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-governor.h"
47 #include "ofproto-dpif-sflow.h"
48 #include "poll-loop.h"
50 #include "unaligned.h"
52 #include "vlan-bitmap.h"
55 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
57 COVERAGE_DEFINE(ofproto_dpif_ctlr_action);
58 COVERAGE_DEFINE(ofproto_dpif_expired);
59 COVERAGE_DEFINE(ofproto_dpif_no_packet_in);
60 COVERAGE_DEFINE(ofproto_dpif_xlate);
61 COVERAGE_DEFINE(facet_changed_rule);
62 COVERAGE_DEFINE(facet_invalidated);
63 COVERAGE_DEFINE(facet_revalidate);
64 COVERAGE_DEFINE(facet_unexpected);
65 COVERAGE_DEFINE(facet_suppress);
67 /* Maximum depth of flow table recursion (due to resubmit actions) in a
68 * flow translation. */
69 #define MAX_RESUBMIT_RECURSION 32
71 /* Number of implemented OpenFlow tables. */
72 enum { N_TABLES = 255 };
73 BUILD_ASSERT_DECL(N_TABLES >= 1 && N_TABLES <= 255);
83 * - Do include packets and bytes from facets that have been deleted or
84 * whose own statistics have been folded into the rule.
86 * - Do include packets and bytes sent "by hand" that were accounted to
87 * the rule without any facet being involved (this is a rare corner
88 * case in rule_execute()).
90 * - Do not include packet or bytes that can be obtained from any facet's
91 * packet_count or byte_count member or that can be obtained from the
92 * datapath by, e.g., dpif_flow_get() for any subfacet.
94 uint64_t packet_count; /* Number of packets received. */
95 uint64_t byte_count; /* Number of bytes received. */
97 tag_type tag; /* Caches rule_calculate_tag() result. */
99 struct list facets; /* List of "struct facet"s. */
102 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
104 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
107 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
108 const struct flow *, uint8_t table);
110 static void rule_credit_stats(struct rule_dpif *,
111 const struct dpif_flow_stats *);
112 static void flow_push_stats(struct rule_dpif *, const struct flow *,
113 const struct dpif_flow_stats *);
114 static tag_type rule_calculate_tag(const struct flow *,
115 const struct flow_wildcards *,
117 static void rule_invalidate(const struct rule_dpif *);
119 #define MAX_MIRRORS 32
120 typedef uint32_t mirror_mask_t;
121 #define MIRROR_MASK_C(X) UINT32_C(X)
122 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
124 struct ofproto_dpif *ofproto; /* Owning ofproto. */
125 size_t idx; /* In ofproto's "mirrors" array. */
126 void *aux; /* Key supplied by ofproto's client. */
127 char *name; /* Identifier for log messages. */
129 /* Selection criteria. */
130 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
131 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
132 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
134 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
135 struct ofbundle *out; /* Output port or NULL. */
136 int out_vlan; /* Output VLAN or -1. */
137 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
140 int64_t packet_count; /* Number of packets sent. */
141 int64_t byte_count; /* Number of bytes sent. */
144 static void mirror_destroy(struct ofmirror *);
145 static void update_mirror_stats(struct ofproto_dpif *ofproto,
146 mirror_mask_t mirrors,
147 uint64_t packets, uint64_t bytes);
150 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
151 struct ofproto_dpif *ofproto; /* Owning ofproto. */
152 void *aux; /* Key supplied by ofproto's client. */
153 char *name; /* Identifier for log messages. */
156 struct list ports; /* Contains "struct ofport"s. */
157 enum port_vlan_mode vlan_mode; /* VLAN mode */
158 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
159 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
160 * NULL if all VLANs are trunked. */
161 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
162 struct bond *bond; /* Nonnull iff more than one port. */
163 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
166 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
168 /* Port mirroring info. */
169 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
170 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
171 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
174 static void bundle_remove(struct ofport *);
175 static void bundle_update(struct ofbundle *);
176 static void bundle_destroy(struct ofbundle *);
177 static void bundle_del_port(struct ofport_dpif *);
178 static void bundle_run(struct ofbundle *);
179 static void bundle_wait(struct ofbundle *);
180 static struct ofbundle *lookup_input_bundle(struct ofproto_dpif *,
181 uint16_t in_port, bool warn,
182 struct ofport_dpif **in_ofportp);
184 /* A controller may use OFPP_NONE as the ingress port to indicate that
185 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
186 * when an input bundle is needed for validation (e.g., mirroring or
187 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
188 * any 'port' structs, so care must be taken when dealing with it. */
189 static struct ofbundle ofpp_none_bundle = {
191 .vlan_mode = PORT_VLAN_TRUNK
194 static void stp_run(struct ofproto_dpif *ofproto);
195 static void stp_wait(struct ofproto_dpif *ofproto);
196 static int set_stp_port(struct ofport *,
197 const struct ofproto_port_stp_settings *);
199 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
201 struct action_xlate_ctx {
202 /* action_xlate_ctx_init() initializes these members. */
205 struct ofproto_dpif *ofproto;
207 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
208 * this flow when actions change header fields. */
211 /* The packet corresponding to 'flow', or a null pointer if we are
212 * revalidating without a packet to refer to. */
213 const struct ofpbuf *packet;
215 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
216 * actions update the flow table?
218 * We want to update these tables if we are actually processing a packet,
219 * or if we are accounting for packets that the datapath has processed, but
220 * not if we are just revalidating. */
223 /* The rule that we are currently translating, or NULL. */
224 struct rule_dpif *rule;
226 /* Union of the set of TCP flags seen so far in this flow. (Used only by
227 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
231 /* If nonnull, flow translation calls this function just before executing a
232 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
233 * when the recursion depth is exceeded.
235 * 'rule' is the rule being submitted into. It will be null if the
236 * resubmit or OFPP_TABLE action didn't find a matching rule.
238 * This is normally null so the client has to set it manually after
239 * calling action_xlate_ctx_init(). */
240 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule);
242 /* If nonnull, flow translation credits the specified statistics to each
243 * rule reached through a resubmit or OFPP_TABLE action.
245 * This is normally null so the client has to set it manually after
246 * calling action_xlate_ctx_init(). */
247 const struct dpif_flow_stats *resubmit_stats;
249 /* xlate_actions() initializes and uses these members. The client might want
250 * to look at them after it returns. */
252 struct ofpbuf *odp_actions; /* Datapath actions. */
253 tag_type tags; /* Tags associated with actions. */
254 bool may_set_up_flow; /* True ordinarily; false if the actions must
255 * be reassessed for every packet. */
256 bool has_learn; /* Actions include NXAST_LEARN? */
257 bool has_normal; /* Actions output to OFPP_NORMAL? */
258 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
259 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
260 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
262 /* xlate_actions() initializes and uses these members, but the client has no
263 * reason to look at them. */
265 int recurse; /* Recursion level, via xlate_table_action. */
266 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
267 struct flow base_flow; /* Flow at the last commit. */
268 uint32_t orig_skb_priority; /* Priority when packet arrived. */
269 uint8_t table_id; /* OpenFlow table ID where flow was found. */
270 uint32_t sflow_n_outputs; /* Number of output ports. */
271 uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
272 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
273 bool exit; /* No further actions should be processed. */
276 static void action_xlate_ctx_init(struct action_xlate_ctx *,
277 struct ofproto_dpif *, const struct flow *,
278 ovs_be16 initial_tci, struct rule_dpif *,
279 uint8_t tcp_flags, const struct ofpbuf *);
280 static void xlate_actions(struct action_xlate_ctx *,
281 const union ofp_action *in, size_t n_in,
282 struct ofpbuf *odp_actions);
283 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
284 const union ofp_action *in,
287 /* A dpif flow and actions associated with a facet.
289 * See also the large comment on struct facet. */
292 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
293 struct list list_node; /* In struct facet's 'facets' list. */
294 struct facet *facet; /* Owning facet. */
298 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
299 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
300 * regenerate the ODP flow key from ->facet->flow. */
301 enum odp_key_fitness key_fitness;
305 long long int used; /* Time last used; time created if not used. */
307 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
308 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
312 * These should be essentially identical for every subfacet in a facet, but
313 * may differ in trivial ways due to VLAN splinters. */
314 size_t actions_len; /* Number of bytes in actions[]. */
315 struct nlattr *actions; /* Datapath actions. */
317 bool installed; /* Installed in datapath? */
319 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
320 * splinters can cause it to differ. This value should be removed when
321 * the VLAN splinters feature is no longer needed. */
322 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
325 static struct subfacet *subfacet_create(struct facet *, enum odp_key_fitness,
326 const struct nlattr *key,
327 size_t key_len, ovs_be16 initial_tci);
328 static struct subfacet *subfacet_find(struct ofproto_dpif *,
329 const struct nlattr *key, size_t key_len);
330 static void subfacet_destroy(struct subfacet *);
331 static void subfacet_destroy__(struct subfacet *);
332 static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *,
334 static void subfacet_reset_dp_stats(struct subfacet *,
335 struct dpif_flow_stats *);
336 static void subfacet_update_time(struct subfacet *, long long int used);
337 static void subfacet_update_stats(struct subfacet *,
338 const struct dpif_flow_stats *);
339 static void subfacet_make_actions(struct subfacet *,
340 const struct ofpbuf *packet,
341 struct ofpbuf *odp_actions);
342 static int subfacet_install(struct subfacet *,
343 const struct nlattr *actions, size_t actions_len,
344 struct dpif_flow_stats *);
345 static void subfacet_uninstall(struct subfacet *);
347 /* An exact-match instantiation of an OpenFlow flow.
349 * A facet associates a "struct flow", which represents the Open vSwitch
350 * userspace idea of an exact-match flow, with one or more subfacets. Each
351 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
352 * the facet. When the kernel module (or other dpif implementation) and Open
353 * vSwitch userspace agree on the definition of a flow key, there is exactly
354 * one subfacet per facet. If the dpif implementation supports more-specific
355 * flow matching than userspace, however, a facet can have more than one
356 * subfacet, each of which corresponds to some distinction in flow that
357 * userspace simply doesn't understand.
359 * Flow expiration works in terms of subfacets, so a facet must have at least
360 * one subfacet or it will never expire, leaking memory. */
363 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
364 struct list list_node; /* In owning rule's 'facets' list. */
365 struct rule_dpif *rule; /* Owning rule. */
368 struct list subfacets;
369 long long int used; /* Time last used; time created if not used. */
376 * - Do include packets and bytes sent "by hand", e.g. with
379 * - Do include packets and bytes that were obtained from the datapath
380 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
381 * DPIF_FP_ZERO_STATS).
383 * - Do not include packets or bytes that can be obtained from the
384 * datapath for any existing subfacet.
386 uint64_t packet_count; /* Number of packets received. */
387 uint64_t byte_count; /* Number of bytes received. */
389 /* Resubmit statistics. */
390 uint64_t prev_packet_count; /* Number of packets from last stats push. */
391 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
392 long long int prev_used; /* Used time from last stats push. */
395 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
396 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
397 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
399 /* Properties of datapath actions.
401 * Every subfacet has its own actions because actions can differ slightly
402 * between splintered and non-splintered subfacets due to the VLAN tag
403 * being initially different (present vs. absent). All of them have these
404 * properties in common so we just store one copy of them here. */
405 bool may_install; /* Reassess actions for every packet? */
406 bool has_learn; /* Actions include NXAST_LEARN? */
407 bool has_normal; /* Actions output to OFPP_NORMAL? */
408 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
409 tag_type tags; /* Tags that would require revalidation. */
410 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
412 /* Storage for a single subfacet, to reduce malloc() time and space
413 * overhead. (A facet always has at least one subfacet and in the common
414 * case has exactly one subfacet.) */
415 struct subfacet one_subfacet;
418 static struct facet *facet_create(struct rule_dpif *,
419 const struct flow *, uint32_t hash);
420 static void facet_remove(struct facet *);
421 static void facet_free(struct facet *);
423 static struct facet *facet_find(struct ofproto_dpif *,
424 const struct flow *, uint32_t hash);
425 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
426 const struct flow *, uint32_t hash);
427 static bool facet_revalidate(struct facet *);
428 static bool facet_check_consistency(struct facet *);
430 static void facet_flush_stats(struct facet *);
432 static void facet_update_time(struct facet *, long long int used);
433 static void facet_reset_counters(struct facet *);
434 static void facet_push_stats(struct facet *);
435 static void facet_learn(struct facet *);
436 static void facet_account(struct facet *);
438 static bool facet_is_controller_flow(struct facet *);
444 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
445 struct list bundle_node; /* In struct ofbundle's "ports" list. */
446 struct cfm *cfm; /* Connectivity Fault Management, if any. */
447 tag_type tag; /* Tag associated with this port. */
448 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
449 bool may_enable; /* May be enabled in bonds. */
450 long long int carrier_seq; /* Carrier status changes. */
453 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
454 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
455 long long int stp_state_entered;
457 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
459 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
461 * This is deprecated. It is only for compatibility with broken device
462 * drivers in old versions of Linux that do not properly support VLANs when
463 * VLAN devices are not used. When broken device drivers are no longer in
464 * widespread use, we will delete these interfaces. */
465 uint16_t realdev_ofp_port;
469 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
470 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
471 * traffic egressing the 'ofport' with that priority should be marked with. */
472 struct priority_to_dscp {
473 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
474 uint32_t priority; /* Priority of this queue (see struct flow). */
476 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
479 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
481 * This is deprecated. It is only for compatibility with broken device drivers
482 * in old versions of Linux that do not properly support VLANs when VLAN
483 * devices are not used. When broken device drivers are no longer in
484 * widespread use, we will delete these interfaces. */
485 struct vlan_splinter {
486 struct hmap_node realdev_vid_node;
487 struct hmap_node vlandev_node;
488 uint16_t realdev_ofp_port;
489 uint16_t vlandev_ofp_port;
493 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
494 uint32_t realdev, ovs_be16 vlan_tci);
495 static uint16_t vsp_vlandev_to_realdev(const struct ofproto_dpif *,
496 uint16_t vlandev, int *vid);
497 static void vsp_remove(struct ofport_dpif *);
498 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
500 static struct ofport_dpif *
501 ofport_dpif_cast(const struct ofport *ofport)
503 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
504 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
507 static void port_run(struct ofport_dpif *);
508 static void port_wait(struct ofport_dpif *);
509 static int set_cfm(struct ofport *, const struct cfm_settings *);
510 static void ofport_clear_priorities(struct ofport_dpif *);
512 struct dpif_completion {
513 struct list list_node;
514 struct ofoperation *op;
517 /* Extra information about a classifier table.
518 * Currently used just for optimized flow revalidation. */
520 /* If either of these is nonnull, then this table has a form that allows
521 * flows to be tagged to avoid revalidating most flows for the most common
522 * kinds of flow table changes. */
523 struct cls_table *catchall_table; /* Table that wildcards all fields. */
524 struct cls_table *other_table; /* Table with any other wildcard set. */
525 uint32_t basis; /* Keeps each table's tags separate. */
528 struct ofproto_dpif {
529 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
538 struct netflow *netflow;
539 struct dpif_sflow *sflow;
540 struct hmap bundles; /* Contains "struct ofbundle"s. */
541 struct mac_learning *ml;
542 struct ofmirror *mirrors[MAX_MIRRORS];
543 bool has_bonded_bundles;
546 struct timer next_expiration;
550 struct hmap subfacets;
551 struct governor *governor;
554 struct table_dpif tables[N_TABLES];
555 bool need_revalidate;
556 struct tag_set revalidate_set;
558 /* Support for debugging async flow mods. */
559 struct list completions;
561 bool has_bundle_action; /* True when the first bundle action appears. */
562 struct netdev_stats stats; /* To account packets generated and consumed in
567 long long int stp_last_tick;
569 /* VLAN splinters. */
570 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
571 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
574 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
575 * for debugging the asynchronous flow_mod implementation.) */
578 /* All existing ofproto_dpif instances, indexed by ->up.name. */
579 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
581 static void ofproto_dpif_unixctl_init(void);
583 static struct ofproto_dpif *
584 ofproto_dpif_cast(const struct ofproto *ofproto)
586 assert(ofproto->ofproto_class == &ofproto_dpif_class);
587 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
590 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
592 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
594 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
595 const struct ofpbuf *, ovs_be16 initial_tci,
598 /* Packet processing. */
599 static void update_learning_table(struct ofproto_dpif *,
600 const struct flow *, int vlan,
603 #define FLOW_MISS_MAX_BATCH 50
604 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
606 /* Flow expiration. */
607 static int expire(struct ofproto_dpif *);
610 static void send_netflow_active_timeouts(struct ofproto_dpif *);
613 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
615 compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions,
616 const struct flow *, uint32_t odp_port);
617 static void add_mirror_actions(struct action_xlate_ctx *ctx,
618 const struct flow *flow);
619 /* Global variables. */
620 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
622 /* Factory functions. */
625 enumerate_types(struct sset *types)
627 dp_enumerate_types(types);
631 enumerate_names(const char *type, struct sset *names)
633 return dp_enumerate_names(type, names);
637 del(const char *type, const char *name)
642 error = dpif_open(name, type, &dpif);
644 error = dpif_delete(dpif);
650 /* Basic life-cycle. */
652 static struct ofproto *
655 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
660 dealloc(struct ofproto *ofproto_)
662 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
667 construct(struct ofproto *ofproto_)
669 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
670 const char *name = ofproto->up.name;
674 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
676 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
680 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
681 ofproto->n_matches = 0;
683 dpif_flow_flush(ofproto->dpif);
684 dpif_recv_purge(ofproto->dpif);
686 error = dpif_recv_set(ofproto->dpif, true);
688 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
689 dpif_close(ofproto->dpif);
693 ofproto->netflow = NULL;
694 ofproto->sflow = NULL;
696 hmap_init(&ofproto->bundles);
697 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
698 for (i = 0; i < MAX_MIRRORS; i++) {
699 ofproto->mirrors[i] = NULL;
701 ofproto->has_bonded_bundles = false;
703 timer_set_duration(&ofproto->next_expiration, 1000);
705 hmap_init(&ofproto->facets);
706 hmap_init(&ofproto->subfacets);
707 ofproto->governor = NULL;
709 for (i = 0; i < N_TABLES; i++) {
710 struct table_dpif *table = &ofproto->tables[i];
712 table->catchall_table = NULL;
713 table->other_table = NULL;
714 table->basis = random_uint32();
716 ofproto->need_revalidate = false;
717 tag_set_init(&ofproto->revalidate_set);
719 list_init(&ofproto->completions);
721 ofproto_dpif_unixctl_init();
723 ofproto->has_bundle_action = false;
725 hmap_init(&ofproto->vlandev_map);
726 hmap_init(&ofproto->realdev_vid_map);
728 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
729 hash_string(ofproto->up.name, 0));
730 memset(&ofproto->stats, 0, sizeof ofproto->stats);
732 ofproto_init_tables(ofproto_, N_TABLES);
738 complete_operations(struct ofproto_dpif *ofproto)
740 struct dpif_completion *c, *next;
742 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
743 ofoperation_complete(c->op, 0);
744 list_remove(&c->list_node);
750 destruct(struct ofproto *ofproto_)
752 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
753 struct rule_dpif *rule, *next_rule;
754 struct oftable *table;
757 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
758 complete_operations(ofproto);
760 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
761 struct cls_cursor cursor;
763 cls_cursor_init(&cursor, &table->cls, NULL);
764 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
765 ofproto_rule_destroy(&rule->up);
769 for (i = 0; i < MAX_MIRRORS; i++) {
770 mirror_destroy(ofproto->mirrors[i]);
773 netflow_destroy(ofproto->netflow);
774 dpif_sflow_destroy(ofproto->sflow);
775 hmap_destroy(&ofproto->bundles);
776 mac_learning_destroy(ofproto->ml);
778 hmap_destroy(&ofproto->facets);
779 hmap_destroy(&ofproto->subfacets);
780 governor_destroy(ofproto->governor);
782 hmap_destroy(&ofproto->vlandev_map);
783 hmap_destroy(&ofproto->realdev_vid_map);
785 dpif_close(ofproto->dpif);
789 run_fast(struct ofproto *ofproto_)
791 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
794 /* Handle one or more batches of upcalls, until there's nothing left to do
795 * or until we do a fixed total amount of work.
