 pdscp = get_priority(ofport, ctx->flow.skb_priority);
4476 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4477 ctx->flow.nw_tos |= pdscp->dscp;
4480 /* We may not have an ofport record for this port, but it doesn't hurt
4481 * to allow forwarding to it anyhow. Maybe such a port will appear
4482 * later and we're pre-populating the flow table. */
4485 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4486 ctx->flow.vlan_tci);
4487 if (out_port != odp_port) {
4488 ctx->flow.vlan_tci = htons(0);
4490 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4491 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4493 ctx->sflow_odp_port = odp_port;
4494 ctx->sflow_n_outputs++;
4495 ctx->nf_output_iface = ofp_port;
4496 ctx->flow.vlan_tci = flow_vlan_tci;
4497 ctx->flow.nw_tos = flow_nw_tos;
4501 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4503 compose_output_action__(ctx, ofp_port, true);
4507 xlate_table_action(struct action_xlate_ctx *ctx,
4508 uint16_t in_port, uint8_t table_id)
4510 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4511 struct ofproto_dpif *ofproto = ctx->ofproto;
4512 struct rule_dpif *rule;
4513 uint16_t old_in_port;
4514 uint8_t old_table_id;
4516 old_table_id = ctx->table_id;
4517 ctx->table_id = table_id;
4519 /* Look up a flow with 'in_port' as the input port. */
4520 old_in_port = ctx->flow.in_port;
4521 ctx->flow.in_port = in_port;
4522 rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
4525 if (table_id > 0 && table_id < N_TABLES) {
4526 struct table_dpif *table = &ofproto->tables[table_id];
4527 if (table->other_table) {
4528 ctx->tags |= (rule && rule->tag
4530 : rule_calculate_tag(&ctx->flow,
4531 &table->other_table->wc,
4536 /* Restore the original input port. Otherwise OFPP_NORMAL and
4537 * OFPP_IN_PORT will have surprising behavior. */
4538 ctx->flow.in_port = old_in_port;
4540 if (ctx->resubmit_hook) {
4541 ctx->resubmit_hook(ctx, rule);
4545 struct rule_dpif *old_rule = ctx->rule;
4547 if (ctx->resubmit_stats) {
4548 rule_credit_stats(rule, ctx->resubmit_stats);
4553 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
4554 ctx->rule = old_rule;
4558 ctx->table_id = old_table_id;
4560 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4562 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
4563 MAX_RESUBMIT_RECURSION);
4564 ctx->max_resubmit_trigger = true;
4569 xlate_resubmit_table(struct action_xlate_ctx *ctx,
4570 const struct nx_action_resubmit *nar)
4575 in_port = (nar->in_port == htons(OFPP_IN_PORT)
4577 : ntohs(nar->in_port));
4578 table_id = nar->table == 255 ? ctx->table_id : nar->table;
4580 xlate_table_action(ctx, in_port, table_id);
4584 flood_packets(struct action_xlate_ctx *ctx, bool all)
4586 struct ofport_dpif *ofport;
4588 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
4589 uint16_t ofp_port = ofport->up.ofp_port;
4591 if (ofp_port == ctx->flow.in_port) {
4596 compose_output_action__(ctx, ofp_port, false);
4597 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
4598 compose_output_action(ctx, ofp_port);
4602 ctx->nf_output_iface = NF_OUT_FLOOD;
4606 execute_controller_action(struct action_xlate_ctx *ctx, int len,
4607 enum ofp_packet_in_reason reason,
4608 uint16_t controller_id)
4610 struct ofputil_packet_in pin;
4611 struct ofpbuf *packet;
4613 ctx->may_set_up_flow = false;
4618 packet = ofpbuf_clone(ctx->packet);
4620 if (packet->l2 && packet->l3) {
4621 struct eth_header *eh;
4623 eth_pop_vlan(packet);
4626 /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
4627 * LLC frame. Calculating the Ethernet type of these frames is more
4628 * trouble than seems appropriate for a simple assertion. */
4629 assert(ntohs(eh->eth_type) < ETH_TYPE_MIN
4630 || eh->eth_type == ctx->flow.dl_type);
4632 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
4633 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
4635 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
4636 eth_push_vlan(packet, ctx->flow.vlan_tci);
4640 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4641 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
4642 ctx->flow.nw_tos, ctx->flow.nw_ttl);
4646 if (ctx->flow.nw_proto == IPPROTO_TCP) {
4647 packet_set_tcp_port(packet, ctx->flow.tp_src,
4649 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
4650 packet_set_udp_port(packet, ctx->flow.tp_src,
4657 pin.packet = packet->data;
4658 pin.packet_len = packet->size;
4659 pin.reason = reason;
4660 pin.controller_id = controller_id;
4661 pin.table_id = ctx->table_id;
4662 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
4665 flow_get_metadata(&ctx->flow, &pin.fmd);
4667 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
4668 ofpbuf_delete(packet);
4672 compose_dec_ttl(struct action_xlate_ctx *ctx)
4674 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
4675 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
4679 if (ctx->flow.nw_ttl > 1) {
4683 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
4685 /* Stop processing for current table. */
4691 xlate_output_action__(struct action_xlate_ctx *ctx,
4692 uint16_t port, uint16_t max_len)
4694 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
4696 ctx->nf_output_iface = NF_OUT_DROP;
4700 compose_output_action(ctx, ctx->flow.in_port);
4703 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
4709 flood_packets(ctx, false);
4712 flood_packets(ctx, true);
4714 case OFPP_CONTROLLER:
4715 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
4721 if (port != ctx->flow.in_port) {
4722 compose_output_action(ctx, port);
4727 if (prev_nf_output_iface == NF_OUT_FLOOD) {
4728 ctx->nf_output_iface = NF_OUT_FLOOD;
4729 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
4730 ctx->nf_output_iface = prev_nf_output_iface;
4731 } else if (prev_nf_output_iface != NF_OUT_DROP &&
4732 ctx->nf_output_iface != NF_OUT_FLOOD) {
4733 ctx->nf_output_iface = NF_OUT_MULTI;
4738 xlate_output_reg_action(struct action_xlate_ctx *ctx,
4739 const struct nx_action_output_reg *naor)
4741 struct mf_subfield src;
4744 nxm_decode(&src, naor->src, naor->ofs_nbits);
4745 ofp_port = mf_get_subfield(&src, &ctx->flow);
4747 if (ofp_port <= UINT16_MAX) {
4748 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
4753 xlate_output_action(struct action_xlate_ctx *ctx,
4754 const struct ofp_action_output *oao)
