2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
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"
25 #include "byte-order.h"
30 #include "dynamic-string.h"
31 #include "fail-open.h"
35 #include "mac-learning.h"
36 #include "meta-flow.h"
37 #include "multipath.h"
38 #include "netdev-vport.h"
45 #include "ofp-actions.h"
46 #include "ofp-parse.h"
47 #include "ofp-print.h"
48 #include "ofproto-dpif-governor.h"
49 #include "ofproto-dpif-sflow.h"
50 #include "poll-loop.h"
55 #include "unaligned.h"
57 #include "vlan-bitmap.h"
60 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
62 COVERAGE_DEFINE(ofproto_dpif_expired);
63 COVERAGE_DEFINE(ofproto_dpif_xlate);
64 COVERAGE_DEFINE(facet_changed_rule);
65 COVERAGE_DEFINE(facet_revalidate);
66 COVERAGE_DEFINE(facet_unexpected);
67 COVERAGE_DEFINE(facet_suppress);
69 /* Maximum depth of flow table recursion (due to resubmit actions) in a
70 * flow translation. */
71 #define MAX_RESUBMIT_RECURSION 64
73 /* Number of implemented OpenFlow tables. */
74 enum { N_TABLES = 255 };
75 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
76 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
87 * - Do include packets and bytes from facets that have been deleted or
88 * whose own statistics have been folded into the rule.
90 * - Do include packets and bytes sent "by hand" that were accounted to
91 * the rule without any facet being involved (this is a rare corner
92 * case in rule_execute()).
94 * - Do not include packet or bytes that can be obtained from any facet's
95 * packet_count or byte_count member or that can be obtained from the
96 * datapath by, e.g., dpif_flow_get() for any subfacet.
98 uint64_t packet_count; /* Number of packets received. */
99 uint64_t byte_count; /* Number of bytes received. */
101 tag_type tag; /* Caches rule_calculate_tag() result. */
103 struct list facets; /* List of "struct facet"s. */
106 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
108 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
111 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
112 const struct flow *);
113 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
116 static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto,
117 const struct flow *flow);
119 static void rule_credit_stats(struct rule_dpif *,
120 const struct dpif_flow_stats *);
121 static void flow_push_stats(struct rule_dpif *, const struct flow *,
122 const struct dpif_flow_stats *);
123 static tag_type rule_calculate_tag(const struct flow *,
124 const struct minimask *, uint32_t basis);
125 static void rule_invalidate(const struct rule_dpif *);
127 #define MAX_MIRRORS 32
128 typedef uint32_t mirror_mask_t;
129 #define MIRROR_MASK_C(X) UINT32_C(X)
130 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
132 struct ofproto_dpif *ofproto; /* Owning ofproto. */
133 size_t idx; /* In ofproto's "mirrors" array. */
134 void *aux; /* Key supplied by ofproto's client. */
135 char *name; /* Identifier for log messages. */
137 /* Selection criteria. */
138 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
139 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
140 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
142 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
143 struct ofbundle *out; /* Output port or NULL. */
144 int out_vlan; /* Output VLAN or -1. */
145 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
148 int64_t packet_count; /* Number of packets sent. */
149 int64_t byte_count; /* Number of bytes sent. */
152 static void mirror_destroy(struct ofmirror *);
153 static void update_mirror_stats(struct ofproto_dpif *ofproto,
154 mirror_mask_t mirrors,
155 uint64_t packets, uint64_t bytes);
158 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
159 struct ofproto_dpif *ofproto; /* Owning ofproto. */
160 void *aux; /* Key supplied by ofproto's client. */
161 char *name; /* Identifier for log messages. */
164 struct list ports; /* Contains "struct ofport"s. */
165 enum port_vlan_mode vlan_mode; /* VLAN mode */
166 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
167 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
168 * NULL if all VLANs are trunked. */
169 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
170 struct bond *bond; /* Nonnull iff more than one port. */
171 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
174 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
176 /* Port mirroring info. */
177 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
178 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
179 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
182 static void bundle_remove(struct ofport *);
183 static void bundle_update(struct ofbundle *);
184 static void bundle_destroy(struct ofbundle *);
185 static void bundle_del_port(struct ofport_dpif *);
186 static void bundle_run(struct ofbundle *);
187 static void bundle_wait(struct ofbundle *);
188 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
189 uint16_t in_port, bool warn,
190 struct ofport_dpif **in_ofportp);
192 /* A controller may use OFPP_NONE as the ingress port to indicate that
193 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
194 * when an input bundle is needed for validation (e.g., mirroring or
195 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
196 * any 'port' structs, so care must be taken when dealing with it. */
197 static struct ofbundle ofpp_none_bundle = {
199 .vlan_mode = PORT_VLAN_TRUNK
202 static void stp_run(struct ofproto_dpif *ofproto);
203 static void stp_wait(struct ofproto_dpif *ofproto);
204 static int set_stp_port(struct ofport *,
205 const struct ofproto_port_stp_settings *);
207 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
209 struct action_xlate_ctx {
210 /* action_xlate_ctx_init() initializes these members. */
213 struct ofproto_dpif *ofproto;
215 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
216 * this flow when actions change header fields. */
219 /* The packet corresponding to 'flow', or a null pointer if we are
220 * revalidating without a packet to refer to. */
221 const struct ofpbuf *packet;
223 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
224 * actions update the flow table?
226 * We want to update these tables if we are actually processing a packet,
227 * or if we are accounting for packets that the datapath has processed, but
228 * not if we are just revalidating. */
231 /* The rule that we are currently translating, or NULL. */
232 struct rule_dpif *rule;
234 /* Union of the set of TCP flags seen so far in this flow. (Used only by
235 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
239 /* If nonnull, flow translation calls this function just before executing a
240 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
241 * when the recursion depth is exceeded.
243 * 'rule' is the rule being submitted into. It will be null if the
244 * resubmit or OFPP_TABLE action didn't find a matching rule.
246 * This is normally null so the client has to set it manually after
247 * calling action_xlate_ctx_init(). */
248 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule);
250 /* If nonnull, flow translation calls this function to report some
251 * significant decision, e.g. to explain why OFPP_NORMAL translation
252 * dropped a packet. */
253 void (*report_hook)(struct action_xlate_ctx *, const char *s);
255 /* If nonnull, flow translation credits the specified statistics to each
256 * rule reached through a resubmit or OFPP_TABLE action.
258 * This is normally null so the client has to set it manually after
259 * calling action_xlate_ctx_init(). */
260 const struct dpif_flow_stats *resubmit_stats;
262 /* xlate_actions() initializes and uses these members. The client might want
263 * to look at them after it returns. */
265 struct ofpbuf *odp_actions; /* Datapath actions. */
266 tag_type tags; /* Tags associated with actions. */
267 enum slow_path_reason slow; /* 0 if fast path may be used. */
268 bool has_learn; /* Actions include NXAST_LEARN? */
269 bool has_normal; /* Actions output to OFPP_NORMAL? */
270 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
271 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
272 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
274 /* xlate_actions() initializes and uses these members, but the client has no
275 * reason to look at them. */
277 int recurse; /* Recursion level, via xlate_table_action. */
278 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
279 struct flow base_flow; /* Flow at the last commit. */
280 uint32_t orig_skb_priority; /* Priority when packet arrived. */
281 uint8_t table_id; /* OpenFlow table ID where flow was found. */
282 uint32_t sflow_n_outputs; /* Number of output ports. */
283 uint32_t sflow_odp_port; /* Output port for composing sFlow action. */
284 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
285 bool exit; /* No further actions should be processed. */
288 static void action_xlate_ctx_init(struct action_xlate_ctx *,
289 struct ofproto_dpif *, const struct flow *,
290 ovs_be16 initial_tci, struct rule_dpif *,
291 uint8_t tcp_flags, const struct ofpbuf *);
292 static void xlate_actions(struct action_xlate_ctx *,
293 const struct ofpact *ofpacts, size_t ofpacts_len,
294 struct ofpbuf *odp_actions);
295 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
296 const struct ofpact *ofpacts,
298 static void xlate_table_action(struct action_xlate_ctx *, uint16_t in_port,
299 uint8_t table_id, bool may_packet_in);
301 static size_t put_userspace_action(const struct ofproto_dpif *,
302 struct ofpbuf *odp_actions,
304 const union user_action_cookie *);
306 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
307 enum slow_path_reason,
308 uint64_t *stub, size_t stub_size,
309 const struct nlattr **actionsp,
310 size_t *actions_lenp);
312 static void xlate_report(struct action_xlate_ctx *ctx, const char *s);
314 /* A subfacet (see "struct subfacet" below) has three possible installation
317 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
318 * case just after the subfacet is created, just before the subfacet is
319 * destroyed, or if the datapath returns an error when we try to install a
322 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
324 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
325 * ofproto_dpif is installed in the datapath.
328 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
329 SF_FAST_PATH, /* Full actions are installed. */
330 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
333 static const char *subfacet_path_to_string(enum subfacet_path);
335 /* A dpif flow and actions associated with a facet.
337 * See also the large comment on struct facet. */
340 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
341 struct list list_node; /* In struct facet's 'facets' list. */
342 struct facet *facet; /* Owning facet. */
346 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
347 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
348 * regenerate the ODP flow key from ->facet->flow. */
349 enum odp_key_fitness key_fitness;
353 long long int used; /* Time last used; time created if not used. */
355 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
356 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
360 * These should be essentially identical for every subfacet in a facet, but
361 * may differ in trivial ways due to VLAN splinters. */
362 size_t actions_len; /* Number of bytes in actions[]. */
363 struct nlattr *actions; /* Datapath actions. */
365 enum slow_path_reason slow; /* 0 if fast path may be used. */
366 enum subfacet_path path; /* Installed in datapath? */
368 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
369 * splinters can cause it to differ. This value should be removed when
370 * the VLAN splinters feature is no longer needed. */
371 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
373 /* Datapath port the packet arrived on. This is needed to remove
374 * flows for ports that are no longer part of the bridge. Since the
375 * flow definition only has the OpenFlow port number and the port is
376 * no longer part of the bridge, we can't determine the datapath port
377 * number needed to delete the flow from the datapath. */
378 uint32_t odp_in_port;
381 #define SUBFACET_DESTROY_MAX_BATCH 50
383 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
385 static struct subfacet *subfacet_find(struct ofproto_dpif *,
386 const struct nlattr *key, size_t key_len,
388 const struct flow *flow);
389 static void subfacet_destroy(struct subfacet *);
390 static void subfacet_destroy__(struct subfacet *);
391 static void subfacet_destroy_batch(struct ofproto_dpif *,
392 struct subfacet **, int n);
393 static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *,
395 static void subfacet_reset_dp_stats(struct subfacet *,
396 struct dpif_flow_stats *);
397 static void subfacet_update_time(struct subfacet *, long long int used);
398 static void subfacet_update_stats(struct subfacet *,
399 const struct dpif_flow_stats *);
400 static void subfacet_make_actions(struct subfacet *,
401 const struct ofpbuf *packet,
402 struct ofpbuf *odp_actions);
403 static int subfacet_install(struct subfacet *,
404 const struct nlattr *actions, size_t actions_len,
405 struct dpif_flow_stats *, enum slow_path_reason);
406 static void subfacet_uninstall(struct subfacet *);
408 static enum subfacet_path subfacet_want_path(enum slow_path_reason);
410 /* An exact-match instantiation of an OpenFlow flow.
412 * A facet associates a "struct flow", which represents the Open vSwitch
413 * userspace idea of an exact-match flow, with one or more subfacets. Each
414 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
415 * the facet. When the kernel module (or other dpif implementation) and Open
416 * vSwitch userspace agree on the definition of a flow key, there is exactly
417 * one subfacet per facet. If the dpif implementation supports more-specific
418 * flow matching than userspace, however, a facet can have more than one
419 * subfacet, each of which corresponds to some distinction in flow that
420 * userspace simply doesn't understand.
422 * Flow expiration works in terms of subfacets, so a facet must have at least
423 * one subfacet or it will never expire, leaking memory. */
426 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
427 struct list list_node; /* In owning rule's 'facets' list. */
428 struct rule_dpif *rule; /* Owning rule. */
431 struct list subfacets;
432 long long int used; /* Time last used; time created if not used. */
439 * - Do include packets and bytes sent "by hand", e.g. with
442 * - Do include packets and bytes that were obtained from the datapath
443 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
444 * DPIF_FP_ZERO_STATS).
446 * - Do not include packets or bytes that can be obtained from the
447 * datapath for any existing subfacet.
449 uint64_t packet_count; /* Number of packets received. */
450 uint64_t byte_count; /* Number of bytes received. */
452 /* Resubmit statistics. */
453 uint64_t prev_packet_count; /* Number of packets from last stats push. */
454 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
455 long long int prev_used; /* Used time from last stats push. */
458 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
459 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
460 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
462 /* Properties of datapath actions.
464 * Every subfacet has its own actions because actions can differ slightly
465 * between splintered and non-splintered subfacets due to the VLAN tag
466 * being initially different (present vs. absent). All of them have these
467 * properties in common so we just store one copy of them here. */
468 bool has_learn; /* Actions include NXAST_LEARN? */
469 bool has_normal; /* Actions output to OFPP_NORMAL? */
470 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
471 tag_type tags; /* Tags that would require revalidation. */
472 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
474 /* Storage for a single subfacet, to reduce malloc() time and space
475 * overhead. (A facet always has at least one subfacet and in the common
476 * case has exactly one subfacet.) */
477 struct subfacet one_subfacet;
480 static struct facet *facet_create(struct rule_dpif *,
481 const struct flow *, uint32_t hash);
482 static void facet_remove(struct facet *);
483 static void facet_free(struct facet *);
485 static struct facet *facet_find(struct ofproto_dpif *,
486 const struct flow *, uint32_t hash);
487 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
488 const struct flow *, uint32_t hash);
489 static void facet_revalidate(struct facet *);
490 static bool facet_check_consistency(struct facet *);
492 static void facet_flush_stats(struct facet *);
494 static void facet_update_time(struct facet *, long long int used);
495 static void facet_reset_counters(struct facet *);
496 static void facet_push_stats(struct facet *);
497 static void facet_learn(struct facet *);
498 static void facet_account(struct facet *);
500 static bool facet_is_controller_flow(struct facet *);
503 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
507 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
508 struct list bundle_node; /* In struct ofbundle's "ports" list. */
509 struct cfm *cfm; /* Connectivity Fault Management, if any. */
510 tag_type tag; /* Tag associated with this port. */
511 bool may_enable; /* May be enabled in bonds. */
512 long long int carrier_seq; /* Carrier status changes. */
513 struct tnl_port *tnl_port; /* Tunnel handle, or null. */
516 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
517 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
518 long long int stp_state_entered;
520 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
522 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
524 * This is deprecated. It is only for compatibility with broken device
525 * drivers in old versions of Linux that do not properly support VLANs when
526 * VLAN devices are not used. When broken device drivers are no longer in
527 * widespread use, we will delete these interfaces. */
528 uint16_t realdev_ofp_port;
532 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
533 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
534 * traffic egressing the 'ofport' with that priority should be marked with. */
535 struct priority_to_dscp {
536 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
537 uint32_t priority; /* Priority of this queue (see struct flow). */
539 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
542 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
544 * This is deprecated. It is only for compatibility with broken device drivers
545 * in old versions of Linux that do not properly support VLANs when VLAN
546 * devices are not used. When broken device drivers are no longer in
547 * widespread use, we will delete these interfaces. */
548 struct vlan_splinter {
549 struct hmap_node realdev_vid_node;
550 struct hmap_node vlandev_node;
551 uint16_t realdev_ofp_port;
552 uint16_t vlandev_ofp_port;
556 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
557 uint32_t realdev, ovs_be16 vlan_tci);
558 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
559 static void vsp_remove(struct ofport_dpif *);
560 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
562 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
564 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
567 static struct ofport_dpif *
568 ofport_dpif_cast(const struct ofport *ofport)
570 ovs_assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
571 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
574 static void port_run(struct ofport_dpif *);
575 static void port_run_fast(struct ofport_dpif *);
576 static void port_wait(struct ofport_dpif *);
577 static int set_cfm(struct ofport *, const struct cfm_settings *);
578 static void ofport_clear_priorities(struct ofport_dpif *);
580 struct dpif_completion {
581 struct list list_node;
582 struct ofoperation *op;
585 /* Extra information about a classifier table.
586 * Currently used just for optimized flow revalidation. */
588 /* If either of these is nonnull, then this table has a form that allows
589 * flows to be tagged to avoid revalidating most flows for the most common
590 * kinds of flow table changes. */
591 struct cls_table *catchall_table; /* Table that wildcards all fields. */
592 struct cls_table *other_table; /* Table with any other wildcard set. */
593 uint32_t basis; /* Keeps each table's tags separate. */
596 /* Reasons that we might need to revalidate every facet, and corresponding
599 * A value of 0 means that there is no need to revalidate.
601 * It would be nice to have some cleaner way to integrate with coverage
602 * counters, but with only a few reasons I guess this is good enough for
604 enum revalidate_reason {
605 REV_RECONFIGURE = 1, /* Switch configuration changed. */
606 REV_STP, /* Spanning tree protocol port status change. */
607 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
608 REV_FLOW_TABLE, /* Flow table changed. */
609 REV_INCONSISTENCY /* Facet self-check failed. */
611 COVERAGE_DEFINE(rev_reconfigure);
612 COVERAGE_DEFINE(rev_stp);
613 COVERAGE_DEFINE(rev_port_toggled);
614 COVERAGE_DEFINE(rev_flow_table);
615 COVERAGE_DEFINE(rev_inconsistency);
617 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
618 * These are datapath flows which have no associated ofproto, if they did we
619 * would use facets. */
621 struct hmap_node hmap_node;
626 /* All datapaths of a given type share a single dpif backer instance. */
631 struct timer next_expiration;
632 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
634 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
636 /* Facet revalidation flags applying to facets which use this backer. */
637 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
638 struct tag_set revalidate_set; /* Revalidate only matching facets. */
640 struct hmap drop_keys; /* Set of dropped odp keys. */
643 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
644 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
646 static void drop_key_clear(struct dpif_backer *);
647 static struct ofport_dpif *
648 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
650 struct ofproto_dpif {
651 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
653 struct dpif_backer *backer;
655 /* Special OpenFlow rules. */
656 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
657 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
663 struct netflow *netflow;
664 struct dpif_sflow *sflow;
665 struct hmap bundles; /* Contains "struct ofbundle"s. */
666 struct mac_learning *ml;
667 struct ofmirror *mirrors[MAX_MIRRORS];
669 bool has_bonded_bundles;
673 struct hmap subfacets;
674 struct governor *governor;
677 struct table_dpif tables[N_TABLES];
679 /* Support for debugging async flow mods. */
680 struct list completions;
682 bool has_bundle_action; /* True when the first bundle action appears. */
683 struct netdev_stats stats; /* To account packets generated and consumed in
688 long long int stp_last_tick;
690 /* VLAN splinters. */
691 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
692 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
695 struct sset ports; /* Set of standard port names. */
696 struct sset ghost_ports; /* Ports with no datapath port. */
697 struct sset port_poll_set; /* Queued names for port_poll() reply. */
698 int port_poll_errno; /* Last errno for port_poll() reply. */
701 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
702 * for debugging the asynchronous flow_mod implementation.) */
705 /* All existing ofproto_dpif instances, indexed by ->up.name. */
706 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
708 static void ofproto_dpif_unixctl_init(void);
710 static struct ofproto_dpif *
711 ofproto_dpif_cast(const struct ofproto *ofproto)
713 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
714 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
717 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
719 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
721 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
722 const struct ofpbuf *, ovs_be16 initial_tci,
724 static bool may_dpif_port_del(struct ofport_dpif *);
726 /* Packet processing. */
727 static void update_learning_table(struct ofproto_dpif *,
728 const struct flow *, int vlan,
731 #define FLOW_MISS_MAX_BATCH 50
732 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
734 /* Flow expiration. */
735 static int expire(struct dpif_backer *);
738 static void send_netflow_active_timeouts(struct ofproto_dpif *);
741 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
742 static size_t compose_sflow_action(const struct ofproto_dpif *,
743 struct ofpbuf *odp_actions,
744 const struct flow *, uint32_t odp_port);
745 static void add_mirror_actions(struct action_xlate_ctx *ctx,
746 const struct flow *flow);
747 /* Global variables. */
748 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
750 /* Initial mappings of port to bridge mappings. */
751 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
753 /* Factory functions. */
756 init(const struct shash *iface_hints)
758 struct shash_node *node;
760 /* Make a local copy, since we don't own 'iface_hints' elements. */
761 SHASH_FOR_EACH(node, iface_hints) {
762 const struct iface_hint *orig_hint = node->data;
763 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
765 new_hint->br_name = xstrdup(orig_hint->br_name);
766 new_hint->br_type = xstrdup(orig_hint->br_type);
767 new_hint->ofp_port = orig_hint->ofp_port;
769 shash_add(&init_ofp_ports, node->name, new_hint);
774 enumerate_types(struct sset *types)
776 dp_enumerate_types(types);
780 enumerate_names(const char *type, struct sset *names)
782 struct ofproto_dpif *ofproto;
785 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
786 if (strcmp(type, ofproto->up.type)) {
789 sset_add(names, ofproto->up.name);
796 del(const char *type, const char *name)
801 error = dpif_open(name, type, &dpif);
803 error = dpif_delete(dpif);
810 port_open_type(const char *datapath_type, const char *port_type)
812 return dpif_port_open_type(datapath_type, port_type);
815 /* Type functions. */
817 static struct ofproto_dpif *
818 lookup_ofproto_dpif_by_port_name(const char *name)
820 struct ofproto_dpif *ofproto;
822 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
823 if (sset_contains(&ofproto->ports, name)) {
832 type_run(const char *type)
834 struct dpif_backer *backer;
838 backer = shash_find_data(&all_dpif_backers, type);
840 /* This is not necessarily a problem, since backers are only
841 * created on demand. */
845 dpif_run(backer->dpif);
847 if (backer->need_revalidate
848 || !tag_set_is_empty(&backer->revalidate_set)) {
849 struct tag_set revalidate_set = backer->revalidate_set;
850 bool need_revalidate = backer->need_revalidate;
851 struct ofproto_dpif *ofproto;
853 switch (backer->need_revalidate) {
854 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
855 case REV_STP: COVERAGE_INC(rev_stp); break;
856 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
857 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
858 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
861 if (backer->need_revalidate) {
862 /* Clear the drop_keys in case we should now be accepting some
863 * formerly dropped flows. */
864 drop_key_clear(backer);
867 /* Clear the revalidation flags. */
868 tag_set_init(&backer->revalidate_set);
869 backer->need_revalidate = 0;
871 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
874 if (ofproto->backer != backer) {
878 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
880 || tag_set_intersects(&revalidate_set, facet->tags)) {
881 facet_revalidate(facet);
887 if (timer_expired(&backer->next_expiration)) {
888 int delay = expire(backer);
889 timer_set_duration(&backer->next_expiration, delay);
892 /* Check for port changes in the dpif. */
893 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
894 struct ofproto_dpif *ofproto;
895 struct dpif_port port;
897 /* Don't report on the datapath's device. */
898 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
902 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
903 &all_ofproto_dpifs) {
904 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
909 ofproto = lookup_ofproto_dpif_by_port_name(devname);
910 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
911 /* The port was removed. If we know the datapath,
912 * report it through poll_set(). If we don't, it may be
913 * notifying us of a removal we initiated, so ignore it.