797 * We do work in batches because it can be much cheaper to set up a number
798 * of flows and fire off their patches all at once. We do multiple batches
799 * because in some cases handling a packet can cause another packet to be
800 * queued almost immediately as part of the return flow. Both
801 * optimizations can make major improvements on some benchmarks and
802 * presumably for real traffic as well. */
804 while (work < FLOW_MISS_MAX_BATCH) {
805 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
815 run(struct ofproto *ofproto_)
817 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
818 struct ofport_dpif *ofport;
819 struct ofbundle *bundle;
823 complete_operations(ofproto);
825 dpif_run(ofproto->dpif);
827 error = run_fast(ofproto_);
832 if (timer_expired(&ofproto->next_expiration)) {
833 int delay = expire(ofproto);
834 timer_set_duration(&ofproto->next_expiration, delay);
837 if (ofproto->netflow) {
838 if (netflow_run(ofproto->netflow)) {
839 send_netflow_active_timeouts(ofproto);
842 if (ofproto->sflow) {
843 dpif_sflow_run(ofproto->sflow);
846 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
849 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
854 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
856 /* Now revalidate if there's anything to do. */
857 if (ofproto->need_revalidate
858 || !tag_set_is_empty(&ofproto->revalidate_set)) {
859 struct tag_set revalidate_set = ofproto->revalidate_set;
860 bool revalidate_all = ofproto->need_revalidate;
861 struct facet *facet, *next;
863 /* Clear the revalidation flags. */
864 tag_set_init(&ofproto->revalidate_set);
865 ofproto->need_revalidate = false;
867 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
869 || tag_set_intersects(&revalidate_set, facet->tags)) {
870 facet_revalidate(facet);
875 /* Check the consistency of a random facet, to aid debugging. */
876 if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) {
879 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
880 struct facet, hmap_node);
881 if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) {
882 if (!facet_check_consistency(facet)) {
883 ofproto->need_revalidate = true;
888 if (ofproto->governor) {
891 governor_run(ofproto->governor);
893 /* If the governor has shrunk to its minimum size and the number of
894 * subfacets has dwindled, then drop the governor entirely.
896 * For hysteresis, the number of subfacets to drop the governor is
897 * smaller than the number needed to trigger its creation. */
898 n_subfacets = hmap_count(&ofproto->subfacets);
899 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
900 && governor_is_idle(ofproto->governor)) {
901 governor_destroy(ofproto->governor);
902 ofproto->governor = NULL;
910 wait(struct ofproto *ofproto_)
912 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
913 struct ofport_dpif *ofport;
914 struct ofbundle *bundle;
916 if (!clogged && !list_is_empty(&ofproto->completions)) {
917 poll_immediate_wake();
920 dpif_wait(ofproto->dpif);
921 dpif_recv_wait(ofproto->dpif);
922 if (ofproto->sflow) {
923 dpif_sflow_wait(ofproto->sflow);
925 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
926 poll_immediate_wake();
928 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
931 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
934 if (ofproto->netflow) {
935 netflow_wait(ofproto->netflow);
937 mac_learning_wait(ofproto->ml);
939 if (ofproto->need_revalidate) {
940 /* Shouldn't happen, but if it does just go around again. */
941 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
942 poll_immediate_wake();
944 timer_wait(&ofproto->next_expiration);
946 if (ofproto->governor) {
947 governor_wait(ofproto->governor);
952 flush(struct ofproto *ofproto_)
954 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
955 struct facet *facet, *next_facet;
957 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
958 /* Mark the facet as not installed so that facet_remove() doesn't
959 * bother trying to uninstall it. There is no point in uninstalling it
960 * individually since we are about to blow away all the facets with
961 * dpif_flow_flush(). */
962 struct subfacet *subfacet;
964 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
965 subfacet->installed = false;
966 subfacet->dp_packet_count = 0;
967 subfacet->dp_byte_count = 0;
971 dpif_flow_flush(ofproto->dpif);
975 get_features(struct ofproto *ofproto_ OVS_UNUSED,
976 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
978 *arp_match_ip = true;
979 *actions = (OFPUTIL_A_OUTPUT |
980 OFPUTIL_A_SET_VLAN_VID |
981 OFPUTIL_A_SET_VLAN_PCP |
982 OFPUTIL_A_STRIP_VLAN |
983 OFPUTIL_A_SET_DL_SRC |
984 OFPUTIL_A_SET_DL_DST |
985 OFPUTIL_A_SET_NW_SRC |
986 OFPUTIL_A_SET_NW_DST |
987 OFPUTIL_A_SET_NW_TOS |
988 OFPUTIL_A_SET_TP_SRC |
989 OFPUTIL_A_SET_TP_DST |
994 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
996 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
997 struct dpif_dp_stats s;
999 strcpy(ots->name, "classifier");
1001 dpif_get_dp_stats(ofproto->dpif, &s);
1002 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
1003 put_32aligned_be64(&ots->matched_count,
1004 htonll(s.n_hit + ofproto->n_matches));
1007 static struct ofport *
1010 struct ofport_dpif *port = xmalloc(sizeof *port);
1015 port_dealloc(struct ofport *port_)
1017 struct ofport_dpif *port = ofport_dpif_cast(port_);
1022 port_construct(struct ofport *port_)
1024 struct ofport_dpif *port = ofport_dpif_cast(port_);
1025 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1027 ofproto->need_revalidate = true;
1028 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
1029 port->bundle = NULL;
1031 port->tag = tag_create_random();
1032 port->may_enable = true;
1033 port->stp_port = NULL;
1034 port->stp_state = STP_DISABLED;
1035 hmap_init(&port->priorities);
1036 port->realdev_ofp_port = 0;
1037 port->vlandev_vid = 0;
1038 port->carrier_seq = netdev_get_carrier_resets(port->up.netdev);
1040 if (ofproto->sflow) {
1041 dpif_sflow_add_port(ofproto->sflow, port_);
1048 port_destruct(struct ofport *port_)
1050 struct ofport_dpif *port = ofport_dpif_cast(port_);
1051 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1053 ofproto->need_revalidate = true;
1054 bundle_remove(port_);
1055 set_cfm(port_, NULL);
1056 if (ofproto->sflow) {
1057 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1060 ofport_clear_priorities(port);
1061 hmap_destroy(&port->priorities);
1065 port_modified(struct ofport *port_)
1067 struct ofport_dpif *port = ofport_dpif_cast(port_);
1069 if (port->bundle && port->bundle->bond) {
1070 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1075 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1077 struct ofport_dpif *port = ofport_dpif_cast(port_);
1078 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1079 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1081 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1082 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD)) {
1083 ofproto->need_revalidate = true;
1085 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1086 bundle_update(port->bundle);
1092 set_sflow(struct ofproto *ofproto_,
1093 const struct ofproto_sflow_options *sflow_options)
1095 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1096 struct dpif_sflow *ds = ofproto->sflow;
1098 if (sflow_options) {
1100 struct ofport_dpif *ofport;
1102 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1103 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1104 dpif_sflow_add_port(ds, &ofport->up);
1106 ofproto->need_revalidate = true;
1108 dpif_sflow_set_options(ds, sflow_options);
1111 dpif_sflow_destroy(ds);
1112 ofproto->need_revalidate = true;
1113 ofproto->sflow = NULL;
1120 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1122 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1129 struct ofproto_dpif *ofproto;
1131 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1132 ofproto->need_revalidate = true;
1133 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1136 if (cfm_configure(ofport->cfm, s)) {
1142 cfm_destroy(ofport->cfm);
1148 get_cfm_fault(const struct ofport *ofport_)
1150 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1152 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1156 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1159 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1162 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1170 get_cfm_health(const struct ofport *ofport_)
1172 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1174 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1177 /* Spanning Tree. */
1180 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1182 struct ofproto_dpif *ofproto = ofproto_;
1183 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1184 struct ofport_dpif *ofport;
1186 ofport = stp_port_get_aux(sp);
1188 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1189 ofproto->up.name, port_num);
1191 struct eth_header *eth = pkt->l2;
1193 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1194 if (eth_addr_is_zero(eth->eth_src)) {
1195 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1196 "with unknown MAC", ofproto->up.name, port_num);
1198 send_packet(ofport, pkt);
1204 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1206 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1208 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1210 /* Only revalidate flows if the configuration changed. */
1211 if (!s != !ofproto->stp) {
1212 ofproto->need_revalidate = true;
1216 if (!ofproto->stp) {
1217 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1218 send_bpdu_cb, ofproto);
1219 ofproto->stp_last_tick = time_msec();
1222 stp_set_bridge_id(ofproto->stp, s->system_id);
1223 stp_set_bridge_priority(ofproto->stp, s->priority);
1224 stp_set_hello_time(ofproto->stp, s->hello_time);
1225 stp_set_max_age(ofproto->stp, s->max_age);
1226 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1228 struct ofport *ofport;
1230 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1231 set_stp_port(ofport, NULL);
1234 stp_destroy(ofproto->stp);
1235 ofproto->stp = NULL;
1242 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1244 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1248 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1249 s->designated_root = stp_get_designated_root(ofproto->stp);
1250 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1259 update_stp_port_state(struct ofport_dpif *ofport)
1261 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1262 enum stp_state state;
1264 /* Figure out new state. */
1265 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1269 if (ofport->stp_state != state) {
1270 enum ofputil_port_state of_state;
1273 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1274 netdev_get_name(ofport->up.netdev),
1275 stp_state_name(ofport->stp_state),
1276 stp_state_name(state));
1277 if (stp_learn_in_state(ofport->stp_state)
1278 != stp_learn_in_state(state)) {
1279 /* xxx Learning action flows should also be flushed. */
1280 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1282 fwd_change = stp_forward_in_state(ofport->stp_state)
1283 != stp_forward_in_state(state);
1285 ofproto->need_revalidate = true;
1286 ofport->stp_state = state;
1287 ofport->stp_state_entered = time_msec();
1289 if (fwd_change && ofport->bundle) {
1290 bundle_update(ofport->bundle);
1293 /* Update the STP state bits in the OpenFlow port description. */
1294 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1295 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1296 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1297 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1298 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1300 ofproto_port_set_state(&ofport->up, of_state);
1304 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1305 * caller is responsible for assigning STP port numbers and ensuring
1306 * there are no duplicates. */
1308 set_stp_port(struct ofport *ofport_,
1309 const struct ofproto_port_stp_settings *s)
1311 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1312 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1313 struct stp_port *sp = ofport->stp_port;
1315 if (!s || !s->enable) {
1317 ofport->stp_port = NULL;
1318 stp_port_disable(sp);
1319 update_stp_port_state(ofport);
1322 } else if (sp && stp_port_no(sp) != s->port_num
1323 && ofport == stp_port_get_aux(sp)) {
1324 /* The port-id changed, so disable the old one if it's not
1325 * already in use by another port. */
1326 stp_port_disable(sp);
1329 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1330 stp_port_enable(sp);
1332 stp_port_set_aux(sp, ofport);
1333 stp_port_set_priority(sp, s->priority);
1334 stp_port_set_path_cost(sp, s->path_cost);
1336 update_stp_port_state(ofport);
1342 get_stp_port_status(struct ofport *ofport_,
1343 struct ofproto_port_stp_status *s)
1345 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1346 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1347 struct stp_port *sp = ofport->stp_port;
1349 if (!ofproto->stp || !sp) {
1355 s->port_id = stp_port_get_id(sp);
1356 s->state = stp_port_get_state(sp);
1357 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1358 s->role = stp_port_get_role(sp);
1359 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1365 stp_run(struct ofproto_dpif *ofproto)
1368 long long int now = time_msec();
1369 long long int elapsed = now - ofproto->stp_last_tick;
1370 struct stp_port *sp;
1373 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1374 ofproto->stp_last_tick = now;
1376 while (stp_get_changed_port(ofproto->stp, &sp)) {
1377 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1380 update_stp_port_state(ofport);
1384 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1385 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1391 stp_wait(struct ofproto_dpif *ofproto)
1394 poll_timer_wait(1000);
1398 /* Returns true if STP should process 'flow'. */
1400 stp_should_process_flow(const struct flow *flow)
1402 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1406 stp_process_packet(const struct ofport_dpif *ofport,
1407 const struct ofpbuf *packet)
1409 struct ofpbuf payload = *packet;
1410 struct eth_header *eth = payload.data;
1411 struct stp_port *sp = ofport->stp_port;
1413 /* Sink packets on ports that have STP disabled when the bridge has
1415 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1419 /* Trim off padding on payload. */
1420 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1421 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1424 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1425 stp_received_bpdu(sp, payload.data, payload.size);
1429 static struct priority_to_dscp *
1430 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1432 struct priority_to_dscp *pdscp;
1435 hash = hash_int(priority, 0);
1436 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1437 if (pdscp->priority == priority) {
1445 ofport_clear_priorities(struct ofport_dpif *ofport)
1447 struct priority_to_dscp *pdscp, *next;
1449 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1450 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1456 set_queues(struct ofport *ofport_,
1457 const struct ofproto_port_queue *qdscp_list,
1460 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1461 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1462 struct hmap new = HMAP_INITIALIZER(&new);
1465 for (i = 0; i < n_qdscp; i++) {
1466 struct priority_to_dscp *pdscp;
1470 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1471 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1476 pdscp = get_priority(ofport, priority);
1478 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1480 pdscp = xmalloc(sizeof *pdscp);
1481 pdscp->priority = priority;
1483 ofproto->need_revalidate = true;
1486 if (pdscp->dscp != dscp) {
1488 ofproto->need_revalidate = true;
1491 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1494 if (!hmap_is_empty(&ofport->priorities)) {
1495 ofport_clear_priorities(ofport);
1496 ofproto->need_revalidate = true;
1499 hmap_swap(&new, &ofport->priorities);
1507 /* Expires all MAC learning entries associated with 'bundle' and forces its
1508 * ofproto to revalidate every flow.