4756 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
4760 xlate_enqueue_action(struct action_xlate_ctx *ctx,
4761 const struct ofp_action_enqueue *oae)
4764 uint32_t flow_priority, priority;
4767 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
4770 /* Fall back to ordinary output action. */
4771 xlate_output_action__(ctx, ntohs(oae->port), 0);
4775 /* Figure out datapath output port. */
4776 ofp_port = ntohs(oae->port);
4777 if (ofp_port == OFPP_IN_PORT) {
4778 ofp_port = ctx->flow.in_port;
4779 } else if (ofp_port == ctx->flow.in_port) {
4783 /* Add datapath actions. */
4784 flow_priority = ctx->flow.skb_priority;
4785 ctx->flow.skb_priority = priority;
4786 compose_output_action(ctx, ofp_port);
4787 ctx->flow.skb_priority = flow_priority;
4789 /* Update NetFlow output port. */
4790 if (ctx->nf_output_iface == NF_OUT_DROP) {
4791 ctx->nf_output_iface = ofp_port;
4792 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
4793 ctx->nf_output_iface = NF_OUT_MULTI;
4798 xlate_set_queue_action(struct action_xlate_ctx *ctx,
4799 const struct nx_action_set_queue *nasq)
4804 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
4807 /* Couldn't translate queue to a priority, so ignore. A warning
4808 * has already been logged. */
4812 ctx->flow.skb_priority = priority;
4815 struct xlate_reg_state {
4821 xlate_autopath(struct action_xlate_ctx *ctx,
4822 const struct nx_action_autopath *naa)
4824 uint16_t ofp_port = ntohl(naa->id);
4825 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
4827 if (!port || !port->bundle) {
4828 ofp_port = OFPP_NONE;
4829 } else if (port->bundle->bond) {
4830 /* Autopath does not support VLAN hashing. */
4831 struct ofport_dpif *slave = bond_choose_output_slave(
4832 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
4834 ofp_port = slave->up.ofp_port;
4837 autopath_execute(naa, &ctx->flow, ofp_port);
4841 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
4843 struct ofproto_dpif *ofproto = ofproto_;
4844 struct ofport_dpif *port;
4854 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
4857 port = get_ofp_port(ofproto, ofp_port);
4858 return port ? port->may_enable : false;
4863 xlate_learn_action(struct action_xlate_ctx *ctx,
4864 const struct nx_action_learn *learn)
4866 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
4867 struct ofputil_flow_mod fm;
4870 learn_execute(learn, &ctx->flow, &fm);
4872 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
4873 if (error && !VLOG_DROP_WARN(&rl)) {
4874 VLOG_WARN("learning action failed to modify flow table (%s)",
4875 ofperr_get_name(error));
4881 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
4882 * means "infinite". */
4884 reduce_timeout(uint16_t max, uint16_t *timeout)
4886 if (max && (!*timeout || *timeout > max)) {
4892 xlate_fin_timeout(struct action_xlate_ctx *ctx,
4893 const struct nx_action_fin_timeout *naft)
4895 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
4896 struct rule_dpif *rule = ctx->rule;
4898 reduce_timeout(ntohs(naft->fin_idle_timeout), &rule->up.idle_timeout);
4899 reduce_timeout(ntohs(naft->fin_hard_timeout), &rule->up.hard_timeout);
4904 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
4906 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
4907 ? OFPUTIL_PC_NO_RECV_STP
4908 : OFPUTIL_PC_NO_RECV)) {
4912 /* Only drop packets here if both forwarding and learning are
4913 * disabled. If just learning is enabled, we need to have
4914 * OFPP_NORMAL and the learning action have a look at the packet
4915 * before we can drop it. */
4916 if (!stp_forward_in_state(port->stp_state)
4917 && !stp_learn_in_state(port->stp_state)) {
4925 do_xlate_actions(const union ofp_action *in, size_t n_in,
4926 struct action_xlate_ctx *ctx)
4928 const struct ofport_dpif *port;
4929 const union ofp_action *ia;
4930 bool was_evictable = true;
4933 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
4934 if (port && !may_receive(port, ctx)) {
4935 /* Drop this flow. */
4940 /* Don't let the rule we're working on get evicted underneath us. */
4941 was_evictable = ctx->rule->up.evictable;
4942 ctx->rule->up.evictable = false;
4944 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
4945 const struct ofp_action_dl_addr *oada;
4946 const struct nx_action_resubmit *nar;
4947 const struct nx_action_set_tunnel *nast;
4948 const struct nx_action_set_queue *nasq;
4949 const struct nx_action_multipath *nam;
4950 const struct nx_action_autopath *naa;
4951 const struct nx_action_bundle *nab;
4952 const struct nx_action_output_reg *naor;
4953 const struct nx_action_controller *nac;
4954 enum ofputil_action_code code;
4961 code = ofputil_decode_action_unsafe(ia);
4963 case OFPUTIL_OFPAT10_OUTPUT:
4964 xlate_output_action(ctx, &ia->output);
4967 case OFPUTIL_OFPAT10_SET_VLAN_VID:
4968 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
4969 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
4972 case OFPUTIL_OFPAT10_SET_VLAN_PCP:
4973 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
4974 ctx->flow.vlan_tci |= htons(
4975 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
4978 case OFPUTIL_OFPAT10_STRIP_VLAN:
4979 ctx->flow.vlan_tci = htons(0);
4982 case OFPUTIL_OFPAT10_SET_DL_SRC:
4983 oada = ((struct ofp_action_dl_addr *) ia);
4984 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
4987 case OFPUTIL_OFPAT10_SET_DL_DST:
4988 oada = ((struct ofp_action_dl_addr *) ia);
4989 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
4992 case OFPUTIL_OFPAT10_SET_NW_SRC:
4993 ctx->flow.nw_src = ia->nw_addr.nw_addr;
4996 case OFPUTIL_OFPAT10_SET_NW_DST:
4997 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
5000 case OFPUTIL_OFPAT10_SET_NW_TOS:
5001 /* OpenFlow 1.0 only supports IPv4. */
5002 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5003 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5004 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
5008 case OFPUTIL_OFPAT10_SET_TP_SRC:
5009 ctx->flow.tp_src = ia->tp_port.tp_port;
5012 case OFPUTIL_OFPAT10_SET_TP_DST:
5013 ctx->flow.tp_dst = ia->tp_port.tp_port;
5016 case OFPUTIL_OFPAT10_ENQUEUE:
5017 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
5020 case OFPUTIL_NXAST_RESUBMIT:
5021 nar = (const struct nx_action_resubmit *) ia;
5022 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
5025 case OFPUTIL_NXAST_RESUBMIT_TABLE:
5026 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
5029 case OFPUTIL_NXAST_SET_TUNNEL:
5030 nast = (const struct nx_action_set_tunnel *) ia;
5031 tun_id = htonll(ntohl(nast->tun_id));
5032 ctx->flow.tun_id = tun_id;
5035 case OFPUTIL_NXAST_SET_QUEUE:
5036 nasq = (const struct nx_action_set_queue *) ia;
5037 xlate_set_queue_action(ctx, nasq);
5040 case OFPUTIL_NXAST_POP_QUEUE:
5041 ctx->flow.skb_priority = ctx->orig_skb_priority;
5044 case OFPUTIL_NXAST_REG_MOVE:
5045 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
5049 case OFPUTIL_NXAST_REG_LOAD:
5050 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
5054 case OFPUTIL_NXAST_NOTE:
5055 /* Nothing to do. */
5058 case OFPUTIL_NXAST_SET_TUNNEL64:
5059 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
5060 ctx->flow.tun_id = tun_id;
5063 case OFPUTIL_NXAST_MULTIPATH:
5064 nam = (const struct nx_action_multipath *) ia;
5065 multipath_execute(nam, &ctx->flow);
5068 case OFPUTIL_NXAST_AUTOPATH:
5069 naa = (const struct nx_action_autopath *) ia;
5070 xlate_autopath(ctx, naa);
5073 case OFPUTIL_NXAST_BUNDLE:
5074 ctx->ofproto->has_bundle_action = true;
5075 nab = (const struct nx_action_bundle *) ia;
5076 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
5081 case OFPUTIL_NXAST_BUNDLE_LOAD:
5082 ctx->ofproto->has_bundle_action = true;
5083 nab = (const struct nx_action_bundle *) ia;
5084 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
5088 case OFPUTIL_NXAST_OUTPUT_REG:
5089 naor = (const struct nx_action_output_reg *) ia;
5090 xlate_output_reg_action(ctx, naor);
5093 case OFPUTIL_NXAST_LEARN:
5094 ctx->has_learn = true;
5095 if (ctx->may_learn) {
5096 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
5100 case OFPUTIL_NXAST_DEC_TTL:
5101 if (compose_dec_ttl(ctx)) {
5106 case OFPUTIL_NXAST_EXIT:
5110 case OFPUTIL_NXAST_FIN_TIMEOUT:
5111 ctx->has_fin_timeout = true;
5112 xlate_fin_timeout(ctx, (const struct nx_action_fin_timeout *) ia);
5115 case OFPUTIL_NXAST_CONTROLLER:
5116 nac = (const struct nx_action_controller *) ia;
5117 execute_controller_action(ctx, ntohs(nac->max_len), nac->reason,
5118 ntohs(nac->controller_id));
5124 /* We've let OFPP_NORMAL and the learning action look at the packet,
5125 * so drop it now if forwarding is disabled. */
5126 if (port && !stp_forward_in_state(port->stp_state)) {
5127 ofpbuf_clear(ctx->odp_actions);
5128 add_sflow_action(ctx);
5131 ctx->rule->up.evictable = was_evictable;
5136 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
5137 struct ofproto_dpif *ofproto, const struct flow *flow,
5138 ovs_be16 initial_tci, struct rule_dpif *rule,
5139 uint8_t tcp_flags, const struct ofpbuf *packet)
5141 ctx->ofproto = ofproto;
5143 ctx->base_flow = ctx->flow;
5144 ctx->base_flow.tun_id = 0;
5145 ctx->base_flow.vlan_tci = initial_tci;
5147 ctx->packet = packet;
5148 ctx->may_learn = packet != NULL;
5149 ctx->tcp_flags = tcp_flags;
5150 ctx->resubmit_hook = NULL;
5151 ctx->resubmit_stats = NULL;
5154 /* Translates the 'n_in' "union ofp_action"s in 'in' into datapath actions in
5155 * 'odp_actions', using 'ctx'. */
5157 xlate_actions(struct action_xlate_ctx *ctx,
5158 const union ofp_action *in, size_t n_in,
5159 struct ofpbuf *odp_actions)
5161 struct flow orig_flow = ctx->flow;
5163 COVERAGE_INC(ofproto_dpif_xlate);
5165 ofpbuf_clear(odp_actions);
5166 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
5168 ctx->odp_actions = odp_actions;
5170 ctx->may_set_up_flow = true;
5171 ctx->has_learn = false;
5172 ctx->has_normal = false;
5173 ctx->has_fin_timeout = false;
5174 ctx->nf_output_iface = NF_OUT_DROP;
5177 ctx->max_resubmit_trigger = false;
5178 ctx->orig_skb_priority = ctx->flow.skb_priority;
5182 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
5183 switch (ctx->ofproto->up.frag_handling) {
5184 case OFPC_FRAG_NORMAL:
5185 /* We must pretend that transport ports are unavailable. */
5186 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
5187 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
5190 case OFPC_FRAG_DROP:
5193 case OFPC_FRAG_REASM:
5196 case OFPC_FRAG_NX_MATCH:
5197 /* Nothing to do. */
5200 case OFPC_INVALID_TTL_TO_CONTROLLER:
5205 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
5206 ctx->may_set_up_flow = false;
5208 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5209 struct flow original_flow = ctx->flow;
5210 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
5212 add_sflow_action(ctx);
5213 do_xlate_actions(in, n_in, ctx);
5215 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook
5216 && !VLOG_DROP_ERR(&trace_rl)) {
5217 struct ds ds = DS_EMPTY_INITIALIZER;
5219 ofproto_trace(ctx->ofproto, &original_flow, ctx->packet,
5221 VLOG_ERR("Trace triggered by excessive resubmit recursion:\n%s",
5226 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
5227 ctx->odp_actions->data,
5228 ctx->odp_actions->size)) {
5229 ctx->may_set_up_flow = false;
5231 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
5233 compose_output_action(ctx, OFPP_LOCAL);
5236 add_mirror_actions(ctx, &orig_flow);
5237 fix_sflow_action(ctx);
5241 /* Translates the 'n_in' "union ofp_action"s in 'in' into datapath actions,
5242 * using 'ctx', and discards the datapath actions. */
5244 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
5245 const union ofp_action *in, size_t n_in)
5247 uint64_t odp_actions_stub[1024 / 8];
5248 struct ofpbuf odp_actions;
5250 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5251 xlate_actions(ctx, in, n_in, &odp_actions);
5252 ofpbuf_uninit(&odp_actions);
5255 /* OFPP_NORMAL implementation. */
5257 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
5259 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
5260 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
5261 * the bundle on which the packet was received, returns the VLAN to which the
5264 * Both 'vid' and the return value are in the range 0...4095. */
5266 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
5268 switch (in_bundle->vlan_mode) {
5269 case PORT_VLAN_ACCESS:
5270 return in_bundle->vlan;
5273 case PORT_VLAN_TRUNK:
5276 case PORT_VLAN_NATIVE_UNTAGGED:
5277 case PORT_VLAN_NATIVE_TAGGED:
5278 return vid ? vid : in_bundle->vlan;
5285 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
5286 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