914 * If there's a pending ENOBUFS, let it stand, since
915 * everything will be reevaluated. */
916 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
917 sset_add(&ofproto->port_poll_set, devname);
918 ofproto->port_poll_errno = 0;
920 } else if (!ofproto) {
921 /* The port was added, but we don't know with which
922 * ofproto we should associate it. Delete it. */
923 dpif_port_del(backer->dpif, port.port_no);
925 dpif_port_destroy(&port);
931 if (error != EAGAIN) {
932 struct ofproto_dpif *ofproto;
934 /* There was some sort of error, so propagate it to all
935 * ofprotos that use this backer. */
936 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
937 &all_ofproto_dpifs) {
938 if (ofproto->backer == backer) {
939 sset_clear(&ofproto->port_poll_set);
940 ofproto->port_poll_errno = error;
949 type_run_fast(const char *type)
951 struct dpif_backer *backer;
954 backer = shash_find_data(&all_dpif_backers, type);
956 /* This is not necessarily a problem, since backers are only
957 * created on demand. */
961 /* Handle one or more batches of upcalls, until there's nothing left to do
962 * or until we do a fixed total amount of work.
964 * We do work in batches because it can be much cheaper to set up a number
965 * of flows and fire off their patches all at once. We do multiple batches
966 * because in some cases handling a packet can cause another packet to be
967 * queued almost immediately as part of the return flow. Both
968 * optimizations can make major improvements on some benchmarks and
969 * presumably for real traffic as well. */
971 while (work < FLOW_MISS_MAX_BATCH) {
972 int retval = handle_upcalls(backer, FLOW_MISS_MAX_BATCH - work);
983 type_wait(const char *type)
985 struct dpif_backer *backer;
987 backer = shash_find_data(&all_dpif_backers, type);
989 /* This is not necessarily a problem, since backers are only
990 * created on demand. */
994 timer_wait(&backer->next_expiration);
997 /* Basic life-cycle. */
999 static int add_internal_flows(struct ofproto_dpif *);
1001 static struct ofproto *
1004 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1005 return &ofproto->up;
1009 dealloc(struct ofproto *ofproto_)
1011 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1016 close_dpif_backer(struct dpif_backer *backer)
1018 struct shash_node *node;
1020 ovs_assert(backer->refcount > 0);
1022 if (--backer->refcount) {
1026 drop_key_clear(backer);
1027 hmap_destroy(&backer->drop_keys);
1029 simap_destroy(&backer->tnl_backers);
1030 hmap_destroy(&backer->odp_to_ofport_map);
1031 node = shash_find(&all_dpif_backers, backer->type);
1033 shash_delete(&all_dpif_backers, node);
1034 dpif_close(backer->dpif);
1039 /* Datapath port slated for removal from datapath. */
1040 struct odp_garbage {
1041 struct list list_node;
1046 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1048 struct dpif_backer *backer;
1049 struct dpif_port_dump port_dump;
1050 struct dpif_port port;
1051 struct shash_node *node;
1052 struct list garbage_list;
1053 struct odp_garbage *garbage, *next;
1059 backer = shash_find_data(&all_dpif_backers, type);
1066 backer_name = xasprintf("ovs-%s", type);
1068 /* Remove any existing datapaths, since we assume we're the only
1069 * userspace controlling the datapath. */
1071 dp_enumerate_names(type, &names);
1072 SSET_FOR_EACH(name, &names) {
1073 struct dpif *old_dpif;
1075 /* Don't remove our backer if it exists. */
1076 if (!strcmp(name, backer_name)) {
1080 if (dpif_open(name, type, &old_dpif)) {
1081 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1083 dpif_delete(old_dpif);
1084 dpif_close(old_dpif);
1087 sset_destroy(&names);
1089 backer = xmalloc(sizeof *backer);
1091 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1094 VLOG_ERR("failed to open datapath of type %s: %s", type,
1100 backer->type = xstrdup(type);
1101 backer->refcount = 1;
1102 hmap_init(&backer->odp_to_ofport_map);
1103 hmap_init(&backer->drop_keys);
1104 timer_set_duration(&backer->next_expiration, 1000);
1105 backer->need_revalidate = 0;
1106 simap_init(&backer->tnl_backers);
1107 tag_set_init(&backer->revalidate_set);
1110 dpif_flow_flush(backer->dpif);
1112 /* Loop through the ports already on the datapath and remove any
1113 * that we don't need anymore. */
1114 list_init(&garbage_list);
1115 dpif_port_dump_start(&port_dump, backer->dpif);
1116 while (dpif_port_dump_next(&port_dump, &port)) {
1117 node = shash_find(&init_ofp_ports, port.name);
1118 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1119 garbage = xmalloc(sizeof *garbage);
1120 garbage->odp_port = port.port_no;
1121 list_push_front(&garbage_list, &garbage->list_node);
1124 dpif_port_dump_done(&port_dump);
1126 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1127 dpif_port_del(backer->dpif, garbage->odp_port);
1128 list_remove(&garbage->list_node);
1132 shash_add(&all_dpif_backers, type, backer);
1134 error = dpif_recv_set(backer->dpif, true);
1136 VLOG_ERR("failed to listen on datapath of type %s: %s",
1137 type, strerror(error));
1138 close_dpif_backer(backer);
1146 construct(struct ofproto *ofproto_)
1148 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1149 struct shash_node *node, *next;
1154 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1159 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1160 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1162 ofproto->n_matches = 0;
1164 ofproto->netflow = NULL;
1165 ofproto->sflow = NULL;
1166 ofproto->stp = NULL;
1167 hmap_init(&ofproto->bundles);
1168 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1169 for (i = 0; i < MAX_MIRRORS; i++) {
1170 ofproto->mirrors[i] = NULL;
1172 ofproto->has_bonded_bundles = false;
1174 hmap_init(&ofproto->facets);
1175 hmap_init(&ofproto->subfacets);
1176 ofproto->governor = NULL;
1178 for (i = 0; i < N_TABLES; i++) {
1179 struct table_dpif *table = &ofproto->tables[i];
1181 table->catchall_table = NULL;
1182 table->other_table = NULL;
1183 table->basis = random_uint32();
1186 list_init(&ofproto->completions);
1188 ofproto_dpif_unixctl_init();
1190 ofproto->has_mirrors = false;
1191 ofproto->has_bundle_action = false;
1193 hmap_init(&ofproto->vlandev_map);
1194 hmap_init(&ofproto->realdev_vid_map);
1196 sset_init(&ofproto->ports);
1197 sset_init(&ofproto->ghost_ports);
1198 sset_init(&ofproto->port_poll_set);
1199 ofproto->port_poll_errno = 0;
1201 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1202 struct iface_hint *iface_hint = node->data;
1204 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1205 /* Check if the datapath already has this port. */
1206 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1207 sset_add(&ofproto->ports, node->name);
1210 free(iface_hint->br_name);
1211 free(iface_hint->br_type);
1213 shash_delete(&init_ofp_ports, node);
1217 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1218 hash_string(ofproto->up.name, 0));
1219 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1221 ofproto_init_tables(ofproto_, N_TABLES);
1222 error = add_internal_flows(ofproto);
1223 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1229 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1230 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1232 struct ofputil_flow_mod fm;
1235 match_init_catchall(&fm.match);
1237 match_set_reg(&fm.match, 0, id);
1238 fm.new_cookie = htonll(0);
1239 fm.cookie = htonll(0);
1240 fm.cookie_mask = htonll(0);
1241 fm.table_id = TBL_INTERNAL;
1242 fm.command = OFPFC_ADD;
1243 fm.idle_timeout = 0;
1244 fm.hard_timeout = 0;
1248 fm.ofpacts = ofpacts->data;
1249 fm.ofpacts_len = ofpacts->size;
1251 error = ofproto_flow_mod(&ofproto->up, &fm);
1253 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1254 id, ofperr_to_string(error));
1258 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1259 ovs_assert(*rulep != NULL);
1265 add_internal_flows(struct ofproto_dpif *ofproto)
1267 struct ofpact_controller *controller;
1268 uint64_t ofpacts_stub[128 / 8];
1269 struct ofpbuf ofpacts;
1273 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1276 controller = ofpact_put_CONTROLLER(&ofpacts);
1277 controller->max_len = UINT16_MAX;
1278 controller->controller_id = 0;
1279 controller->reason = OFPR_NO_MATCH;
1280 ofpact_pad(&ofpacts);
1282 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1287 ofpbuf_clear(&ofpacts);
1288 error = add_internal_flow(ofproto, id++, &ofpacts,
1289 &ofproto->no_packet_in_rule);
1294 complete_operations(struct ofproto_dpif *ofproto)
1296 struct dpif_completion *c, *next;
1298 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1299 ofoperation_complete(c->op, 0);
1300 list_remove(&c->list_node);
1306 destruct(struct ofproto *ofproto_)
1308 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1309 struct rule_dpif *rule, *next_rule;
1310 struct oftable *table;
1313 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1314 complete_operations(ofproto);
1316 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1317 struct cls_cursor cursor;
1319 cls_cursor_init(&cursor, &table->cls, NULL);
1320 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1321 ofproto_rule_destroy(&rule->up);
1325 for (i = 0; i < MAX_MIRRORS; i++) {
1326 mirror_destroy(ofproto->mirrors[i]);
1329 netflow_destroy(ofproto->netflow);
1330 dpif_sflow_destroy(ofproto->sflow);
1331 hmap_destroy(&ofproto->bundles);
1332 mac_learning_destroy(ofproto->ml);
1334 hmap_destroy(&ofproto->facets);
1335 hmap_destroy(&ofproto->subfacets);
1336 governor_destroy(ofproto->governor);
1338 hmap_destroy(&ofproto->vlandev_map);
1339 hmap_destroy(&ofproto->realdev_vid_map);
1341 sset_destroy(&ofproto->ports);
1342 sset_destroy(&ofproto->ghost_ports);
1343 sset_destroy(&ofproto->port_poll_set);
1345 close_dpif_backer(ofproto->backer);
1349 run_fast(struct ofproto *ofproto_)
1351 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1352 struct ofport_dpif *ofport;
1354 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1355 port_run_fast(ofport);
1362 run(struct ofproto *ofproto_)
1364 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1365 struct ofport_dpif *ofport;
1366 struct ofbundle *bundle;
1370 complete_operations(ofproto);
1373 error = run_fast(ofproto_);
1378 if (ofproto->netflow) {
1379 if (netflow_run(ofproto->netflow)) {
1380 send_netflow_active_timeouts(ofproto);
1383 if (ofproto->sflow) {
1384 dpif_sflow_run(ofproto->sflow);
1387 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1390 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1395 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1397 /* Check the consistency of a random facet, to aid debugging. */
1398 if (!hmap_is_empty(&ofproto->facets)
1399 && !ofproto->backer->need_revalidate) {
1400 struct facet *facet;
1402 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1403 struct facet, hmap_node);
1404 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1406 if (!facet_check_consistency(facet)) {
1407 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1412 if (ofproto->governor) {
1415 governor_run(ofproto->governor);
1417 /* If the governor has shrunk to its minimum size and the number of
1418 * subfacets has dwindled, then drop the governor entirely.
1420 * For hysteresis, the number of subfacets to drop the governor is
1421 * smaller than the number needed to trigger its creation. */
1422 n_subfacets = hmap_count(&ofproto->subfacets);
1423 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1424 && governor_is_idle(ofproto->governor)) {
1425 governor_destroy(ofproto->governor);
1426 ofproto->governor = NULL;
1434 wait(struct ofproto *ofproto_)
1436 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1437 struct ofport_dpif *ofport;
1438 struct ofbundle *bundle;
1440 if (!clogged && !list_is_empty(&ofproto->completions)) {
1441 poll_immediate_wake();
1444 dpif_wait(ofproto->backer->dpif);
1445 dpif_recv_wait(ofproto->backer->dpif);
1446 if (ofproto->sflow) {
1447 dpif_sflow_wait(ofproto->sflow);
1449 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1450 poll_immediate_wake();
1452 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1455 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1456 bundle_wait(bundle);
1458 if (ofproto->netflow) {
1459 netflow_wait(ofproto->netflow);
1461 mac_learning_wait(ofproto->ml);
1463 if (ofproto->backer->need_revalidate) {
1464 /* Shouldn't happen, but if it does just go around again. */
1465 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1466 poll_immediate_wake();
1468 if (ofproto->governor) {
1469 governor_wait(ofproto->governor);
1474 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1476 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1478 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1479 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1483 flush(struct ofproto *ofproto_)
1485 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1486 struct subfacet *subfacet, *next_subfacet;
1487 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1491 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1492 &ofproto->subfacets) {
1493 if (subfacet->path != SF_NOT_INSTALLED) {
1494 batch[n_batch++] = subfacet;
1495 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1496 subfacet_destroy_batch(ofproto, batch, n_batch);
1500 subfacet_destroy(subfacet);
1505 subfacet_destroy_batch(ofproto, batch, n_batch);
1510 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1511 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1513 *arp_match_ip = true;
1514 *actions = (OFPUTIL_A_OUTPUT |
1515 OFPUTIL_A_SET_VLAN_VID |
1516 OFPUTIL_A_SET_VLAN_PCP |
1517 OFPUTIL_A_STRIP_VLAN |
1518 OFPUTIL_A_SET_DL_SRC |
1519 OFPUTIL_A_SET_DL_DST |
1520 OFPUTIL_A_SET_NW_SRC |
1521 OFPUTIL_A_SET_NW_DST |
1522 OFPUTIL_A_SET_NW_TOS |
1523 OFPUTIL_A_SET_TP_SRC |
1524 OFPUTIL_A_SET_TP_DST |
1529 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1531 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1532 struct dpif_dp_stats s;
1534 strcpy(ots->name, "classifier");
1536 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1538 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1539 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1542 static struct ofport *
1545 struct ofport_dpif *port = xmalloc(sizeof *port);
1550 port_dealloc(struct ofport *port_)
1552 struct ofport_dpif *port = ofport_dpif_cast(port_);
1557 port_construct(struct ofport *port_)
1559 struct ofport_dpif *port = ofport_dpif_cast(port_);
1560 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1561 const struct netdev *netdev = port->up.netdev;
1562 struct dpif_port dpif_port;
1565 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1566 port->bundle = NULL;
1568 port->tag = tag_create_random();
1569 port->may_enable = true;
1570 port->stp_port = NULL;
1571 port->stp_state = STP_DISABLED;
1572 port->tnl_port = NULL;
1573 hmap_init(&port->priorities);
1574 port->realdev_ofp_port = 0;
1575 port->vlandev_vid = 0;
1576 port->carrier_seq = netdev_get_carrier_resets(netdev);
1578 if (netdev_vport_is_patch(netdev)) {
1579 /* XXX By bailing out here, we don't do required sFlow work. */
1580 port->odp_port = OVSP_NONE;
1584 error = dpif_port_query_by_name(ofproto->backer->dpif,
1585 netdev_vport_get_dpif_port(netdev),
1591 port->odp_port = dpif_port.port_no;
1593 if (netdev_get_tunnel_config(netdev)) {
1594 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1596 /* Sanity-check that a mapping doesn't already exist. This
1597 * shouldn't happen for non-tunnel ports. */
1598 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1599 VLOG_ERR("port %s already has an OpenFlow port number",
1601 dpif_port_destroy(&dpif_port);
1605 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1606 hash_int(port->odp_port, 0));
1608 dpif_port_destroy(&dpif_port);
1610 if (ofproto->sflow) {
1611 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1618 port_destruct(struct ofport *port_)
1620 struct ofport_dpif *port = ofport_dpif_cast(port_);
1621 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1622 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1623 const char *devname = netdev_get_name(port->up.netdev);
1625 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)
1626 && may_dpif_port_del(port)) {
1627 /* The underlying device is still there, so delete it. This
1628 * happens when the ofproto is being destroyed, since the caller
1629 * assumes that removal of attached ports will happen as part of
1631 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1632 simap_find_and_delete(&ofproto->backer->tnl_backers, dp_port_name);
1635 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1636 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1639 tnl_port_del(port->tnl_port);
1640 sset_find_and_delete(&ofproto->ports, devname);
1641 sset_find_and_delete(&ofproto->ghost_ports, devname);
1642 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1643 bundle_remove(port_);
1644 set_cfm(port_, NULL);
1645 if (ofproto->sflow) {
1646 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1649 ofport_clear_priorities(port);
1650 hmap_destroy(&port->priorities);
1654 port_modified(struct ofport *port_)
1656 struct ofport_dpif *port = ofport_dpif_cast(port_);
1658 if (port->bundle && port->bundle->bond) {
1659 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1664 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1666 struct ofport_dpif *port = ofport_dpif_cast(port_);
1667 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1668 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1670 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1671 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1672 OFPUTIL_PC_NO_PACKET_IN)) {
1673 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1675 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1676 bundle_update(port->bundle);
1682 set_sflow(struct ofproto *ofproto_,
1683 const struct ofproto_sflow_options *sflow_options)
1685 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1686 struct dpif_sflow *ds = ofproto->sflow;
1688 if (sflow_options) {
1690 struct ofport_dpif *ofport;
1692 ds = ofproto->sflow = dpif_sflow_create();
1693 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1694 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1696 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1698 dpif_sflow_set_options(ds, sflow_options);
1701 dpif_sflow_destroy(ds);
1702 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1703 ofproto->sflow = NULL;
1710 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1712 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1719 struct ofproto_dpif *ofproto;
1721 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1722 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1723 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1726 if (cfm_configure(ofport->cfm, s)) {
1732 cfm_destroy(ofport->cfm);
1738 get_cfm_fault(const struct ofport *ofport_)
1740 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1742 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1746 get_cfm_opup(const struct ofport *ofport_)
1748 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1750 return ofport->cfm ? cfm_get_opup(ofport->cfm) : -1;
1754 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1757 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1760 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1768 get_cfm_health(const struct ofport *ofport_)
1770 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1772 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1775 /* Spanning Tree. */
1778 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1780 struct ofproto_dpif *ofproto = ofproto_;
1781 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1782 struct ofport_dpif *ofport;
1784 ofport = stp_port_get_aux(sp);
1786 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1787 ofproto->up.name, port_num);
1789 struct eth_header *eth = pkt->l2;
1791 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1792 if (eth_addr_is_zero(eth->eth_src)) {
1793 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1794 "with unknown MAC", ofproto->up.name, port_num);
1796 send_packet(ofport, pkt);
1802 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1804 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1806 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1808 /* Only revalidate flows if the configuration changed. */
1809 if (!s != !ofproto->stp) {
1810 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1814 if (!ofproto->stp) {
1815 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1816 send_bpdu_cb, ofproto);
1817 ofproto->stp_last_tick = time_msec();
1820 stp_set_bridge_id(ofproto->stp, s->system_id);
1821 stp_set_bridge_priority(ofproto->stp, s->priority);
1822 stp_set_hello_time(ofproto->stp, s->hello_time);
1823 stp_set_max_age(ofproto->stp, s->max_age);
1824 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1826 struct ofport *ofport;
1828 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1829 set_stp_port(ofport, NULL);
1832 stp_destroy(ofproto->stp);
1833 ofproto->stp = NULL;
1840 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1842 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1846 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1847 s->designated_root = stp_get_designated_root(ofproto->stp);
1848 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1857 update_stp_port_state(struct ofport_dpif *ofport)
1859 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1860 enum stp_state state;
1862 /* Figure out new state. */
1863 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1867 if (ofport->stp_state != state) {
1868 enum ofputil_port_state of_state;
1871 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1872 netdev_get_name(ofport->up.netdev),
1873 stp_state_name(ofport->stp_state),
1874 stp_state_name(state));
1875 if (stp_learn_in_state(ofport->stp_state)
1876 != stp_learn_in_state(state)) {
1877 /* xxx Learning action flows should also be flushed. */
1878 mac_learning_flush(ofproto->ml,
1879 &ofproto->backer->revalidate_set);
1881 fwd_change = stp_forward_in_state(ofport->stp_state)
1882 != stp_forward_in_state(state);
1884 ofproto->backer->need_revalidate = REV_STP;
1885 ofport->stp_state = state;
1886 ofport->stp_state_entered = time_msec();
1888 if (fwd_change && ofport->bundle) {
1889 bundle_update(ofport->bundle);
1892 /* Update the STP state bits in the OpenFlow port description. */
1893 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1894 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1895 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1896 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1897 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1899 ofproto_port_set_state(&ofport->up, of_state);
1903 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1904 * caller is responsible for assigning STP port numbers and ensuring
1905 * there are no duplicates. */
1907 set_stp_port(struct ofport *ofport_,
1908 const struct ofproto_port_stp_settings *s)
1910 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1911 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1912 struct stp_port *sp = ofport->stp_port;
1914 if (!s || !s->enable) {
1916 ofport->stp_port = NULL;
1917 stp_port_disable(sp);
1918 update_stp_port_state(ofport);
1921 } else if (sp && stp_port_no(sp) != s->port_num
1922 && ofport == stp_port_get_aux(sp)) {
1923 /* The port-id changed, so disable the old one if it's not
1924 * already in use by another port. */
1925 stp_port_disable(sp);
1928 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1929 stp_port_enable(sp);
1931 stp_port_set_aux(sp, ofport);
1932 stp_port_set_priority(sp, s->priority);
1933 stp_port_set_path_cost(sp, s->path_cost);
1935 update_stp_port_state(ofport);
1941 get_stp_port_status(struct ofport *ofport_,
1942 struct ofproto_port_stp_status *s)
1944 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1945 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1946 struct stp_port *sp = ofport->stp_port;
1948 if (!ofproto->stp || !sp) {
1954 s->port_id = stp_port_get_id(sp);
1955 s->state = stp_port_get_state(sp);
1956 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1957 s->role = stp_port_get_role(sp);
1958 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1964 stp_run(struct ofproto_dpif *ofproto)
1967 long long int now = time_msec();
1968 long long int elapsed = now - ofproto->stp_last_tick;
1969 struct stp_port *sp;
1972 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1973 ofproto->stp_last_tick = now;
1975 while (stp_get_changed_port(ofproto->stp, &sp)) {
1976 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1979 update_stp_port_state(ofport);
1983 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1984 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
1990 stp_wait(struct ofproto_dpif *ofproto)
1993 poll_timer_wait(1000);
1997 /* Returns true if STP should process 'flow'. */
1999 stp_should_process_flow(const struct flow *flow)
2001 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2005 stp_process_packet(const struct ofport_dpif *ofport,
2006 const struct ofpbuf *packet)
2008 struct ofpbuf payload = *packet;
2009 struct eth_header *eth = payload.data;
2010 struct stp_port *sp = ofport->stp_port;
2012 /* Sink packets on ports that have STP disabled when the bridge has
2014 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2018 /* Trim off padding on payload. */
2019 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2020 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2023 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2024 stp_received_bpdu(sp, payload.data, payload.size);
2028 static struct priority_to_dscp *
2029 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2031 struct priority_to_dscp *pdscp;
2034 hash = hash_int(priority, 0);
2035 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2036 if (pdscp->priority == priority) {
2044 ofport_clear_priorities(struct ofport_dpif *ofport)
2046 struct priority_to_dscp *pdscp, *next;
2048 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2049 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2055 set_queues(struct ofport *ofport_,
2056 const struct ofproto_port_queue *qdscp_list,
2059 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2060 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2061 struct hmap new = HMAP_INITIALIZER(&new);
2064 for (i = 0; i < n_qdscp; i++) {
2065 struct priority_to_dscp *pdscp;
2069 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2070 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2075 pdscp = get_priority(ofport, priority);
2077 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2079 pdscp = xmalloc(sizeof *pdscp);
2080 pdscp->priority = priority;
2082 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2085 if (pdscp->dscp != dscp) {
2087 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2090 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2093 if (!hmap_is_empty(&ofport->priorities)) {
2094 ofport_clear_priorities(ofport);
2095 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2098 hmap_swap(&new, &ofport->priorities);
2106 /* Expires all MAC learning entries associated with 'bundle' and forces its
2107 * ofproto to revalidate every flow.