1510 * Normally MAC learning entries are removed only from the ofproto associated
1511 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1512 * are removed from every ofproto. When patch ports and SLB bonds are in use
1513 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1514 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1515 * with the host from which it migrated. */
1517 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1519 struct ofproto_dpif *ofproto = bundle->ofproto;
1520 struct mac_learning *ml = ofproto->ml;
1521 struct mac_entry *mac, *next_mac;
1523 ofproto->need_revalidate = true;
1524 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1525 if (mac->port.p == bundle) {
1527 struct ofproto_dpif *o;
1529 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1531 struct mac_entry *e;
1533 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1536 tag_set_add(&o->revalidate_set, e->tag);
1537 mac_learning_expire(o->ml, e);
1543 mac_learning_expire(ml, mac);
1548 static struct ofbundle *
1549 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1551 struct ofbundle *bundle;
1553 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1554 &ofproto->bundles) {
1555 if (bundle->aux == aux) {
1562 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1563 * ones that are found to 'bundles'. */
1565 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1566 void **auxes, size_t n_auxes,
1567 struct hmapx *bundles)
1571 hmapx_init(bundles);
1572 for (i = 0; i < n_auxes; i++) {
1573 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1575 hmapx_add(bundles, bundle);
1581 bundle_update(struct ofbundle *bundle)
1583 struct ofport_dpif *port;
1585 bundle->floodable = true;
1586 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1587 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1588 || !stp_forward_in_state(port->stp_state)) {
1589 bundle->floodable = false;
1596 bundle_del_port(struct ofport_dpif *port)
1598 struct ofbundle *bundle = port->bundle;
1600 bundle->ofproto->need_revalidate = true;
1602 list_remove(&port->bundle_node);
1603 port->bundle = NULL;
1606 lacp_slave_unregister(bundle->lacp, port);
1609 bond_slave_unregister(bundle->bond, port);
1612 bundle_update(bundle);
1616 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1617 struct lacp_slave_settings *lacp,
1618 uint32_t bond_stable_id)
1620 struct ofport_dpif *port;
1622 port = get_ofp_port(bundle->ofproto, ofp_port);
1627 if (port->bundle != bundle) {
1628 bundle->ofproto->need_revalidate = true;
1630 bundle_del_port(port);
1633 port->bundle = bundle;
1634 list_push_back(&bundle->ports, &port->bundle_node);
1635 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1636 || !stp_forward_in_state(port->stp_state)) {
1637 bundle->floodable = false;
1641 port->bundle->ofproto->need_revalidate = true;
1642 lacp_slave_register(bundle->lacp, port, lacp);
1645 port->bond_stable_id = bond_stable_id;
1651 bundle_destroy(struct ofbundle *bundle)
1653 struct ofproto_dpif *ofproto;
1654 struct ofport_dpif *port, *next_port;
1661 ofproto = bundle->ofproto;
1662 for (i = 0; i < MAX_MIRRORS; i++) {
1663 struct ofmirror *m = ofproto->mirrors[i];
1665 if (m->out == bundle) {
1667 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1668 || hmapx_find_and_delete(&m->dsts, bundle)) {
1669 ofproto->need_revalidate = true;
1674 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1675 bundle_del_port(port);
1678 bundle_flush_macs(bundle, true);
1679 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1681 free(bundle->trunks);
1682 lacp_destroy(bundle->lacp);
1683 bond_destroy(bundle->bond);
1688 bundle_set(struct ofproto *ofproto_, void *aux,
1689 const struct ofproto_bundle_settings *s)
1691 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1692 bool need_flush = false;
1693 struct ofport_dpif *port;
1694 struct ofbundle *bundle;
1695 unsigned long *trunks;
1701 bundle_destroy(bundle_lookup(ofproto, aux));
1705 assert(s->n_slaves == 1 || s->bond != NULL);
1706 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1708 bundle = bundle_lookup(ofproto, aux);
1710 bundle = xmalloc(sizeof *bundle);
1712 bundle->ofproto = ofproto;
1713 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1714 hash_pointer(aux, 0));
1716 bundle->name = NULL;
1718 list_init(&bundle->ports);
1719 bundle->vlan_mode = PORT_VLAN_TRUNK;
1721 bundle->trunks = NULL;
1722 bundle->use_priority_tags = s->use_priority_tags;
1723 bundle->lacp = NULL;
1724 bundle->bond = NULL;
1726 bundle->floodable = true;
1728 bundle->src_mirrors = 0;
1729 bundle->dst_mirrors = 0;
1730 bundle->mirror_out = 0;
1733 if (!bundle->name || strcmp(s->name, bundle->name)) {
1735 bundle->name = xstrdup(s->name);
1740 if (!bundle->lacp) {
1741 ofproto->need_revalidate = true;
1742 bundle->lacp = lacp_create();
1744 lacp_configure(bundle->lacp, s->lacp);
1746 lacp_destroy(bundle->lacp);
1747 bundle->lacp = NULL;
1750 /* Update set of ports. */
1752 for (i = 0; i < s->n_slaves; i++) {
1753 if (!bundle_add_port(bundle, s->slaves[i],
1754 s->lacp ? &s->lacp_slaves[i] : NULL,
1755 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1759 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1760 struct ofport_dpif *next_port;
1762 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1763 for (i = 0; i < s->n_slaves; i++) {
1764 if (s->slaves[i] == port->up.ofp_port) {
1769 bundle_del_port(port);
1773 assert(list_size(&bundle->ports) <= s->n_slaves);
1775 if (list_is_empty(&bundle->ports)) {
1776 bundle_destroy(bundle);
1780 /* Set VLAN tagging mode */
1781 if (s->vlan_mode != bundle->vlan_mode
1782 || s->use_priority_tags != bundle->use_priority_tags) {
1783 bundle->vlan_mode = s->vlan_mode;
1784 bundle->use_priority_tags = s->use_priority_tags;
1789 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1790 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1792 if (vlan != bundle->vlan) {
1793 bundle->vlan = vlan;
1797 /* Get trunked VLANs. */
1798 switch (s->vlan_mode) {
1799 case PORT_VLAN_ACCESS:
1803 case PORT_VLAN_TRUNK:
1804 trunks = (unsigned long *) s->trunks;
1807 case PORT_VLAN_NATIVE_UNTAGGED:
1808 case PORT_VLAN_NATIVE_TAGGED:
1809 if (vlan != 0 && (!s->trunks
1810 || !bitmap_is_set(s->trunks, vlan)
1811 || bitmap_is_set(s->trunks, 0))) {
1812 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1814 trunks = bitmap_clone(s->trunks, 4096);
1816 trunks = bitmap_allocate1(4096);
1818 bitmap_set1(trunks, vlan);
1819 bitmap_set0(trunks, 0);
1821 trunks = (unsigned long *) s->trunks;
1828 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1829 free(bundle->trunks);
1830 if (trunks == s->trunks) {
1831 bundle->trunks = vlan_bitmap_clone(trunks);
1833 bundle->trunks = trunks;
1838 if (trunks != s->trunks) {
1843 if (!list_is_short(&bundle->ports)) {
1844 bundle->ofproto->has_bonded_bundles = true;
1846 if (bond_reconfigure(bundle->bond, s->bond)) {
1847 ofproto->need_revalidate = true;
1850 bundle->bond = bond_create(s->bond);
1851 ofproto->need_revalidate = true;
1854 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1855 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1859 bond_destroy(bundle->bond);
1860 bundle->bond = NULL;
1863 /* If we changed something that would affect MAC learning, un-learn
1864 * everything on this port and force flow revalidation. */
1866 bundle_flush_macs(bundle, false);
1873 bundle_remove(struct ofport *port_)
1875 struct ofport_dpif *port = ofport_dpif_cast(port_);
1876 struct ofbundle *bundle = port->bundle;
1879 bundle_del_port(port);
1880 if (list_is_empty(&bundle->ports)) {
1881 bundle_destroy(bundle);
1882 } else if (list_is_short(&bundle->ports)) {
1883 bond_destroy(bundle->bond);
1884 bundle->bond = NULL;
1890 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
1892 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1893 struct ofport_dpif *port = port_;
1894 uint8_t ea[ETH_ADDR_LEN];
1897 error = netdev_get_etheraddr(port->up.netdev, ea);
1899 struct ofpbuf packet;
1902 ofpbuf_init(&packet, 0);
1903 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1905 memcpy(packet_pdu, pdu, pdu_size);
1907 send_packet(port, &packet);
1908 ofpbuf_uninit(&packet);
1910 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1911 "%s (%s)", port->bundle->name,
1912 netdev_get_name(port->up.netdev), strerror(error));
1917 bundle_send_learning_packets(struct ofbundle *bundle)
1919 struct ofproto_dpif *ofproto = bundle->ofproto;
1920 int error, n_packets, n_errors;
1921 struct mac_entry *e;
1923 error = n_packets = n_errors = 0;
1924 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1925 if (e->port.p != bundle) {
1926 struct ofpbuf *learning_packet;
1927 struct ofport_dpif *port;
1931 /* The assignment to "port" is unnecessary but makes "grep"ing for
1932 * struct ofport_dpif more effective. */
1933 learning_packet = bond_compose_learning_packet(bundle->bond,
1937 ret = send_packet(port, learning_packet);
1938 ofpbuf_delete(learning_packet);
1948 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1949 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1950 "packets, last error was: %s",
1951 bundle->name, n_errors, n_packets, strerror(error));
1953 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1954 bundle->name, n_packets);
1959 bundle_run(struct ofbundle *bundle)
1962 lacp_run(bundle->lacp, send_pdu_cb);
1965 struct ofport_dpif *port;
1967 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1968 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1971 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1972 lacp_status(bundle->lacp));
1973 if (bond_should_send_learning_packets(bundle->bond)) {
1974 bundle_send_learning_packets(bundle);
1980 bundle_wait(struct ofbundle *bundle)
1983 lacp_wait(bundle->lacp);
1986 bond_wait(bundle->bond);
1993 mirror_scan(struct ofproto_dpif *ofproto)
1997 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1998 if (!ofproto->mirrors[idx]) {
2005 static struct ofmirror *
2006 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2010 for (i = 0; i < MAX_MIRRORS; i++) {
2011 struct ofmirror *mirror = ofproto->mirrors[i];
2012 if (mirror && mirror->aux == aux) {
2020 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2022 mirror_update_dups(struct ofproto_dpif *ofproto)
2026 for (i = 0; i < MAX_MIRRORS; i++) {
2027 struct ofmirror *m = ofproto->mirrors[i];
2030 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2034 for (i = 0; i < MAX_MIRRORS; i++) {
2035 struct ofmirror *m1 = ofproto->mirrors[i];
2042 for (j = i + 1; j < MAX_MIRRORS; j++) {
2043 struct ofmirror *m2 = ofproto->mirrors[j];
2045 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2046 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2047 m2->dup_mirrors |= m1->dup_mirrors;
2054 mirror_set(struct ofproto *ofproto_, void *aux,
2055 const struct ofproto_mirror_settings *s)
2057 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2058 mirror_mask_t mirror_bit;
2059 struct ofbundle *bundle;
2060 struct ofmirror *mirror;
2061 struct ofbundle *out;
2062 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2063 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2066 mirror = mirror_lookup(ofproto, aux);
2068 mirror_destroy(mirror);
2074 idx = mirror_scan(ofproto);
2076 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2078 ofproto->up.name, MAX_MIRRORS, s->name);
2082 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2083 mirror->ofproto = ofproto;
2086 mirror->out_vlan = -1;
2087 mirror->name = NULL;
2090 if (!mirror->name || strcmp(s->name, mirror->name)) {
2092 mirror->name = xstrdup(s->name);
2095 /* Get the new configuration. */
2096 if (s->out_bundle) {
2097 out = bundle_lookup(ofproto, s->out_bundle);
2099 mirror_destroy(mirror);
2105 out_vlan = s->out_vlan;
2107 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2108 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2110 /* If the configuration has not changed, do nothing. */
2111 if (hmapx_equals(&srcs, &mirror->srcs)
2112 && hmapx_equals(&dsts, &mirror->dsts)
2113 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2114 && mirror->out == out
2115 && mirror->out_vlan == out_vlan)
2117 hmapx_destroy(&srcs);
2118 hmapx_destroy(&dsts);
2122 hmapx_swap(&srcs, &mirror->srcs);
2123 hmapx_destroy(&srcs);
2125 hmapx_swap(&dsts, &mirror->dsts);
2126 hmapx_destroy(&dsts);
2128 free(mirror->vlans);
2129 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2132 mirror->out_vlan = out_vlan;
2134 /* Update bundles. */
2135 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2136 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2137 if (hmapx_contains(&mirror->srcs, bundle)) {
2138 bundle->src_mirrors |= mirror_bit;
2140 bundle->src_mirrors &= ~mirror_bit;
2143 if (hmapx_contains(&mirror->dsts, bundle)) {
2144 bundle->dst_mirrors |= mirror_bit;
2146 bundle->dst_mirrors &= ~mirror_bit;
2149 if (mirror->out == bundle) {
2150 bundle->mirror_out |= mirror_bit;
2152 bundle->mirror_out &= ~mirror_bit;
2156 ofproto->need_revalidate = true;
2157 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2158 mirror_update_dups(ofproto);
2164 mirror_destroy(struct ofmirror *mirror)
2166 struct ofproto_dpif *ofproto;
2167 mirror_mask_t mirror_bit;
2168 struct ofbundle *bundle;
2174 ofproto = mirror->ofproto;
2175 ofproto->need_revalidate = true;
2176 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2178 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2179 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2180 bundle->src_mirrors &= ~mirror_bit;
2181 bundle->dst_mirrors &= ~mirror_bit;
2182 bundle->mirror_out &= ~mirror_bit;
2185 hmapx_destroy(&mirror->srcs);
2186 hmapx_destroy(&mirror->dsts);
2187 free(mirror->vlans);
2189 ofproto->mirrors[mirror->idx] = NULL;
2193 mirror_update_dups(ofproto);
2197 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2198 uint64_t *packets, uint64_t *bytes)
2200 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2201 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2204 *packets = *bytes = UINT64_MAX;
2208 *packets = mirror->packet_count;
2209 *bytes = mirror->byte_count;
2215 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2217 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2218 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2219 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2225 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2227 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2228 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2229 return bundle && bundle->mirror_out != 0;
2233 forward_bpdu_changed(struct ofproto *ofproto_)
2235 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2236 /* Revalidate cached flows whenever forward_bpdu option changes. */
2237 ofproto->need_revalidate = true;
2241 set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time)
2243 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2244 mac_learning_set_idle_time(ofproto->ml, idle_time);
2249 static struct ofport_dpif *
2250 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
2252 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2253 return ofport ? ofport_dpif_cast(ofport) : NULL;
2256 static struct ofport_dpif *
2257 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
2259 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2263 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2264 struct dpif_port *dpif_port)
2266 ofproto_port->name = dpif_port->name;
2267 ofproto_port->type = dpif_port->type;
2268 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2272 port_run(struct ofport_dpif *ofport)
2274 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2275 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2276 bool enable = netdev_get_carrier(ofport->up.netdev);
2278 ofport->carrier_seq = carrier_seq;
2281 cfm_run(ofport->cfm);
2283 if (cfm_should_send_ccm(ofport->cfm)) {
2284 struct ofpbuf packet;
2286 ofpbuf_init(&packet, 0);
2287 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2288 send_packet(ofport, &packet);
2289 ofpbuf_uninit(&packet);
2292 enable = enable && !cfm_get_fault(ofport->cfm)
2293 && cfm_get_opup(ofport->cfm);
2296 if (ofport->bundle) {
2297 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2298 if (carrier_changed) {
2299 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2303 if (ofport->may_enable != enable) {
2304 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2306 if (ofproto->has_bundle_action) {
2307 ofproto->need_revalidate = true;
2311 ofport->may_enable = enable;
2315 port_wait(struct ofport_dpif *ofport)
2318 cfm_wait(ofport->cfm);
2323 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2324 struct ofproto_port *ofproto_port)
2326 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2327 struct dpif_port dpif_port;
2330 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2332 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2338 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2340 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2344 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2346 *ofp_portp = odp_port_to_ofp_port(odp_port);
2352 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2354 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2357 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2359 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2361 /* The caller is going to close ofport->up.netdev. If this is a
2362 * bonded port, then the bond is using that netdev, so remove it
2363 * from the bond. The client will need to reconfigure everything
2364 * after deleting ports, so then the slave will get re-added. */
2365 bundle_remove(&ofport->up);
2372 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2374 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2377 error = netdev_get_stats(ofport->up.netdev, stats);
2379 if (!error && ofport->odp_port == OVSP_LOCAL) {
2380 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2382 /* ofproto->stats.tx_packets represents packets that we created
2383 * internally and sent to some port (e.g. packets sent with
2384 * send_packet()). Account for them as if they had come from
2385 * OFPP_LOCAL and got forwarded. */
2387 if (stats->rx_packets != UINT64_MAX) {
2388 stats->rx_packets += ofproto->stats.tx_packets;
2391 if (stats->rx_bytes != UINT64_MAX) {
2392 stats->rx_bytes += ofproto->stats.tx_bytes;
2395 /* ofproto->stats.rx_packets represents packets that were received on
2396 * some port and we processed internally and dropped (e.g. STP).
2397 * Account fro them as if they had been forwarded to OFPP_LOCAL. */
2399 if (stats->tx_packets != UINT64_MAX) {
2400 stats->tx_packets += ofproto->stats.rx_packets;
2403 if (stats->tx_bytes != UINT64_MAX) {
2404 stats->tx_bytes += ofproto->stats.rx_bytes;
2411 /* Account packets for LOCAL port. */
2413 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2414 size_t tx_size, size_t rx_size)
2416 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2419 ofproto->stats.rx_packets++;
2420 ofproto->stats.rx_bytes += rx_size;
2423 ofproto->stats.tx_packets++;
2424 ofproto->stats.tx_bytes += tx_size;
2428 struct port_dump_state {
2429 struct dpif_port_dump dump;
2434 port_dump_start(const struct ofproto *ofproto_, void **statep)
2436 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2437 struct port_dump_state *state;
2439 *statep = state = xmalloc(sizeof *state);
2440 dpif_port_dump_start(&state->dump, ofproto->dpif);
2441 state->done = false;
2446 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2447 struct ofproto_port *port)
2449 struct port_dump_state *state = state_;
2450 struct dpif_port dpif_port;
2452 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2453 ofproto_port_from_dpif_port(port, &dpif_port);
2456 int error = dpif_port_dump_done(&state->dump);
2458 return error ? error : EOF;
2463 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2465 struct port_dump_state *state = state_;
2468 dpif_port_dump_done(&state->dump);
2475 port_poll(const struct ofproto *ofproto_, char **devnamep)
2477 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2478 return dpif_port_poll(ofproto->dpif, devnamep);
2482 port_poll_wait(const struct ofproto *ofproto_)
2484 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2485 dpif_port_poll_wait(ofproto->dpif);
2489 port_is_lacp_current(const struct ofport *ofport_)
2491 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2492 return (ofport->bundle && ofport->bundle->lacp
2493 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2497 /* Upcall handling. */
2499 /* Flow miss batching.
2501 * Some dpifs implement operations faster when you hand them off in a batch.
2502 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2503 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2504 * more packets, plus possibly installing the flow in the dpif.
2506 * So far we only batch the operations that affect flow setup time the most.
2507 * It's possible to batch more than that, but the benefit might be minimal. */
2509 struct hmap_node hmap_node;
2511 enum odp_key_fitness key_fitness;
2512 const struct nlattr *key;
2514 ovs_be16 initial_tci;
2515 struct list packets;
2518 struct flow_miss_op {
2519 struct dpif_op dpif_op;
2520 struct subfacet *subfacet; /* Subfacet */
2521 void *garbage; /* Pointer to pass to free(), NULL if none. */
2522 uint64_t stub[1024 / 8]; /* Temporary buffer. */
2525 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2526 * OpenFlow controller as necessary according to their individual
2527 * configurations. */
2529 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
2530 const struct flow *flow)
2532 struct ofputil_packet_in pin;
2534 pin.packet = packet->data;
2535 pin.packet_len = packet->size;
2536 pin.reason = OFPR_NO_MATCH;
2537 pin.controller_id = 0;
2542 pin.send_len = 0; /* not used for flow table misses */
2544 flow_get_metadata(flow, &pin.fmd);
2546 /* Registers aren't meaningful on a miss. */
2547 memset(pin.fmd.reg_masks, 0, sizeof pin.fmd.reg_masks);
2549 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
2553 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2554 const struct ofpbuf *packet)
2556 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2562 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2564 cfm_process_heartbeat(ofport->cfm, packet);
2567 } else if (ofport->bundle && ofport->bundle->lacp
2568 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2570 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2573 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2575 stp_process_packet(ofport, packet);
2582 static struct flow_miss *
2583 flow_miss_find(struct hmap *todo, const struct flow *flow, uint32_t hash)
2585 struct flow_miss *miss;
2587 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2588 if (flow_equal(&miss->flow, flow)) {
2596 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
2597 * 'packet'. The caller must initialize op->actions and op->actions_len. If
2598 * 'miss' is associated with a subfacet the caller must also initialize the
2599 * returned op->subfacet, and if anything needs to be freed after processing
2600 * the op, the caller must initialize op->garbage also. */
2602 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
2603 struct flow_miss_op *op)
2605 if (miss->flow.vlan_tci != miss->initial_tci) {
2606 /* This packet was received on a VLAN splinter port. We
2607 * added a VLAN to the packet to make the packet resemble
2608 * the flow, but the actions were composed assuming that
2609 * the packet contained no VLAN. So, we must remove the
2610 * VLAN header from the packet before trying to execute the
2612 eth_pop_vlan(packet);
2615 op->subfacet = NULL;
2617 op->dpif_op.type = DPIF_OP_EXECUTE;
2618 op->dpif_op.u.execute.key = miss->key;
2619 op->dpif_op.u.execute.key_len = miss->key_len;
2620 op->dpif_op.u.execute.packet = packet;
2623 /* Helper for handle_flow_miss_without_facet() and
2624 * handle_flow_miss_with_facet(). */
2626 handle_flow_miss_common(struct rule_dpif *rule,
2627 struct ofpbuf *packet, const struct flow *flow)
2629 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2631 ofproto->n_matches++;
2633 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2635 * Extra-special case for fail-open mode.
2637 * We are in fail-open mode and the packet matched the fail-open
2638 * rule, but we are connected to a controller too. We should send
2639 * the packet up to the controller in the hope that it will try to
2640 * set up a flow and thereby allow us to exit fail-open.
2642 * See the top-level comment in fail-open.c for more information.