5289 * 'vid' should be the VID obtained from the 802.1Q header that was received as
5290 * part of a packet (specify 0 if there was no 802.1Q header), in the range
5293 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
5295 /* Allow any VID on the OFPP_NONE port. */
5296 if (in_bundle == &ofpp_none_bundle) {
5300 switch (in_bundle->vlan_mode) {
5301 case PORT_VLAN_ACCESS:
5304 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5305 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
5306 "packet received on port %s configured as VLAN "
5307 "%"PRIu16" access port",
5308 in_bundle->ofproto->up.name, vid,
5309 in_bundle->name, in_bundle->vlan);
5315 case PORT_VLAN_NATIVE_UNTAGGED:
5316 case PORT_VLAN_NATIVE_TAGGED:
5318 /* Port must always carry its native VLAN. */
5322 case PORT_VLAN_TRUNK:
5323 if (!ofbundle_includes_vlan(in_bundle, vid)) {
5325 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5326 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
5327 "received on port %s not configured for trunking "
5329 in_bundle->ofproto->up.name, vid,
5330 in_bundle->name, vid);
5342 /* Given 'vlan', the VLAN that a packet belongs to, and
5343 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
5344 * that should be included in the 802.1Q header. (If the return value is 0,
5345 * then the 802.1Q header should only be included in the packet if there is a
5348 * Both 'vlan' and the return value are in the range 0...4095. */
5350 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
5352 switch (out_bundle->vlan_mode) {
5353 case PORT_VLAN_ACCESS:
5356 case PORT_VLAN_TRUNK:
5357 case PORT_VLAN_NATIVE_TAGGED:
5360 case PORT_VLAN_NATIVE_UNTAGGED:
5361 return vlan == out_bundle->vlan ? 0 : vlan;
5369 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
5372 struct ofport_dpif *port;
5374 ovs_be16 tci, old_tci;
5376 vid = output_vlan_to_vid(out_bundle, vlan);
5377 if (!out_bundle->bond) {
5378 port = ofbundle_get_a_port(out_bundle);
5380 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
5383 /* No slaves enabled, so drop packet. */
5388 old_tci = ctx->flow.vlan_tci;
5390 if (tci || out_bundle->use_priority_tags) {
5391 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
5393 tci |= htons(VLAN_CFI);
5396 ctx->flow.vlan_tci = tci;
5398 compose_output_action(ctx, port->up.ofp_port);
5399 ctx->flow.vlan_tci = old_tci;
5403 mirror_mask_ffs(mirror_mask_t mask)
5405 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
5410 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
5412 return (bundle->vlan_mode != PORT_VLAN_ACCESS
5413 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
5417 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
5419 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
5422 /* Returns an arbitrary interface within 'bundle'. */
5423 static struct ofport_dpif *
5424 ofbundle_get_a_port(const struct ofbundle *bundle)
5426 return CONTAINER_OF(list_front(&bundle->ports),
5427 struct ofport_dpif, bundle_node);
5431 vlan_is_mirrored(const struct ofmirror *m, int vlan)
5433 return !m->vlans || bitmap_is_set(m->vlans, vlan);
5436 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
5437 * to a VLAN. In general most packets may be mirrored but we want to drop
5438 * protocols that may confuse switches. */
5440 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
5442 /* If you change this function's behavior, please update corresponding
5443 * documentation in vswitch.xml at the same time. */
5444 if (dst[0] != 0x01) {
5445 /* All the currently banned MACs happen to start with 01 currently, so
5446 * this is a quick way to eliminate most of the good ones. */
5448 if (eth_addr_is_reserved(dst)) {
5449 /* Drop STP, IEEE pause frames, and other reserved protocols
5450 * (01-80-c2-00-00-0x). */
5454 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
5456 if ((dst[3] & 0xfe) == 0xcc &&
5457 (dst[4] & 0xfe) == 0xcc &&
5458 (dst[5] & 0xfe) == 0xcc) {
5459 /* Drop the following protocols plus others following the same
5462 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
5463 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
5464 STP Uplink Fast (01-00-0c-cd-cd-cd) */
5468 if (!(dst[3] | dst[4] | dst[5])) {
5469 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
5478 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5480 struct ofproto_dpif *ofproto = ctx->ofproto;
5481 mirror_mask_t mirrors;
5482 struct ofbundle *in_bundle;
5485 const struct nlattr *a;
5488 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5489 ctx->packet != NULL);
5493 mirrors = in_bundle->src_mirrors;
5495 /* Drop frames on bundles reserved for mirroring. */
5496 if (in_bundle->mirror_out) {
5497 if (ctx->packet != NULL) {
5498 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5499 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5500 "%s, which is reserved exclusively for mirroring",
5501 ctx->ofproto->up.name, in_bundle->name);
5507 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5508 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5511 vlan = input_vid_to_vlan(in_bundle, vid);
5513 /* Look at the output ports to check for destination selections. */
5515 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5516 ctx->odp_actions->size) {
5517 enum ovs_action_attr type = nl_attr_type(a);
5518 struct ofport_dpif *ofport;
5520 if (type != OVS_ACTION_ATTR_OUTPUT) {
5524 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5525 if (ofport && ofport->bundle) {
5526 mirrors |= ofport->bundle->dst_mirrors;
5534 /* Restore the original packet before adding the mirror actions. */
5535 ctx->flow = *orig_flow;
5540 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5542 if (!vlan_is_mirrored(m, vlan)) {
5543 mirrors &= mirrors - 1;
5547 mirrors &= ~m->dup_mirrors;
5548 ctx->mirrors |= m->dup_mirrors;
5550 output_normal(ctx, m->out, vlan);
5551 } else if (eth_dst_may_rspan(orig_flow->dl_dst)
5552 && vlan != m->out_vlan) {
5553 struct ofbundle *bundle;
5555 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5556 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5557 && !bundle->mirror_out) {
5558 output_normal(ctx, bundle, m->out_vlan);
5566 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5567 uint64_t packets, uint64_t bytes)
5573 for (; mirrors; mirrors &= mirrors - 1) {