2109 * Normally MAC learning entries are removed only from the ofproto associated
2110 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2111 * are removed from every ofproto. When patch ports and SLB bonds are in use
2112 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2113 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2114 * with the host from which it migrated. */
2116 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2118 struct ofproto_dpif *ofproto = bundle->ofproto;
2119 struct mac_learning *ml = ofproto->ml;
2120 struct mac_entry *mac, *next_mac;
2122 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2123 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2124 if (mac->port.p == bundle) {
2126 struct ofproto_dpif *o;
2128 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2130 struct mac_entry *e;
2132 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2135 mac_learning_expire(o->ml, e);
2141 mac_learning_expire(ml, mac);
2146 static struct ofbundle *
2147 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2149 struct ofbundle *bundle;
2151 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2152 &ofproto->bundles) {
2153 if (bundle->aux == aux) {
2160 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2161 * ones that are found to 'bundles'. */
2163 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2164 void **auxes, size_t n_auxes,
2165 struct hmapx *bundles)
2169 hmapx_init(bundles);
2170 for (i = 0; i < n_auxes; i++) {
2171 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2173 hmapx_add(bundles, bundle);
2179 bundle_update(struct ofbundle *bundle)
2181 struct ofport_dpif *port;
2183 bundle->floodable = true;
2184 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2185 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2186 || !stp_forward_in_state(port->stp_state)) {
2187 bundle->floodable = false;
2194 bundle_del_port(struct ofport_dpif *port)
2196 struct ofbundle *bundle = port->bundle;
2198 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2200 list_remove(&port->bundle_node);
2201 port->bundle = NULL;
2204 lacp_slave_unregister(bundle->lacp, port);
2207 bond_slave_unregister(bundle->bond, port);
2210 bundle_update(bundle);
2214 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2215 struct lacp_slave_settings *lacp)
2217 struct ofport_dpif *port;
2219 port = get_ofp_port(bundle->ofproto, ofp_port);
2224 if (port->bundle != bundle) {
2225 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2227 bundle_del_port(port);
2230 port->bundle = bundle;
2231 list_push_back(&bundle->ports, &port->bundle_node);
2232 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2233 || !stp_forward_in_state(port->stp_state)) {
2234 bundle->floodable = false;
2238 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2239 lacp_slave_register(bundle->lacp, port, lacp);
2246 bundle_destroy(struct ofbundle *bundle)
2248 struct ofproto_dpif *ofproto;
2249 struct ofport_dpif *port, *next_port;
2256 ofproto = bundle->ofproto;
2257 for (i = 0; i < MAX_MIRRORS; i++) {
2258 struct ofmirror *m = ofproto->mirrors[i];
2260 if (m->out == bundle) {
2262 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2263 || hmapx_find_and_delete(&m->dsts, bundle)) {
2264 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2269 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2270 bundle_del_port(port);
2273 bundle_flush_macs(bundle, true);
2274 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2276 free(bundle->trunks);
2277 lacp_destroy(bundle->lacp);
2278 bond_destroy(bundle->bond);
2283 bundle_set(struct ofproto *ofproto_, void *aux,
2284 const struct ofproto_bundle_settings *s)
2286 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2287 bool need_flush = false;
2288 struct ofport_dpif *port;
2289 struct ofbundle *bundle;
2290 unsigned long *trunks;
2296 bundle_destroy(bundle_lookup(ofproto, aux));
2300 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2301 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2303 bundle = bundle_lookup(ofproto, aux);
2305 bundle = xmalloc(sizeof *bundle);
2307 bundle->ofproto = ofproto;
2308 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2309 hash_pointer(aux, 0));
2311 bundle->name = NULL;
2313 list_init(&bundle->ports);
2314 bundle->vlan_mode = PORT_VLAN_TRUNK;
2316 bundle->trunks = NULL;
2317 bundle->use_priority_tags = s->use_priority_tags;
2318 bundle->lacp = NULL;
2319 bundle->bond = NULL;
2321 bundle->floodable = true;
2323 bundle->src_mirrors = 0;
2324 bundle->dst_mirrors = 0;
2325 bundle->mirror_out = 0;
2328 if (!bundle->name || strcmp(s->name, bundle->name)) {
2330 bundle->name = xstrdup(s->name);
2335 if (!bundle->lacp) {
2336 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2337 bundle->lacp = lacp_create();
2339 lacp_configure(bundle->lacp, s->lacp);
2341 lacp_destroy(bundle->lacp);
2342 bundle->lacp = NULL;
2345 /* Update set of ports. */
2347 for (i = 0; i < s->n_slaves; i++) {
2348 if (!bundle_add_port(bundle, s->slaves[i],
2349 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2353 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2354 struct ofport_dpif *next_port;
2356 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2357 for (i = 0; i < s->n_slaves; i++) {
2358 if (s->slaves[i] == port->up.ofp_port) {
2363 bundle_del_port(port);
2367 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2369 if (list_is_empty(&bundle->ports)) {
2370 bundle_destroy(bundle);
2374 /* Set VLAN tagging mode */
2375 if (s->vlan_mode != bundle->vlan_mode
2376 || s->use_priority_tags != bundle->use_priority_tags) {
2377 bundle->vlan_mode = s->vlan_mode;
2378 bundle->use_priority_tags = s->use_priority_tags;
2383 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2384 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2386 if (vlan != bundle->vlan) {
2387 bundle->vlan = vlan;
2391 /* Get trunked VLANs. */
2392 switch (s->vlan_mode) {
2393 case PORT_VLAN_ACCESS:
2397 case PORT_VLAN_TRUNK:
2398 trunks = CONST_CAST(unsigned long *, s->trunks);
2401 case PORT_VLAN_NATIVE_UNTAGGED:
2402 case PORT_VLAN_NATIVE_TAGGED:
2403 if (vlan != 0 && (!s->trunks
2404 || !bitmap_is_set(s->trunks, vlan)
2405 || bitmap_is_set(s->trunks, 0))) {
2406 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2408 trunks = bitmap_clone(s->trunks, 4096);
2410 trunks = bitmap_allocate1(4096);
2412 bitmap_set1(trunks, vlan);
2413 bitmap_set0(trunks, 0);
2415 trunks = CONST_CAST(unsigned long *, s->trunks);
2422 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2423 free(bundle->trunks);
2424 if (trunks == s->trunks) {
2425 bundle->trunks = vlan_bitmap_clone(trunks);
2427 bundle->trunks = trunks;
2432 if (trunks != s->trunks) {
2437 if (!list_is_short(&bundle->ports)) {
2438 bundle->ofproto->has_bonded_bundles = true;
2440 if (bond_reconfigure(bundle->bond, s->bond)) {
2441 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2444 bundle->bond = bond_create(s->bond);
2445 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2448 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2449 bond_slave_register(bundle->bond, port, port->up.netdev);
2452 bond_destroy(bundle->bond);
2453 bundle->bond = NULL;
2456 /* If we changed something that would affect MAC learning, un-learn
2457 * everything on this port and force flow revalidation. */
2459 bundle_flush_macs(bundle, false);
2466 bundle_remove(struct ofport *port_)
2468 struct ofport_dpif *port = ofport_dpif_cast(port_);
2469 struct ofbundle *bundle = port->bundle;
2472 bundle_del_port(port);
2473 if (list_is_empty(&bundle->ports)) {
2474 bundle_destroy(bundle);
2475 } else if (list_is_short(&bundle->ports)) {
2476 bond_destroy(bundle->bond);
2477 bundle->bond = NULL;
2483 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2485 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2486 struct ofport_dpif *port = port_;
2487 uint8_t ea[ETH_ADDR_LEN];
2490 error = netdev_get_etheraddr(port->up.netdev, ea);
2492 struct ofpbuf packet;
2495 ofpbuf_init(&packet, 0);
2496 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2498 memcpy(packet_pdu, pdu, pdu_size);
2500 send_packet(port, &packet);
2501 ofpbuf_uninit(&packet);
2503 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2504 "%s (%s)", port->bundle->name,
2505 netdev_get_name(port->up.netdev), strerror(error));
2510 bundle_send_learning_packets(struct ofbundle *bundle)
2512 struct ofproto_dpif *ofproto = bundle->ofproto;
2513 int error, n_packets, n_errors;
2514 struct mac_entry *e;
2516 error = n_packets = n_errors = 0;
2517 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2518 if (e->port.p != bundle) {
2519 struct ofpbuf *learning_packet;
2520 struct ofport_dpif *port;
2524 /* The assignment to "port" is unnecessary but makes "grep"ing for
2525 * struct ofport_dpif more effective. */
2526 learning_packet = bond_compose_learning_packet(bundle->bond,
2530 ret = send_packet(port, learning_packet);
2531 ofpbuf_delete(learning_packet);
2541 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2542 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2543 "packets, last error was: %s",
2544 bundle->name, n_errors, n_packets, strerror(error));
2546 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2547 bundle->name, n_packets);
2552 bundle_run(struct ofbundle *bundle)
2555 lacp_run(bundle->lacp, send_pdu_cb);
2558 struct ofport_dpif *port;
2560 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2561 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2564 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2565 lacp_status(bundle->lacp));
2566 if (bond_should_send_learning_packets(bundle->bond)) {
2567 bundle_send_learning_packets(bundle);
2573 bundle_wait(struct ofbundle *bundle)
2576 lacp_wait(bundle->lacp);
2579 bond_wait(bundle->bond);
2586 mirror_scan(struct ofproto_dpif *ofproto)
2590 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2591 if (!ofproto->mirrors[idx]) {
2598 static struct ofmirror *
2599 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2603 for (i = 0; i < MAX_MIRRORS; i++) {
2604 struct ofmirror *mirror = ofproto->mirrors[i];
2605 if (mirror && mirror->aux == aux) {
2613 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2615 mirror_update_dups(struct ofproto_dpif *ofproto)
2619 for (i = 0; i < MAX_MIRRORS; i++) {
2620 struct ofmirror *m = ofproto->mirrors[i];
2623 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2627 for (i = 0; i < MAX_MIRRORS; i++) {
2628 struct ofmirror *m1 = ofproto->mirrors[i];
2635 for (j = i + 1; j < MAX_MIRRORS; j++) {
2636 struct ofmirror *m2 = ofproto->mirrors[j];
2638 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2639 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2640 m2->dup_mirrors |= m1->dup_mirrors;
2647 mirror_set(struct ofproto *ofproto_, void *aux,
2648 const struct ofproto_mirror_settings *s)
2650 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2651 mirror_mask_t mirror_bit;
2652 struct ofbundle *bundle;
2653 struct ofmirror *mirror;
2654 struct ofbundle *out;
2655 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2656 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2659 mirror = mirror_lookup(ofproto, aux);
2661 mirror_destroy(mirror);
2667 idx = mirror_scan(ofproto);
2669 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2671 ofproto->up.name, MAX_MIRRORS, s->name);
2675 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2676 mirror->ofproto = ofproto;
2679 mirror->out_vlan = -1;
2680 mirror->name = NULL;
2683 if (!mirror->name || strcmp(s->name, mirror->name)) {
2685 mirror->name = xstrdup(s->name);
2688 /* Get the new configuration. */
2689 if (s->out_bundle) {
2690 out = bundle_lookup(ofproto, s->out_bundle);
2692 mirror_destroy(mirror);
2698 out_vlan = s->out_vlan;
2700 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2701 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2703 /* If the configuration has not changed, do nothing. */
2704 if (hmapx_equals(&srcs, &mirror->srcs)
2705 && hmapx_equals(&dsts, &mirror->dsts)
2706 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2707 && mirror->out == out
2708 && mirror->out_vlan == out_vlan)
2710 hmapx_destroy(&srcs);
2711 hmapx_destroy(&dsts);
2715 hmapx_swap(&srcs, &mirror->srcs);
2716 hmapx_destroy(&srcs);
2718 hmapx_swap(&dsts, &mirror->dsts);
2719 hmapx_destroy(&dsts);
2721 free(mirror->vlans);
2722 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2725 mirror->out_vlan = out_vlan;
2727 /* Update bundles. */
2728 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2729 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2730 if (hmapx_contains(&mirror->srcs, bundle)) {
2731 bundle->src_mirrors |= mirror_bit;
2733 bundle->src_mirrors &= ~mirror_bit;
2736 if (hmapx_contains(&mirror->dsts, bundle)) {
2737 bundle->dst_mirrors |= mirror_bit;
2739 bundle->dst_mirrors &= ~mirror_bit;
2742 if (mirror->out == bundle) {
2743 bundle->mirror_out |= mirror_bit;
2745 bundle->mirror_out &= ~mirror_bit;
2749 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2750 ofproto->has_mirrors = true;
2751 mac_learning_flush(ofproto->ml,
2752 &ofproto->backer->revalidate_set);
2753 mirror_update_dups(ofproto);
2759 mirror_destroy(struct ofmirror *mirror)
2761 struct ofproto_dpif *ofproto;
2762 mirror_mask_t mirror_bit;
2763 struct ofbundle *bundle;
2770 ofproto = mirror->ofproto;
2771 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2772 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2774 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2775 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2776 bundle->src_mirrors &= ~mirror_bit;
2777 bundle->dst_mirrors &= ~mirror_bit;
2778 bundle->mirror_out &= ~mirror_bit;
2781 hmapx_destroy(&mirror->srcs);
2782 hmapx_destroy(&mirror->dsts);
2783 free(mirror->vlans);
2785 ofproto->mirrors[mirror->idx] = NULL;
2789 mirror_update_dups(ofproto);
2791 ofproto->has_mirrors = false;
2792 for (i = 0; i < MAX_MIRRORS; i++) {
2793 if (ofproto->mirrors[i]) {
2794 ofproto->has_mirrors = true;
2801 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2802 uint64_t *packets, uint64_t *bytes)
2804 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2805 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2808 *packets = *bytes = UINT64_MAX;
2812 *packets = mirror->packet_count;
2813 *bytes = mirror->byte_count;
2819 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2821 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2822 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2823 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2829 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2831 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2832 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2833 return bundle && bundle->mirror_out != 0;
2837 forward_bpdu_changed(struct ofproto *ofproto_)
2839 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2840 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2844 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2847 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2848 mac_learning_set_idle_time(ofproto->ml, idle_time);
2849 mac_learning_set_max_entries(ofproto->ml, max_entries);
2854 static struct ofport_dpif *
2855 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2857 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2858 return ofport ? ofport_dpif_cast(ofport) : NULL;
2861 static struct ofport_dpif *
2862 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2864 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2865 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2869 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2870 struct ofproto_port *ofproto_port,
2871 struct dpif_port *dpif_port)
2873 ofproto_port->name = dpif_port->name;
2874 ofproto_port->type = dpif_port->type;
2875 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2878 static struct ofport_dpif *
2879 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
2881 const struct ofproto_dpif *ofproto;
2884 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
2889 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2890 struct ofport *ofport;
2892 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
2893 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
2894 return ofport_dpif_cast(ofport);
2901 port_run_fast(struct ofport_dpif *ofport)
2903 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2904 struct ofpbuf packet;
2906 ofpbuf_init(&packet, 0);
2907 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2908 send_packet(ofport, &packet);
2909 ofpbuf_uninit(&packet);
2914 port_run(struct ofport_dpif *ofport)
2916 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2917 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2918 bool enable = netdev_get_carrier(ofport->up.netdev);
2920 ofport->carrier_seq = carrier_seq;
2922 port_run_fast(ofport);
2924 if (ofport->tnl_port
2925 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
2926 &ofport->tnl_port)) {
2927 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
2931 int cfm_opup = cfm_get_opup(ofport->cfm);
2933 cfm_run(ofport->cfm);
2934 enable = enable && !cfm_get_fault(ofport->cfm);
2936 if (cfm_opup >= 0) {
2937 enable = enable && cfm_opup;
2941 if (ofport->bundle) {
2942 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2943 if (carrier_changed) {
2944 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2948 if (ofport->may_enable != enable) {
2949 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2951 if (ofproto->has_bundle_action) {
2952 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2956 ofport->may_enable = enable;
2960 port_wait(struct ofport_dpif *ofport)
2963 cfm_wait(ofport->cfm);
2968 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2969 struct ofproto_port *ofproto_port)
2971 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2972 struct dpif_port dpif_port;
2975 if (sset_contains(&ofproto->ghost_ports, devname)) {
2976 const char *type = netdev_get_type_from_name(devname);
2978 /* We may be called before ofproto->up.port_by_name is populated with
2979 * the appropriate ofport. For this reason, we must get the name and
2980 * type from the netdev layer directly. */
2982 const struct ofport *ofport;
2984 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2985 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2986 ofproto_port->name = xstrdup(devname);
2987 ofproto_port->type = xstrdup(type);
2993 if (!sset_contains(&ofproto->ports, devname)) {
2996 error = dpif_port_query_by_name(ofproto->backer->dpif,
2997 devname, &dpif_port);
2999 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3005 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3007 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3008 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3009 const char *devname = netdev_get_name(netdev);
3011 if (netdev_vport_is_patch(netdev)) {
3012 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3016 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3017 uint32_t port_no = UINT32_MAX;
3020 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3024 if (netdev_get_tunnel_config(netdev)) {
3025 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3029 if (netdev_get_tunnel_config(netdev)) {
3030 sset_add(&ofproto->ghost_ports, devname);
3032 sset_add(&ofproto->ports, devname);
3037 /* Returns true if the odp_port backing 'ofport' may be deleted from the
3038 * datapath. In most cases, this function simply returns true. However, for
3039 * tunnels it's possible that multiple ofports use the same odp_port, in which
3040 * case we need to keep the odp_port backer around until the last ofport is
3043 may_dpif_port_del(struct ofport_dpif *ofport)
3045 struct dpif_backer *backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
3046 struct ofproto_dpif *ofproto_iter;
3048 if (!ofport->tnl_port) {
3052 HMAP_FOR_EACH (ofproto_iter, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3053 struct ofport_dpif *iter;
3055 if (backer != ofproto_iter->backer) {
3059 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto_iter->up.ports) {
3060 if (ofport == iter) {
3064 if (!strcmp(netdev_vport_get_dpif_port(ofport->up.netdev),
3065 netdev_vport_get_dpif_port(iter->up.netdev))) {
3075 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3077 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3078 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3085 sset_find_and_delete(&ofproto->ghost_ports,
3086 netdev_get_name(ofport->up.netdev));
3087 if (may_dpif_port_del(ofport)) {
3088 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3090 const char *dpif_port;
3092 /* The caller is going to close ofport->up.netdev. If this is a
3093 * bonded port, then the bond is using that netdev, so remove it
3094 * from the bond. The client will need to reconfigure everything
3095 * after deleting ports, so then the slave will get re-added. */
3096 dpif_port = netdev_vport_get_dpif_port(ofport->up.netdev);
3097 simap_find_and_delete(&ofproto->backer->tnl_backers, dpif_port);
3098 bundle_remove(&ofport->up);
3105 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3107 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3110 error = netdev_get_stats(ofport->up.netdev, stats);
3112 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3113 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3115 /* ofproto->stats.tx_packets represents packets that we created
3116 * internally and sent to some port (e.g. packets sent with
3117 * send_packet()). Account for them as if they had come from
3118 * OFPP_LOCAL and got forwarded. */
3120 if (stats->rx_packets != UINT64_MAX) {
3121 stats->rx_packets += ofproto->stats.tx_packets;
3124 if (stats->rx_bytes != UINT64_MAX) {
3125 stats->rx_bytes += ofproto->stats.tx_bytes;
3128 /* ofproto->stats.rx_packets represents packets that were received on
3129 * some port and we processed internally and dropped (e.g. STP).