2644 send_packet_in_miss(ofproto, packet, flow);
2648 /* Figures out whether a flow that missed in 'ofproto', whose details are in
2649 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
2650 * installing a datapath flow. The answer is usually "yes" (a return value of
2651 * true). However, for short flows the cost of bookkeeping is much higher than
2652 * the benefits, so when the datapath holds a large number of flows we impose
2653 * some heuristics to decide which flows are likely to be worth tracking. */
2655 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
2656 struct flow_miss *miss, uint32_t hash)
2658 if (!ofproto->governor) {
2661 n_subfacets = hmap_count(&ofproto->subfacets);
2662 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
2666 ofproto->governor = governor_create(ofproto->up.name);
2669 return governor_should_install_flow(ofproto->governor, hash,
2670 list_size(&miss->packets));
2673 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
2674 * or creating any datapath flow. May add an "execute" operation to 'ops' and
2675 * increment '*n_ops'. */
2677 handle_flow_miss_without_facet(struct flow_miss *miss,
2678 struct rule_dpif *rule,
2679 struct flow_miss_op *ops, size_t *n_ops)
2681 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2682 struct action_xlate_ctx ctx;
2683 struct ofpbuf *packet;
2685 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2686 struct flow_miss_op *op = &ops[*n_ops];
2687 struct dpif_flow_stats stats;
2688 struct ofpbuf odp_actions;
2690 COVERAGE_INC(facet_suppress);
2692 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
2694 dpif_flow_stats_extract(&miss->flow, packet, &stats);
2695 rule_credit_stats(rule, &stats);
2697 action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci,
2699 ctx.resubmit_stats = &stats;
2700 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions,
2703 if (odp_actions.size) {
2704 struct dpif_execute *execute = &op->dpif_op.u.execute;
2706 init_flow_miss_execute_op(miss, packet, op);
2707 execute->actions = odp_actions.data;
2708 execute->actions_len = odp_actions.size;
2709 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
2713 ofpbuf_uninit(&odp_actions);
2718 /* Handles 'miss', which matches 'facet'. May add any required datapath
2719 * operations to 'ops', incrementing '*n_ops' for each new op. */
2721 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
2722 struct flow_miss_op *ops, size_t *n_ops)
2724 struct subfacet *subfacet;
2725 struct ofpbuf *packet;
2727 subfacet = subfacet_create(facet,
2728 miss->key_fitness, miss->key, miss->key_len,
2731 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2732 struct flow_miss_op *op = &ops[*n_ops];
2733 struct dpif_flow_stats stats;
2734 struct ofpbuf odp_actions;
2736 handle_flow_miss_common(facet->rule, packet, &miss->flow);
2738 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
2739 if (!facet->may_install || !subfacet->actions) {
2740 subfacet_make_actions(subfacet, packet, &odp_actions);
2743 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2744 subfacet_update_stats(subfacet, &stats);
2746 if (subfacet->actions_len) {
2747 struct dpif_execute *execute = &op->dpif_op.u.execute;
2749 init_flow_miss_execute_op(miss, packet, op);
2750 op->subfacet = subfacet;
2751 if (facet->may_install) {
2752 execute->actions = subfacet->actions;
2753 execute->actions_len = subfacet->actions_len;
2754 ofpbuf_uninit(&odp_actions);
2756 execute->actions = odp_actions.data;
2757 execute->actions_len = odp_actions.size;
2758 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
2763 ofpbuf_uninit(&odp_actions);
2767 if (facet->may_install && subfacet->key_fitness != ODP_FIT_TOO_LITTLE) {
2768 struct flow_miss_op *op = &ops[(*n_ops)++];
2769 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
2771 op->subfacet = subfacet;
2773 op->dpif_op.type = DPIF_OP_FLOW_PUT;
2774 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2775 put->key = miss->key;
2776 put->key_len = miss->key_len;
2777 put->actions = subfacet->actions;
2778 put->actions_len = subfacet->actions_len;
2783 /* Handles flow miss 'miss' on 'ofproto'. The flow does not match any flow in
2784 * the OpenFlow flow table. */
2786 handle_flow_miss_no_rule(struct ofproto_dpif *ofproto, struct flow_miss *miss)
2788 uint16_t in_port = miss->flow.in_port;
2789 struct ofport_dpif *port = get_ofp_port(ofproto, in_port);
2792 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, in_port);
2795 if (port && port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
2796 /* XXX install 'drop' flow entry */
2797 COVERAGE_INC(ofproto_dpif_no_packet_in);
2799 const struct ofpbuf *packet;
2801 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2802 send_packet_in_miss(ofproto, packet, &miss->flow);
2807 /* Handles flow miss 'miss' on 'ofproto'. May add any required datapath
2808 * operations to 'ops', incrementing '*n_ops' for each new op. */
2810 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2811 struct flow_miss_op *ops, size_t *n_ops)
2813 struct facet *facet;
2816 /* The caller must ensure that miss->hmap_node.hash contains
2817 * flow_hash(miss->flow, 0). */
2818 hash = miss->hmap_node.hash;
2820 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
2822 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow, 0);
2824 handle_flow_miss_no_rule(ofproto, miss);
2826 } else if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
2827 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
2831 facet = facet_create(rule, &miss->flow, hash);
2833 handle_flow_miss_with_facet(miss, facet, ops, n_ops);
2836 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
2837 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
2838 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
2839 * what a flow key should contain.
2841 * This function also includes some logic to help make VLAN splinters
2842 * transparent to the rest of the upcall processing logic. In particular, if
2843 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
2844 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
2845 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
2847 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
2848 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
2849 * (This differs from the value returned in flow->vlan_tci only for packets
2850 * received on VLAN splinters.)
2852 static enum odp_key_fitness
2853 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
2854 const struct nlattr *key, size_t key_len,
2855 struct flow *flow, ovs_be16 *initial_tci,
2856 struct ofpbuf *packet)
2858 enum odp_key_fitness fitness;
2862 fitness = odp_flow_key_to_flow(key, key_len, flow);
2863 if (fitness == ODP_FIT_ERROR) {
2866 *initial_tci = flow->vlan_tci;
2868 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
2870 /* Cause the flow to be processed as if it came in on the real device
2871 * with the VLAN device's VLAN ID. */
2872 flow->in_port = realdev;
2873 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
2875 /* Make the packet resemble the flow, so that it gets sent to an
2876 * OpenFlow controller properly, so that it looks correct for
2877 * sFlow, and so that flow_extract() will get the correct vlan_tci
2878 * if it is called on 'packet'.
2880 * The allocated space inside 'packet' probably also contains
2881 * 'key', that is, both 'packet' and 'key' are probably part of a
2882 * struct dpif_upcall (see the large comment on that structure
2883 * definition), so pushing data on 'packet' is in general not a
2884 * good idea since it could overwrite 'key' or free it as a side
2885 * effect. However, it's OK in this special case because we know
2886 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
2887 * will just overwrite the 4-byte "struct nlattr", which is fine
2888 * since we don't need that header anymore. */
2889 eth_push_vlan(packet, flow->vlan_tci);
2892 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
2893 if (fitness == ODP_FIT_PERFECT) {
2894 fitness = ODP_FIT_TOO_MUCH;
2902 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
2905 struct dpif_upcall *upcall;
2906 struct flow_miss *miss;
2907 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
2908 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
2909 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
2919 /* Construct the to-do list.
2921 * This just amounts to extracting the flow from each packet and sticking
2922 * the packets that have the same flow in the same "flow_miss" structure so
2923 * that we can process them together. */
2926 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
2927 struct flow_miss *miss = &misses[n_misses];
2928 struct flow_miss *existing_miss;
2931 /* Obtain metadata and check userspace/kernel agreement on flow match,
2932 * then set 'flow''s header pointers. */
2933 miss->key_fitness = ofproto_dpif_extract_flow_key(
2934 ofproto, upcall->key, upcall->key_len,
2935 &miss->flow, &miss->initial_tci, upcall->packet);
2936 if (miss->key_fitness == ODP_FIT_ERROR) {
2939 flow_extract(upcall->packet, miss->flow.skb_priority,
2940 miss->flow.tun_id, miss->flow.in_port, &miss->flow);
2942 /* Handle 802.1ag, LACP, and STP specially. */
2943 if (process_special(ofproto, &miss->flow, upcall->packet)) {
2944 ofproto_update_local_port_stats(&ofproto->up,
2945 0, upcall->packet->size);
2946 ofproto->n_matches++;
2950 /* Add other packets to a to-do list. */
2951 hash = flow_hash(&miss->flow, 0);
2952 existing_miss = flow_miss_find(&todo, &miss->flow, hash);
2953 if (!existing_miss) {
2954 hmap_insert(&todo, &miss->hmap_node, hash);
2955 miss->key = upcall->key;
2956 miss->key_len = upcall->key_len;
2957 list_init(&miss->packets);
2961 miss = existing_miss;
2963 list_push_back(&miss->packets, &upcall->packet->list_node);
2966 /* Process each element in the to-do list, constructing the set of
2967 * operations to batch. */
2969 HMAP_FOR_EACH (miss, hmap_node, &todo) {
2970 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
2972 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
2974 /* Execute batch. */
2975 for (i = 0; i < n_ops; i++) {
2976 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
2978 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
2980 /* Free memory and update facets. */
2981 for (i = 0; i < n_ops; i++) {
2982 struct flow_miss_op *op = &flow_miss_ops[i];
2984 switch (op->dpif_op.type) {
2985 case DPIF_OP_EXECUTE:
2988 case DPIF_OP_FLOW_PUT:
2989 if (!op->dpif_op.error) {
2990 op->subfacet->installed = true;
2994 case DPIF_OP_FLOW_DEL:
3000 hmap_destroy(&todo);
3004 handle_userspace_upcall(struct ofproto_dpif *ofproto,
3005 struct dpif_upcall *upcall)
3007 struct user_action_cookie cookie;
3008 enum odp_key_fitness fitness;
3009 ovs_be16 initial_tci;
3012 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3014 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
3015 upcall->key_len, &flow,
3016 &initial_tci, upcall->packet);
3017 if (fitness == ODP_FIT_ERROR) {
3021 if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
3022 if (ofproto->sflow) {
3023 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3027 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
3032 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
3034 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3035 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3036 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3041 assert(max_batch <= FLOW_MISS_MAX_BATCH);
3045 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3046 struct dpif_upcall *upcall = &misses[n_misses];
3047 struct ofpbuf *buf = &miss_bufs[n_misses];
3050 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3051 sizeof miss_buf_stubs[n_misses]);
3052 error = dpif_recv(ofproto->dpif, upcall, buf);
3058 switch (upcall->type) {
3059 case DPIF_UC_ACTION:
3060 handle_userspace_upcall(ofproto, upcall);
3065 /* Handle it later. */
3069 case DPIF_N_UC_TYPES:
3071 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32,
3077 handle_miss_upcalls(ofproto, misses, n_misses);
3078 for (i = 0; i < n_misses; i++) {
3079 ofpbuf_uninit(&miss_bufs[i]);
3085 /* Flow expiration. */
3087 static int subfacet_max_idle(const struct ofproto_dpif *);
3088 static void update_stats(struct ofproto_dpif *);
3089 static void rule_expire(struct rule_dpif *);
3090 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
3092 /* This function is called periodically by run(). Its job is to collect
3093 * updates for the flows that have been installed into the datapath, most
3094 * importantly when they last were used, and then use that information to
3095 * expire flows that have not been used recently.
3097 * Returns the number of milliseconds after which it should be called again. */
3099 expire(struct ofproto_dpif *ofproto)
3101 struct rule_dpif *rule, *next_rule;
3102 struct oftable *table;
3105 /* Update stats for each flow in the datapath. */
3106 update_stats(ofproto);
3108 /* Expire subfacets that have been idle too long. */
3109 dp_max_idle = subfacet_max_idle(ofproto);
3110 expire_subfacets(ofproto, dp_max_idle);
3112 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
3113 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
3114 struct cls_cursor cursor;
3116 cls_cursor_init(&cursor, &table->cls, NULL);
3117 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
3122 /* All outstanding data in existing flows has been accounted, so it's a
3123 * good time to do bond rebalancing. */
3124 if (ofproto->has_bonded_bundles) {
3125 struct ofbundle *bundle;
3127 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3129 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
3134 return MIN(dp_max_idle, 1000);
3137 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3139 * This function also pushes statistics updates to rules which each facet
3140 * resubmits into. Generally these statistics will be accurate. However, if a
3141 * facet changes the rule it resubmits into at some time in between
3142 * update_stats() runs, it is possible that statistics accrued to the
3143 * old rule will be incorrectly attributed to the new rule. This could be
3144 * avoided by calling update_stats() whenever rules are created or
3145 * deleted. However, the performance impact of making so many calls to the
3146 * datapath do not justify the benefit of having perfectly accurate statistics.
3149 update_stats(struct ofproto_dpif *p)
3151 const struct dpif_flow_stats *stats;
3152 struct dpif_flow_dump dump;
3153 const struct nlattr *key;
3156 dpif_flow_dump_start(&dump, p->dpif);
3157 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
3158 struct subfacet *subfacet;
3160 subfacet = subfacet_find(p, key, key_len);
3161 if (subfacet && subfacet->installed) {
3162 struct facet *facet = subfacet->facet;
3164 if (stats->n_packets >= subfacet->dp_packet_count) {
3165 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
3166 facet->packet_count += extra;
3168 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3171 if (stats->n_bytes >= subfacet->dp_byte_count) {
3172 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
3174 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3177 subfacet->dp_packet_count = stats->n_packets;
3178 subfacet->dp_byte_count = stats->n_bytes;
3180 facet->tcp_flags |= stats->tcp_flags;
3182 subfacet_update_time(subfacet, stats->used);
3183 if (facet->accounted_bytes < facet->byte_count) {
3185 facet_account(facet);
3186 facet->accounted_bytes = facet->byte_count;
3188 facet_push_stats(facet);
3190 if (!VLOG_DROP_WARN(&rl)) {
3194 odp_flow_key_format(key, key_len, &s);
3195 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
3199 COVERAGE_INC(facet_unexpected);
3200 /* There's a flow in the datapath that we know nothing about, or a
3201 * flow that shouldn't be installed but was anyway. Delete it. */
3202 dpif_flow_del(p->dpif, key, key_len, NULL);
3205 dpif_flow_dump_done(&dump);
3208 /* Calculates and returns the number of milliseconds of idle time after which
3209 * subfacets should expire from the datapath. When a subfacet expires, we fold
3210 * its statistics into its facet, and when a facet's last subfacet expires, we
3211 * fold its statistic into its rule. */
3213 subfacet_max_idle(const struct ofproto_dpif *ofproto)
3216 * Idle time histogram.
3218 * Most of the time a switch has a relatively small number of subfacets.
3219 * When this is the case we might as well keep statistics for all of them
3220 * in userspace and to cache them in the kernel datapath for performance as
3223 * As the number of subfacets increases, the memory required to maintain
3224 * statistics about them in userspace and in the kernel becomes
3225 * significant. However, with a large number of subfacets it is likely
3226 * that only a few of them are "heavy hitters" that consume a large amount
3227 * of bandwidth. At this point, only heavy hitters are worth caching in
3228 * the kernel and maintaining in userspaces; other subfacets we can
3231 * The technique used to compute the idle time is to build a histogram with
3232 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3233 * that is installed in the kernel gets dropped in the appropriate bucket.
3234 * After the histogram has been built, we compute the cutoff so that only
3235 * the most-recently-used 1% of subfacets (but at least
3236 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3237 * the most-recently-used bucket of subfacets is kept, so actually an
3238 * arbitrary number of subfacets can be kept in any given expiration run
3239 * (though the next run will delete most of those unless they receive
3242 * This requires a second pass through the subfacets, in addition to the
3243 * pass made by update_stats(), because the former function never looks at
3244 * uninstallable subfacets.
3246 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3247 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3248 int buckets[N_BUCKETS] = { 0 };
3249 int total, subtotal, bucket;
3250 struct subfacet *subfacet;
3254 total = hmap_count(&ofproto->subfacets);
3255 if (total <= ofproto->up.flow_eviction_threshold) {
3256 return N_BUCKETS * BUCKET_WIDTH;
3259 /* Build histogram. */
3261 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3262 long long int idle = now - subfacet->used;
3263 int bucket = (idle <= 0 ? 0
3264 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3265 : (unsigned int) idle / BUCKET_WIDTH);
3269 /* Find the first bucket whose flows should be expired. */
3270 subtotal = bucket = 0;
3272 subtotal += buckets[bucket++];
3273 } while (bucket < N_BUCKETS &&
3274 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3276 if (VLOG_IS_DBG_ENABLED()) {
3280 ds_put_cstr(&s, "keep");
3281 for (i = 0; i < N_BUCKETS; i++) {
3283 ds_put_cstr(&s, ", drop");
3286 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3289 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3293 return bucket * BUCKET_WIDTH;
3296 enum { EXPIRE_MAX_BATCH = 50 };
3299 expire_batch(struct ofproto_dpif *ofproto, struct subfacet **subfacets, int n)
3301 struct odputil_keybuf keybufs[EXPIRE_MAX_BATCH];
3302 struct dpif_op ops[EXPIRE_MAX_BATCH];
3303 struct dpif_op *opsp[EXPIRE_MAX_BATCH];
3304 struct ofpbuf keys[EXPIRE_MAX_BATCH];
3305 struct dpif_flow_stats stats[EXPIRE_MAX_BATCH];
3308 for (i = 0; i < n; i++) {
3309 ops[i].type = DPIF_OP_FLOW_DEL;
3310 subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]);
3311 ops[i].u.flow_del.key = keys[i].data;
3312 ops[i].u.flow_del.key_len = keys[i].size;
3313 ops[i].u.flow_del.stats = &stats[i];
3317 dpif_operate(ofproto->dpif, opsp, n);
3318 for (i = 0; i < n; i++) {
3319 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
3320 subfacets[i]->installed = false;
3321 subfacet_destroy(subfacets[i]);
3326 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3328 long long int cutoff = time_msec() - dp_max_idle;
3330 struct subfacet *subfacet, *next_subfacet;
3331 struct subfacet *batch[EXPIRE_MAX_BATCH];
3335 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3336 &ofproto->subfacets) {
3337 if (subfacet->used < cutoff) {
3338 if (subfacet->installed) {
3339 batch[n_batch++] = subfacet;
3340 if (n_batch >= EXPIRE_MAX_BATCH) {
3341 expire_batch(ofproto, batch, n_batch);
3345 subfacet_destroy(subfacet);
3351 expire_batch(ofproto, batch, n_batch);
3355 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3356 * then delete it entirely. */
3358 rule_expire(struct rule_dpif *rule)
3360 struct facet *facet, *next_facet;
3364 /* Has 'rule' expired? */
3366 if (rule->up.hard_timeout
3367 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3368 reason = OFPRR_HARD_TIMEOUT;
3369 } else if (rule->up.idle_timeout
3370 && now > rule->up.used + rule->up.idle_timeout * 1000) {
3371 reason = OFPRR_IDLE_TIMEOUT;
3376 COVERAGE_INC(ofproto_dpif_expired);
3378 /* Update stats. (This is a no-op if the rule expired due to an idle
3379 * timeout, because that only happens when the rule has no facets left.) */
3380 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3381 facet_remove(facet);
3384 /* Get rid of the rule. */
3385 ofproto_rule_expire(&rule->up, reason);
3390 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3392 * The caller must already have determined that no facet with an identical
3393 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3394 * the ofproto's classifier table.
3396 * 'hash' must be the return value of flow_hash(flow, 0).
3398 * The facet will initially have no subfacets. The caller should create (at
3399 * least) one subfacet with subfacet_create(). */
3400 static struct facet *
3401 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
3403 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3404 struct facet *facet;
3406 facet = xzalloc(sizeof *facet);
3407 facet->used = time_msec();
3408 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
3409 list_push_back(&rule->facets, &facet->list_node);
3411 facet->flow = *flow;
3412 list_init(&facet->subfacets);
3413 netflow_flow_init(&facet->nf_flow);
3414 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3420 facet_free(struct facet *facet)
3425 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3426 * 'packet', which arrived on 'in_port'.
3428 * Takes ownership of 'packet'. */
3430 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3431 const struct nlattr *odp_actions, size_t actions_len,
3432 struct ofpbuf *packet)
3434 struct odputil_keybuf keybuf;
3438 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3439 odp_flow_key_from_flow(&key, flow);
3441 error = dpif_execute(ofproto->dpif, key.data, key.size,
3442 odp_actions, actions_len, packet);
3444 ofpbuf_delete(packet);
3448 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3450 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3451 * rule's statistics, via subfacet_uninstall().
3453 * - Removes 'facet' from its rule and from ofproto->facets.
3456 facet_remove(struct facet *facet)
3458 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3459 struct subfacet *subfacet, *next_subfacet;
3461 assert(!list_is_empty(&facet->subfacets));
3463 /* First uninstall all of the subfacets to get final statistics. */
3464 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3465 subfacet_uninstall(subfacet);
3468 /* Flush the final stats to the rule.