5576 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5579 /* In normal circumstances 'm' will not be NULL. However,
5580 * if mirrors are reconfigured, we can temporarily get out
5581 * of sync in facet_revalidate(). We could "correct" the
5582 * mirror list before reaching here, but doing that would
5583 * not properly account the traffic stats we've currently
5584 * accumulated for previous mirror configuration. */
5588 m->packet_count += packets;
5589 m->byte_count += bytes;
5593 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
5594 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
5595 * indicate this; newer upstream kernels use gratuitous ARP requests. */
5597 is_gratuitous_arp(const struct flow *flow)
5599 return (flow->dl_type == htons(ETH_TYPE_ARP)
5600 && eth_addr_is_broadcast(flow->dl_dst)
5601 && (flow->nw_proto == ARP_OP_REPLY
5602 || (flow->nw_proto == ARP_OP_REQUEST
5603 && flow->nw_src == flow->nw_dst)));
5607 update_learning_table(struct ofproto_dpif *ofproto,
5608 const struct flow *flow, int vlan,
5609 struct ofbundle *in_bundle)
5611 struct mac_entry *mac;
5613 /* Don't learn the OFPP_NONE port. */
5614 if (in_bundle == &ofpp_none_bundle) {
5618 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
5622 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
5623 if (is_gratuitous_arp(flow)) {
5624 /* We don't want to learn from gratuitous ARP packets that are
5625 * reflected back over bond slaves so we lock the learning table. */
5626 if (!in_bundle->bond) {
5627 mac_entry_set_grat_arp_lock(mac);
5628 } else if (mac_entry_is_grat_arp_locked(mac)) {
5633 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
5634 /* The log messages here could actually be useful in debugging,
5635 * so keep the rate limit relatively high. */
5636 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5637 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
5638 "on port %s in VLAN %d",
5639 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
5640 in_bundle->name, vlan);
5642 mac->port.p = in_bundle;
5643 tag_set_add(&ofproto->revalidate_set,
5644 mac_learning_changed(ofproto->ml, mac));
5648 static struct ofbundle *
5649 lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn)
5651 struct ofport_dpif *ofport;
5653 /* Special-case OFPP_NONE, which a controller may use as the ingress
5654 * port for traffic that it is sourcing. */
5655 if (in_port == OFPP_NONE) {
5656 return &ofpp_none_bundle;
5659 /* Find the port and bundle for the received packet. */
5660 ofport = get_ofp_port(ofproto, in_port);
5661 if (ofport && ofport->bundle) {
5662 return ofport->bundle;
5665 /* Odd. A few possible reasons here:
5667 * - We deleted a port but there are still a few packets queued up
5670 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
5671 * we don't know about.
5673 * - The ofproto client didn't configure the port as part of a bundle.
5676 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5678 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
5679 "port %"PRIu16, ofproto->up.name, in_port);
5684 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
5685 * dropped. Returns true if they may be forwarded, false if they should be
5688 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
5689 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
5691 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
5692 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
5693 * checked by input_vid_is_valid().
5695 * May also add tags to '*tags', although the current implementation only does
5696 * so in one special case.
5699 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
5700 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
5702 struct ofbundle *in_bundle = in_port->bundle;
5704 /* Drop frames for reserved multicast addresses
5705 * only if forward_bpdu option is absent. */
5706 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
5710 if (in_bundle->bond) {
5711 struct mac_entry *mac;
5713 switch (bond_check_admissibility(in_bundle->bond, in_port,
5714 flow->dl_dst, tags)) {
5721 case BV_DROP_IF_MOVED:
5722 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
5723 if (mac && mac->port.p != in_bundle &&
5724 (!is_gratuitous_arp(flow)
5725 || mac_entry_is_grat_arp_locked(mac))) {
5736 xlate_normal(struct action_xlate_ctx *ctx)
5738 struct ofport_dpif *in_port;
5739 struct ofbundle *in_bundle;
5740 struct mac_entry *mac;
5744 ctx->has_normal = true;
5746 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
5747 ctx->packet != NULL);
5752 /* We know 'in_port' exists unless it is "ofpp_none_bundle",
5753 * since lookup_input_bundle() succeeded. */
5754 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5756 /* Drop malformed frames. */
5757 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
5758 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
5759 if (ctx->packet != NULL) {
5760 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5761 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
5762 "VLAN tag received on port %s",
5763 ctx->ofproto->up.name, in_bundle->name);
5768 /* Drop frames on bundles reserved for mirroring. */
5769 if (in_bundle->mirror_out) {
5770 if (ctx->packet != NULL) {
5771 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5772 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5773 "%s, which is reserved exclusively for mirroring",
5774 ctx->ofproto->up.name, in_bundle->name);
5780 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
5781 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5784 vlan = input_vid_to_vlan(in_bundle, vid);
5786 /* Check other admissibility requirements. */
5788 !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
5792 /* Learn source MAC. */
5793 if (ctx->may_learn) {
5794 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
5797 /* Determine output bundle. */
5798 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
5801 if (mac->port.p != in_bundle) {
5802 output_normal(ctx, mac->port.p, vlan);
5805 struct ofbundle *bundle;
5807 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
5808 if (bundle != in_bundle
5809 && ofbundle_includes_vlan(bundle, vlan)
5810 && bundle->floodable
5811 && !bundle->mirror_out) {
5812 output_normal(ctx, bundle, vlan);
5815 ctx->nf_output_iface = NF_OUT_FLOOD;
5819 /* Optimized flow revalidation.