3130 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3132 if (stats->tx_packets != UINT64_MAX) {
3133 stats->tx_packets += ofproto->stats.rx_packets;
3136 if (stats->tx_bytes != UINT64_MAX) {
3137 stats->tx_bytes += ofproto->stats.rx_bytes;
3144 /* Account packets for LOCAL port. */
3146 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
3147 size_t tx_size, size_t rx_size)
3149 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3152 ofproto->stats.rx_packets++;
3153 ofproto->stats.rx_bytes += rx_size;
3156 ofproto->stats.tx_packets++;
3157 ofproto->stats.tx_bytes += tx_size;
3161 struct port_dump_state {
3166 struct ofproto_port port;
3171 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3173 *statep = xzalloc(sizeof(struct port_dump_state));
3178 port_dump_next(const struct ofproto *ofproto_, void *state_,
3179 struct ofproto_port *port)
3181 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3182 struct port_dump_state *state = state_;
3183 const struct sset *sset;
3184 struct sset_node *node;
3186 if (state->has_port) {
3187 ofproto_port_destroy(&state->port);
3188 state->has_port = false;
3190 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3191 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3194 error = port_query_by_name(ofproto_, node->name, &state->port);
3196 *port = state->port;
3197 state->has_port = true;
3199 } else if (error != ENODEV) {
3204 if (!state->ghost) {
3205 state->ghost = true;
3208 return port_dump_next(ofproto_, state_, port);
3215 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3217 struct port_dump_state *state = state_;
3219 if (state->has_port) {
3220 ofproto_port_destroy(&state->port);
3227 port_poll(const struct ofproto *ofproto_, char **devnamep)
3229 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3231 if (ofproto->port_poll_errno) {
3232 int error = ofproto->port_poll_errno;
3233 ofproto->port_poll_errno = 0;
3237 if (sset_is_empty(&ofproto->port_poll_set)) {
3241 *devnamep = sset_pop(&ofproto->port_poll_set);
3246 port_poll_wait(const struct ofproto *ofproto_)
3248 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3249 dpif_port_poll_wait(ofproto->backer->dpif);
3253 port_is_lacp_current(const struct ofport *ofport_)
3255 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3256 return (ofport->bundle && ofport->bundle->lacp
3257 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3261 /* Upcall handling. */
3263 /* Flow miss batching.
3265 * Some dpifs implement operations faster when you hand them off in a batch.
3266 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3267 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3268 * more packets, plus possibly installing the flow in the dpif.
3270 * So far we only batch the operations that affect flow setup time the most.
3271 * It's possible to batch more than that, but the benefit might be minimal. */
3273 struct hmap_node hmap_node;
3274 struct ofproto_dpif *ofproto;
3276 enum odp_key_fitness key_fitness;
3277 const struct nlattr *key;
3279 ovs_be16 initial_tci;
3280 struct list packets;
3281 enum dpif_upcall_type upcall_type;
3282 uint32_t odp_in_port;
3285 struct flow_miss_op {
3286 struct dpif_op dpif_op;
3287 struct subfacet *subfacet; /* Subfacet */
3288 void *garbage; /* Pointer to pass to free(), NULL if none. */
3289 uint64_t stub[1024 / 8]; /* Temporary buffer. */
3292 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3293 * OpenFlow controller as necessary according to their individual
3294 * configurations. */
3296 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3297 const struct flow *flow)
3299 struct ofputil_packet_in pin;
3301 pin.packet = packet->data;
3302 pin.packet_len = packet->size;
3303 pin.reason = OFPR_NO_MATCH;
3304 pin.controller_id = 0;
3309 pin.send_len = 0; /* not used for flow table misses */
3311 flow_get_metadata(flow, &pin.fmd);
3313 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3316 static enum slow_path_reason
3317 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3318 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3322 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3324 cfm_process_heartbeat(ofport->cfm, packet);
3327 } else if (ofport->bundle && ofport->bundle->lacp
3328 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3330 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3333 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3335 stp_process_packet(ofport, packet);
3343 static struct flow_miss *
3344 flow_miss_find(struct hmap *todo, const struct flow *flow, uint32_t hash)
3346 struct flow_miss *miss;
3348 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3349 if (flow_equal(&miss->flow, flow)) {
3357 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3358 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3359 * 'miss' is associated with a subfacet the caller must also initialize the
3360 * returned op->subfacet, and if anything needs to be freed after processing
3361 * the op, the caller must initialize op->garbage also. */
3363 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3364 struct flow_miss_op *op)
3366 if (miss->flow.vlan_tci != miss->initial_tci) {
3367 /* This packet was received on a VLAN splinter port. We
3368 * added a VLAN to the packet to make the packet resemble
3369 * the flow, but the actions were composed assuming that
3370 * the packet contained no VLAN. So, we must remove the
3371 * VLAN header from the packet before trying to execute the
3373 eth_pop_vlan(packet);
3376 op->subfacet = NULL;
3378 op->dpif_op.type = DPIF_OP_EXECUTE;
3379 op->dpif_op.u.execute.key = miss->key;
3380 op->dpif_op.u.execute.key_len = miss->key_len;
3381 op->dpif_op.u.execute.packet = packet;
3384 /* Helper for handle_flow_miss_without_facet() and
3385 * handle_flow_miss_with_facet(). */
3387 handle_flow_miss_common(struct rule_dpif *rule,
3388 struct ofpbuf *packet, const struct flow *flow)
3390 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3392 ofproto->n_matches++;
3394 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3396 * Extra-special case for fail-open mode.
3398 * We are in fail-open mode and the packet matched the fail-open
3399 * rule, but we are connected to a controller too. We should send
3400 * the packet up to the controller in the hope that it will try to
3401 * set up a flow and thereby allow us to exit fail-open.
3403 * See the top-level comment in fail-open.c for more information.
3405 send_packet_in_miss(ofproto, packet, flow);
3409 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3410 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3411 * installing a datapath flow. The answer is usually "yes" (a return value of
3412 * true). However, for short flows the cost of bookkeeping is much higher than
3413 * the benefits, so when the datapath holds a large number of flows we impose
3414 * some heuristics to decide which flows are likely to be worth tracking. */
3416 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3417 struct flow_miss *miss, uint32_t hash)
3419 if (!ofproto->governor) {
3422 n_subfacets = hmap_count(&ofproto->subfacets);
3423 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3427 ofproto->governor = governor_create(ofproto->up.name);
3430 return governor_should_install_flow(ofproto->governor, hash,
3431 list_size(&miss->packets));
3434 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3435 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3436 * increment '*n_ops'. */
3438 handle_flow_miss_without_facet(struct flow_miss *miss,
3439 struct rule_dpif *rule,
3440 struct flow_miss_op *ops, size_t *n_ops)
3442 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3443 long long int now = time_msec();
3444 struct action_xlate_ctx ctx;
3445 struct ofpbuf *packet;
3447 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3448 struct flow_miss_op *op = &ops[*n_ops];
3449 struct dpif_flow_stats stats;
3450 struct ofpbuf odp_actions;
3452 COVERAGE_INC(facet_suppress);
3454 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3456 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3457 rule_credit_stats(rule, &stats);
3459 action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci,
3461 ctx.resubmit_stats = &stats;
3462 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3465 if (odp_actions.size) {
3466 struct dpif_execute *execute = &op->dpif_op.u.execute;
3468 init_flow_miss_execute_op(miss, packet, op);
3469 execute->actions = odp_actions.data;
3470 execute->actions_len = odp_actions.size;
3471 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3475 ofpbuf_uninit(&odp_actions);
3480 /* Handles 'miss', which matches 'facet'. May add any required datapath
3481 * operations to 'ops', incrementing '*n_ops' for each new op.
3483 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3484 * This is really important only for new facets: if we just called time_msec()
3485 * here, then the new subfacet or its packets could look (occasionally) as
3486 * though it was used some time after the facet was used. That can make a
3487 * one-packet flow look like it has a nonzero duration, which looks odd in
3488 * e.g. NetFlow statistics. */
3490 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3492 struct flow_miss_op *ops, size_t *n_ops)
3494 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3495 enum subfacet_path want_path;
3496 struct subfacet *subfacet;
3497 struct ofpbuf *packet;
3499 subfacet = subfacet_create(facet, miss, now);
3501 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3502 struct flow_miss_op *op = &ops[*n_ops];
3503 struct dpif_flow_stats stats;
3504 struct ofpbuf odp_actions;
3506 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3508 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3509 if (!subfacet->actions || subfacet->slow) {
3510 subfacet_make_actions(subfacet, packet, &odp_actions);
3513 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3514 subfacet_update_stats(subfacet, &stats);
3516 if (subfacet->actions_len) {
3517 struct dpif_execute *execute = &op->dpif_op.u.execute;
3519 init_flow_miss_execute_op(miss, packet, op);
3520 op->subfacet = subfacet;
3521 if (!subfacet->slow) {
3522 execute->actions = subfacet->actions;
3523 execute->actions_len = subfacet->actions_len;
3524 ofpbuf_uninit(&odp_actions);
3526 execute->actions = odp_actions.data;
3527 execute->actions_len = odp_actions.size;
3528 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3533 ofpbuf_uninit(&odp_actions);
3537 want_path = subfacet_want_path(subfacet->slow);
3538 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3539 struct flow_miss_op *op = &ops[(*n_ops)++];
3540 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3542 op->subfacet = subfacet;
3544 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3545 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3546 put->key = miss->key;
3547 put->key_len = miss->key_len;
3548 if (want_path == SF_FAST_PATH) {
3549 put->actions = subfacet->actions;
3550 put->actions_len = subfacet->actions_len;
3552 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
3553 op->stub, sizeof op->stub,
3554 &put->actions, &put->actions_len);
3560 /* Handles flow miss 'miss'. May add any required datapath operations
3561 * to 'ops', incrementing '*n_ops' for each new op. */
3563 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3566 struct ofproto_dpif *ofproto = miss->ofproto;
3567 struct facet *facet;
3571 /* The caller must ensure that miss->hmap_node.hash contains
3572 * flow_hash(miss->flow, 0). */
3573 hash = miss->hmap_node.hash;
3575 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3577 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
3579 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
3580 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
3584 facet = facet_create(rule, &miss->flow, hash);
3589 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3592 static struct drop_key *
3593 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3596 struct drop_key *drop_key;
3598 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3599 &backer->drop_keys) {
3600 if (drop_key->key_len == key_len
3601 && !memcmp(drop_key->key, key, key_len)) {
3609 drop_key_clear(struct dpif_backer *backer)
3611 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3612 struct drop_key *drop_key, *next;
3614 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3617 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3619 if (error && !VLOG_DROP_WARN(&rl)) {
3620 struct ds ds = DS_EMPTY_INITIALIZER;
3621 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3622 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3627 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3628 free(drop_key->key);
3633 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3634 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3635 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3636 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3637 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3638 * 'packet' ingressed.
3640 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3641 * 'flow''s in_port to OFPP_NONE.
3643 * This function does post-processing on data returned from
3644 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3645 * of the upcall processing logic. In particular, if the extracted in_port is
3646 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3647 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3648 * a VLAN header onto 'packet' (if it is nonnull).
3650 * Optionally, if nonnull, sets '*initial_tci' to the VLAN TCI with which the
3651 * packet was really received, that is, the actual VLAN TCI extracted by
3652 * odp_flow_key_to_flow(). (This differs from the value returned in
3653 * flow->vlan_tci only for packets received on VLAN splinters.)
3655 * Similarly, this function also includes some logic to help with tunnels. It
3656 * may modify 'flow' as necessary to make the tunneling implementation
3657 * transparent to the upcall processing logic.
3659 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3660 * or some other positive errno if there are other problems. */
3662 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3663 const struct nlattr *key, size_t key_len,
3664 struct flow *flow, enum odp_key_fitness *fitnessp,
3665 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3666 ovs_be16 *initial_tci)
3668 const struct ofport_dpif *port;
3669 enum odp_key_fitness fitness;
3672 fitness = odp_flow_key_to_flow(key, key_len, flow);
3673 if (fitness == ODP_FIT_ERROR) {
3679 *initial_tci = flow->vlan_tci;
3683 *odp_in_port = flow->in_port;
3686 if (tnl_port_should_receive(flow)) {
3687 const struct ofport *ofport = tnl_port_receive(flow);
3689 flow->in_port = OFPP_NONE;
3692 port = ofport_dpif_cast(ofport);
3694 /* We can't reproduce 'key' from 'flow'. */
3695 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3697 /* XXX: Since the tunnel module is not scoped per backer, it's
3698 * theoretically possible that we'll receive an ofport belonging to an
3699 * entirely different datapath. In practice, this can't happen because
3700 * no platforms has two separate datapaths which each support
3702 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3704 port = odp_port_to_ofport(backer, flow->in_port);
3706 flow->in_port = OFPP_NONE;
3710 flow->in_port = port->up.ofp_port;
3711 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3713 /* Make the packet resemble the flow, so that it gets sent to
3714 * an OpenFlow controller properly, so that it looks correct
3715 * for sFlow, and so that flow_extract() will get the correct
3716 * vlan_tci if it is called on 'packet'.
3718 * The allocated space inside 'packet' probably also contains
3719 * 'key', that is, both 'packet' and 'key' are probably part of
3720 * a struct dpif_upcall (see the large comment on that
3721 * structure definition), so pushing data on 'packet' is in
3722 * general not a good idea since it could overwrite 'key' or
3723 * free it as a side effect. However, it's OK in this special
3724 * case because we know that 'packet' is inside a Netlink
3725 * attribute: pushing 4 bytes will just overwrite the 4-byte
3726 * "struct nlattr", which is fine since we don't need that
3727 * header anymore. */
3728 eth_push_vlan(packet, flow->vlan_tci);
3730 /* We can't reproduce 'key' from 'flow'. */
3731 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3737 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3742 *fitnessp = fitness;
3748 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3751 struct dpif_upcall *upcall;
3752 struct flow_miss *miss;
3753 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3754 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3755 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3765 /* Construct the to-do list.
3767 * This just amounts to extracting the flow from each packet and sticking
3768 * the packets that have the same flow in the same "flow_miss" structure so
3769 * that we can process them together. */
3772 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3773 struct flow_miss *miss = &misses[n_misses];
3774 struct flow_miss *existing_miss;
3775 struct ofproto_dpif *ofproto;
3776 uint32_t odp_in_port;
3781 error = ofproto_receive(backer, upcall->packet, upcall->key,
3782 upcall->key_len, &flow, &miss->key_fitness,
3783 &ofproto, &odp_in_port, &miss->initial_tci);
3784 if (error == ENODEV) {
3785 struct drop_key *drop_key;
3787 /* Received packet on port for which we couldn't associate
3788 * an ofproto. This can happen if a port is removed while
3789 * traffic is being received. Print a rate-limited message
3790 * in case it happens frequently. Install a drop flow so
3791 * that future packets of the flow are inexpensively dropped
3793 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
3796 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3798 drop_key = xmalloc(sizeof *drop_key);
3799 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3800 drop_key->key_len = upcall->key_len;
3802 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3803 hash_bytes(drop_key->key, drop_key->key_len, 0));
3804 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3805 drop_key->key, drop_key->key_len, NULL, 0, NULL);
3812 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3813 &flow.tunnel, flow.in_port, &miss->flow);
3815 /* Add other packets to a to-do list. */
3816 hash = flow_hash(&miss->flow, 0);
3817 existing_miss = flow_miss_find(&todo, &miss->flow, hash);
3818 if (!existing_miss) {
3819 hmap_insert(&todo, &miss->hmap_node, hash);
3820 miss->ofproto = ofproto;
3821 miss->key = upcall->key;
3822 miss->key_len = upcall->key_len;
3823 miss->upcall_type = upcall->type;
3824 miss->odp_in_port = odp_in_port;
3825 list_init(&miss->packets);
3829 miss = existing_miss;
3831 list_push_back(&miss->packets, &upcall->packet->list_node);
3834 /* Process each element in the to-do list, constructing the set of
3835 * operations to batch. */
3837 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3838 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3840 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3842 /* Execute batch. */
3843 for (i = 0; i < n_ops; i++) {
3844 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3846 dpif_operate(backer->dpif, dpif_ops, n_ops);
3848 /* Free memory and update facets. */
3849 for (i = 0; i < n_ops; i++) {
3850 struct flow_miss_op *op = &flow_miss_ops[i];
3852 switch (op->dpif_op.type) {
3853 case DPIF_OP_EXECUTE:
3856 case DPIF_OP_FLOW_PUT:
3857 if (!op->dpif_op.error) {
3858 op->subfacet->path = subfacet_want_path(op->subfacet->slow);
3862 case DPIF_OP_FLOW_DEL:
3868 hmap_destroy(&todo);
3871 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
3872 classify_upcall(const struct dpif_upcall *upcall)
3874 union user_action_cookie cookie;
3876 /* First look at the upcall type. */
3877 switch (upcall->type) {
3878 case DPIF_UC_ACTION:
3884 case DPIF_N_UC_TYPES:
3886 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3890 /* "action" upcalls need a closer look. */
3891 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3892 switch (cookie.type) {
3893 case USER_ACTION_COOKIE_SFLOW:
3894 return SFLOW_UPCALL;
3896 case USER_ACTION_COOKIE_SLOW_PATH:
3899 case USER_ACTION_COOKIE_UNSPEC:
3901 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
3907 handle_sflow_upcall(struct dpif_backer *backer,
3908 const struct dpif_upcall *upcall)
3910 struct ofproto_dpif *ofproto;
3911 union user_action_cookie cookie;
3913 uint32_t odp_in_port;
3915 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3916 &flow, NULL, &ofproto, &odp_in_port, NULL)
3917 || !ofproto->sflow) {
3921 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3922 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3923 odp_in_port, &cookie);
3927 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3929 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3930 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3931 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3936 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
3939 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3940 struct dpif_upcall *upcall = &misses[n_misses];
3941 struct ofpbuf *buf = &miss_bufs[n_misses];
3944 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3945 sizeof miss_buf_stubs[n_misses]);
3946 error = dpif_recv(backer->dpif, upcall, buf);
3952 switch (classify_upcall(upcall)) {
3954 /* Handle it later. */
3959 handle_sflow_upcall(backer, upcall);
3969 /* Handle deferred MISS_UPCALL processing. */
3970 handle_miss_upcalls(backer, misses, n_misses);
3971 for (i = 0; i < n_misses; i++) {
3972 ofpbuf_uninit(&miss_bufs[i]);
3978 /* Flow expiration. */
3980 static int subfacet_max_idle(const struct ofproto_dpif *);
3981 static void update_stats(struct dpif_backer *);
3982 static void rule_expire(struct rule_dpif *);
3983 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
3985 /* This function is called periodically by run(). Its job is to collect
3986 * updates for the flows that have been installed into the datapath, most
3987 * importantly when they last were used, and then use that information to
3988 * expire flows that have not been used recently.