3470 * This might require us to have at least one subfacet around so that we
3471 * can use its actions for accounting in facet_account(), which is why we
3472 * have uninstalled but not yet destroyed the subfacets. */
3473 facet_flush_stats(facet);
3475 /* Now we're really all done so destroy everything. */
3476 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3477 &facet->subfacets) {
3478 subfacet_destroy__(subfacet);
3480 hmap_remove(&ofproto->facets, &facet->hmap_node);
3481 list_remove(&facet->list_node);
3485 /* Feed information from 'facet' back into the learning table to keep it in
3486 * sync with what is actually flowing through the datapath. */
3488 facet_learn(struct facet *facet)
3490 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3491 struct action_xlate_ctx ctx;
3493 if (!facet->has_learn
3494 && !facet->has_normal
3495 && (!facet->has_fin_timeout
3496 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
3500 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3501 facet->flow.vlan_tci,
3502 facet->rule, facet->tcp_flags, NULL);
3503 ctx.may_learn = true;
3504 xlate_actions_for_side_effects(&ctx, facet->rule->up.actions,
3505 facet->rule->up.n_actions);
3509 facet_account(struct facet *facet)
3511 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3512 struct subfacet *subfacet;
3513 const struct nlattr *a;
3518 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3521 n_bytes = facet->byte_count - facet->accounted_bytes;
3523 /* This loop feeds byte counters to bond_account() for rebalancing to use
3524 * as a basis. We also need to track the actual VLAN on which the packet
3525 * is going to be sent to ensure that it matches the one passed to
3526 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3529 * We use the actions from an arbitrary subfacet because they should all
3530 * be equally valid for our purpose. */
3531 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3532 struct subfacet, list_node);
3533 vlan_tci = facet->flow.vlan_tci;
3534 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3535 subfacet->actions, subfacet->actions_len) {
3536 const struct ovs_action_push_vlan *vlan;
3537 struct ofport_dpif *port;
3539 switch (nl_attr_type(a)) {
3540 case OVS_ACTION_ATTR_OUTPUT:
3541 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3542 if (port && port->bundle && port->bundle->bond) {
3543 bond_account(port->bundle->bond, &facet->flow,
3544 vlan_tci_to_vid(vlan_tci), n_bytes);
3548 case OVS_ACTION_ATTR_POP_VLAN:
3549 vlan_tci = htons(0);
3552 case OVS_ACTION_ATTR_PUSH_VLAN:
3553 vlan = nl_attr_get(a);
3554 vlan_tci = vlan->vlan_tci;
3560 /* Returns true if the only action for 'facet' is to send to the controller.
3561 * (We don't report NetFlow expiration messages for such facets because they
3562 * are just part of the control logic for the network, not real traffic). */
3564 facet_is_controller_flow(struct facet *facet)
3567 && facet->rule->up.n_actions == 1
3568 && action_outputs_to_port(&facet->rule->up.actions[0],
3569 htons(OFPP_CONTROLLER)));
3572 /* Folds all of 'facet''s statistics into its rule. Also updates the
3573 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3574 * 'facet''s statistics in the datapath should have been zeroed and folded into
3575 * its packet and byte counts before this function is called. */
3577 facet_flush_stats(struct facet *facet)
3579 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3580 struct subfacet *subfacet;
3582 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3583 assert(!subfacet->dp_byte_count);
3584 assert(!subfacet->dp_packet_count);
3587 facet_push_stats(facet);
3588 if (facet->accounted_bytes < facet->byte_count) {
3589 facet_account(facet);
3590 facet->accounted_bytes = facet->byte_count;
3593 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3594 struct ofexpired expired;
3595 expired.flow = facet->flow;
3596 expired.packet_count = facet->packet_count;
3597 expired.byte_count = facet->byte_count;
3598 expired.used = facet->used;
3599 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3602 facet->rule->packet_count += facet->packet_count;
3603 facet->rule->byte_count += facet->byte_count;
3605 /* Reset counters to prevent double counting if 'facet' ever gets
3607 facet_reset_counters(facet);
3609 netflow_flow_clear(&facet->nf_flow);
3610 facet->tcp_flags = 0;
3613 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3614 * Returns it if found, otherwise a null pointer.
3616 * 'hash' must be the return value of flow_hash(flow, 0).
3618 * The returned facet might need revalidation; use facet_lookup_valid()
3619 * instead if that is important. */
3620 static struct facet *
3621 facet_find(struct ofproto_dpif *ofproto,
3622 const struct flow *flow, uint32_t hash)
3624 struct facet *facet;
3626 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
3627 if (flow_equal(flow, &facet->flow)) {
3635 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3636 * Returns it if found, otherwise a null pointer.
3638 * 'hash' must be the return value of flow_hash(flow, 0).
3640 * The returned facet is guaranteed to be valid. */
3641 static struct facet *
3642 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
3645 struct facet *facet = facet_find(ofproto, flow, hash);
3647 /* The facet we found might not be valid, since we could be in need of
3648 * revalidation. If it is not valid, don't return it. */
3650 && (ofproto->need_revalidate
3651 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))
3652 && !facet_revalidate(facet)) {
3653 COVERAGE_INC(facet_invalidated);
3661 facet_check_consistency(struct facet *facet)
3663 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3665 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3667 uint64_t odp_actions_stub[1024 / 8];
3668 struct ofpbuf odp_actions;
3670 struct rule_dpif *rule;
3671 struct subfacet *subfacet;
3672 bool may_log = false;
3675 /* Check the rule for consistency. */
3676 rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3678 if (!VLOG_DROP_WARN(&rl)) {
3679 char *s = flow_to_string(&facet->flow);
3680 VLOG_WARN("%s: facet should not exist", s);
3684 } else if (rule != facet->rule) {
3685 may_log = !VLOG_DROP_WARN(&rl);
3691 flow_format(&s, &facet->flow);
3692 ds_put_format(&s, ": facet associated with wrong rule (was "
3693 "table=%"PRIu8",", facet->rule->up.table_id);
3694 cls_rule_format(&facet->rule->up.cr, &s);
3695 ds_put_format(&s, ") (should have been table=%"PRIu8",",
3697 cls_rule_format(&rule->up.cr, &s);
3698 ds_put_char(&s, ')');
3700 VLOG_WARN("%s", ds_cstr(&s));
3707 /* Check the datapath actions for consistency. */
3708 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
3709 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3710 struct action_xlate_ctx ctx;
3711 bool actions_changed;
3712 bool should_install;
3714 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3715 subfacet->initial_tci, rule, 0, NULL);
3716 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions,
3719 should_install = (ctx.may_set_up_flow
3720 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3721 if (!should_install && !subfacet->installed) {
3722 /* The actions for uninstallable flows may vary from one packet to
3723 * the next, so don't compare the actions. */
3727 actions_changed = (subfacet->actions_len != odp_actions.size
3728 || memcmp(subfacet->actions, odp_actions.data,
3729 subfacet->actions_len));
3730 if (should_install != subfacet->installed || actions_changed) {
3732 may_log = !VLOG_DROP_WARN(&rl);
3737 struct odputil_keybuf keybuf;
3742 subfacet_get_key(subfacet, &keybuf, &key);
3743 odp_flow_key_format(key.data, key.size, &s);
3745 ds_put_cstr(&s, ": inconsistency in subfacet");
3746 if (should_install != subfacet->installed) {
3747 enum odp_key_fitness fitness = subfacet->key_fitness;
3749 ds_put_format(&s, " (should%s have been installed)",
3750 should_install ? "" : " not");
3751 ds_put_format(&s, " (may_set_up_flow=%s, fitness=%s)",
3752 ctx.may_set_up_flow ? "true" : "false",
3753 odp_key_fitness_to_string(fitness));
3755 if (actions_changed) {
3756 ds_put_cstr(&s, " (actions were: ");
3757 format_odp_actions(&s, subfacet->actions,
3758 subfacet->actions_len);
3759 ds_put_cstr(&s, ") (correct actions: ");
3760 format_odp_actions(&s, odp_actions.data, odp_actions.size);
3761 ds_put_char(&s, ')');
3763 ds_put_cstr(&s, " (actions: ");
3764 format_odp_actions(&s, subfacet->actions,
3765 subfacet->actions_len);
3766 ds_put_char(&s, ')');
3768 VLOG_WARN("%s", ds_cstr(&s));
3773 ofpbuf_uninit(&odp_actions);
3778 /* Re-searches the classifier for 'facet':
3780 * - If the rule found is different from 'facet''s current rule, moves
3781 * 'facet' to the new rule and recompiles its actions.
3783 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3784 * where it is and recompiles its actions anyway.
3786 * - If there is none, destroys 'facet'.
3788 * Returns true if 'facet' still exists, false if it has been destroyed. */
3790 facet_revalidate(struct facet *facet)
3792 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3794 struct nlattr *odp_actions;
3797 struct actions *new_actions;
3799 struct action_xlate_ctx ctx;
3800 uint64_t odp_actions_stub[1024 / 8];
3801 struct ofpbuf odp_actions;
3803 struct rule_dpif *new_rule;
3804 struct subfacet *subfacet;
3805 bool actions_changed;
3808 COVERAGE_INC(facet_revalidate);
3810 /* Determine the new rule. */
3811 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3813 /* No new rule, so delete the facet. */
3814 facet_remove(facet);
3818 /* Calculate new datapath actions.
3820 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3821 * emit a NetFlow expiration and, if so, we need to have the old state
3822 * around to properly compose it. */
3824 /* If the datapath actions changed or the installability changed,
3825 * then we need to talk to the datapath. */
3828 memset(&ctx, 0, sizeof ctx);
3829 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
3830 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3831 bool should_install;
3833 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3834 subfacet->initial_tci, new_rule, 0, NULL);
3835 xlate_actions(&ctx, new_rule->up.actions, new_rule->up.n_actions,
3837 actions_changed = (subfacet->actions_len != odp_actions.size
3838 || memcmp(subfacet->actions, odp_actions.data,
3839 subfacet->actions_len));
3841 should_install = (ctx.may_set_up_flow
3842 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3843 if (actions_changed || should_install != subfacet->installed) {
3844 if (should_install) {
3845 struct dpif_flow_stats stats;
3847 subfacet_install(subfacet,
3848 odp_actions.data, odp_actions.size, &stats);
3849 subfacet_update_stats(subfacet, &stats);
3851 subfacet_uninstall(subfacet);
3855 new_actions = xcalloc(list_size(&facet->subfacets),
3856 sizeof *new_actions);
3858 new_actions[i].odp_actions = xmemdup(odp_actions.data,
3860 new_actions[i].actions_len = odp_actions.size;
3865 ofpbuf_uninit(&odp_actions);
3868 facet_flush_stats(facet);
3871 /* Update 'facet' now that we've taken care of all the old state. */
3872 facet->tags = ctx.tags;
3873 facet->nf_flow.output_iface = ctx.nf_output_iface;
3874 facet->may_install = ctx.may_set_up_flow;
3875 facet->has_learn = ctx.has_learn;
3876 facet->has_normal = ctx.has_normal;
3877 facet->has_fin_timeout = ctx.has_fin_timeout;
3878 facet->mirrors = ctx.mirrors;
3881 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3882 if (new_actions[i].odp_actions) {
3883 free(subfacet->actions);
3884 subfacet->actions = new_actions[i].odp_actions;
3885 subfacet->actions_len = new_actions[i].actions_len;
3891 if (facet->rule != new_rule) {
3892 COVERAGE_INC(facet_changed_rule);
3893 list_remove(&facet->list_node);
3894 list_push_back(&new_rule->facets, &facet->list_node);
3895 facet->rule = new_rule;
3896 facet->used = new_rule->up.created;
3897 facet->prev_used = facet->used;
3903 /* Updates 'facet''s used time. Caller is responsible for calling
3904 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3906 facet_update_time(struct facet *facet, long long int used)
3908 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3909 if (used > facet->used) {
3911 ofproto_rule_update_used(&facet->rule->up, used);
3912 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3917 facet_reset_counters(struct facet *facet)
3919 facet->packet_count = 0;
3920 facet->byte_count = 0;
3921 facet->prev_packet_count = 0;
3922 facet->prev_byte_count = 0;
3923 facet->accounted_bytes = 0;
3927 facet_push_stats(struct facet *facet)
3929 struct dpif_flow_stats stats;
3931 assert(facet->packet_count >= facet->prev_packet_count);
3932 assert(facet->byte_count >= facet->prev_byte_count);
3933 assert(facet->used >= facet->prev_used);
3935 stats.n_packets = facet->packet_count - facet->prev_packet_count;
3936 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
3937 stats.used = facet->used;
3938 stats.tcp_flags = 0;
3940 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
3941 facet->prev_packet_count = facet->packet_count;
3942 facet->prev_byte_count = facet->byte_count;
3943 facet->prev_used = facet->used;
3945 flow_push_stats(facet->rule, &facet->flow, &stats);
3947 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
3948 facet->mirrors, stats.n_packets, stats.n_bytes);
3953 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
3955 rule->packet_count += stats->n_packets;
3956 rule->byte_count += stats->n_bytes;
3957 ofproto_rule_update_used(&rule->up, stats->used);
3960 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3961 * 'rule''s actions and mirrors. */
3963 flow_push_stats(struct rule_dpif *rule,
3964 const struct flow *flow, const struct dpif_flow_stats *stats)
3966 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3967 struct action_xlate_ctx ctx;
3969 ofproto_rule_update_used(&rule->up, stats->used);
3971 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule,
3973 ctx.resubmit_stats = stats;
3974 xlate_actions_for_side_effects(&ctx, rule->up.actions, rule->up.n_actions);
3979 static struct subfacet *
3980 subfacet_find__(struct ofproto_dpif *ofproto,
3981 const struct nlattr *key, size_t key_len, uint32_t key_hash,
3982 const struct flow *flow)
3984 struct subfacet *subfacet;
3986 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
3987 &ofproto->subfacets) {
3989 ? (subfacet->key_len == key_len
3990 && !memcmp(key, subfacet->key, key_len))
3991 : flow_equal(flow, &subfacet->facet->flow)) {
3999 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4000 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
4001 * there is one, otherwise creates and returns a new subfacet.
4003 * If the returned subfacet is new, then subfacet->actions will be NULL, in
4004 * which case the caller must populate the actions with
4005 * subfacet_make_actions(). */
4006 static struct subfacet *
4007 subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness,
4008 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
4010 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4011 uint32_t key_hash = odp_flow_key_hash(key, key_len);
4012 struct subfacet *subfacet;
4014 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
4016 if (subfacet->facet == facet) {
4020 /* This shouldn't happen. */
4021 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4022 subfacet_destroy(subfacet);
4025 subfacet = (list_is_empty(&facet->subfacets)
4026 ? &facet->one_subfacet
4027 : xmalloc(sizeof *subfacet));
4028 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
4029 list_push_back(&facet->subfacets, &subfacet->list_node);
4030 subfacet->facet = facet;
4031 subfacet->key_fitness = key_fitness;
4032 if (key_fitness != ODP_FIT_PERFECT) {
4033 subfacet->key = xmemdup(key, key_len);
4034 subfacet->key_len = key_len;
4036 subfacet->key = NULL;
4037 subfacet->key_len = 0;
4039 subfacet->used = time_msec();
4040 subfacet->dp_packet_count = 0;
4041 subfacet->dp_byte_count = 0;
4042 subfacet->actions_len = 0;
4043 subfacet->actions = NULL;
4044 subfacet->installed = false;
4045 subfacet->initial_tci = initial_tci;
4050 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
4051 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
4052 static struct subfacet *
4053 subfacet_find(struct ofproto_dpif *ofproto,
4054 const struct nlattr *key, size_t key_len)
4056 uint32_t key_hash = odp_flow_key_hash(key, key_len);
4057 enum odp_key_fitness fitness;
4060 fitness = odp_flow_key_to_flow(key, key_len, &flow);
4061 if (fitness == ODP_FIT_ERROR) {
4065 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
4068 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4069 * its facet within 'ofproto', and frees it. */
4071 subfacet_destroy__(struct subfacet *subfacet)
4073 struct facet *facet = subfacet->facet;
4074 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4076 subfacet_uninstall(subfacet);
4077 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
4078 list_remove(&subfacet->list_node);
4079 free(subfacet->key);
4080 free(subfacet->actions);
4081 if (subfacet != &facet->one_subfacet) {
4086 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4087 * last remaining subfacet in its facet destroys the facet too. */
4089 subfacet_destroy(struct subfacet *subfacet)
4091 struct facet *facet = subfacet->facet;
4093 if (list_is_singleton(&facet->subfacets)) {
4094 /* facet_remove() needs at least one subfacet (it will remove it). */
4095 facet_remove(facet);
4097 subfacet_destroy__(subfacet);
4101 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
4102 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
4103 * for use as temporary storage. */
4105 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
4108 if (!subfacet->key) {
4109 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
4110 odp_flow_key_from_flow(key, &subfacet->facet->flow);
4112 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
4116 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
4117 * Translates the actions into 'odp_actions', which the caller must have
4118 * initialized and is responsible for uninitializing. */
4120 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
4121 struct ofpbuf *odp_actions)
4123 struct facet *facet = subfacet->facet;
4124 struct rule_dpif *rule = facet->rule;
4125 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4127 struct action_xlate_ctx ctx;
4129 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
4131 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions, odp_actions);
4132 facet->tags = ctx.tags;
4133 facet->may_install = ctx.may_set_up_flow;
4134 facet->has_learn = ctx.has_learn;
4135 facet->has_normal = ctx.has_normal;
4136 facet->has_fin_timeout = ctx.has_fin_timeout;
4137 facet->nf_flow.output_iface = ctx.nf_output_iface;
4138 facet->mirrors = ctx.mirrors;
4140 if (subfacet->actions_len != odp_actions->size
4141 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
4142 free(subfacet->actions);
4143 subfacet->actions_len = odp_actions->size;
4144 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
4148 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4149 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4150 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4151 * since 'subfacet' was last updated.