5821 * It's a difficult problem, in general, to tell which facets need to have
5822 * their actions recalculated whenever the OpenFlow flow table changes. We
5823 * don't try to solve that general problem: for most kinds of OpenFlow flow
5824 * table changes, we recalculate the actions for every facet. This is
5825 * relatively expensive, but it's good enough if the OpenFlow flow table
5826 * doesn't change very often.
5828 * However, we can expect one particular kind of OpenFlow flow table change to
5829 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5830 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5831 * table, we add a special case that applies to flow tables in which every rule
5832 * has the same form (that is, the same wildcards), except that the table is
5833 * also allowed to have a single "catch-all" flow that matches all packets. We
5834 * optimize this case by tagging all of the facets that resubmit into the table
5835 * and invalidating the same tag whenever a flow changes in that table. The
5836 * end result is that we revalidate just the facets that need it (and sometimes
5837 * a few more, but not all of the facets or even all of the facets that
5838 * resubmit to the table modified by MAC learning). */
5840 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
5841 * into an OpenFlow table with the given 'basis'. */
5843 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
5846 if (flow_wildcards_is_catchall(wc)) {
5849 struct flow tag_flow = *flow;
5850 flow_zero_wildcards(&tag_flow, wc);
5851 return tag_create_deterministic(flow_hash(&tag_flow, secret));
5855 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5856 * taggability of that table.
5858 * This function must be called after *each* change to a flow table. If you
5859 * skip calling it on some changes then the pointer comparisons at the end can
5860 * be invalid if you get unlucky. For example, if a flow removal causes a
5861 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5862 * different wildcards to be created with the same address, then this function
5863 * will incorrectly skip revalidation. */
5865 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5867 struct table_dpif *table = &ofproto->tables[table_id];
5868 const struct oftable *oftable = &ofproto->up.tables[table_id];
5869 struct cls_table *catchall, *other;
5870 struct cls_table *t;
5872 catchall = other = NULL;
5874 switch (hmap_count(&oftable->cls.tables)) {
5876 /* We could tag this OpenFlow table but it would make the logic a
5877 * little harder and it's a corner case that doesn't seem worth it
5883 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
5884 if (cls_table_is_catchall(t)) {
5886 } else if (!other) {
5889 /* Indicate that we can't tag this by setting both tables to
5890 * NULL. (We know that 'catchall' is already NULL.) */
5897 /* Can't tag this table. */
5901 if (table->catchall_table != catchall || table->other_table != other) {
5902 table->catchall_table = catchall;
5903 table->other_table = other;
5904 ofproto->need_revalidate = true;
5908 /* Given 'rule' that has changed in some way (either it is a rule being
5909 * inserted, a rule being deleted, or a rule whose actions are being
5910 * modified), marks facets for revalidation to ensure that packets will be
5911 * forwarded correctly according to the new state of the flow table.
5913 * This function must be called after *each* change to a flow table. See
5914 * the comment on table_update_taggable() for more information. */
5916 rule_invalidate(const struct rule_dpif *rule)
5918 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5920 table_update_taggable(ofproto, rule->up.table_id);
5922 if (!ofproto->need_revalidate) {
5923 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5925 if (table->other_table && rule->tag) {
5926 tag_set_add(&ofproto->revalidate_set, rule->tag);
5928 ofproto->need_revalidate = true;
5934 set_frag_handling(struct ofproto *ofproto_,
5935 enum ofp_config_flags frag_handling)
5937 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5939 if (frag_handling != OFPC_FRAG_REASM) {
5940 ofproto->need_revalidate = true;
5948 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5949 const struct flow *flow,
5950 const union ofp_action *ofp_actions, size_t n_ofp_actions)
5952 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5955 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
5956 return OFPERR_NXBRC_BAD_IN_PORT;
5959 error = validate_actions(ofp_actions, n_ofp_actions, flow,
5960 ofproto->max_ports);
5962 struct odputil_keybuf keybuf;
5963 struct dpif_flow_stats stats;
5967 struct action_xlate_ctx ctx;
5968 uint64_t odp_actions_stub[1024 / 8];
5969 struct ofpbuf odp_actions;
5971 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5972 odp_flow_key_from_flow(&key, flow);
5974 dpif_flow_stats_extract(flow, packet, &stats);
5976 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL,
5977 packet_get_tcp_flags(packet, flow), packet);
5978 ctx.resubmit_stats = &stats;
5980 ofpbuf_use_stub(&odp_actions,
5981 odp_actions_stub, sizeof odp_actions_stub);
5982 xlate_actions(&ctx, ofp_actions, n_ofp_actions, &odp_actions);
5983 dpif_execute(ofproto->dpif, key.data, key.size,
5984 odp_actions.data, odp_actions.size, packet);
5985 ofpbuf_uninit(&odp_actions);
5993 set_netflow(struct ofproto *ofproto_,
5994 const struct netflow_options *netflow_options)
5996 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5998 if (netflow_options) {
5999 if (!ofproto->netflow) {
6000 ofproto->netflow = netflow_create();
6002 return netflow_set_options(ofproto->netflow, netflow_options);
6004 netflow_destroy(ofproto->netflow);
6005 ofproto->netflow = NULL;
6011 get_netflow_ids(const struct ofproto *ofproto_,
6012 uint8_t *engine_type, uint8_t *engine_id)
6014 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6016 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
6020 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
6022 if (!facet_is_controller_flow(facet) &&
6023 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
6024 struct subfacet *subfacet;
6025 struct ofexpired expired;
6027 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
6028 if (subfacet->installed) {
6029 struct dpif_flow_stats stats;
6031 subfacet_install(subfacet, subfacet->actions,
6032 subfacet->actions_len, &stats);
6033 subfacet_update_stats(subfacet, &stats);
6037 expired.flow = facet->flow;
6038 expired.packet_count = facet->packet_count;
6039 expired.byte_count = facet->byte_count;
6040 expired.used = facet->used;
6041 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
6046 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