3990 * Returns the number of milliseconds after which it should be called again. */
3992 expire(struct dpif_backer *backer)
3994 struct ofproto_dpif *ofproto;
3995 int max_idle = INT32_MAX;
3997 /* Periodically clear out the drop keys in an effort to keep them
3998 * relatively few. */
3999 drop_key_clear(backer);
4001 /* Update stats for each flow in the backer. */
4002 update_stats(backer);
4004 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4005 struct rule *rule, *next_rule;
4008 if (ofproto->backer != backer) {
4012 /* Expire subfacets that have been idle too long. */
4013 dp_max_idle = subfacet_max_idle(ofproto);
4014 expire_subfacets(ofproto, dp_max_idle);
4016 max_idle = MIN(max_idle, dp_max_idle);
4018 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4020 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4021 &ofproto->up.expirable) {
4022 rule_expire(rule_dpif_cast(rule));
4025 /* All outstanding data in existing flows has been accounted, so it's a
4026 * good time to do bond rebalancing. */
4027 if (ofproto->has_bonded_bundles) {
4028 struct ofbundle *bundle;
4030 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4032 bond_rebalance(bundle->bond, &backer->revalidate_set);
4038 return MIN(max_idle, 1000);
4041 /* Updates flow table statistics given that the datapath just reported 'stats'
4042 * as 'subfacet''s statistics. */
4044 update_subfacet_stats(struct subfacet *subfacet,
4045 const struct dpif_flow_stats *stats)
4047 struct facet *facet = subfacet->facet;
4049 if (stats->n_packets >= subfacet->dp_packet_count) {
4050 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
4051 facet->packet_count += extra;
4053 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4056 if (stats->n_bytes >= subfacet->dp_byte_count) {
4057 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
4059 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4062 subfacet->dp_packet_count = stats->n_packets;
4063 subfacet->dp_byte_count = stats->n_bytes;
4065 facet->tcp_flags |= stats->tcp_flags;
4067 subfacet_update_time(subfacet, stats->used);
4068 if (facet->accounted_bytes < facet->byte_count) {
4070 facet_account(facet);
4071 facet->accounted_bytes = facet->byte_count;
4073 facet_push_stats(facet);
4076 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4077 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4079 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4080 const struct nlattr *key, size_t key_len)
4082 if (!VLOG_DROP_WARN(&rl)) {
4086 odp_flow_key_format(key, key_len, &s);
4087 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4091 COVERAGE_INC(facet_unexpected);
4092 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4095 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4097 * This function also pushes statistics updates to rules which each facet
4098 * resubmits into. Generally these statistics will be accurate. However, if a
4099 * facet changes the rule it resubmits into at some time in between
4100 * update_stats() runs, it is possible that statistics accrued to the
4101 * old rule will be incorrectly attributed to the new rule. This could be
4102 * avoided by calling update_stats() whenever rules are created or
4103 * deleted. However, the performance impact of making so many calls to the
4104 * datapath do not justify the benefit of having perfectly accurate statistics.
4107 update_stats(struct dpif_backer *backer)
4109 const struct dpif_flow_stats *stats;
4110 struct dpif_flow_dump dump;
4111 const struct nlattr *key;
4114 dpif_flow_dump_start(&dump, backer->dpif);
4115 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4117 struct subfacet *subfacet;
4118 struct ofproto_dpif *ofproto;
4119 struct ofport_dpif *ofport;
4122 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4127 ofport = get_ofp_port(ofproto, flow.in_port);
4128 if (ofport && ofport->tnl_port) {
4129 netdev_vport_inc_rx(ofport->up.netdev, stats);
4132 key_hash = odp_flow_key_hash(key, key_len);
4133 subfacet = subfacet_find(ofproto, key, key_len, key_hash, &flow);
4134 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4136 update_subfacet_stats(subfacet, stats);
4140 /* Stats are updated per-packet. */
4143 case SF_NOT_INSTALLED:
4145 delete_unexpected_flow(ofproto, key, key_len);
4149 dpif_flow_dump_done(&dump);
4152 /* Calculates and returns the number of milliseconds of idle time after which
4153 * subfacets should expire from the datapath. When a subfacet expires, we fold
4154 * its statistics into its facet, and when a facet's last subfacet expires, we
4155 * fold its statistic into its rule. */
4157 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4160 * Idle time histogram.
4162 * Most of the time a switch has a relatively small number of subfacets.
4163 * When this is the case we might as well keep statistics for all of them
4164 * in userspace and to cache them in the kernel datapath for performance as
4167 * As the number of subfacets increases, the memory required to maintain
4168 * statistics about them in userspace and in the kernel becomes
4169 * significant. However, with a large number of subfacets it is likely
4170 * that only a few of them are "heavy hitters" that consume a large amount
4171 * of bandwidth. At this point, only heavy hitters are worth caching in
4172 * the kernel and maintaining in userspaces; other subfacets we can
4175 * The technique used to compute the idle time is to build a histogram with
4176 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4177 * that is installed in the kernel gets dropped in the appropriate bucket.
4178 * After the histogram has been built, we compute the cutoff so that only
4179 * the most-recently-used 1% of subfacets (but at least
4180 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4181 * the most-recently-used bucket of subfacets is kept, so actually an
4182 * arbitrary number of subfacets can be kept in any given expiration run
4183 * (though the next run will delete most of those unless they receive
4186 * This requires a second pass through the subfacets, in addition to the
4187 * pass made by update_stats(), because the former function never looks at
4188 * uninstallable subfacets.
4190 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4191 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4192 int buckets[N_BUCKETS] = { 0 };
4193 int total, subtotal, bucket;
4194 struct subfacet *subfacet;
4198 total = hmap_count(&ofproto->subfacets);
4199 if (total <= ofproto->up.flow_eviction_threshold) {
4200 return N_BUCKETS * BUCKET_WIDTH;
4203 /* Build histogram. */
4205 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4206 long long int idle = now - subfacet->used;
4207 int bucket = (idle <= 0 ? 0
4208 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4209 : (unsigned int) idle / BUCKET_WIDTH);
4213 /* Find the first bucket whose flows should be expired. */
4214 subtotal = bucket = 0;
4216 subtotal += buckets[bucket++];
4217 } while (bucket < N_BUCKETS &&
4218 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4220 if (VLOG_IS_DBG_ENABLED()) {
4224 ds_put_cstr(&s, "keep");
4225 for (i = 0; i < N_BUCKETS; i++) {
4227 ds_put_cstr(&s, ", drop");
4230 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4233 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4237 return bucket * BUCKET_WIDTH;
4241 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4243 /* Cutoff time for most flows. */
4244 long long int normal_cutoff = time_msec() - dp_max_idle;
4246 /* We really want to keep flows for special protocols around, so use a more
4247 * conservative cutoff. */
4248 long long int special_cutoff = time_msec() - 10000;
4250 struct subfacet *subfacet, *next_subfacet;
4251 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4255 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4256 &ofproto->subfacets) {
4257 long long int cutoff;
4259 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
4262 if (subfacet->used < cutoff) {
4263 if (subfacet->path != SF_NOT_INSTALLED) {
4264 batch[n_batch++] = subfacet;
4265 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4266 subfacet_destroy_batch(ofproto, batch, n_batch);
4270 subfacet_destroy(subfacet);
4276 subfacet_destroy_batch(ofproto, batch, n_batch);
4280 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4281 * then delete it entirely. */
4283 rule_expire(struct rule_dpif *rule)
4285 struct facet *facet, *next_facet;
4289 if (rule->up.pending) {
4290 /* We'll have to expire it later. */
4294 /* Has 'rule' expired? */
4296 if (rule->up.hard_timeout
4297 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4298 reason = OFPRR_HARD_TIMEOUT;
4299 } else if (rule->up.idle_timeout
4300 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4301 reason = OFPRR_IDLE_TIMEOUT;
4306 COVERAGE_INC(ofproto_dpif_expired);
4308 /* Update stats. (This is a no-op if the rule expired due to an idle
4309 * timeout, because that only happens when the rule has no facets left.) */
4310 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4311 facet_remove(facet);
4314 /* Get rid of the rule. */
4315 ofproto_rule_expire(&rule->up, reason);
4320 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
4322 * The caller must already have determined that no facet with an identical
4323 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
4324 * the ofproto's classifier table.
4326 * 'hash' must be the return value of flow_hash(flow, 0).
4328 * The facet will initially have no subfacets. The caller should create (at
4329 * least) one subfacet with subfacet_create(). */
4330 static struct facet *
4331 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
4333 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4334 struct facet *facet;
4336 facet = xzalloc(sizeof *facet);
4337 facet->used = time_msec();
4338 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4339 list_push_back(&rule->facets, &facet->list_node);
4341 facet->flow = *flow;
4342 list_init(&facet->subfacets);
4343 netflow_flow_init(&facet->nf_flow);
4344 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4350 facet_free(struct facet *facet)
4355 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4356 * 'packet', which arrived on 'in_port'. */
4358 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4359 const struct nlattr *odp_actions, size_t actions_len,
4360 struct ofpbuf *packet)
4362 struct odputil_keybuf keybuf;
4366 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4367 odp_flow_key_from_flow(&key, flow,
4368 ofp_port_to_odp_port(ofproto, flow->in_port));
4370 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4371 odp_actions, actions_len, packet);
4375 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4377 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4378 * rule's statistics, via subfacet_uninstall().
4380 * - Removes 'facet' from its rule and from ofproto->facets.
4383 facet_remove(struct facet *facet)
4385 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4386 struct subfacet *subfacet, *next_subfacet;
4388 ovs_assert(!list_is_empty(&facet->subfacets));
4390 /* First uninstall all of the subfacets to get final statistics. */
4391 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4392 subfacet_uninstall(subfacet);
4395 /* Flush the final stats to the rule.
4397 * This might require us to have at least one subfacet around so that we
4398 * can use its actions for accounting in facet_account(), which is why we
4399 * have uninstalled but not yet destroyed the subfacets. */
4400 facet_flush_stats(facet);
4402 /* Now we're really all done so destroy everything. */
4403 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4404 &facet->subfacets) {
4405 subfacet_destroy__(subfacet);
4407 hmap_remove(&ofproto->facets, &facet->hmap_node);
4408 list_remove(&facet->list_node);
4412 /* Feed information from 'facet' back into the learning table to keep it in
4413 * sync with what is actually flowing through the datapath. */
4415 facet_learn(struct facet *facet)
4417 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4418 struct action_xlate_ctx ctx;
4420 if (!facet->has_learn
4421 && !facet->has_normal
4422 && (!facet->has_fin_timeout
4423 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4427 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4428 facet->flow.vlan_tci,
4429 facet->rule, facet->tcp_flags, NULL);
4430 ctx.may_learn = true;
4431 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
4432 facet->rule->up.ofpacts_len);
4436 facet_account(struct facet *facet)
4438 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4439 struct subfacet *subfacet;
4440 const struct nlattr *a;
4445 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
4448 n_bytes = facet->byte_count - facet->accounted_bytes;
4450 /* This loop feeds byte counters to bond_account() for rebalancing to use
4451 * as a basis. We also need to track the actual VLAN on which the packet
4452 * is going to be sent to ensure that it matches the one passed to
4453 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4456 * We use the actions from an arbitrary subfacet because they should all
4457 * be equally valid for our purpose. */
4458 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
4459 struct subfacet, list_node);
4460 vlan_tci = facet->flow.vlan_tci;
4461 NL_ATTR_FOR_EACH_UNSAFE (a, left,
4462 subfacet->actions, subfacet->actions_len) {
4463 const struct ovs_action_push_vlan *vlan;
4464 struct ofport_dpif *port;
4466 switch (nl_attr_type(a)) {
4467 case OVS_ACTION_ATTR_OUTPUT:
4468 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4469 if (port && port->bundle && port->bundle->bond) {
4470 bond_account(port->bundle->bond, &facet->flow,
4471 vlan_tci_to_vid(vlan_tci), n_bytes);
4475 case OVS_ACTION_ATTR_POP_VLAN:
4476 vlan_tci = htons(0);
4479 case OVS_ACTION_ATTR_PUSH_VLAN:
4480 vlan = nl_attr_get(a);
4481 vlan_tci = vlan->vlan_tci;
4487 /* Returns true if the only action for 'facet' is to send to the controller.
4488 * (We don't report NetFlow expiration messages for such facets because they
4489 * are just part of the control logic for the network, not real traffic). */
4491 facet_is_controller_flow(struct facet *facet)
4494 const struct rule *rule = &facet->rule->up;
4495 const struct ofpact *ofpacts = rule->ofpacts;
4496 size_t ofpacts_len = rule->ofpacts_len;
4498 if (ofpacts_len > 0 &&
4499 ofpacts->type == OFPACT_CONTROLLER &&
4500 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4507 /* Folds all of 'facet''s statistics into its rule. Also updates the
4508 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4509 * 'facet''s statistics in the datapath should have been zeroed and folded into
4510 * its packet and byte counts before this function is called. */
4512 facet_flush_stats(struct facet *facet)
4514 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4515 struct subfacet *subfacet;
4517 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4518 ovs_assert(!subfacet->dp_byte_count);
4519 ovs_assert(!subfacet->dp_packet_count);
4522 facet_push_stats(facet);
4523 if (facet->accounted_bytes < facet->byte_count) {
4524 facet_account(facet);
4525 facet->accounted_bytes = facet->byte_count;
4528 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4529 struct ofexpired expired;
4530 expired.flow = facet->flow;
4531 expired.packet_count = facet->packet_count;
4532 expired.byte_count = facet->byte_count;
4533 expired.used = facet->used;
4534 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4537 facet->rule->packet_count += facet->packet_count;
4538 facet->rule->byte_count += facet->byte_count;
4540 /* Reset counters to prevent double counting if 'facet' ever gets
4542 facet_reset_counters(facet);
4544 netflow_flow_clear(&facet->nf_flow);
4545 facet->tcp_flags = 0;
4548 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4549 * Returns it if found, otherwise a null pointer.
4551 * 'hash' must be the return value of flow_hash(flow, 0).
4553 * The returned facet might need revalidation; use facet_lookup_valid()
4554 * instead if that is important. */
4555 static struct facet *
4556 facet_find(struct ofproto_dpif *ofproto,
4557 const struct flow *flow, uint32_t hash)
4559 struct facet *facet;
4561 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4562 if (flow_equal(flow, &facet->flow)) {
4570 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4571 * Returns it if found, otherwise a null pointer.
4573 * 'hash' must be the return value of flow_hash(flow, 0).
4575 * The returned facet is guaranteed to be valid. */
4576 static struct facet *
4577 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4580 struct facet *facet;
4582 facet = facet_find(ofproto, flow, hash);
4584 && (ofproto->backer->need_revalidate
4585 || tag_set_intersects(&ofproto->backer->revalidate_set,
4587 facet_revalidate(facet);
4594 subfacet_path_to_string(enum subfacet_path path)
4597 case SF_NOT_INSTALLED:
4598 return "not installed";
4600 return "in fast path";
4602 return "in slow path";
4608 /* Returns the path in which a subfacet should be installed if its 'slow'
4609 * member has the specified value. */
4610 static enum subfacet_path
4611 subfacet_want_path(enum slow_path_reason slow)
4613 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
4616 /* Returns true if 'subfacet' needs to have its datapath flow updated,
4617 * supposing that its actions have been recalculated as 'want_actions' and that
4618 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
4620 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
4621 const struct ofpbuf *want_actions)
4623 enum subfacet_path want_path = subfacet_want_path(slow);
4624 return (want_path != subfacet->path
4625 || (want_path == SF_FAST_PATH
4626 && (subfacet->actions_len != want_actions->size
4627 || memcmp(subfacet->actions, want_actions->data,
4628 subfacet->actions_len))));
4632 facet_check_consistency(struct facet *facet)
4634 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4636 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4638 uint64_t odp_actions_stub[1024 / 8];
4639 struct ofpbuf odp_actions;
4641 struct rule_dpif *rule;
4642 struct subfacet *subfacet;
4643 bool may_log = false;
4646 /* Check the rule for consistency. */
4647 rule = rule_dpif_lookup(ofproto, &facet->flow);
4648 ok = rule == facet->rule;
4650 may_log = !VLOG_DROP_WARN(&rl);
4655 flow_format(&s, &facet->flow);
4656 ds_put_format(&s, ": facet associated with wrong rule (was "
4657 "table=%"PRIu8",", facet->rule->up.table_id);
4658 cls_rule_format(&facet->rule->up.cr, &s);
4659 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4661 cls_rule_format(&rule->up.cr, &s);
4662 ds_put_char(&s, ')');
4664 VLOG_WARN("%s", ds_cstr(&s));
4669 /* Check the datapath actions for consistency. */
4670 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4671 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4672 enum subfacet_path want_path;
4673 struct odputil_keybuf keybuf;
4674 struct action_xlate_ctx ctx;
4678 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4679 subfacet->initial_tci, rule, 0, NULL);
4680 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
4683 if (subfacet->path == SF_NOT_INSTALLED) {
4684 /* This only happens if the datapath reported an error when we
4685 * tried to install the flow. Don't flag another error here. */
4689 want_path = subfacet_want_path(subfacet->slow);
4690 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
4691 /* The actions for slow-path flows may legitimately vary from one
4692 * packet to the next. We're done. */
4696 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
4700 /* Inconsistency! */
4702 may_log = !VLOG_DROP_WARN(&rl);
4706 /* Rate-limited, skip reporting. */
4711 subfacet_get_key(subfacet, &keybuf, &key);
4712 odp_flow_key_format(key.data, key.size, &s);
4714 ds_put_cstr(&s, ": inconsistency in subfacet");
4715 if (want_path != subfacet->path) {
4716 enum odp_key_fitness fitness = subfacet->key_fitness;
4718 ds_put_format(&s, " (%s, fitness=%s)",
4719 subfacet_path_to_string(subfacet->path),
4720 odp_key_fitness_to_string(fitness));
4721 ds_put_format(&s, " (should have been %s)",
4722 subfacet_path_to_string(want_path));
4723 } else if (want_path == SF_FAST_PATH) {
4724 ds_put_cstr(&s, " (actions were: ");
4725 format_odp_actions(&s, subfacet->actions,
4726 subfacet->actions_len);
4727 ds_put_cstr(&s, ") (correct actions: ");
4728 format_odp_actions(&s, odp_actions.data, odp_actions.size);
4729 ds_put_char(&s, ')');
4731 ds_put_cstr(&s, " (actions: ");
4732 format_odp_actions(&s, subfacet->actions,
4733 subfacet->actions_len);
4734 ds_put_char(&s, ')');
4736 VLOG_WARN("%s", ds_cstr(&s));
4739 ofpbuf_uninit(&odp_actions);
4744 /* Re-searches the classifier for 'facet':
4746 * - If the rule found is different from 'facet''s current rule, moves
4747 * 'facet' to the new rule and recompiles its actions.
4749 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4750 * where it is and recompiles its actions anyway. */
4752 facet_revalidate(struct facet *facet)
4754 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4756 struct nlattr *odp_actions;
4759 struct actions *new_actions;
4761 struct action_xlate_ctx ctx;
4762 uint64_t odp_actions_stub[1024 / 8];
4763 struct ofpbuf odp_actions;
4765 struct rule_dpif *new_rule;
4766 struct subfacet *subfacet;
4769 COVERAGE_INC(facet_revalidate);
4771 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
4773 /* Calculate new datapath actions.