4153 * Returns 0 if successful, otherwise a positive errno value. */
4155 subfacet_install(struct subfacet *subfacet,
4156 const struct nlattr *actions, size_t actions_len,
4157 struct dpif_flow_stats *stats)
4159 struct facet *facet = subfacet->facet;
4160 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4161 struct odputil_keybuf keybuf;
4162 enum dpif_flow_put_flags flags;
4166 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
4168 flags |= DPIF_FP_ZERO_STATS;
4171 subfacet_get_key(subfacet, &keybuf, &key);
4172 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
4173 actions, actions_len, stats);
4176 subfacet_reset_dp_stats(subfacet, stats);
4182 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4184 subfacet_uninstall(struct subfacet *subfacet)
4186 if (subfacet->installed) {
4187 struct rule_dpif *rule = subfacet->facet->rule;
4188 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4189 struct odputil_keybuf keybuf;
4190 struct dpif_flow_stats stats;
4194 subfacet_get_key(subfacet, &keybuf, &key);
4195 error = dpif_flow_del(ofproto->dpif, key.data, key.size, &stats);
4196 subfacet_reset_dp_stats(subfacet, &stats);
4198 subfacet_update_stats(subfacet, &stats);
4200 subfacet->installed = false;
4202 assert(subfacet->dp_packet_count == 0);
4203 assert(subfacet->dp_byte_count == 0);
4207 /* Resets 'subfacet''s datapath statistics counters. This should be called
4208 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4209 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4210 * was reset in the datapath. 'stats' will be modified to include only
4211 * statistics new since 'subfacet' was last updated. */
4213 subfacet_reset_dp_stats(struct subfacet *subfacet,
4214 struct dpif_flow_stats *stats)
4217 && subfacet->dp_packet_count <= stats->n_packets
4218 && subfacet->dp_byte_count <= stats->n_bytes) {
4219 stats->n_packets -= subfacet->dp_packet_count;
4220 stats->n_bytes -= subfacet->dp_byte_count;
4223 subfacet->dp_packet_count = 0;
4224 subfacet->dp_byte_count = 0;
4227 /* Updates 'subfacet''s used time. The caller is responsible for calling
4228 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
4230 subfacet_update_time(struct subfacet *subfacet, long long int used)
4232 if (used > subfacet->used) {
4233 subfacet->used = used;
4234 facet_update_time(subfacet->facet, used);
4238 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4240 * Because of the meaning of a subfacet's counters, it only makes sense to do
4241 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4242 * represents a packet that was sent by hand or if it represents statistics
4243 * that have been cleared out of the datapath. */
4245 subfacet_update_stats(struct subfacet *subfacet,
4246 const struct dpif_flow_stats *stats)
4248 if (stats->n_packets || stats->used > subfacet->used) {
4249 struct facet *facet = subfacet->facet;
4251 subfacet_update_time(subfacet, stats->used);
4252 facet->packet_count += stats->n_packets;
4253 facet->byte_count += stats->n_bytes;
4254 facet->tcp_flags |= stats->tcp_flags;
4255 facet_push_stats(facet);
4256 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
4262 static struct rule_dpif *
4263 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4266 struct cls_rule *cls_rule;
4267 struct classifier *cls;
4269 if (table_id >= N_TABLES) {
4273 cls = &ofproto->up.tables[table_id].cls;
4274 if (flow->nw_frag & FLOW_NW_FRAG_ANY
4275 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4276 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
4277 * are unavailable. */
4278 struct flow ofpc_normal_flow = *flow;
4279 ofpc_normal_flow.tp_src = htons(0);
4280 ofpc_normal_flow.tp_dst = htons(0);
4281 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
4283 cls_rule = classifier_lookup(cls, flow);
4285 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
4289 complete_operation(struct rule_dpif *rule)
4291 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4293 rule_invalidate(rule);
4295 struct dpif_completion *c = xmalloc(sizeof *c);
4296 c->op = rule->up.pending;
4297 list_push_back(&ofproto->completions, &c->list_node);
4299 ofoperation_complete(rule->up.pending, 0);
4303 static struct rule *
4306 struct rule_dpif *rule = xmalloc(sizeof *rule);
4311 rule_dealloc(struct rule *rule_)
4313 struct rule_dpif *rule = rule_dpif_cast(rule_);
4318 rule_construct(struct rule *rule_)
4320 struct rule_dpif *rule = rule_dpif_cast(rule_);
4321 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4322 struct rule_dpif *victim;
4326 error = validate_actions(rule->up.actions, rule->up.n_actions,
4327 &rule->up.cr.flow, ofproto->max_ports);
4332 rule->packet_count = 0;
4333 rule->byte_count = 0;
4335 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
4336 if (victim && !list_is_empty(&victim->facets)) {
4337 struct facet *facet;
4339 rule->facets = victim->facets;
4340 list_moved(&rule->facets);
4341 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4342 /* XXX: We're only clearing our local counters here. It's possible
4343 * that quite a few packets are unaccounted for in the datapath
4344 * statistics. These will be accounted to the new rule instead of
4345 * cleared as required. This could be fixed by clearing out the
4346 * datapath statistics for this facet, but currently it doesn't
4348 facet_reset_counters(facet);
4352 /* Must avoid list_moved() in this case. */
4353 list_init(&rule->facets);
4356 table_id = rule->up.table_id;
4357 rule->tag = (victim ? victim->tag
4359 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
4360 ofproto->tables[table_id].basis));
4362 complete_operation(rule);
4367 rule_destruct(struct rule *rule_)
4369 struct rule_dpif *rule = rule_dpif_cast(rule_);
4370 struct facet *facet, *next_facet;
4372 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4373 facet_revalidate(facet);
4376 complete_operation(rule);
4380 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4382 struct rule_dpif *rule = rule_dpif_cast(rule_);
4383 struct facet *facet;
4385 /* Start from historical data for 'rule' itself that are no longer tracked
4386 * in facets. This counts, for example, facets that have expired. */
4387 *packets = rule->packet_count;
4388 *bytes = rule->byte_count;
4390 /* Add any statistics that are tracked by facets. This includes
4391 * statistical data recently updated by ofproto_update_stats() as well as
4392 * stats for packets that were executed "by hand" via dpif_execute(). */
4393 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4394 *packets += facet->packet_count;
4395 *bytes += facet->byte_count;
4400 rule_execute(struct rule *rule_, const struct flow *flow,
4401 struct ofpbuf *packet)
4403 struct rule_dpif *rule = rule_dpif_cast(rule_);
4404 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4406 struct dpif_flow_stats stats;
4408 struct action_xlate_ctx ctx;
4409 uint64_t odp_actions_stub[1024 / 8];
4410 struct ofpbuf odp_actions;
4412 dpif_flow_stats_extract(flow, packet, &stats);
4413 rule_credit_stats(rule, &stats);
4415 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4416 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
4417 rule, stats.tcp_flags, packet);
4418 ctx.resubmit_stats = &stats;
4419 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions, &odp_actions);
4421 execute_odp_actions(ofproto, flow, odp_actions.data,
4422 odp_actions.size, packet);
4424 ofpbuf_uninit(&odp_actions);
4430 rule_modify_actions(struct rule *rule_)
4432 struct rule_dpif *rule = rule_dpif_cast(rule_);
4433 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4436 error = validate_actions(rule->up.actions, rule->up.n_actions,
4437 &rule->up.cr.flow, ofproto->max_ports);
4439 ofoperation_complete(rule->up.pending, error);
4443 complete_operation(rule);
4446 /* Sends 'packet' out 'ofport'.
4447 * May modify 'packet'.
4448 * Returns 0 if successful, otherwise a positive errno value. */
4450 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4452 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4453 struct ofpbuf key, odp_actions;
4454 struct odputil_keybuf keybuf;
4459 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
4460 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4462 if (odp_port != ofport->odp_port) {
4463 eth_pop_vlan(packet);
4464 flow.vlan_tci = htons(0);
4467 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4468 odp_flow_key_from_flow(&key, &flow);
4470 ofpbuf_init(&odp_actions, 32);
4471 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4473 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4474 error = dpif_execute(ofproto->dpif,
4476 odp_actions.data, odp_actions.size,
4478 ofpbuf_uninit(&odp_actions);
4481 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4482 ofproto->up.name, odp_port, strerror(error));
4484 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
4488 /* OpenFlow to datapath action translation. */
4490 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
4491 struct action_xlate_ctx *ctx);
4492 static void xlate_normal(struct action_xlate_ctx *);
4495 put_userspace_action(const struct ofproto_dpif *ofproto,
4496 struct ofpbuf *odp_actions,
4497 const struct flow *flow,
4498 const struct user_action_cookie *cookie)
4502 pid = dpif_port_get_pid(ofproto->dpif,
4503 ofp_port_to_odp_port(flow->in_port));
4505 return odp_put_userspace_action(pid, cookie, odp_actions);
4508 /* Compose SAMPLE action for sFlow. */
4510 compose_sflow_action(const struct ofproto_dpif *ofproto,
4511 struct ofpbuf *odp_actions,
4512 const struct flow *flow,
4515 uint32_t port_ifindex;
4516 uint32_t probability;
4517 struct user_action_cookie cookie;
4518 size_t sample_offset, actions_offset;
4519 int cookie_offset, n_output;
4521 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4525 if (odp_port == OVSP_NONE) {
4529 port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4533 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4535 /* Number of packets out of UINT_MAX to sample. */
4536 probability = dpif_sflow_get_probability(ofproto->sflow);
4537 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4539 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4541 cookie.type = USER_ACTION_COOKIE_SFLOW;
4542 cookie.data = port_ifindex;
4543 cookie.n_output = n_output;
4544 cookie.vlan_tci = 0;
4545 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4547 nl_msg_end_nested(odp_actions, actions_offset);
4548 nl_msg_end_nested(odp_actions, sample_offset);
4549 return cookie_offset;
4552 /* SAMPLE action must be first action in any given list of actions.
4553 * At this point we do not have all information required to build it. So try to
4554 * build sample action as complete as possible. */
4556 add_sflow_action(struct action_xlate_ctx *ctx)
4558 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4560 &ctx->flow, OVSP_NONE);
4561 ctx->sflow_odp_port = 0;
4562 ctx->sflow_n_outputs = 0;
4565 /* Fix SAMPLE action according to data collected while composing ODP actions.
4566 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4567 * USERSPACE action's user-cookie which is required for sflow. */
4569 fix_sflow_action(struct action_xlate_ctx *ctx)
4571 const struct flow *base = &ctx->base_flow;
4572 struct user_action_cookie *cookie;
4574 if (!ctx->user_cookie_offset) {
4578 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4580 assert(cookie != NULL);
4581 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4583 if (ctx->sflow_n_outputs) {
4584 cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow,
4585 ctx->sflow_odp_port);
4587 if (ctx->sflow_n_outputs >= 255) {
4588 cookie->n_output = 255;
4590 cookie->n_output = ctx->sflow_n_outputs;
4592 cookie->vlan_tci = base->vlan_tci;
4596 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4599 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4600 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4601 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4602 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4606 struct priority_to_dscp *pdscp;
4608 if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD
4609 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4613 pdscp = get_priority(ofport, ctx->flow.skb_priority);
4615 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4616 ctx->flow.nw_tos |= pdscp->dscp;
4619 /* We may not have an ofport record for this port, but it doesn't hurt
4620 * to allow forwarding to it anyhow. Maybe such a port will appear
4621 * later and we're pre-populating the flow table. */
4624 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4625 ctx->flow.vlan_tci);
4626 if (out_port != odp_port) {
4627 ctx->flow.vlan_tci = htons(0);
4629 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4630 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4632 ctx->sflow_odp_port = odp_port;
4633 ctx->sflow_n_outputs++;
4634 ctx->nf_output_iface = ofp_port;
4635 ctx->flow.vlan_tci = flow_vlan_tci;
4636 ctx->flow.nw_tos = flow_nw_tos;
4640 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4642 compose_output_action__(ctx, ofp_port, true);
4646 xlate_table_action(struct action_xlate_ctx *ctx,
4647 uint16_t in_port, uint8_t table_id)
4649 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4650 struct ofproto_dpif *ofproto = ctx->ofproto;
4651 struct rule_dpif *rule;
4652 uint16_t old_in_port;
4653 uint8_t old_table_id;
4655 old_table_id = ctx->table_id;
4656 ctx->table_id = table_id;
4658 /* Look up a flow with 'in_port' as the input port. */
4659 old_in_port = ctx->flow.in_port;
4660 ctx->flow.in_port = in_port;
4661 rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
4664 if (table_id > 0 && table_id < N_TABLES) {
4665 struct table_dpif *table = &ofproto->tables[table_id];
4666 if (table->other_table) {
4667 ctx->tags |= (rule && rule->tag
4669 : rule_calculate_tag(&ctx->flow,
4670 &table->other_table->wc,
4675 /* Restore the original input port. Otherwise OFPP_NORMAL and
4676 * OFPP_IN_PORT will have surprising behavior. */
4677 ctx->flow.in_port = old_in_port;
4679 if (ctx->resubmit_hook) {
4680 ctx->resubmit_hook(ctx, rule);
4684 struct rule_dpif *old_rule = ctx->rule;
4686 if (ctx->resubmit_stats) {
4687 rule_credit_stats(rule, ctx->resubmit_stats);
4692 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
4693 ctx->rule = old_rule;
4697 ctx->table_id = old_table_id;
4699 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4701 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
4702 MAX_RESUBMIT_RECURSION);
4703 ctx->max_resubmit_trigger = true;
4708 xlate_resubmit_table(struct action_xlate_ctx *ctx,
4709 const struct nx_action_resubmit *nar)
4714 in_port = (nar->in_port == htons(OFPP_IN_PORT)
4716 : ntohs(nar->in_port));
4717 table_id = nar->table == 255 ? ctx->table_id : nar->table;
4719 xlate_table_action(ctx, in_port, table_id);
4723 flood_packets(struct action_xlate_ctx *ctx, bool all)
4725 struct ofport_dpif *ofport;
4727 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
4728 uint16_t ofp_port = ofport->up.ofp_port;
4730 if (ofp_port == ctx->flow.in_port) {
4735 compose_output_action__(ctx, ofp_port, false);
4736 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
4737 compose_output_action(ctx, ofp_port);
4741 ctx->nf_output_iface = NF_OUT_FLOOD;
4745 execute_controller_action(struct action_xlate_ctx *ctx, int len,
4746 enum ofp_packet_in_reason reason,
4747 uint16_t controller_id)
4749 struct ofputil_packet_in pin;
4750 struct ofpbuf *packet;
4752 ctx->may_set_up_flow = false;
4757 packet = ofpbuf_clone(ctx->packet);
4759 if (packet->l2 && packet->l3) {
4760 struct eth_header *eh;
4762 eth_pop_vlan(packet);
4765 /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
4766 * LLC frame. Calculating the Ethernet type of these frames is more
4767 * trouble than seems appropriate for a simple assertion. */
4768 assert(ntohs(eh->eth_type) < ETH_TYPE_MIN
4769 || eh->eth_type == ctx->flow.dl_type);
4771 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
4772 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
4774 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
4775 eth_push_vlan(packet, ctx->flow.vlan_tci);
4779 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4780 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
4781 ctx->flow.nw_tos, ctx->flow.nw_ttl);
4785 if (ctx->flow.nw_proto == IPPROTO_TCP) {
4786 packet_set_tcp_port(packet, ctx->flow.tp_src,
4788 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
4789 packet_set_udp_port(packet, ctx->flow.tp_src,
4796 pin.packet = packet->data;
4797 pin.packet_len = packet->size;
4798 pin.reason = reason;
4799 pin.controller_id = controller_id;
4800 pin.table_id = ctx->table_id;
4801 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
4804 flow_get_metadata(&ctx->flow, &pin.fmd);
4806 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
4807 ofpbuf_delete(packet);
4811 compose_dec_ttl(struct action_xlate_ctx *ctx)
4813 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
4814 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
4818 if (ctx->flow.nw_ttl > 1) {
4822 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
4824 /* Stop processing for current table. */
4830 xlate_output_action__(struct action_xlate_ctx *ctx,
4831 uint16_t port, uint16_t max_len)
4833 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
4835 ctx->nf_output_iface = NF_OUT_DROP;
4839 compose_output_action(ctx, ctx->flow.in_port);
4842 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
4848 flood_packets(ctx, false);
4851 flood_packets(ctx, true);
4853 case OFPP_CONTROLLER:
4854 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
4860 if (port != ctx->flow.in_port) {
4861 compose_output_action(ctx, port);
4866 if (prev_nf_output_iface == NF_OUT_FLOOD) {
4867 ctx->nf_output_iface = NF_OUT_FLOOD;
4868 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
4869 ctx->nf_output_iface = prev_nf_output_iface;
4870 } else if (prev_nf_output_iface != NF_OUT_DROP &&
4871 ctx->nf_output_iface != NF_OUT_FLOOD) {
4872 ctx->nf_output_iface = NF_OUT_MULTI;
4877 xlate_output_reg_action(struct action_xlate_ctx *ctx,
4878 const struct nx_action_output_reg *naor)
4880 struct mf_subfield src;
4883 nxm_decode(&src, naor->src, naor->ofs_nbits);
4884 ofp_port = mf_get_subfield(&src, &ctx->flow);
4886 if (ofp_port <= UINT16_MAX) {
4887 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
4892 xlate_output_action(struct action_xlate_ctx *ctx,
4893 const struct ofp_action_output *oao)
4895 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
4899 xlate_enqueue_action(struct action_xlate_ctx *ctx,
4900 const struct ofp_action_enqueue *oae)
4903 uint32_t flow_priority, priority;
4906 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
4909 /* Fall back to ordinary output action. */
4910 xlate_output_action__(ctx, ntohs(oae->port), 0);
4914 /* Figure out datapath output port. */
4915 ofp_port = ntohs(oae->port);
4916 if (ofp_port == OFPP_IN_PORT) {
4917 ofp_port = ctx->flow.in_port;
4918 } else if (ofp_port == ctx->flow.in_port) {
4922 /* Add datapath actions. */
4923 flow_priority = ctx->flow.skb_priority;
4924 ctx->flow.skb_priority = priority;
4925 compose_output_action(ctx, ofp_port);
4926 ctx->flow.skb_priority = flow_priority;
4928 /* Update NetFlow output port. */
4929 if (ctx->nf_output_iface == NF_OUT_DROP) {
4930 ctx->nf_output_iface = ofp_port;
4931 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
4932 ctx->nf_output_iface = NF_OUT_MULTI;
4937 xlate_set_queue_action(struct action_xlate_ctx *ctx,
4938 const struct nx_action_set_queue *nasq)
4943 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
4946 /* Couldn't translate queue to a priority, so ignore. A warning
4947 * has already been logged. */
4951 ctx->flow.skb_priority = priority;
4954 struct xlate_reg_state {
4960 xlate_autopath(struct action_xlate_ctx *ctx,
4961 const struct nx_action_autopath *naa)
4963 uint16_t ofp_port = ntohl(naa->id);
4964 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
4966 if (!port || !port->bundle) {
4967 ofp_port = OFPP_NONE;
4968 } else if (port->bundle->bond) {
4969 /* Autopath does not support VLAN hashing. */
4970 struct ofport_dpif *slave = bond_choose_output_slave(
4971 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
4973 ofp_port = slave->up.ofp_port;
4976 autopath_execute(naa, &ctx->flow, ofp_port);
4980 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
4982 struct ofproto_dpif *ofproto = ofproto_;
4983 struct ofport_dpif *port;
4993 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