6048 struct facet *facet;
6050 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6051 send_active_timeout(ofproto, facet);
6055 static struct ofproto_dpif *
6056 ofproto_dpif_lookup(const char *name)
6058 struct ofproto_dpif *ofproto;
6060 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
6061 hash_string(name, 0), &all_ofproto_dpifs) {
6062 if (!strcmp(ofproto->up.name, name)) {
6070 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
6071 const char *argv[], void *aux OVS_UNUSED)
6073 struct ofproto_dpif *ofproto;
6076 ofproto = ofproto_dpif_lookup(argv[1]);
6078 unixctl_command_reply_error(conn, "no such bridge");
6081 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6083 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6084 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6088 unixctl_command_reply(conn, "table successfully flushed");
6092 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6093 const char *argv[], void *aux OVS_UNUSED)
6095 struct ds ds = DS_EMPTY_INITIALIZER;
6096 const struct ofproto_dpif *ofproto;
6097 const struct mac_entry *e;
6099 ofproto = ofproto_dpif_lookup(argv[1]);
6101 unixctl_command_reply_error(conn, "no such bridge");
6105 ds_put_cstr(&ds, " port VLAN MAC Age\n");
6106 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
6107 struct ofbundle *bundle = e->port.p;
6108 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
6109 ofbundle_get_a_port(bundle)->odp_port,
6110 e->vlan, ETH_ADDR_ARGS(e->mac),
6111 mac_entry_age(ofproto->ml, e));
6113 unixctl_command_reply(conn, ds_cstr(&ds));
6118 struct action_xlate_ctx ctx;
6124 trace_format_rule(struct ds *result, uint8_t table_id, int level,
6125 const struct rule_dpif *rule)
6127 ds_put_char_multiple(result, '\t', level);
6129 ds_put_cstr(result, "No match\n");
6133 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
6134 table_id, ntohll(rule->up.flow_cookie));
6135 cls_rule_format(&rule->up.cr, result);
6136 ds_put_char(result, '\n');
6138 ds_put_char_multiple(result, '\t', level);
6139 ds_put_cstr(result, "OpenFlow ");
6140 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
6141 ds_put_char(result, '\n');
6145 trace_format_flow(struct ds *result, int level, const char *title,
6146 struct trace_ctx *trace)
6148 ds_put_char_multiple(result, '\t', level);
6149 ds_put_format(result, "%s: ", title);
6150 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
6151 ds_put_cstr(result, "unchanged");
6153 flow_format(result, &trace->ctx.flow);
6154 trace->flow = trace->ctx.flow;
6156 ds_put_char(result, '\n');
6160 trace_format_regs(struct ds *result, int level, const char *title,
6161 struct trace_ctx *trace)
6165 ds_put_char_multiple(result, '\t', level);
6166 ds_put_format(result, "%s:", title);
6167 for (i = 0; i < FLOW_N_REGS; i++) {
6168 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
6170 ds_put_char(result, '\n');
6174 trace_format_odp(struct ds *result, int level, const char *title,
6175 struct trace_ctx *trace)
6177 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
6179 ds_put_char_multiple(result, '\t', level);
6180 ds_put_format(result, "%s: ", title);
6181 format_odp_actions(result, odp_actions->data, odp_actions->size);
6182 ds_put_char(result, '\n');
6186 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
6188 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
6189 struct ds *result = trace->result;
6191 ds_put_char(result, '\n');
6192 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
6193 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
6194 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
6195 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
6199 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
6200 void *aux OVS_UNUSED)
6202 const char *dpname = argv[1];
6203 struct ofproto_dpif *ofproto;
6204 struct ofpbuf odp_key;
6205 struct ofpbuf *packet;
6206 ovs_be16 initial_tci;
6212 ofpbuf_init(&odp_key, 0);
6215 ofproto = ofproto_dpif_lookup(dpname);
6217 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
6221 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
6222 /* ofproto/trace dpname flow [-generate] */
6223 const char *flow_s = argv[2];
6224 const char *generate_s = argv[3];
6227 /* Convert string to datapath key. */
6228 ofpbuf_init(&odp_key, 0);
6229 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
6231 unixctl_command_reply_error(conn, "Bad flow syntax");
6235 /* Convert odp_key to flow. */
6236 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
6237 odp_key.size, &flow,
6238 &initial_tci, NULL);
6239 if (error == ODP_FIT_ERROR) {
6240 unixctl_command_reply_error(conn, "Invalid flow");
6244 /* Generate a packet, if requested. */
6246 packet = ofpbuf_new(0);
6247 flow_compose(packet, &flow);
6249 } else if (argc == 6) {
6250 /* ofproto/trace dpname priority tun_id in_port packet */
6251 const char *priority_s = argv[2];
6252 const char *tun_id_s = argv[3];
6253 const char *in_port_s = argv[4];
6254 const char *packet_s = argv[5];
6255 uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s));
6256 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
6257 uint32_t priority = atoi(priority_s);
6260 msg = eth_from_hex(packet_s, &packet);
6262 unixctl_command_reply_error(conn, msg);
6266 ds_put_cstr(&result, "Packet: ");
6267 s = ofp_packet_to_string(packet->data, packet->size);
6268 ds_put_cstr(&result, s);
6271 flow_extract(packet, priority, tun_id, in_port, &flow);
6272 initial_tci = flow.vlan_tci;
6274 unixctl_command_reply_error(conn, "Bad command syntax");
6278 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
6279 unixctl_command_reply(conn, ds_cstr(&result));
6282 ds_destroy(&result);
6283 ofpbuf_delete(packet);
6284 ofpbuf_uninit(&odp_key);
6288 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
6289 const struct ofpbuf *packet, ovs_be16 initial_tci,
6292 struct rule_dpif *rule;
6294 ds_put_cstr(ds, "Flow: ");
6295 flow_format(ds, flow);
6296 ds_put_char(ds, '\n');
6298 rule = rule_dpif_lookup(ofproto, flow, 0);
6299 trace_format_rule(ds, 0, 0, rule);
6301 uint64_t odp_actions_stub[1024 / 8];
6302 struct ofpbuf odp_actions;
6304 struct trace_ctx trace;
6307 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
6310 ofpbuf_use_stub(&odp_actions,
6311 odp_actions_stub, sizeof odp_actions_stub);
6312 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
6313 rule, tcp_flags, packet);
6314 trace.ctx.resubmit_hook = trace_resubmit;
6315 xlate_actions(&trace.ctx, rule->up.actions, rule->up.n_actions,
6318 ds_put_char(ds, '\n');
6319 trace_format_flow(ds, 0, "Final flow", &trace);
6320 ds_put_cstr(ds, "Datapath actions: ");
6321 format_odp_actions(ds, odp_actions.data, odp_actions.size);