4775 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4776 * emit a NetFlow expiration and, if so, we need to have the old state
4777 * around to properly compose it. */
4779 /* If the datapath actions changed or the installability changed,
4780 * then we need to talk to the datapath. */
4783 memset(&ctx, 0, sizeof ctx);
4784 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4785 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4786 enum slow_path_reason slow;
4788 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4789 subfacet->initial_tci, new_rule, 0, NULL);
4790 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
4793 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4794 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
4795 struct dpif_flow_stats stats;
4797 subfacet_install(subfacet,
4798 odp_actions.data, odp_actions.size, &stats, slow);
4799 subfacet_update_stats(subfacet, &stats);
4802 new_actions = xcalloc(list_size(&facet->subfacets),
4803 sizeof *new_actions);
4805 new_actions[i].odp_actions = xmemdup(odp_actions.data,
4807 new_actions[i].actions_len = odp_actions.size;
4812 ofpbuf_uninit(&odp_actions);
4815 facet_flush_stats(facet);
4818 /* Update 'facet' now that we've taken care of all the old state. */
4819 facet->tags = ctx.tags;
4820 facet->nf_flow.output_iface = ctx.nf_output_iface;
4821 facet->has_learn = ctx.has_learn;
4822 facet->has_normal = ctx.has_normal;
4823 facet->has_fin_timeout = ctx.has_fin_timeout;
4824 facet->mirrors = ctx.mirrors;
4827 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4828 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4830 if (new_actions && new_actions[i].odp_actions) {
4831 free(subfacet->actions);
4832 subfacet->actions = new_actions[i].odp_actions;
4833 subfacet->actions_len = new_actions[i].actions_len;
4839 if (facet->rule != new_rule) {
4840 COVERAGE_INC(facet_changed_rule);
4841 list_remove(&facet->list_node);
4842 list_push_back(&new_rule->facets, &facet->list_node);
4843 facet->rule = new_rule;
4844 facet->used = new_rule->up.created;
4845 facet->prev_used = facet->used;
4849 /* Updates 'facet''s used time. Caller is responsible for calling
4850 * facet_push_stats() to update the flows which 'facet' resubmits into. */
4852 facet_update_time(struct facet *facet, long long int used)
4854 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4855 if (used > facet->used) {
4857 ofproto_rule_update_used(&facet->rule->up, used);
4858 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
4863 facet_reset_counters(struct facet *facet)
4865 facet->packet_count = 0;
4866 facet->byte_count = 0;
4867 facet->prev_packet_count = 0;
4868 facet->prev_byte_count = 0;
4869 facet->accounted_bytes = 0;
4873 facet_push_stats(struct facet *facet)
4875 struct dpif_flow_stats stats;
4877 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4878 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4879 ovs_assert(facet->used >= facet->prev_used);
4881 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4882 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4883 stats.used = facet->used;
4884 stats.tcp_flags = 0;
4886 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
4887 facet->prev_packet_count = facet->packet_count;
4888 facet->prev_byte_count = facet->byte_count;
4889 facet->prev_used = facet->used;
4891 flow_push_stats(facet->rule, &facet->flow, &stats);
4893 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
4894 facet->mirrors, stats.n_packets, stats.n_bytes);
4899 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4901 rule->packet_count += stats->n_packets;
4902 rule->byte_count += stats->n_bytes;
4903 ofproto_rule_update_used(&rule->up, stats->used);
4906 /* Pushes flow statistics to the rules which 'flow' resubmits into given
4907 * 'rule''s actions and mirrors. */
4909 flow_push_stats(struct rule_dpif *rule,
4910 const struct flow *flow, const struct dpif_flow_stats *stats)
4912 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4913 struct action_xlate_ctx ctx;
4915 ofproto_rule_update_used(&rule->up, stats->used);
4917 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule,
4919 ctx.resubmit_stats = stats;
4920 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
4921 rule->up.ofpacts_len);
4926 static struct subfacet *
4927 subfacet_find(struct ofproto_dpif *ofproto,
4928 const struct nlattr *key, size_t key_len, uint32_t key_hash,
4929 const struct flow *flow)
4931 struct subfacet *subfacet;
4933 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4934 &ofproto->subfacets) {
4936 ? (subfacet->key_len == key_len
4937 && !memcmp(key, subfacet->key, key_len))
4938 : flow_equal(flow, &subfacet->facet->flow)) {
4946 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4947 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4948 * existing subfacet if there is one, otherwise creates and returns a
4951 * If the returned subfacet is new, then subfacet->actions will be NULL, in
4952 * which case the caller must populate the actions with
4953 * subfacet_make_actions(). */
4954 static struct subfacet *
4955 subfacet_create(struct facet *facet, struct flow_miss *miss,
4958 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4959 enum odp_key_fitness key_fitness = miss->key_fitness;
4960 const struct nlattr *key = miss->key;
4961 size_t key_len = miss->key_len;
4963 struct subfacet *subfacet;
4965 key_hash = odp_flow_key_hash(key, key_len);
4967 if (list_is_empty(&facet->subfacets)) {
4968 subfacet = &facet->one_subfacet;
4970 subfacet = subfacet_find(ofproto, key, key_len, key_hash,
4973 if (subfacet->facet == facet) {
4977 /* This shouldn't happen. */
4978 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4979 subfacet_destroy(subfacet);
4982 subfacet = xmalloc(sizeof *subfacet);
4985 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
4986 list_push_back(&facet->subfacets, &subfacet->list_node);
4987 subfacet->facet = facet;
4988 subfacet->key_fitness = key_fitness;
4989 if (key_fitness != ODP_FIT_PERFECT) {
4990 subfacet->key = xmemdup(key, key_len);
4991 subfacet->key_len = key_len;
4993 subfacet->key = NULL;
4994 subfacet->key_len = 0;
4996 subfacet->used = now;
4997 subfacet->dp_packet_count = 0;
4998 subfacet->dp_byte_count = 0;
4999 subfacet->actions_len = 0;
5000 subfacet->actions = NULL;
5001 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
5004 subfacet->path = SF_NOT_INSTALLED;
5005 subfacet->initial_tci = miss->initial_tci;
5006 subfacet->odp_in_port = miss->odp_in_port;
5011 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5012 * its facet within 'ofproto', and frees it. */
5014 subfacet_destroy__(struct subfacet *subfacet)
5016 struct facet *facet = subfacet->facet;
5017 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5019 subfacet_uninstall(subfacet);
5020 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5021 list_remove(&subfacet->list_node);
5022 free(subfacet->key);
5023 free(subfacet->actions);
5024 if (subfacet != &facet->one_subfacet) {
5029 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5030 * last remaining subfacet in its facet destroys the facet too. */
5032 subfacet_destroy(struct subfacet *subfacet)
5034 struct facet *facet = subfacet->facet;
5036 if (list_is_singleton(&facet->subfacets)) {
5037 /* facet_remove() needs at least one subfacet (it will remove it). */
5038 facet_remove(facet);
5040 subfacet_destroy__(subfacet);
5045 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5046 struct subfacet **subfacets, int n)
5048 struct odputil_keybuf keybufs[SUBFACET_DESTROY_MAX_BATCH];
5049 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5050 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5051 struct ofpbuf keys[SUBFACET_DESTROY_MAX_BATCH];
5052 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5055 for (i = 0; i < n; i++) {
5056 ops[i].type = DPIF_OP_FLOW_DEL;
5057 subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]);
5058 ops[i].u.flow_del.key = keys[i].data;
5059 ops[i].u.flow_del.key_len = keys[i].size;
5060 ops[i].u.flow_del.stats = &stats[i];
5064 dpif_operate(ofproto->backer->dpif, opsp, n);
5065 for (i = 0; i < n; i++) {
5066 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5067 subfacets[i]->path = SF_NOT_INSTALLED;
5068 subfacet_destroy(subfacets[i]);
5072 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
5073 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
5074 * for use as temporary storage. */
5076 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
5080 if (!subfacet->key) {
5081 struct flow *flow = &subfacet->facet->flow;
5083 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
5084 odp_flow_key_from_flow(key, flow, subfacet->odp_in_port);
5086 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
5090 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
5091 * Translates the actions into 'odp_actions', which the caller must have
5092 * initialized and is responsible for uninitializing. */
5094 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
5095 struct ofpbuf *odp_actions)
5097 struct facet *facet = subfacet->facet;
5098 struct rule_dpif *rule = facet->rule;
5099 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5101 struct action_xlate_ctx ctx;
5103 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
5105 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
5106 facet->tags = ctx.tags;
5107 facet->has_learn = ctx.has_learn;
5108 facet->has_normal = ctx.has_normal;
5109 facet->has_fin_timeout = ctx.has_fin_timeout;
5110 facet->nf_flow.output_iface = ctx.nf_output_iface;
5111 facet->mirrors = ctx.mirrors;
5113 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
5114 if (subfacet->actions_len != odp_actions->size
5115 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
5116 free(subfacet->actions);
5117 subfacet->actions_len = odp_actions->size;
5118 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
5122 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5123 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5124 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5125 * since 'subfacet' was last updated.
5127 * Returns 0 if successful, otherwise a positive errno value. */
5129 subfacet_install(struct subfacet *subfacet,
5130 const struct nlattr *actions, size_t actions_len,
5131 struct dpif_flow_stats *stats,
5132 enum slow_path_reason slow)
5134 struct facet *facet = subfacet->facet;
5135 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5136 enum subfacet_path path = subfacet_want_path(slow);
5137 uint64_t slow_path_stub[128 / 8];
5138 struct odputil_keybuf keybuf;
5139 enum dpif_flow_put_flags flags;
5143 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5145 flags |= DPIF_FP_ZERO_STATS;
5148 if (path == SF_SLOW_PATH) {
5149 compose_slow_path(ofproto, &facet->flow, slow,
5150 slow_path_stub, sizeof slow_path_stub,
5151 &actions, &actions_len);
5154 subfacet_get_key(subfacet, &keybuf, &key);
5155 ret = dpif_flow_put(ofproto->backer->dpif, flags, key.data, key.size,
5156 actions, actions_len, stats);
5159 subfacet_reset_dp_stats(subfacet, stats);
5163 subfacet->path = path;
5169 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
5171 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
5172 stats, subfacet->slow);
5175 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5177 subfacet_uninstall(struct subfacet *subfacet)
5179 if (subfacet->path != SF_NOT_INSTALLED) {
5180 struct rule_dpif *rule = subfacet->facet->rule;
5181 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5182 struct odputil_keybuf keybuf;
5183 struct dpif_flow_stats stats;
5187 subfacet_get_key(subfacet, &keybuf, &key);
5188 error = dpif_flow_del(ofproto->backer->dpif,
5189 key.data, key.size, &stats);
5190 subfacet_reset_dp_stats(subfacet, &stats);
5192 subfacet_update_stats(subfacet, &stats);
5194 subfacet->path = SF_NOT_INSTALLED;
5196 ovs_assert(subfacet->dp_packet_count == 0);
5197 ovs_assert(subfacet->dp_byte_count == 0);
5201 /* Resets 'subfacet''s datapath statistics counters. This should be called
5202 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5203 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5204 * was reset in the datapath. 'stats' will be modified to include only
5205 * statistics new since 'subfacet' was last updated. */
5207 subfacet_reset_dp_stats(struct subfacet *subfacet,
5208 struct dpif_flow_stats *stats)
5211 && subfacet->dp_packet_count <= stats->n_packets
5212 && subfacet->dp_byte_count <= stats->n_bytes) {
5213 stats->n_packets -= subfacet->dp_packet_count;
5214 stats->n_bytes -= subfacet->dp_byte_count;
5217 subfacet->dp_packet_count = 0;
5218 subfacet->dp_byte_count = 0;
5221 /* Updates 'subfacet''s used time. The caller is responsible for calling
5222 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
5224 subfacet_update_time(struct subfacet *subfacet, long long int used)
5226 if (used > subfacet->used) {
5227 subfacet->used = used;
5228 facet_update_time(subfacet->facet, used);
5232 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5234 * Because of the meaning of a subfacet's counters, it only makes sense to do
5235 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5236 * represents a packet that was sent by hand or if it represents statistics
5237 * that have been cleared out of the datapath. */
5239 subfacet_update_stats(struct subfacet *subfacet,
5240 const struct dpif_flow_stats *stats)
5242 if (stats->n_packets || stats->used > subfacet->used) {
5243 struct facet *facet = subfacet->facet;
5245 subfacet_update_time(subfacet, stats->used);
5246 facet->packet_count += stats->n_packets;
5247 facet->byte_count += stats->n_bytes;
5248 facet->tcp_flags |= stats->tcp_flags;
5249 facet_push_stats(facet);
5250 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
5256 static struct rule_dpif *
5257 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5259 struct rule_dpif *rule;
5261 rule = rule_dpif_lookup__(ofproto, flow, 0);
5266 return rule_dpif_miss_rule(ofproto, flow);
5269 static struct rule_dpif *
5270 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5273 struct cls_rule *cls_rule;
5274 struct classifier *cls;
5276 if (table_id >= N_TABLES) {
5280 cls = &ofproto->up.tables[table_id].cls;
5281 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5282 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5283 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5284 * are unavailable. */
5285 struct flow ofpc_normal_flow = *flow;
5286 ofpc_normal_flow.tp_src = htons(0);
5287 ofpc_normal_flow.tp_dst = htons(0);
5288 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5290 cls_rule = classifier_lookup(cls, flow);
5292 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5295 static struct rule_dpif *
5296 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5298 struct ofport_dpif *port;
5300 port = get_ofp_port(ofproto, flow->in_port);
5302 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5303 return ofproto->miss_rule;
5306 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5307 return ofproto->no_packet_in_rule;
5309 return ofproto->miss_rule;
5313 complete_operation(struct rule_dpif *rule)
5315 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5317 rule_invalidate(rule);
5319 struct dpif_completion *c = xmalloc(sizeof *c);
5320 c->op = rule->up.pending;
5321 list_push_back(&ofproto->completions, &c->list_node);
5323 ofoperation_complete(rule->up.pending, 0);
5327 static struct rule *
5330 struct rule_dpif *rule = xmalloc(sizeof *rule);
5335 rule_dealloc(struct rule *rule_)
5337 struct rule_dpif *rule = rule_dpif_cast(rule_);
5342 rule_construct(struct rule *rule_)
5344 struct rule_dpif *rule = rule_dpif_cast(rule_);
5345 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5346 struct rule_dpif *victim;
5349 rule->packet_count = 0;
5350 rule->byte_count = 0;
5352 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5353 if (victim && !list_is_empty(&victim->facets)) {
5354 struct facet *facet;
5356 rule->facets = victim->facets;
5357 list_moved(&rule->facets);
5358 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5359 /* XXX: We're only clearing our local counters here. It's possible
5360 * that quite a few packets are unaccounted for in the datapath
5361 * statistics. These will be accounted to the new rule instead of
5362 * cleared as required. This could be fixed by clearing out the
5363 * datapath statistics for this facet, but currently it doesn't
5365 facet_reset_counters(facet);
5369 /* Must avoid list_moved() in this case. */
5370 list_init(&rule->facets);
5373 table_id = rule->up.table_id;
5375 rule->tag = victim->tag;
5376 } else if (table_id == 0) {
5381 miniflow_expand(&rule->up.cr.match.flow, &flow);
5382 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5383 ofproto->tables[table_id].basis);
5386 complete_operation(rule);
5391 rule_destruct(struct rule *rule_)
5393 struct rule_dpif *rule = rule_dpif_cast(rule_);
5394 struct facet *facet, *next_facet;
5396 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5397 facet_revalidate(facet);
5400 complete_operation(rule);
5404 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5406 struct rule_dpif *rule = rule_dpif_cast(rule_);
5407 struct facet *facet;
5409 /* Start from historical data for 'rule' itself that are no longer tracked
5410 * in facets. This counts, for example, facets that have expired. */
5411 *packets = rule->packet_count;
5412 *bytes = rule->byte_count;
5414 /* Add any statistics that are tracked by facets. This includes
5415 * statistical data recently updated by ofproto_update_stats() as well as
5416 * stats for packets that were executed "by hand" via dpif_execute(). */
5417 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5418 *packets += facet->packet_count;
5419 *bytes += facet->byte_count;
5424 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5425 struct ofpbuf *packet)
5427 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5429 struct dpif_flow_stats stats;
5431 struct action_xlate_ctx ctx;
5432 uint64_t odp_actions_stub[1024 / 8];
5433 struct ofpbuf odp_actions;
5435 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5436 rule_credit_stats(rule, &stats);
5438 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5439 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
5440 rule, stats.tcp_flags, packet);
5441 ctx.resubmit_stats = &stats;
5442 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
5444 execute_odp_actions(ofproto, flow, odp_actions.data,
5445 odp_actions.size, packet);
5447 ofpbuf_uninit(&odp_actions);
5451 rule_execute(struct rule *rule, const struct flow *flow,
5452 struct ofpbuf *packet)
5454 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5455 ofpbuf_delete(packet);
5460 rule_modify_actions(struct rule *rule_)
5462 struct rule_dpif *rule = rule_dpif_cast(rule_);
5464 complete_operation(rule);
5467 /* Sends 'packet' out 'ofport'.
5468 * May modify 'packet'.
5469 * Returns 0 if successful, otherwise a positive errno value. */
5471 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5473 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5474 uint64_t odp_actions_stub[1024 / 8];
5475 struct ofpbuf key, odp_actions;
5476 struct odputil_keybuf keybuf;
5481 flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow);
5482 if (netdev_vport_is_patch(ofport->up.netdev)) {
5483 struct ofproto_dpif *peer_ofproto;
5484 struct dpif_flow_stats stats;
5485 struct ofport_dpif *peer;
5486 struct rule_dpif *rule;
5488 peer = ofport_get_peer(ofport);
5493 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5494 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5495 netdev_vport_inc_rx(peer->up.netdev, &stats);
5497 flow.in_port = peer->up.ofp_port;
5498 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5499 rule = rule_dpif_lookup(peer_ofproto, &flow);
5500 rule_dpif_execute(rule, &flow, packet);
5505 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5507 if (ofport->tnl_port) {
5508 struct dpif_flow_stats stats;
5510 odp_port = tnl_port_send(ofport->tnl_port, &flow);
5511 if (odp_port == OVSP_NONE) {
5515 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5516 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5517 odp_put_tunnel_action(&flow.tunnel, &odp_actions);
5519 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5521 if (odp_port != ofport->odp_port) {
5522 eth_pop_vlan(packet);
5523 flow.vlan_tci = htons(0);
5527 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5528 odp_flow_key_from_flow(&key, &flow,
5529 ofp_port_to_odp_port(ofproto, flow.in_port));
5531 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5533 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5534 error = dpif_execute(ofproto->backer->dpif,
5536 odp_actions.data, odp_actions.size,
5538 ofpbuf_uninit(&odp_actions);
5541 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5542 ofproto->up.name, odp_port, strerror(error));
5544 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5548 /* OpenFlow to datapath action translation. */
5550 static bool may_receive(const struct ofport_dpif *, struct action_xlate_ctx *);
5551 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5552 struct action_xlate_ctx *);
5553 static void xlate_normal(struct action_xlate_ctx *);
5555 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5556 * The action will state 'slow' as the reason that the action is in the slow
5557 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5558 * dump-flows" output to see why a flow is in the slow path.)
5560 * The 'stub_size' bytes in 'stub' will be used to store the action.
5561 * 'stub_size' must be large enough for the action.
5563 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5566 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5567 enum slow_path_reason slow,
5568 uint64_t *stub, size_t stub_size,
5569 const struct nlattr **actionsp, size_t *actions_lenp)
5571 union user_action_cookie cookie;
5574 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5575 cookie.slow_path.unused = 0;
5576 cookie.slow_path.reason = slow;
5578 ofpbuf_use_stack(&buf, stub, stub_size);
5579 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
5580 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5581 odp_put_userspace_action(pid, &cookie, &buf);
5583 put_userspace_action(ofproto, &buf, flow, &cookie);
5585 *actionsp = buf.data;
5586 *actions_lenp = buf.size;
5590 put_userspace_action(const struct ofproto_dpif *ofproto,
5591 struct ofpbuf *odp_actions,
5592 const struct flow *flow,
5593 const union user_action_cookie *cookie)
5597 pid = dpif_port_get_pid(ofproto->backer->dpif,
5598 ofp_port_to_odp_port(ofproto, flow->in_port));
5600 return odp_put_userspace_action(pid, cookie, odp_actions);
5604 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5605 ovs_be16 vlan_tci, uint32_t odp_port,
5606 unsigned int n_outputs, union user_action_cookie *cookie)
5610 cookie->type = USER_ACTION_COOKIE_SFLOW;
5611 cookie->sflow.vlan_tci = vlan_tci;
5613 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5614 * port information") for the interpretation of cookie->output. */
5615 switch (n_outputs) {
5617 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5618 cookie->sflow.output = 0x40000000 | 256;
5622 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5624 cookie->sflow.output = ifindex;
5629 /* 0x80000000 means "multiple output ports. */
5630 cookie->sflow.output = 0x80000000 | n_outputs;
5635 /* Compose SAMPLE action for sFlow. */
5637 compose_sflow_action(const struct ofproto_dpif *ofproto,
5638 struct ofpbuf *odp_actions,
5639 const struct flow *flow,
5642 uint32_t probability;
5643 union user_action_cookie cookie;
5644 size_t sample_offset, actions_offset;
5647 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5651 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5653 /* Number of packets out of UINT_MAX to sample. */
5654 probability = dpif_sflow_get_probability(ofproto->sflow);
5655 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5657 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5658 compose_sflow_cookie(ofproto, htons(0), odp_port,
5659 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5660 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
5662 nl_msg_end_nested(odp_actions, actions_offset);
5663 nl_msg_end_nested(odp_actions, sample_offset);
5664 return cookie_offset;
5667 /* SAMPLE action must be first action in any given list of actions.
5668 * At this point we do not have all information required to build it. So try to
5669 * build sample action as complete as possible. */
5671 add_sflow_action(struct action_xlate_ctx *ctx)
5673 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5675 &ctx->flow, OVSP_NONE);
5676 ctx->sflow_odp_port = 0;
5677 ctx->sflow_n_outputs = 0;
5680 /* Fix SAMPLE action according to data collected while composing ODP actions.