4996 port = get_ofp_port(ofproto, ofp_port);
4997 return port ? port->may_enable : false;
5002 xlate_learn_action(struct action_xlate_ctx *ctx,
5003 const struct nx_action_learn *learn)
5005 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
5006 struct ofputil_flow_mod fm;
5009 learn_execute(learn, &ctx->flow, &fm);
5011 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
5012 if (error && !VLOG_DROP_WARN(&rl)) {
5013 VLOG_WARN("learning action failed to modify flow table (%s)",
5014 ofperr_get_name(error));
5020 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
5021 * means "infinite". */
5023 reduce_timeout(uint16_t max, uint16_t *timeout)
5025 if (max && (!*timeout || *timeout > max)) {
5031 xlate_fin_timeout(struct action_xlate_ctx *ctx,
5032 const struct nx_action_fin_timeout *naft)
5034 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
5035 struct rule_dpif *rule = ctx->rule;
5037 reduce_timeout(ntohs(naft->fin_idle_timeout), &rule->up.idle_timeout);
5038 reduce_timeout(ntohs(naft->fin_hard_timeout), &rule->up.hard_timeout);
5043 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
5045 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
5046 ? OFPUTIL_PC_NO_RECV_STP
5047 : OFPUTIL_PC_NO_RECV)) {
5051 /* Only drop packets here if both forwarding and learning are
5052 * disabled. If just learning is enabled, we need to have
5053 * OFPP_NORMAL and the learning action have a look at the packet
5054 * before we can drop it. */
5055 if (!stp_forward_in_state(port->stp_state)
5056 && !stp_learn_in_state(port->stp_state)) {
5064 do_xlate_actions(const union ofp_action *in, size_t n_in,
5065 struct action_xlate_ctx *ctx)
5067 const struct ofport_dpif *port;
5068 const union ofp_action *ia;
5069 bool was_evictable = true;
5072 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5073 if (port && !may_receive(port, ctx)) {
5074 /* Drop this flow. */
5079 /* Don't let the rule we're working on get evicted underneath us. */
5080 was_evictable = ctx->rule->up.evictable;
5081 ctx->rule->up.evictable = false;
5083 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
5084 const struct ofp_action_dl_addr *oada;
5085 const struct nx_action_resubmit *nar;
5086 const struct nx_action_set_tunnel *nast;
5087 const struct nx_action_set_queue *nasq;
5088 const struct nx_action_multipath *nam;
5089 const struct nx_action_autopath *naa;
5090 const struct nx_action_bundle *nab;
5091 const struct nx_action_output_reg *naor;
5092 const struct nx_action_controller *nac;
5093 enum ofputil_action_code code;
5100 code = ofputil_decode_action_unsafe(ia);
5102 case OFPUTIL_OFPAT10_OUTPUT:
5103 xlate_output_action(ctx, &ia->output);
5106 case OFPUTIL_OFPAT10_SET_VLAN_VID:
5107 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
5108 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
5111 case OFPUTIL_OFPAT10_SET_VLAN_PCP:
5112 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
5113 ctx->flow.vlan_tci |= htons(
5114 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
5117 case OFPUTIL_OFPAT10_STRIP_VLAN:
5118 ctx->flow.vlan_tci = htons(0);
5121 case OFPUTIL_OFPAT10_SET_DL_SRC:
5122 oada = ((struct ofp_action_dl_addr *) ia);
5123 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
5126 case OFPUTIL_OFPAT10_SET_DL_DST:
5127 oada = ((struct ofp_action_dl_addr *) ia);
5128 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
5131 case OFPUTIL_OFPAT10_SET_NW_SRC:
5132 ctx->flow.nw_src = ia->nw_addr.nw_addr;
5135 case OFPUTIL_OFPAT10_SET_NW_DST:
5136 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
5139 case OFPUTIL_OFPAT10_SET_NW_TOS:
5140 /* OpenFlow 1.0 only supports IPv4. */
5141 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5142 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5143 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
5147 case OFPUTIL_OFPAT10_SET_TP_SRC:
5148 ctx->flow.tp_src = ia->tp_port.tp_port;
5151 case OFPUTIL_OFPAT10_SET_TP_DST:
5152 ctx->flow.tp_dst = ia->tp_port.tp_port;
5155 case OFPUTIL_OFPAT10_ENQUEUE:
5156 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
5159 case OFPUTIL_NXAST_RESUBMIT:
5160 nar = (const struct nx_action_resubmit *) ia;
5161 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
5164 case OFPUTIL_NXAST_RESUBMIT_TABLE:
5165 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
5168 case OFPUTIL_NXAST_SET_TUNNEL:
5169 nast = (const struct nx_action_set_tunnel *) ia;
5170 tun_id = htonll(ntohl(nast->tun_id));
5171 ctx->flow.tun_id = tun_id;
5174 case OFPUTIL_NXAST_SET_QUEUE:
5175 nasq = (const struct nx_action_set_queue *) ia;
5176 xlate_set_queue_action(ctx, nasq);
5179 case OFPUTIL_NXAST_POP_QUEUE:
5180 ctx->flow.skb_priority = ctx->orig_skb_priority;
5183 case OFPUTIL_NXAST_REG_MOVE:
5184 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
5188 case OFPUTIL_NXAST_REG_LOAD:
5189 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
5193 case OFPUTIL_NXAST_NOTE:
5194 /* Nothing to do. */
5197 case OFPUTIL_NXAST_SET_TUNNEL64:
5198 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
5199 ctx->flow.tun_id = tun_id;
5202 case OFPUTIL_NXAST_MULTIPATH:
5203 nam = (const struct nx_action_multipath *) ia;
5204 multipath_execute(nam, &ctx->flow);
5207 case OFPUTIL_NXAST_AUTOPATH:
5208 naa = (const struct nx_action_autopath *) ia;
5209 xlate_autopath(ctx, naa);
5212 case OFPUTIL_NXAST_BUNDLE:
5213 ctx->ofproto->has_bundle_action = true;
5214 nab = (const struct nx_action_bundle *) ia;
5215 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
5220 case OFPUTIL_NXAST_BUNDLE_LOAD:
5221 ctx->ofproto->has_bundle_action = true;
5222 nab = (const struct nx_action_bundle *) ia;
5223 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
5227 case OFPUTIL_NXAST_OUTPUT_REG:
5228 naor = (const struct nx_action_output_reg *) ia;
5229 xlate_output_reg_action(ctx, naor);
5232 case OFPUTIL_NXAST_LEARN:
5233 ctx->has_learn = true;
5234 if (ctx->may_learn) {
5235 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
5239 case OFPUTIL_NXAST_DEC_TTL:
5240 if (compose_dec_ttl(ctx)) {
5245 case OFPUTIL_NXAST_EXIT:
5249 case OFPUTIL_NXAST_FIN_TIMEOUT:
5250 ctx->has_fin_timeout = true;
5251 xlate_fin_timeout(ctx, (const struct nx_action_fin_timeout *) ia);
5254 case OFPUTIL_NXAST_CONTROLLER:
5255 nac = (const struct nx_action_controller *) ia;
5256 execute_controller_action(ctx, ntohs(nac->max_len), nac->reason,
5257 ntohs(nac->controller_id));
5263 /* We've let OFPP_NORMAL and the learning action look at the packet,
5264 * so drop it now if forwarding is disabled. */
5265 if (port && !stp_forward_in_state(port->stp_state)) {
5266 ofpbuf_clear(ctx->odp_actions);
5267 add_sflow_action(ctx);
5270 ctx->rule->up.evictable = was_evictable;
5275 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
5276 struct ofproto_dpif *ofproto, const struct flow *flow,
5277 ovs_be16 initial_tci, struct rule_dpif *rule,
5278 uint8_t tcp_flags, const struct ofpbuf *packet)
5280 ctx->ofproto = ofproto;
5282 ctx->base_flow = ctx->flow;
5283 ctx->base_flow.tun_id = 0;
5284 ctx->base_flow.vlan_tci = initial_tci;
5286 ctx->packet = packet;
5287 ctx->may_learn = packet != NULL;
5288 ctx->tcp_flags = tcp_flags;
5289 ctx->resubmit_hook = NULL;
5290 ctx->resubmit_stats = NULL;
5293 /* Translates the 'n_in' "union ofp_action"s in 'in' into datapath actions in
5294 * 'odp_actions', using 'ctx'. */
5296 xlate_actions(struct action_xlate_ctx *ctx,
5297 const union ofp_action *in, size_t n_in,
5298 struct ofpbuf *odp_actions)
5300 struct flow orig_flow = ctx->flow;
5302 COVERAGE_INC(ofproto_dpif_xlate);
5304 ofpbuf_clear(odp_actions);
5305 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
5307 ctx->odp_actions = odp_actions;
5309 ctx->may_set_up_flow = true;
5310 ctx->has_learn = false;
5311 ctx->has_normal = false;
5312 ctx->has_fin_timeout = false;
5313 ctx->nf_output_iface = NF_OUT_DROP;
5316 ctx->max_resubmit_trigger = false;
5317 ctx->orig_skb_priority = ctx->flow.skb_priority;
5321 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
5322 switch (ctx->ofproto->up.frag_handling) {
5323 case OFPC_FRAG_NORMAL:
5324 /* We must pretend that transport ports are unavailable. */
5325 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
5326 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
5329 case OFPC_FRAG_DROP:
5332 case OFPC_FRAG_REASM:
5335 case OFPC_FRAG_NX_MATCH:
5336 /* Nothing to do. */
5339 case OFPC_INVALID_TTL_TO_CONTROLLER:
5344 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
5345 ctx->may_set_up_flow = false;
5347 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5348 struct flow original_flow = ctx->flow;
5349 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
5351 add_sflow_action(ctx);
5352 do_xlate_actions(in, n_in, ctx);
5354 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook
5355 && !VLOG_DROP_ERR(&trace_rl)) {
5356 struct ds ds = DS_EMPTY_INITIALIZER;
5358 ofproto_trace(ctx->ofproto, &original_flow, ctx->packet,
5360 VLOG_ERR("Trace triggered by excessive resubmit recursion:\n%s",
5365 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
5366 ctx->odp_actions->data,
5367 ctx->odp_actions->size)) {
5368 ctx->may_set_up_flow = false;
5370 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
5372 compose_output_action(ctx, OFPP_LOCAL);
5375 add_mirror_actions(ctx, &orig_flow);
5376 fix_sflow_action(ctx);
5380 /* Translates the 'n_in' "union ofp_action"s in 'in' into datapath actions,
5381 * using 'ctx', and discards the datapath actions. */
5383 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
5384 const union ofp_action *in, size_t n_in)
5386 uint64_t odp_actions_stub[1024 / 8];
5387 struct ofpbuf odp_actions;
5389 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5390 xlate_actions(ctx, in, n_in, &odp_actions);
5391 ofpbuf_uninit(&odp_actions);
5394 /* OFPP_NORMAL implementation. */
5396 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
5398 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
5399 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
5400 * the bundle on which the packet was received, returns the VLAN to which the
5403 * Both 'vid' and the return value are in the range 0...4095. */
5405 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
5407 switch (in_bundle->vlan_mode) {
5408 case PORT_VLAN_ACCESS:
5409 return in_bundle->vlan;
5412 case PORT_VLAN_TRUNK:
5415 case PORT_VLAN_NATIVE_UNTAGGED:
5416 case PORT_VLAN_NATIVE_TAGGED:
5417 return vid ? vid : in_bundle->vlan;
5424 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
5425 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
5428 * 'vid' should be the VID obtained from the 802.1Q header that was received as
5429 * part of a packet (specify 0 if there was no 802.1Q header), in the range
5432 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
5434 /* Allow any VID on the OFPP_NONE port. */
5435 if (in_bundle == &ofpp_none_bundle) {
5439 switch (in_bundle->vlan_mode) {
5440 case PORT_VLAN_ACCESS:
5443 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5444 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
5445 "packet received on port %s configured as VLAN "
5446 "%"PRIu16" access port",
5447 in_bundle->ofproto->up.name, vid,
5448 in_bundle->name, in_bundle->vlan);
5454 case PORT_VLAN_NATIVE_UNTAGGED:
5455 case PORT_VLAN_NATIVE_TAGGED:
5457 /* Port must always carry its native VLAN. */
5461 case PORT_VLAN_TRUNK:
5462 if (!ofbundle_includes_vlan(in_bundle, vid)) {
5464 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5465 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
5466 "received on port %s not configured for trunking "
5468 in_bundle->ofproto->up.name, vid,
5469 in_bundle->name, vid);
5481 /* Given 'vlan', the VLAN that a packet belongs to, and
5482 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
5483 * that should be included in the 802.1Q header. (If the return value is 0,
5484 * then the 802.1Q header should only be included in the packet if there is a
5487 * Both 'vlan' and the return value are in the range 0...4095. */
5489 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
5491 switch (out_bundle->vlan_mode) {
5492 case PORT_VLAN_ACCESS:
5495 case PORT_VLAN_TRUNK:
5496 case PORT_VLAN_NATIVE_TAGGED:
5499 case PORT_VLAN_NATIVE_UNTAGGED:
5500 return vlan == out_bundle->vlan ? 0 : vlan;
5508 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
5511 struct ofport_dpif *port;
5513 ovs_be16 tci, old_tci;
5515 vid = output_vlan_to_vid(out_bundle, vlan);
5516 if (!out_bundle->bond) {
5517 port = ofbundle_get_a_port(out_bundle);
5519 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
5522 /* No slaves enabled, so drop packet. */
5527 old_tci = ctx->flow.vlan_tci;
5529 if (tci || out_bundle->use_priority_tags) {
5530 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
5532 tci |= htons(VLAN_CFI);
5535 ctx->flow.vlan_tci = tci;
5537 compose_output_action(ctx, port->up.ofp_port);
5538 ctx->flow.vlan_tci = old_tci;
5542 mirror_mask_ffs(mirror_mask_t mask)
5544 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
5549 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
5551 return (bundle->vlan_mode != PORT_VLAN_ACCESS
5552 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
5556 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
5558 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
5561 /* Returns an arbitrary interface within 'bundle'. */
5562 static struct ofport_dpif *
5563 ofbundle_get_a_port(const struct ofbundle *bundle)
5565 return CONTAINER_OF(list_front(&bundle->ports),
5566 struct ofport_dpif, bundle_node);
5570 vlan_is_mirrored(const struct ofmirror *m, int vlan)
5572 return !m->vlans || bitmap_is_set(m->vlans, vlan);
5575 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
5576 * to a VLAN. In general most packets may be mirrored but we want to drop
5577 * protocols that may confuse switches. */
5579 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
5581 /* If you change this function's behavior, please update corresponding
5582 * documentation in vswitch.xml at the same time. */
5583 if (dst[0] != 0x01) {
5584 /* All the currently banned MACs happen to start with 01 currently, so
5585 * this is a quick way to eliminate most of the good ones. */
5587 if (eth_addr_is_reserved(dst)) {
5588 /* Drop STP, IEEE pause frames, and other reserved protocols
5589 * (01-80-c2-00-00-0x). */
5593 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
5595 if ((dst[3] & 0xfe) == 0xcc &&
5596 (dst[4] & 0xfe) == 0xcc &&
5597 (dst[5] & 0xfe) == 0xcc) {
5598 /* Drop the following protocols plus others following the same
5601 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
5602 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
5603 STP Uplink Fast (01-00-0c-cd-cd-cd) */
5607 if (!(dst[3] | dst[4] | dst[5])) {
5608 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
5617 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5619 struct ofproto_dpif *ofproto = ctx->ofproto;
5620 mirror_mask_t mirrors;
5621 struct ofbundle *in_bundle;
5624 const struct nlattr *a;
5627 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5628 ctx->packet != NULL, NULL);
5632 mirrors = in_bundle->src_mirrors;
5634 /* Drop frames on bundles reserved for mirroring. */
5635 if (in_bundle->mirror_out) {
5636 if (ctx->packet != NULL) {
5637 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5638 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5639 "%s, which is reserved exclusively for mirroring",
5640 ctx->ofproto->up.name, in_bundle->name);
5646 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5647 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5650 vlan = input_vid_to_vlan(in_bundle, vid);
5652 /* Look at the output ports to check for destination selections. */
5654 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5655 ctx->odp_actions->size) {
5656 enum ovs_action_attr type = nl_attr_type(a);
5657 struct ofport_dpif *ofport;
5659 if (type != OVS_ACTION_ATTR_OUTPUT) {
5663 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5664 if (ofport && ofport->bundle) {
5665 mirrors |= ofport->bundle->dst_mirrors;
5673 /* Restore the original packet before adding the mirror actions. */
5674 ctx->flow = *orig_flow;
5679 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5681 if (!vlan_is_mirrored(m, vlan)) {
5682 mirrors &= mirrors - 1;
5686 mirrors &= ~m->dup_mirrors;
5687 ctx->mirrors |= m->dup_mirrors;
5689 output_normal(ctx, m->out, vlan);
5690 } else if (eth_dst_may_rspan(orig_flow->dl_dst)
5691 && vlan != m->out_vlan) {
5692 struct ofbundle *bundle;
5694 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5695 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5696 && !bundle->mirror_out) {
5697 output_normal(ctx, bundle, m->out_vlan);
5705 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5706 uint64_t packets, uint64_t bytes)
5712 for (; mirrors; mirrors &= mirrors - 1) {
5715 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5718 /* In normal circumstances 'm' will not be NULL. However,
5719 * if mirrors are reconfigured, we can temporarily get out
5720 * of sync in facet_revalidate(). We could "correct" the
5721 * mirror list before reaching here, but doing that would
5722 * not properly account the traffic stats we've currently
5723 * accumulated for previous mirror configuration. */
5727 m->packet_count += packets;
5728 m->byte_count += bytes;
5732 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
5733 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
5734 * indicate this; newer upstream kernels use gratuitous ARP requests. */
5736 is_gratuitous_arp(const struct flow *flow)
5738 return (flow->dl_type == htons(ETH_TYPE_ARP)
5739 && eth_addr_is_broadcast(flow->dl_dst)
5740 && (flow->nw_proto == ARP_OP_REPLY
5741 || (flow->nw_proto == ARP_OP_REQUEST
5742 && flow->nw_src == flow->nw_dst)));
5746 update_learning_table(struct ofproto_dpif *ofproto,
5747 const struct flow *flow, int vlan,
5748 struct ofbundle *in_bundle)
5750 struct mac_entry *mac;
5752 /* Don't learn the OFPP_NONE port. */
5753 if (in_bundle == &ofpp_none_bundle) {
5757 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
5761 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
5762 if (is_gratuitous_arp(flow)) {
5763 /* We don't want to learn from gratuitous ARP packets that are
5764 * reflected back over bond slaves so we lock the learning table. */
5765 if (!in_bundle->bond) {
5766 mac_entry_set_grat_arp_lock(mac);
5767 } else if (mac_entry_is_grat_arp_locked(mac)) {
5772 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
5773 /* The log messages here could actually be useful in debugging,
5774 * so keep the rate limit relatively high. */
5775 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5776 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
5777 "on port %s in VLAN %d",
5778 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
5779 in_bundle->name, vlan);
5781 mac->port.p = in_bundle;
5782 tag_set_add(&ofproto->revalidate_set,
5783 mac_learning_changed(ofproto->ml, mac));
5787 static struct ofbundle *
5788 lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn,
5789 struct ofport_dpif **in_ofportp)
5791 struct ofport_dpif *ofport;
5793 /* Find the port and bundle for the received packet. */
5794 ofport = get_ofp_port(ofproto, in_port);
5796 *in_ofportp = ofport;
5798 if (ofport && ofport->bundle) {
5799 return ofport->bundle;
5802 /* Special-case OFPP_NONE, which a controller may use as the ingress
5803 * port for traffic that it is sourcing. */
5804 if (in_port == OFPP_NONE) {
5805 return &ofpp_none_bundle;
5808 /* Odd. A few possible reasons here:
5810 * - We deleted a port but there are still a few packets queued up
5813 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
5814 * we don't know about.
5816 * - The ofproto client didn't configure the port as part of a bundle.
5819 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5821 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
5822 "port %"PRIu16, ofproto->up.name, in_port);
5827 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
5828 * dropped. Returns true if they may be forwarded, false if they should be
5831 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
5832 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
5834 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
5835 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
5836 * checked by input_vid_is_valid().
5838 * May also add tags to '*tags', although the current implementation only does
5839 * so in one special case.