6322 ofpbuf_uninit(&odp_actions);
6324 if (!trace.ctx.may_set_up_flow) {
6326 ds_put_cstr(ds, "\nThis flow is not cachable.");
6328 ds_put_cstr(ds, "\nThe datapath actions are incomplete--"
6329 "for complete actions, please supply a packet.");
6336 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6337 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6340 unixctl_command_reply(conn, NULL);
6344 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6345 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6348 unixctl_command_reply(conn, NULL);
6351 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
6352 * 'reply' describing the results. */
6354 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
6356 struct facet *facet;
6360 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6361 if (!facet_check_consistency(facet)) {
6366 ofproto->need_revalidate = true;
6370 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
6371 ofproto->up.name, errors);
6373 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
6378 ofproto_dpif_self_check(struct unixctl_conn *conn,
6379 int argc, const char *argv[], void *aux OVS_UNUSED)
6381 struct ds reply = DS_EMPTY_INITIALIZER;
6382 struct ofproto_dpif *ofproto;
6385 ofproto = ofproto_dpif_lookup(argv[1]);
6387 unixctl_command_reply_error(conn, "Unknown ofproto (use "
6388 "ofproto/list for help)");
6391 ofproto_dpif_self_check__(ofproto, &reply);
6393 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6394 ofproto_dpif_self_check__(ofproto, &reply);
6398 unixctl_command_reply(conn, ds_cstr(&reply));
6403 ofproto_dpif_unixctl_init(void)
6405 static bool registered;
6411 unixctl_command_register(
6413 "bridge {tun_id in_port packet | odp_flow [-generate]}",
6414 2, 5, ofproto_unixctl_trace, NULL);
6415 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6416 ofproto_unixctl_fdb_flush, NULL);
6417 unixctl_command_register("fdb/show", "bridge", 1, 1,
6418 ofproto_unixctl_fdb_show, NULL);
6419 unixctl_command_register("ofproto/clog", "", 0, 0,
6420 ofproto_dpif_clog, NULL);
6421 unixctl_command_register("ofproto/unclog", "", 0, 0,
6422 ofproto_dpif_unclog, NULL);
6423 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6424 ofproto_dpif_self_check, NULL);
6427 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6429 * This is deprecated. It is only for compatibility with broken device drivers
6430 * in old versions of Linux that do not properly support VLANs when VLAN
6431 * devices are not used. When broken device drivers are no longer in
6432 * widespread use, we will delete these interfaces. */
6435 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
6437 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6438 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6440 if (realdev_ofp_port == ofport->realdev_ofp_port
6441 && vid == ofport->vlandev_vid) {
6445 ofproto->need_revalidate = true;
6447 if (ofport->realdev_ofp_port) {
6450 if (realdev_ofp_port && ofport->bundle) {
6451 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6452 * themselves be part of a bundle. */
6453 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6456 ofport->realdev_ofp_port = realdev_ofp_port;
6457 ofport->vlandev_vid = vid;
6459 if (realdev_ofp_port) {
6460 vsp_add(ofport, realdev_ofp_port, vid);
6467 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
6469 return hash_2words(realdev_ofp_port, vid);
6472 /* Returns the ODP port number of the Linux VLAN device that corresponds to
6473 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
6474 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
6475 * it would return the port number of eth0.9.
6477 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
6478 * function just returns its 'realdev_odp_port' argument. */
6480 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6481 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
6483 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6484 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
6485 int vid = vlan_tci_to_vid(vlan_tci);
6486 const struct vlan_splinter *vsp;
6488 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6489 hash_realdev_vid(realdev_ofp_port, vid),
6490 &ofproto->realdev_vid_map) {
6491 if (vsp->realdev_ofp_port == realdev_ofp_port
6492 && vsp->vid == vid) {
6493 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
6497 return realdev_odp_port;
6500 static struct vlan_splinter *
6501 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
6503 struct vlan_splinter *vsp;
6505 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
6506 &ofproto->vlandev_map) {
6507 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6515 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6516 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6517 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6518 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6519 * eth0 and store 9 in '*vid'.
6521 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6522 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6525 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6526 uint16_t vlandev_ofp_port, int *vid)
6528 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6529 const struct vlan_splinter *vsp;
6531 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6536 return vsp->realdev_ofp_port;
6543 vsp_remove(struct ofport_dpif *port)
6545 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6546 struct vlan_splinter *vsp;
6548 vsp = vlandev_find(ofproto, port->up.ofp_port);
6550 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6551 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6554 port->realdev_ofp_port = 0;
6556 VLOG_ERR("missing vlan device record");
6561 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
6563 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6565 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6566 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6567 == realdev_ofp_port)) {
6568 struct vlan_splinter *vsp;
6570 vsp = xmalloc(sizeof *vsp);
6571 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6572 hash_int(port->up.ofp_port, 0));
6573 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6574 hash_realdev_vid(realdev_ofp_port, vid));
6575 vsp->realdev_ofp_port = realdev_ofp_port;
6576 vsp->vlandev_ofp_port = port->up.ofp_port;
6579 port->realdev_ofp_port = realdev_ofp_port;
6581 VLOG_ERR("duplicate vlan device record");
6585 const struct ofproto_class ofproto_dpif_class = {
6614 port_is_lacp_current,
6615 NULL, /* rule_choose_table */
6622 rule_modify_actions,
6630 get_cfm_remote_mpids,
6635 get_stp_port_status,
6642 is_mirror_output_bundle,
6643 forward_bpdu_changed,