5681 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5682 * USERSPACE action's user-cookie which is required for sflow. */
5684 fix_sflow_action(struct action_xlate_ctx *ctx)
5686 const struct flow *base = &ctx->base_flow;
5687 union user_action_cookie *cookie;
5689 if (!ctx->user_cookie_offset) {
5693 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
5695 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5697 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5698 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5702 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
5705 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5706 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
5707 ovs_be64 flow_tun_id = ctx->flow.tunnel.tun_id;
5708 uint8_t flow_nw_tos = ctx->flow.nw_tos;
5709 struct priority_to_dscp *pdscp;
5710 uint32_t out_port, odp_port;
5712 /* If 'struct flow' gets additional metadata, we'll need to zero it out
5713 * before traversing a patch port. */
5714 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 19);
5717 xlate_report(ctx, "Nonexistent output port");
5719 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5720 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5722 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5723 xlate_report(ctx, "STP not in forwarding state, skipping output");
5727 if (netdev_vport_is_patch(ofport->up.netdev)) {
5728 struct ofport_dpif *peer = ofport_get_peer(ofport);
5729 struct flow old_flow = ctx->flow;
5730 const struct ofproto_dpif *peer_ofproto;
5731 struct ofport_dpif *in_port;
5734 xlate_report(ctx, "Nonexistent patch port peer");
5738 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5739 if (peer_ofproto->backer != ctx->ofproto->backer) {
5740 xlate_report(ctx, "Patch port peer on a different datapath");
5744 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
5745 ctx->flow.in_port = peer->up.ofp_port;
5746 ctx->flow.metadata = htonll(0);
5747 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
5748 memset(ctx->flow.regs, 0, sizeof ctx->flow.regs);
5750 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5751 if (!in_port || may_receive(in_port, ctx)) {
5752 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
5753 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5755 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
5756 * learning action look at the packet, then drop it. */
5757 struct flow old_base_flow = ctx->base_flow;
5758 size_t old_size = ctx->odp_actions->size;
5759 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5760 ctx->base_flow = old_base_flow;
5761 ctx->odp_actions->size = old_size;
5765 ctx->flow = old_flow;
5766 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5768 if (ctx->resubmit_stats) {
5769 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
5770 netdev_vport_inc_rx(peer->up.netdev, ctx->resubmit_stats);
5776 pdscp = get_priority(ofport, ctx->flow.skb_priority);
5778 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5779 ctx->flow.nw_tos |= pdscp->dscp;
5782 odp_port = ofp_port_to_odp_port(ctx->ofproto, ofp_port);
5783 if (ofport->tnl_port) {
5784 odp_port = tnl_port_send(ofport->tnl_port, &ctx->flow);
5785 if (odp_port == OVSP_NONE) {
5786 xlate_report(ctx, "Tunneling decided against output");
5790 if (ctx->resubmit_stats) {
5791 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
5793 out_port = odp_port;
5794 commit_odp_tunnel_action(&ctx->flow, &ctx->base_flow,
5797 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
5798 ctx->flow.vlan_tci);
5799 if (out_port != odp_port) {
5800 ctx->flow.vlan_tci = htons(0);
5803 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
5804 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
5806 ctx->sflow_odp_port = odp_port;
5807 ctx->sflow_n_outputs++;
5808 ctx->nf_output_iface = ofp_port;
5809 ctx->flow.tunnel.tun_id = flow_tun_id;
5810 ctx->flow.vlan_tci = flow_vlan_tci;
5811 ctx->flow.nw_tos = flow_nw_tos;
5815 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
5817 compose_output_action__(ctx, ofp_port, true);
5821 xlate_table_action(struct action_xlate_ctx *ctx,
5822 uint16_t in_port, uint8_t table_id, bool may_packet_in)
5824 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
5825 struct ofproto_dpif *ofproto = ctx->ofproto;
5826 struct rule_dpif *rule;
5827 uint16_t old_in_port;
5828 uint8_t old_table_id;
5830 old_table_id = ctx->table_id;
5831 ctx->table_id = table_id;
5833 /* Look up a flow with 'in_port' as the input port. */
5834 old_in_port = ctx->flow.in_port;
5835 ctx->flow.in_port = in_port;
5836 rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id);
5839 if (table_id > 0 && table_id < N_TABLES) {
5840 struct table_dpif *table = &ofproto->tables[table_id];
5841 if (table->other_table) {
5842 ctx->tags |= (rule && rule->tag
5844 : rule_calculate_tag(&ctx->flow,
5845 &table->other_table->mask,
5850 /* Restore the original input port. Otherwise OFPP_NORMAL and
5851 * OFPP_IN_PORT will have surprising behavior. */
5852 ctx->flow.in_port = old_in_port;
5854 if (ctx->resubmit_hook) {
5855 ctx->resubmit_hook(ctx, rule);
5858 if (rule == NULL && may_packet_in) {
5860 * check if table configuration flags
5861 * OFPTC_TABLE_MISS_CONTROLLER, default.
5862 * OFPTC_TABLE_MISS_CONTINUE,
5863 * OFPTC_TABLE_MISS_DROP
5864 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
5866 rule = rule_dpif_miss_rule(ofproto, &ctx->flow);
5870 struct rule_dpif *old_rule = ctx->rule;
5872 if (ctx->resubmit_stats) {
5873 rule_credit_stats(rule, ctx->resubmit_stats);
5878 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
5879 ctx->rule = old_rule;
5883 ctx->table_id = old_table_id;
5885 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5887 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
5888 MAX_RESUBMIT_RECURSION);
5889 ctx->max_resubmit_trigger = true;
5894 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
5895 const struct ofpact_resubmit *resubmit)
5900 in_port = resubmit->in_port;
5901 if (in_port == OFPP_IN_PORT) {
5902 in_port = ctx->flow.in_port;
5905 table_id = resubmit->table_id;
5906 if (table_id == 255) {
5907 table_id = ctx->table_id;
5910 xlate_table_action(ctx, in_port, table_id, false);
5914 flood_packets(struct action_xlate_ctx *ctx, bool all)
5916 struct ofport_dpif *ofport;
5918 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
5919 uint16_t ofp_port = ofport->up.ofp_port;
5921 if (ofp_port == ctx->flow.in_port) {
5926 compose_output_action__(ctx, ofp_port, false);
5927 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
5928 compose_output_action(ctx, ofp_port);
5932 ctx->nf_output_iface = NF_OUT_FLOOD;
5936 execute_controller_action(struct action_xlate_ctx *ctx, int len,
5937 enum ofp_packet_in_reason reason,
5938 uint16_t controller_id)
5940 struct ofputil_packet_in pin;
5941 struct ofpbuf *packet;
5943 ctx->slow |= SLOW_CONTROLLER;
5948 packet = ofpbuf_clone(ctx->packet);
5950 if (packet->l2 && packet->l3) {
5951 struct eth_header *eh;
5952 uint16_t mpls_depth;
5954 eth_pop_vlan(packet);
5957 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
5958 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
5960 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
5961 eth_push_vlan(packet, ctx->flow.vlan_tci);
5964 mpls_depth = eth_mpls_depth(packet);
5966 if (mpls_depth < ctx->flow.mpls_depth) {
5967 push_mpls(packet, ctx->flow.dl_type, ctx->flow.mpls_lse);
5968 } else if (mpls_depth > ctx->flow.mpls_depth) {
5969 pop_mpls(packet, ctx->flow.dl_type);
5970 } else if (mpls_depth) {
5971 set_mpls_lse(packet, ctx->flow.mpls_lse);
5975 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5976 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
5977 ctx->flow.nw_tos, ctx->flow.nw_ttl);
5981 if (ctx->flow.nw_proto == IPPROTO_TCP) {
5982 packet_set_tcp_port(packet, ctx->flow.tp_src,
5984 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
5985 packet_set_udp_port(packet, ctx->flow.tp_src,
5992 pin.packet = packet->data;
5993 pin.packet_len = packet->size;
5994 pin.reason = reason;
5995 pin.controller_id = controller_id;
5996 pin.table_id = ctx->table_id;
5997 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
6000 flow_get_metadata(&ctx->flow, &pin.fmd);
6002 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6003 ofpbuf_delete(packet);
6007 execute_mpls_push_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
6009 ovs_assert(eth_type_mpls(eth_type));
6011 if (ctx->base_flow.mpls_depth) {
6012 ctx->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK);
6013 ctx->flow.mpls_depth++;
6018 if (ctx->flow.dl_type == htons(ETH_TYPE_IPV6)) {
6019 label = htonl(0x2); /* IPV6 Explicit Null. */
6021 label = htonl(0x0); /* IPV4 Explicit Null. */
6023 tc = (ctx->flow.nw_tos & IP_DSCP_MASK) >> 2;
6024 ttl = ctx->flow.nw_ttl ? ctx->flow.nw_ttl : 0x40;
6025 ctx->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
6026 ctx->flow.encap_dl_type = ctx->flow.dl_type;
6027 ctx->flow.mpls_depth = 1;
6029 ctx->flow.dl_type = eth_type;
6033 execute_mpls_pop_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
6035 ovs_assert(eth_type_mpls(ctx->flow.dl_type));
6036 ovs_assert(!eth_type_mpls(eth_type));
6038 if (ctx->flow.mpls_depth) {
6039 ctx->flow.mpls_depth--;
6040 ctx->flow.mpls_lse = htonl(0);
6041 if (!ctx->flow.mpls_depth) {
6042 ctx->flow.dl_type = eth_type;
6043 ctx->flow.encap_dl_type = htons(0);
6049 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6051 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
6052 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6056 if (ctx->flow.nw_ttl > 1) {
6062 for (i = 0; i < ids->n_controllers; i++) {
6063 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6067 /* Stop processing for current table. */
6073 xlate_output_action(struct action_xlate_ctx *ctx,
6074 uint16_t port, uint16_t max_len, bool may_packet_in)
6076 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
6078 ctx->nf_output_iface = NF_OUT_DROP;
6082 compose_output_action(ctx, ctx->flow.in_port);
6085 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
6091 flood_packets(ctx, false);
6094 flood_packets(ctx, true);
6096 case OFPP_CONTROLLER:
6097 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6103 if (port != ctx->flow.in_port) {
6104 compose_output_action(ctx, port);
6106 xlate_report(ctx, "skipping output to input port");
6111 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6112 ctx->nf_output_iface = NF_OUT_FLOOD;
6113 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
6114 ctx->nf_output_iface = prev_nf_output_iface;
6115 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6116 ctx->nf_output_iface != NF_OUT_FLOOD) {
6117 ctx->nf_output_iface = NF_OUT_MULTI;
6122 xlate_output_reg_action(struct action_xlate_ctx *ctx,
6123 const struct ofpact_output_reg *or)
6125 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
6126 if (port <= UINT16_MAX) {
6127 xlate_output_action(ctx, port, or->max_len, false);
6132 xlate_enqueue_action(struct action_xlate_ctx *ctx,
6133 const struct ofpact_enqueue *enqueue)
6135 uint16_t ofp_port = enqueue->port;
6136 uint32_t queue_id = enqueue->queue;
6137 uint32_t flow_priority, priority;
6140 /* Translate queue to priority. */
6141 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6142 queue_id, &priority);
6144 /* Fall back to ordinary output action. */
6145 xlate_output_action(ctx, enqueue->port, 0, false);
6149 /* Check output port. */
6150 if (ofp_port == OFPP_IN_PORT) {
6151 ofp_port = ctx->flow.in_port;
6152 } else if (ofp_port == ctx->flow.in_port) {
6156 /* Add datapath actions. */
6157 flow_priority = ctx->flow.skb_priority;
6158 ctx->flow.skb_priority = priority;
6159 compose_output_action(ctx, ofp_port);
6160 ctx->flow.skb_priority = flow_priority;
6162 /* Update NetFlow output port. */
6163 if (ctx->nf_output_iface == NF_OUT_DROP) {
6164 ctx->nf_output_iface = ofp_port;
6165 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
6166 ctx->nf_output_iface = NF_OUT_MULTI;
6171 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
6173 uint32_t skb_priority;
6175 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6176 queue_id, &skb_priority)) {
6177 ctx->flow.skb_priority = skb_priority;
6179 /* Couldn't translate queue to a priority. Nothing to do. A warning
6180 * has already been logged. */
6184 struct xlate_reg_state {
6190 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6192 struct ofproto_dpif *ofproto = ofproto_;
6193 struct ofport_dpif *port;
6203 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6206 port = get_ofp_port(ofproto, ofp_port);
6207 return port ? port->may_enable : false;
6212 xlate_bundle_action(struct action_xlate_ctx *ctx,
6213 const struct ofpact_bundle *bundle)
6217 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
6218 if (bundle->dst.field) {
6219 nxm_reg_load(&bundle->dst, port, &ctx->flow);
6221 xlate_output_action(ctx, port, 0, false);
6226 xlate_learn_action(struct action_xlate_ctx *ctx,
6227 const struct ofpact_learn *learn)
6229 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6230 struct ofputil_flow_mod fm;
6231 uint64_t ofpacts_stub[1024 / 8];
6232 struct ofpbuf ofpacts;
6235 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6236 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
6238 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6239 if (error && !VLOG_DROP_WARN(&rl)) {
6240 VLOG_WARN("learning action failed to modify flow table (%s)",
6241 ofperr_get_name(error));
6244 ofpbuf_uninit(&ofpacts);
6247 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6248 * means "infinite". */
6250 reduce_timeout(uint16_t max, uint16_t *timeout)
6252 if (max && (!*timeout || *timeout > max)) {
6258 xlate_fin_timeout(struct action_xlate_ctx *ctx,
6259 const struct ofpact_fin_timeout *oft)
6261 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6262 struct rule_dpif *rule = ctx->rule;
6264 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6265 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6270 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
6272 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
6273 ? OFPUTIL_PC_NO_RECV_STP
6274 : OFPUTIL_PC_NO_RECV)) {
6278 /* Only drop packets here if both forwarding and learning are
6279 * disabled. If just learning is enabled, we need to have
6280 * OFPP_NORMAL and the learning action have a look at the packet
6281 * before we can drop it. */
6282 if (!stp_forward_in_state(port->stp_state)
6283 && !stp_learn_in_state(port->stp_state)) {
6291 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6292 struct action_xlate_ctx *ctx)
6294 bool was_evictable = true;
6295 const struct ofpact *a;
6298 /* Don't let the rule we're working on get evicted underneath us. */
6299 was_evictable = ctx->rule->up.evictable;
6300 ctx->rule->up.evictable = false;
6302 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6303 struct ofpact_controller *controller;
6304 const struct ofpact_metadata *metadata;
6312 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6313 ofpact_get_OUTPUT(a)->max_len, true);
6316 case OFPACT_CONTROLLER:
6317 controller = ofpact_get_CONTROLLER(a);
6318 execute_controller_action(ctx, controller->max_len,
6320 controller->controller_id);
6323 case OFPACT_ENQUEUE:
6324 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6327 case OFPACT_SET_VLAN_VID:
6328 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6329 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6333 case OFPACT_SET_VLAN_PCP:
6334 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6335 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
6340 case OFPACT_STRIP_VLAN:
6341 ctx->flow.vlan_tci = htons(0);
6344 case OFPACT_PUSH_VLAN:
6345 /* XXX 802.1AD(QinQ) */
6346 ctx->flow.vlan_tci = htons(VLAN_CFI);
6349 case OFPACT_SET_ETH_SRC:
6350 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6354 case OFPACT_SET_ETH_DST:
6355 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6359 case OFPACT_SET_IPV4_SRC:
6360 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6363 case OFPACT_SET_IPV4_DST:
6364 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6367 case OFPACT_SET_IPV4_DSCP:
6368 /* OpenFlow 1.0 only supports IPv4. */
6369 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6370 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
6371 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6375 case OFPACT_SET_L4_SRC_PORT:
6376 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6379 case OFPACT_SET_L4_DST_PORT:
6380 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6383 case OFPACT_RESUBMIT:
6384 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6387 case OFPACT_SET_TUNNEL:
6388 ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6391 case OFPACT_SET_QUEUE:
6392 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6395 case OFPACT_POP_QUEUE:
6396 ctx->flow.skb_priority = ctx->orig_skb_priority;
6399 case OFPACT_REG_MOVE:
6400 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
6403 case OFPACT_REG_LOAD:
6404 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
6407 case OFPACT_PUSH_MPLS:
6408 execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
6411 case OFPACT_POP_MPLS:
6412 execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
6415 case OFPACT_DEC_TTL:
6416 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6422 /* Nothing to do. */
6425 case OFPACT_MULTIPATH:
6426 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
6430 ctx->ofproto->has_bundle_action = true;
6431 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6434 case OFPACT_OUTPUT_REG:
6435 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6439 ctx->has_learn = true;
6440 if (ctx->may_learn) {
6441 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6449 case OFPACT_FIN_TIMEOUT:
6450 ctx->has_fin_timeout = true;
6451 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6454 case OFPACT_CLEAR_ACTIONS:
6456 * Nothing to do because writa-actions is not supported for now.
6457 * When writa-actions is supported, clear-actions also must
6458 * be supported at the same time.
6462 case OFPACT_WRITE_METADATA:
6463 metadata = ofpact_get_WRITE_METADATA(a);
6464 ctx->flow.metadata &= ~metadata->mask;
6465 ctx->flow.metadata |= metadata->metadata & metadata->mask;
6468 case OFPACT_GOTO_TABLE: {
6469 /* XXX remove recursion */
6470 /* It is assumed that goto-table is last action */
6471 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6472 ovs_assert(ctx->table_id < ogt->table_id);
6473 xlate_table_action(ctx, ctx->flow.in_port, ogt->table_id, true);
6481 ctx->rule->up.evictable = was_evictable;
6486 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
6487 struct ofproto_dpif *ofproto, const struct flow *flow,
6488 ovs_be16 initial_tci, struct rule_dpif *rule,
6489 uint8_t tcp_flags, const struct ofpbuf *packet)
6491 ovs_be64 initial_tun_id = flow->tunnel.tun_id;
6493 /* Flow initialization rules:
6494 * - 'base_flow' must match the kernel's view of the packet at the
6495 * time that action processing starts. 'flow' represents any
6496 * transformations we wish to make through actions.
6497 * - By default 'base_flow' and 'flow' are the same since the input
6498 * packet matches the output before any actions are applied.
6499 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
6500 * of the received packet as seen by the kernel. If we later output
6501 * to another device without any modifications this will cause us to
6502 * insert a new tag since the original one was stripped off by the
6504 * - Tunnel 'flow' is largely cleared when transitioning between
6505 * the input and output stages since it does not make sense to output
6506 * a packet with the exact headers that it was received with (i.e.