5842 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
5843 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
5845 struct ofbundle *in_bundle = in_port->bundle;
5847 /* Drop frames for reserved multicast addresses
5848 * only if forward_bpdu option is absent. */
5849 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
5853 if (in_bundle->bond) {
5854 struct mac_entry *mac;
5856 switch (bond_check_admissibility(in_bundle->bond, in_port,
5857 flow->dl_dst, tags)) {
5864 case BV_DROP_IF_MOVED:
5865 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
5866 if (mac && mac->port.p != in_bundle &&
5867 (!is_gratuitous_arp(flow)
5868 || mac_entry_is_grat_arp_locked(mac))) {
5879 xlate_normal(struct action_xlate_ctx *ctx)
5881 struct ofport_dpif *in_port;
5882 struct ofbundle *in_bundle;
5883 struct mac_entry *mac;
5887 ctx->has_normal = true;
5889 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
5890 ctx->packet != NULL, &in_port);
5895 /* Drop malformed frames. */
5896 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
5897 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
5898 if (ctx->packet != NULL) {
5899 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5900 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
5901 "VLAN tag received on port %s",
5902 ctx->ofproto->up.name, in_bundle->name);
5907 /* Drop frames on bundles reserved for mirroring. */
5908 if (in_bundle->mirror_out) {
5909 if (ctx->packet != NULL) {
5910 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5911 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5912 "%s, which is reserved exclusively for mirroring",
5913 ctx->ofproto->up.name, in_bundle->name);
5919 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
5920 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5923 vlan = input_vid_to_vlan(in_bundle, vid);
5925 /* Check other admissibility requirements. */
5927 !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
5931 /* Learn source MAC. */
5932 if (ctx->may_learn) {
5933 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
5936 /* Determine output bundle. */
5937 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
5940 if (mac->port.p != in_bundle) {
5941 output_normal(ctx, mac->port.p, vlan);
5944 struct ofbundle *bundle;
5946 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
5947 if (bundle != in_bundle
5948 && ofbundle_includes_vlan(bundle, vlan)
5949 && bundle->floodable
5950 && !bundle->mirror_out) {
5951 output_normal(ctx, bundle, vlan);
5954 ctx->nf_output_iface = NF_OUT_FLOOD;
5958 /* Optimized flow revalidation.
5960 * It's a difficult problem, in general, to tell which facets need to have
5961 * their actions recalculated whenever the OpenFlow flow table changes. We
5962 * don't try to solve that general problem: for most kinds of OpenFlow flow
5963 * table changes, we recalculate the actions for every facet. This is
5964 * relatively expensive, but it's good enough if the OpenFlow flow table
5965 * doesn't change very often.
5967 * However, we can expect one particular kind of OpenFlow flow table change to
5968 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5969 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5970 * table, we add a special case that applies to flow tables in which every rule
5971 * has the same form (that is, the same wildcards), except that the table is
5972 * also allowed to have a single "catch-all" flow that matches all packets. We
5973 * optimize this case by tagging all of the facets that resubmit into the table
5974 * and invalidating the same tag whenever a flow changes in that table. The
5975 * end result is that we revalidate just the facets that need it (and sometimes
5976 * a few more, but not all of the facets or even all of the facets that
5977 * resubmit to the table modified by MAC learning). */
5979 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
5980 * into an OpenFlow table with the given 'basis'. */
5982 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
5985 if (flow_wildcards_is_catchall(wc)) {
5988 struct flow tag_flow = *flow;
5989 flow_zero_wildcards(&tag_flow, wc);
5990 return tag_create_deterministic(flow_hash(&tag_flow, secret));
5994 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5995 * taggability of that table.
5997 * This function must be called after *each* change to a flow table. If you
5998 * skip calling it on some changes then the pointer comparisons at the end can
5999 * be invalid if you get unlucky. For example, if a flow removal causes a
6000 * cls_table to be destroyed and then a flow insertion causes a cls_table with
6001 * different wildcards to be created with the same address, then this function
6002 * will incorrectly skip revalidation. */
6004 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
6006 struct table_dpif *table = &ofproto->tables[table_id];
6007 const struct oftable *oftable = &ofproto->up.tables[table_id];
6008 struct cls_table *catchall, *other;
6009 struct cls_table *t;
6011 catchall = other = NULL;
6013 switch (hmap_count(&oftable->cls.tables)) {
6015 /* We could tag this OpenFlow table but it would make the logic a
6016 * little harder and it's a corner case that doesn't seem worth it
6022 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
6023 if (cls_table_is_catchall(t)) {
6025 } else if (!other) {
6028 /* Indicate that we can't tag this by setting both tables to
6029 * NULL. (We know that 'catchall' is already NULL.) */
6036 /* Can't tag this table. */
6040 if (table->catchall_table != catchall || table->other_table != other) {
6041 table->catchall_table = catchall;
6042 table->other_table = other;
6043 ofproto->need_revalidate = true;
6047 /* Given 'rule' that has changed in some way (either it is a rule being
6048 * inserted, a rule being deleted, or a rule whose actions are being
6049 * modified), marks facets for revalidation to ensure that packets will be
6050 * forwarded correctly according to the new state of the flow table.
6052 * This function must be called after *each* change to a flow table. See
6053 * the comment on table_update_taggable() for more information. */
6055 rule_invalidate(const struct rule_dpif *rule)
6057 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
6059 table_update_taggable(ofproto, rule->up.table_id);
6061 if (!ofproto->need_revalidate) {
6062 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
6064 if (table->other_table && rule->tag) {
6065 tag_set_add(&ofproto->revalidate_set, rule->tag);
6067 ofproto->need_revalidate = true;
6073 set_frag_handling(struct ofproto *ofproto_,
6074 enum ofp_config_flags frag_handling)
6076 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6078 if (frag_handling != OFPC_FRAG_REASM) {
6079 ofproto->need_revalidate = true;
6087 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
6088 const struct flow *flow,
6089 const union ofp_action *ofp_actions, size_t n_ofp_actions)
6091 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6094 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
6095 return OFPERR_NXBRC_BAD_IN_PORT;
6098 error = validate_actions(ofp_actions, n_ofp_actions, flow,
6099 ofproto->max_ports);
6101 struct odputil_keybuf keybuf;
6102 struct dpif_flow_stats stats;
6106 struct action_xlate_ctx ctx;
6107 uint64_t odp_actions_stub[1024 / 8];
6108 struct ofpbuf odp_actions;
6110 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
6111 odp_flow_key_from_flow(&key, flow);
6113 dpif_flow_stats_extract(flow, packet, &stats);
6115 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL,
6116 packet_get_tcp_flags(packet, flow), packet);
6117 ctx.resubmit_stats = &stats;
6119 ofpbuf_use_stub(&odp_actions,
6120 odp_actions_stub, sizeof odp_actions_stub);
6121 xlate_actions(&ctx, ofp_actions, n_ofp_actions, &odp_actions);
6122 dpif_execute(ofproto->dpif, key.data, key.size,
6123 odp_actions.data, odp_actions.size, packet);
6124 ofpbuf_uninit(&odp_actions);
6132 set_netflow(struct ofproto *ofproto_,
6133 const struct netflow_options *netflow_options)
6135 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6137 if (netflow_options) {
6138 if (!ofproto->netflow) {
6139 ofproto->netflow = netflow_create();
6141 return netflow_set_options(ofproto->netflow, netflow_options);
6143 netflow_destroy(ofproto->netflow);
6144 ofproto->netflow = NULL;
6150 get_netflow_ids(const struct ofproto *ofproto_,
6151 uint8_t *engine_type, uint8_t *engine_id)
6153 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6155 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
6159 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
6161 if (!facet_is_controller_flow(facet) &&
6162 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
6163 struct subfacet *subfacet;
6164 struct ofexpired expired;
6166 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
6167 if (subfacet->installed) {
6168 struct dpif_flow_stats stats;
6170 subfacet_install(subfacet, subfacet->actions,
6171 subfacet->actions_len, &stats);
6172 subfacet_update_stats(subfacet, &stats);
6176 expired.flow = facet->flow;
6177 expired.packet_count = facet->packet_count;
6178 expired.byte_count = facet->byte_count;
6179 expired.used = facet->used;
6180 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
6185 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
6187 struct facet *facet;
6189 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6190 send_active_timeout(ofproto, facet);
6194 static struct ofproto_dpif *
6195 ofproto_dpif_lookup(const char *name)
6197 struct ofproto_dpif *ofproto;
6199 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
6200 hash_string(name, 0), &all_ofproto_dpifs) {
6201 if (!strcmp(ofproto->up.name, name)) {
6209 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
6210 const char *argv[], void *aux OVS_UNUSED)
6212 struct ofproto_dpif *ofproto;
6215 ofproto = ofproto_dpif_lookup(argv[1]);
6217 unixctl_command_reply_error(conn, "no such bridge");
6220 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6222 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6223 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6227 unixctl_command_reply(conn, "table successfully flushed");
6231 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6232 const char *argv[], void *aux OVS_UNUSED)
6234 struct ds ds = DS_EMPTY_INITIALIZER;
6235 const struct ofproto_dpif *ofproto;
6236 const struct mac_entry *e;
6238 ofproto = ofproto_dpif_lookup(argv[1]);
6240 unixctl_command_reply_error(conn, "no such bridge");
6244 ds_put_cstr(&ds, " port VLAN MAC Age\n");
6245 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
6246 struct ofbundle *bundle = e->port.p;
6247 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
6248 ofbundle_get_a_port(bundle)->odp_port,
6249 e->vlan, ETH_ADDR_ARGS(e->mac),
6250 mac_entry_age(ofproto->ml, e));
6252 unixctl_command_reply(conn, ds_cstr(&ds));
6257 struct action_xlate_ctx ctx;
6263 trace_format_rule(struct ds *result, uint8_t table_id, int level,
6264 const struct rule_dpif *rule)
6266 ds_put_char_multiple(result, '\t', level);
6268 ds_put_cstr(result, "No match\n");
6272 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
6273 table_id, ntohll(rule->up.flow_cookie));
6274 cls_rule_format(&rule->up.cr, result);
6275 ds_put_char(result, '\n');
6277 ds_put_char_multiple(result, '\t', level);
6278 ds_put_cstr(result, "OpenFlow ");
6279 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
6280 ds_put_char(result, '\n');
6284 trace_format_flow(struct ds *result, int level, const char *title,
6285 struct trace_ctx *trace)
6287 ds_put_char_multiple(result, '\t', level);
6288 ds_put_format(result, "%s: ", title);
6289 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
6290 ds_put_cstr(result, "unchanged");
6292 flow_format(result, &trace->ctx.flow);
6293 trace->flow = trace->ctx.flow;
6295 ds_put_char(result, '\n');
6299 trace_format_regs(struct ds *result, int level, const char *title,
6300 struct trace_ctx *trace)
6304 ds_put_char_multiple(result, '\t', level);
6305 ds_put_format(result, "%s:", title);
6306 for (i = 0; i < FLOW_N_REGS; i++) {
6307 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
6309 ds_put_char(result, '\n');
6313 trace_format_odp(struct ds *result, int level, const char *title,
6314 struct trace_ctx *trace)
6316 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
6318 ds_put_char_multiple(result, '\t', level);
6319 ds_put_format(result, "%s: ", title);
6320 format_odp_actions(result, odp_actions->data, odp_actions->size);
6321 ds_put_char(result, '\n');
6325 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
6327 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
6328 struct ds *result = trace->result;
6330 ds_put_char(result, '\n');
6331 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
6332 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
6333 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
6334 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
6338 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
6339 void *aux OVS_UNUSED)
6341 const char *dpname = argv[1];
6342 struct ofproto_dpif *ofproto;
6343 struct ofpbuf odp_key;
6344 struct ofpbuf *packet;
6345 ovs_be16 initial_tci;
6351 ofpbuf_init(&odp_key, 0);
6354 ofproto = ofproto_dpif_lookup(dpname);
6356 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
6360 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
6361 /* ofproto/trace dpname flow [-generate] */
6362 const char *flow_s = argv[2];
6363 const char *generate_s = argv[3];
6366 /* Convert string to datapath key. */
6367 ofpbuf_init(&odp_key, 0);
6368 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
6370 unixctl_command_reply_error(conn, "Bad flow syntax");
6374 /* Convert odp_key to flow. */
6375 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
6376 odp_key.size, &flow,
6377 &initial_tci, NULL);
6378 if (error == ODP_FIT_ERROR) {
6379 unixctl_command_reply_error(conn, "Invalid flow");
6383 /* Generate a packet, if requested. */
6385 packet = ofpbuf_new(0);
6386 flow_compose(packet, &flow);
6388 } else if (argc == 6) {
6389 /* ofproto/trace dpname priority tun_id in_port packet */
6390 const char *priority_s = argv[2];
6391 const char *tun_id_s = argv[3];
6392 const char *in_port_s = argv[4];
6393 const char *packet_s = argv[5];
6394 uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s));
6395 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
6396 uint32_t priority = atoi(priority_s);
6399 msg = eth_from_hex(packet_s, &packet);
6401 unixctl_command_reply_error(conn, msg);
6405 ds_put_cstr(&result, "Packet: ");
6406 s = ofp_packet_to_string(packet->data, packet->size);
6407 ds_put_cstr(&result, s);
6410 flow_extract(packet, priority, tun_id, in_port, &flow);
6411 initial_tci = flow.vlan_tci;
6413 unixctl_command_reply_error(conn, "Bad command syntax");
6417 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
6418 unixctl_command_reply(conn, ds_cstr(&result));
6421 ds_destroy(&result);
6422 ofpbuf_delete(packet);
6423 ofpbuf_uninit(&odp_key);
6427 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
6428 const struct ofpbuf *packet, ovs_be16 initial_tci,
6431 struct rule_dpif *rule;
6433 ds_put_cstr(ds, "Flow: ");
6434 flow_format(ds, flow);
6435 ds_put_char(ds, '\n');
6437 rule = rule_dpif_lookup(ofproto, flow, 0);
6438 trace_format_rule(ds, 0, 0, rule);
6440 uint64_t odp_actions_stub[1024 / 8];
6441 struct ofpbuf odp_actions;
6443 struct trace_ctx trace;
6446 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
6449 ofpbuf_use_stub(&odp_actions,
6450 odp_actions_stub, sizeof odp_actions_stub);
6451 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
6452 rule, tcp_flags, packet);
6453 trace.ctx.resubmit_hook = trace_resubmit;
6454 xlate_actions(&trace.ctx, rule->up.actions, rule->up.n_actions,
6457 ds_put_char(ds, '\n');
6458 trace_format_flow(ds, 0, "Final flow", &trace);
6459 ds_put_cstr(ds, "Datapath actions: ");
6460 format_odp_actions(ds, odp_actions.data, odp_actions.size);
6461 ofpbuf_uninit(&odp_actions);
6463 if (!trace.ctx.may_set_up_flow) {
6465 ds_put_cstr(ds, "\nThis flow is not cachable.");
6467 ds_put_cstr(ds, "\nThe datapath actions are incomplete--"
6468 "for complete actions, please supply a packet.");
6475 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6476 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6479 unixctl_command_reply(conn, NULL);
6483 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6484 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6487 unixctl_command_reply(conn, NULL);
6490 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
6491 * 'reply' describing the results. */
6493 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
6495 struct facet *facet;
6499 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6500 if (!facet_check_consistency(facet)) {
6505 ofproto->need_revalidate = true;
6509 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
6510 ofproto->up.name, errors);
6512 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
6517 ofproto_dpif_self_check(struct unixctl_conn *conn,
6518 int argc, const char *argv[], void *aux OVS_UNUSED)
6520 struct ds reply = DS_EMPTY_INITIALIZER;
6521 struct ofproto_dpif *ofproto;
6524 ofproto = ofproto_dpif_lookup(argv[1]);
6526 unixctl_command_reply_error(conn, "Unknown ofproto (use "
6527 "ofproto/list for help)");
6530 ofproto_dpif_self_check__(ofproto, &reply);
6532 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6533 ofproto_dpif_self_check__(ofproto, &reply);
6537 unixctl_command_reply(conn, ds_cstr(&reply));
6542 ofproto_dpif_unixctl_init(void)
6544 static bool registered;
6550 unixctl_command_register(
6552 "bridge {tun_id in_port packet | odp_flow [-generate]}",
6553 2, 5, ofproto_unixctl_trace, NULL);
6554 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6555 ofproto_unixctl_fdb_flush, NULL);
6556 unixctl_command_register("fdb/show", "bridge", 1, 1,
6557 ofproto_unixctl_fdb_show, NULL);
6558 unixctl_command_register("ofproto/clog", "", 0, 0,
6559 ofproto_dpif_clog, NULL);
6560 unixctl_command_register("ofproto/unclog", "", 0, 0,
6561 ofproto_dpif_unclog, NULL);
6562 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6563 ofproto_dpif_self_check, NULL);
6566 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6568 * This is deprecated. It is only for compatibility with broken device drivers
6569 * in old versions of Linux that do not properly support VLANs when VLAN
6570 * devices are not used. When broken device drivers are no longer in
6571 * widespread use, we will delete these interfaces. */
6574 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
6576 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6577 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6579 if (realdev_ofp_port == ofport->realdev_ofp_port
6580 && vid == ofport->vlandev_vid) {
6584 ofproto->need_revalidate = true;
6586 if (ofport->realdev_ofp_port) {
6589 if (realdev_ofp_port && ofport->bundle) {
6590 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6591 * themselves be part of a bundle. */
6592 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6595 ofport->realdev_ofp_port = realdev_ofp_port;
6596 ofport->vlandev_vid = vid;
6598 if (realdev_ofp_port) {
6599 vsp_add(ofport, realdev_ofp_port, vid);
6606 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
6608 return hash_2words(realdev_ofp_port, vid);
6611 /* Returns the ODP port number of the Linux VLAN device that corresponds to
6612 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
6613 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
6614 * it would return the port number of eth0.9.
6616 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
6617 * function just returns its 'realdev_odp_port' argument. */
6619 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6620 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
6622 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6623 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
6624 int vid = vlan_tci_to_vid(vlan_tci);
6625 const struct vlan_splinter *vsp;
6627 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6628 hash_realdev_vid(realdev_ofp_port, vid),
6629 &ofproto->realdev_vid_map) {
6630 if (vsp->realdev_ofp_port == realdev_ofp_port
6631 && vsp->vid == vid) {
6632 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
6636 return realdev_odp_port;
6639 static struct vlan_splinter *
6640 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
6642 struct vlan_splinter *vsp;
6644 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
6645 &ofproto->vlandev_map) {
6646 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6654 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6655 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6656 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6657 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6658 * eth0 and store 9 in '*vid'.
6660 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6661 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6664 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6665 uint16_t vlandev_ofp_port, int *vid)
6667 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6668 const struct vlan_splinter *vsp;
6670 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6675 return vsp->realdev_ofp_port;
6682 vsp_remove(struct ofport_dpif *port)
6684 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6685 struct vlan_splinter *vsp;
6687 vsp = vlandev_find(ofproto, port->up.ofp_port);
6689 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6690 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6693 port->realdev_ofp_port = 0;
6695 VLOG_ERR("missing vlan device record");
6700 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
6702 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6704 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6705 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6706 == realdev_ofp_port)) {
6707 struct vlan_splinter *vsp;
6709 vsp = xmalloc(sizeof *vsp);
6710 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6711 hash_int(port->up.ofp_port, 0));
6712 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6713 hash_realdev_vid(realdev_ofp_port, vid));
6714 vsp->realdev_ofp_port = realdev_ofp_port;
6715 vsp->vlandev_ofp_port = port->up.ofp_port;
6718 port->realdev_ofp_port = realdev_ofp_port;
6720 VLOG_ERR("duplicate vlan device record");
6724 const struct ofproto_class ofproto_dpif_class = {
6753 port_is_lacp_current,
6754 NULL, /* rule_choose_table */
6761 rule_modify_actions,
6769 get_cfm_remote_mpids,
6774 get_stp_port_status,
6781 is_mirror_output_bundle,
6782 forward_bpdu_changed,