6507 * the destination IP is us). The one exception is the tun_id, which
6508 * is preserved to allow use in later resubmit lookups and loads into
6510 * - Tunnel 'base_flow' is completely cleared since that is what the
6511 * kernel does. If we wish to maintain the original values an action
6512 * needs to be generated. */
6514 ctx->ofproto = ofproto;
6516 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
6517 ctx->base_flow = ctx->flow;
6518 ctx->base_flow.vlan_tci = initial_tci;
6519 ctx->flow.tunnel.tun_id = initial_tun_id;
6521 ctx->packet = packet;
6522 ctx->may_learn = packet != NULL;
6523 ctx->tcp_flags = tcp_flags;
6524 ctx->resubmit_hook = NULL;
6525 ctx->report_hook = NULL;
6526 ctx->resubmit_stats = NULL;
6529 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6530 * into datapath actions in 'odp_actions', using 'ctx'. */
6532 xlate_actions(struct action_xlate_ctx *ctx,
6533 const struct ofpact *ofpacts, size_t ofpacts_len,
6534 struct ofpbuf *odp_actions)
6536 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6537 * that in the future we always keep a copy of the original flow for
6538 * tracing purposes. */
6539 static bool hit_resubmit_limit;
6541 enum slow_path_reason special;
6542 struct ofport_dpif *in_port;
6543 struct flow orig_flow;
6545 COVERAGE_INC(ofproto_dpif_xlate);
6547 ofpbuf_clear(odp_actions);
6548 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
6550 ctx->odp_actions = odp_actions;
6553 ctx->has_learn = false;
6554 ctx->has_normal = false;
6555 ctx->has_fin_timeout = false;
6556 ctx->nf_output_iface = NF_OUT_DROP;
6559 ctx->max_resubmit_trigger = false;
6560 ctx->orig_skb_priority = ctx->flow.skb_priority;
6564 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
6565 /* Do this conditionally because the copy is expensive enough that it
6566 * shows up in profiles. */
6567 orig_flow = ctx->flow;
6570 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6571 switch (ctx->ofproto->up.frag_handling) {
6572 case OFPC_FRAG_NORMAL:
6573 /* We must pretend that transport ports are unavailable. */
6574 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
6575 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
6578 case OFPC_FRAG_DROP:
6581 case OFPC_FRAG_REASM:
6584 case OFPC_FRAG_NX_MATCH:
6585 /* Nothing to do. */
6588 case OFPC_INVALID_TTL_TO_CONTROLLER:
6593 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
6594 special = process_special(ctx->ofproto, &ctx->flow, in_port, ctx->packet);
6596 ctx->slow |= special;
6598 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6599 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
6600 uint32_t local_odp_port;
6602 add_sflow_action(ctx);
6604 if (!in_port || may_receive(in_port, ctx)) {
6605 do_xlate_actions(ofpacts, ofpacts_len, ctx);
6607 /* We've let OFPP_NORMAL and the learning action look at the
6608 * packet, so drop it now if forwarding is disabled. */
6609 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
6610 ofpbuf_clear(ctx->odp_actions);
6611 add_sflow_action(ctx);
6615 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
6616 if (!hit_resubmit_limit) {
6617 /* We didn't record the original flow. Make sure we do from
6619 hit_resubmit_limit = true;
6620 } else if (!VLOG_DROP_ERR(&trace_rl)) {
6621 struct ds ds = DS_EMPTY_INITIALIZER;
6623 ofproto_trace(ctx->ofproto, &orig_flow, ctx->packet,
6625 VLOG_ERR("Trace triggered by excessive resubmit "
6626 "recursion:\n%s", ds_cstr(&ds));
6631 local_odp_port = ofp_port_to_odp_port(ctx->ofproto, OFPP_LOCAL);
6632 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
6634 ctx->odp_actions->data,
6635 ctx->odp_actions->size)) {
6636 ctx->slow |= SLOW_IN_BAND;
6638 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
6640 compose_output_action(ctx, OFPP_LOCAL);
6643 if (ctx->ofproto->has_mirrors) {
6644 add_mirror_actions(ctx, &orig_flow);
6646 fix_sflow_action(ctx);
6650 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
6651 * into datapath actions, using 'ctx', and discards the datapath actions. */
6653 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
6654 const struct ofpact *ofpacts,
6657 uint64_t odp_actions_stub[1024 / 8];
6658 struct ofpbuf odp_actions;
6660 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
6661 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
6662 ofpbuf_uninit(&odp_actions);
6666 xlate_report(struct action_xlate_ctx *ctx, const char *s)
6668 if (ctx->report_hook) {
6669 ctx->report_hook(ctx, s);
6673 /* OFPP_NORMAL implementation. */
6675 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
6677 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
6678 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
6679 * the bundle on which the packet was received, returns the VLAN to which the
6682 * Both 'vid' and the return value are in the range 0...4095. */
6684 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
6686 switch (in_bundle->vlan_mode) {
6687 case PORT_VLAN_ACCESS:
6688 return in_bundle->vlan;
6691 case PORT_VLAN_TRUNK:
6694 case PORT_VLAN_NATIVE_UNTAGGED:
6695 case PORT_VLAN_NATIVE_TAGGED:
6696 return vid ? vid : in_bundle->vlan;
6703 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
6704 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
6707 * 'vid' should be the VID obtained from the 802.1Q header that was received as
6708 * part of a packet (specify 0 if there was no 802.1Q header), in the range
6711 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
6713 /* Allow any VID on the OFPP_NONE port. */
6714 if (in_bundle == &ofpp_none_bundle) {
6718 switch (in_bundle->vlan_mode) {
6719 case PORT_VLAN_ACCESS:
6722 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6723 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
6724 "packet received on port %s configured as VLAN "
6725 "%"PRIu16" access port",
6726 in_bundle->ofproto->up.name, vid,
6727 in_bundle->name, in_bundle->vlan);
6733 case PORT_VLAN_NATIVE_UNTAGGED:
6734 case PORT_VLAN_NATIVE_TAGGED:
6736 /* Port must always carry its native VLAN. */
6740 case PORT_VLAN_TRUNK:
6741 if (!ofbundle_includes_vlan(in_bundle, vid)) {
6743 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6744 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
6745 "received on port %s not configured for trunking "
6747 in_bundle->ofproto->up.name, vid,
6748 in_bundle->name, vid);
6760 /* Given 'vlan', the VLAN that a packet belongs to, and
6761 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
6762 * that should be included in the 802.1Q header. (If the return value is 0,
6763 * then the 802.1Q header should only be included in the packet if there is a
6766 * Both 'vlan' and the return value are in the range 0...4095. */
6768 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
6770 switch (out_bundle->vlan_mode) {
6771 case PORT_VLAN_ACCESS:
6774 case PORT_VLAN_TRUNK:
6775 case PORT_VLAN_NATIVE_TAGGED:
6778 case PORT_VLAN_NATIVE_UNTAGGED:
6779 return vlan == out_bundle->vlan ? 0 : vlan;
6787 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
6790 struct ofport_dpif *port;
6792 ovs_be16 tci, old_tci;
6794 vid = output_vlan_to_vid(out_bundle, vlan);
6795 if (!out_bundle->bond) {
6796 port = ofbundle_get_a_port(out_bundle);
6798 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
6801 /* No slaves enabled, so drop packet. */
6806 old_tci = ctx->flow.vlan_tci;
6808 if (tci || out_bundle->use_priority_tags) {
6809 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
6811 tci |= htons(VLAN_CFI);
6814 ctx->flow.vlan_tci = tci;
6816 compose_output_action(ctx, port->up.ofp_port);
6817 ctx->flow.vlan_tci = old_tci;
6821 mirror_mask_ffs(mirror_mask_t mask)
6823 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
6828 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
6830 return (bundle->vlan_mode != PORT_VLAN_ACCESS
6831 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
6835 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
6837 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
6840 /* Returns an arbitrary interface within 'bundle'. */
6841 static struct ofport_dpif *
6842 ofbundle_get_a_port(const struct ofbundle *bundle)
6844 return CONTAINER_OF(list_front(&bundle->ports),
6845 struct ofport_dpif, bundle_node);
6849 vlan_is_mirrored(const struct ofmirror *m, int vlan)
6851 return !m->vlans || bitmap_is_set(m->vlans, vlan);
6855 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
6857 struct ofproto_dpif *ofproto = ctx->ofproto;
6858 mirror_mask_t mirrors;
6859 struct ofbundle *in_bundle;
6862 const struct nlattr *a;
6865 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
6866 ctx->packet != NULL, NULL);
6870 mirrors = in_bundle->src_mirrors;
6872 /* Drop frames on bundles reserved for mirroring. */
6873 if (in_bundle->mirror_out) {
6874 if (ctx->packet != NULL) {
6875 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6876 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6877 "%s, which is reserved exclusively for mirroring",
6878 ctx->ofproto->up.name, in_bundle->name);
6884 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
6885 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6888 vlan = input_vid_to_vlan(in_bundle, vid);
6890 /* Look at the output ports to check for destination selections. */
6892 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
6893 ctx->odp_actions->size) {
6894 enum ovs_action_attr type = nl_attr_type(a);
6895 struct ofport_dpif *ofport;
6897 if (type != OVS_ACTION_ATTR_OUTPUT) {
6901 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
6902 if (ofport && ofport->bundle) {
6903 mirrors |= ofport->bundle->dst_mirrors;
6911 /* Restore the original packet before adding the mirror actions. */
6912 ctx->flow = *orig_flow;
6917 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6919 if (!vlan_is_mirrored(m, vlan)) {
6920 mirrors = zero_rightmost_1bit(mirrors);
6924 mirrors &= ~m->dup_mirrors;
6925 ctx->mirrors |= m->dup_mirrors;
6927 output_normal(ctx, m->out, vlan);
6928 } else if (vlan != m->out_vlan
6929 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
6930 struct ofbundle *bundle;
6932 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
6933 if (ofbundle_includes_vlan(bundle, m->out_vlan)
6934 && !bundle->mirror_out) {
6935 output_normal(ctx, bundle, m->out_vlan);
6943 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
6944 uint64_t packets, uint64_t bytes)
6950 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
6953 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6956 /* In normal circumstances 'm' will not be NULL. However,
6957 * if mirrors are reconfigured, we can temporarily get out
6958 * of sync in facet_revalidate(). We could "correct" the
6959 * mirror list before reaching here, but doing that would
6960 * not properly account the traffic stats we've currently
6961 * accumulated for previous mirror configuration. */
6965 m->packet_count += packets;
6966 m->byte_count += bytes;
6970 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
6971 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
6972 * indicate this; newer upstream kernels use gratuitous ARP requests. */
6974 is_gratuitous_arp(const struct flow *flow)
6976 return (flow->dl_type == htons(ETH_TYPE_ARP)
6977 && eth_addr_is_broadcast(flow->dl_dst)
6978 && (flow->nw_proto == ARP_OP_REPLY
6979 || (flow->nw_proto == ARP_OP_REQUEST
6980 && flow->nw_src == flow->nw_dst)));
6984 update_learning_table(struct ofproto_dpif *ofproto,
6985 const struct flow *flow, int vlan,
6986 struct ofbundle *in_bundle)
6988 struct mac_entry *mac;
6990 /* Don't learn the OFPP_NONE port. */
6991 if (in_bundle == &ofpp_none_bundle) {
6995 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
6999 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7000 if (is_gratuitous_arp(flow)) {
7001 /* We don't want to learn from gratuitous ARP packets that are
7002 * reflected back over bond slaves so we lock the learning table. */
7003 if (!in_bundle->bond) {
7004 mac_entry_set_grat_arp_lock(mac);
7005 } else if (mac_entry_is_grat_arp_locked(mac)) {
7010 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7011 /* The log messages here could actually be useful in debugging,
7012 * so keep the rate limit relatively high. */
7013 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7014 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7015 "on port %s in VLAN %d",
7016 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7017 in_bundle->name, vlan);
7019 mac->port.p = in_bundle;
7020 tag_set_add(&ofproto->backer->revalidate_set,
7021 mac_learning_changed(ofproto->ml, mac));
7025 static struct ofbundle *
7026 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7027 bool warn, struct ofport_dpif **in_ofportp)
7029 struct ofport_dpif *ofport;
7031 /* Find the port and bundle for the received packet. */
7032 ofport = get_ofp_port(ofproto, in_port);
7034 *in_ofportp = ofport;
7036 if (ofport && ofport->bundle) {
7037 return ofport->bundle;
7040 /* Special-case OFPP_NONE, which a controller may use as the ingress
7041 * port for traffic that it is sourcing. */
7042 if (in_port == OFPP_NONE) {
7043 return &ofpp_none_bundle;
7046 /* Odd. A few possible reasons here:
7048 * - We deleted a port but there are still a few packets queued up
7051 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7052 * we don't know about.
7054 * - The ofproto client didn't configure the port as part of a bundle.
7055 * This is particularly likely to happen if a packet was received on the
7056 * port after it was created, but before the client had a chance to
7057 * configure its bundle.
7060 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7062 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7063 "port %"PRIu16, ofproto->up.name, in_port);
7068 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7069 * dropped. Returns true if they may be forwarded, false if they should be
7072 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7073 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7075 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7076 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7077 * checked by input_vid_is_valid().
7079 * May also add tags to '*tags', although the current implementation only does
7080 * so in one special case.
7083 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
7086 struct ofproto_dpif *ofproto = ctx->ofproto;
7087 struct flow *flow = &ctx->flow;
7088 struct ofbundle *in_bundle = in_port->bundle;
7090 /* Drop frames for reserved multicast addresses
7091 * only if forward_bpdu option is absent. */
7092 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7093 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7097 if (in_bundle->bond) {
7098 struct mac_entry *mac;
7100 switch (bond_check_admissibility(in_bundle->bond, in_port,
7101 flow->dl_dst, &ctx->tags)) {
7106 xlate_report(ctx, "bonding refused admissibility, dropping");
7109 case BV_DROP_IF_MOVED:
7110 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7111 if (mac && mac->port.p != in_bundle &&
7112 (!is_gratuitous_arp(flow)
7113 || mac_entry_is_grat_arp_locked(mac))) {
7114 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7126 xlate_normal(struct action_xlate_ctx *ctx)
7128 struct ofport_dpif *in_port;
7129 struct ofbundle *in_bundle;
7130 struct mac_entry *mac;
7134 ctx->has_normal = true;
7136 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
7137 ctx->packet != NULL, &in_port);
7139 xlate_report(ctx, "no input bundle, dropping");
7143 /* Drop malformed frames. */
7144 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7145 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
7146 if (ctx->packet != NULL) {
7147 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7148 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7149 "VLAN tag received on port %s",
7150 ctx->ofproto->up.name, in_bundle->name);
7152 xlate_report(ctx, "partial VLAN tag, dropping");
7156 /* Drop frames on bundles reserved for mirroring. */
7157 if (in_bundle->mirror_out) {
7158 if (ctx->packet != NULL) {
7159 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7160 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7161 "%s, which is reserved exclusively for mirroring",
7162 ctx->ofproto->up.name, in_bundle->name);
7164 xlate_report(ctx, "input port is mirror output port, dropping");
7169 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
7170 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7171 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7174 vlan = input_vid_to_vlan(in_bundle, vid);
7176 /* Check other admissibility requirements. */
7177 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7181 /* Learn source MAC. */
7182 if (ctx->may_learn) {
7183 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
7186 /* Determine output bundle. */
7187 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
7190 if (mac->port.p != in_bundle) {
7191 xlate_report(ctx, "forwarding to learned port");
7192 output_normal(ctx, mac->port.p, vlan);
7194 xlate_report(ctx, "learned port is input port, dropping");
7197 struct ofbundle *bundle;
7199 xlate_report(ctx, "no learned MAC for destination, flooding");
7200 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7201 if (bundle != in_bundle
7202 && ofbundle_includes_vlan(bundle, vlan)
7203 && bundle->floodable
7204 && !bundle->mirror_out) {
7205 output_normal(ctx, bundle, vlan);
7208 ctx->nf_output_iface = NF_OUT_FLOOD;
7212 /* Optimized flow revalidation.
7214 * It's a difficult problem, in general, to tell which facets need to have
7215 * their actions recalculated whenever the OpenFlow flow table changes. We
7216 * don't try to solve that general problem: for most kinds of OpenFlow flow
7217 * table changes, we recalculate the actions for every facet. This is
7218 * relatively expensive, but it's good enough if the OpenFlow flow table
7219 * doesn't change very often.
7221 * However, we can expect one particular kind of OpenFlow flow table change to
7222 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7223 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7224 * table, we add a special case that applies to flow tables in which every rule
7225 * has the same form (that is, the same wildcards), except that the table is
7226 * also allowed to have a single "catch-all" flow that matches all packets. We
7227 * optimize this case by tagging all of the facets that resubmit into the table
7228 * and invalidating the same tag whenever a flow changes in that table. The
7229 * end result is that we revalidate just the facets that need it (and sometimes
7230 * a few more, but not all of the facets or even all of the facets that
7231 * resubmit to the table modified by MAC learning). */
7233 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7234 * into an OpenFlow table with the given 'basis'. */
7236 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7239 if (minimask_is_catchall(mask)) {
7242 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7243 return tag_create_deterministic(hash);
7247 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7248 * taggability of that table.
7250 * This function must be called after *each* change to a flow table. If you
7251 * skip calling it on some changes then the pointer comparisons at the end can
7252 * be invalid if you get unlucky. For example, if a flow removal causes a
7253 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7254 * different wildcards to be created with the same address, then this function
7255 * will incorrectly skip revalidation. */
7257 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7259 struct table_dpif *table = &ofproto->tables[table_id];
7260 const struct oftable *oftable = &ofproto->up.tables[table_id];
7261 struct cls_table *catchall, *other;
7262 struct cls_table *t;
7264 catchall = other = NULL;
7266 switch (hmap_count(&oftable->cls.tables)) {
7268 /* We could tag this OpenFlow table but it would make the logic a
7269 * little harder and it's a corner case that doesn't seem worth it
7275 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7276 if (cls_table_is_catchall(t)) {
7278 } else if (!other) {
7281 /* Indicate that we can't tag this by setting both tables to
7282 * NULL. (We know that 'catchall' is already NULL.) */
7289 /* Can't tag this table. */
7293 if (table->catchall_table != catchall || table->other_table != other) {
7294 table->catchall_table = catchall;
7295 table->other_table = other;
7296 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7300 /* Given 'rule' that has changed in some way (either it is a rule being
7301 * inserted, a rule being deleted, or a rule whose actions are being
7302 * modified), marks facets for revalidation to ensure that packets will be
7303 * forwarded correctly according to the new state of the flow table.
7305 * This function must be called after *each* change to a flow table. See
7306 * the comment on table_update_taggable() for more information. */
7308 rule_invalidate(const struct rule_dpif *rule)
7310 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7312 table_update_taggable(ofproto, rule->up.table_id);
7314 if (!ofproto->backer->need_revalidate) {
7315 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7317 if (table->other_table && rule->tag) {
7318 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7320 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7326 set_frag_handling(struct ofproto *ofproto_,
7327 enum ofp_config_flags frag_handling)
7329 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7330 if (frag_handling != OFPC_FRAG_REASM) {
7331 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7339 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7340 const struct flow *flow,
7341 const struct ofpact *ofpacts, size_t ofpacts_len)
7343 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7344 struct odputil_keybuf keybuf;
7345 struct dpif_flow_stats stats;
7349 struct action_xlate_ctx ctx;
7350 uint64_t odp_actions_stub[1024 / 8];
7351 struct ofpbuf odp_actions;
7353 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7354 odp_flow_key_from_flow(&key, flow,
7355 ofp_port_to_odp_port(ofproto, flow->in_port));
7357 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7359 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL,
7360 packet_get_tcp_flags(packet, flow), packet);
7361 ctx.resubmit_stats = &stats;
7363 ofpbuf_use_stub(&odp_actions,
7364 odp_actions_stub, sizeof odp_actions_stub);
7365 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
7366 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7367 odp_actions.data, odp_actions.size, packet);
7368 ofpbuf_uninit(&odp_actions);
7376 set_netflow(struct ofproto *ofproto_,
7377 const struct netflow_options *netflow_options)
7379 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7381 if (netflow_options) {
7382 if (!ofproto->netflow) {
7383 ofproto->netflow = netflow_create();
7385 return netflow_set_options(ofproto->netflow, netflow_options);
7387 netflow_destroy(ofproto->netflow);
7388 ofproto->netflow = NULL;
7394 get_netflow_ids(const struct ofproto *ofproto_,
7395 uint8_t *engine_type, uint8_t *engine_id)
7397 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7399 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7403 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7405 if (!facet_is_controller_flow(facet) &&
7406 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7407 struct subfacet *subfacet;
7408 struct ofexpired expired;
7410 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7411 if (subfacet->path == SF_FAST_PATH) {
7412 struct dpif_flow_stats stats;
7414 subfacet_reinstall(subfacet, &stats);
7415 subfacet_update_stats(subfacet, &stats);
7419 expired.flow = facet->flow;
7420 expired.packet_count = facet->packet_count;
7421 expired.byte_count = facet->byte_count;
7422 expired.used = facet->used;
7423 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
7428 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7430 struct facet *facet;
7432 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7433 send_active_timeout(ofproto, facet);
7437 static struct ofproto_dpif *
7438 ofproto_dpif_lookup(const char *name)
7440 struct ofproto_dpif *ofproto;
7442 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7443 hash_string(name, 0), &all_ofproto_dpifs) {
7444 if (!strcmp(ofproto->up.name, name)) {
7452 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7453 const char *argv[], void *aux OVS_UNUSED)
7455 struct ofproto_dpif *ofproto;
7458 ofproto = ofproto_dpif_lookup(argv[1]);
7460 unixctl_command_reply_error(conn, "no such bridge");
7463 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7465 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7466 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7470 unixctl_command_reply(conn, "table successfully flushed");
7474 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7475 const char *argv[], void *aux OVS_UNUSED)
7477 struct ds ds = DS_EMPTY_INITIALIZER;
7478 const struct ofproto_dpif *ofproto;
7479 const struct mac_entry *e;
7481 ofproto = ofproto_dpif_lookup(argv[1]);
7483 unixctl_command_reply_error(conn, "no such bridge");
7487 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7488 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7489 struct ofbundle *bundle = e->port.p;
7490 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7491 ofbundle_get_a_port(bundle)->odp_port,
7492 e->vlan, ETH_ADDR_ARGS(e->mac),
7493 mac_entry_age(ofproto->ml, e));
7495 unixctl_command_reply(conn, ds_cstr(&ds));
7500 struct action_xlate_ctx ctx;
7506 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7507 const struct rule_dpif *rule)
7509 ds_put_char_multiple(result, '\t', level);
7511 ds_put_cstr(result, "No match\n");
7515 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7516 table_id, ntohll(rule->up.flow_cookie));
7517 cls_rule_format(&rule->up.cr, result);
7518 ds_put_char(result, '\n');
7520 ds_put_char_multiple(result, '\t', level);
7521 ds_put_cstr(result, "OpenFlow ");
7522 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7523 ds_put_char(result, '\n');
7527 trace_format_flow(struct ds *result, int level, const char *title,
7528 struct trace_ctx *trace)
7530 ds_put_char_multiple(result, '\t', level);
7531 ds_put_format(result, "%s: ", title);
7532 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
7533 ds_put_cstr(result, "unchanged");
7535 flow_format(result, &trace->ctx.flow);
7536 trace->flow = trace->ctx.flow;
7538 ds_put_char(result, '\n');
7542 trace_format_regs(struct ds *result, int level, const char *title,
7543 struct trace_ctx *trace)
7547 ds_put_char_multiple(result, '\t', level);
7548 ds_put_format(result, "%s:", title);
7549 for (i = 0; i < FLOW_N_REGS; i++) {
7550 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7552 ds_put_char(result, '\n');
7556 trace_format_odp(struct ds *result, int level, const char *title,
7557 struct trace_ctx *trace)
7559 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
7561 ds_put_char_multiple(result, '\t', level);
7562 ds_put_format(result, "%s: ", title);
7563 format_odp_actions(result, odp_actions->data, odp_actions->size);
7564 ds_put_char(result, '\n');
7568 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
7570 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7571 struct ds *result = trace->result;
7573 ds_put_char(result, '\n');
7574 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7575 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
7576 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
7577 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
7581 trace_report(struct action_xlate_ctx *ctx, const char *s)
7583 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7584 struct ds *result = trace->result;
7586 ds_put_char_multiple(result, '\t', ctx->recurse);
7587 ds_put_cstr(result, s);
7588 ds_put_char(result, '\n');
7592 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
7593 void *aux OVS_UNUSED)
7595 const char *dpname = argv[1];
7596 struct ofproto_dpif *ofproto;
7597 struct ofpbuf odp_key;
7598 struct ofpbuf *packet;
7599 ovs_be16 initial_tci;
7605 ofpbuf_init(&odp_key, 0);
7608 ofproto = ofproto_dpif_lookup(dpname);
7610 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
7614 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
7615 /* ofproto/trace dpname flow [-generate] */
7616 const char *flow_s = argv[2];
7617 const char *generate_s = argv[3];
7619 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
7620 * flow. We guess which type it is based on whether 'flow_s' contains
7621 * an '(', since a datapath flow always contains '(') but an
7622 * OpenFlow-like flow should not (in fact it's allowed but I believe
7623 * that's not documented anywhere).
7625 * An alternative would be to try to parse 'flow_s' both ways, but then
7626 * it would be tricky giving a sensible error message. After all, do
7627 * you just say "syntax error" or do you present both error messages?
7628 * Both choices seem lousy. */
7629 if (strchr(flow_s, '(')) {
7632 /* Convert string to datapath key. */
7633 ofpbuf_init(&odp_key, 0);
7634 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
7636 unixctl_command_reply_error(conn, "Bad flow syntax");
7640 /* XXX: Since we allow the user to specify an ofproto, it's
7641 * possible they will specify a different ofproto than the one the
7642 * port actually belongs too. Ideally we should simply remove the
7643 * ability to specify the ofproto. */
7644 if (ofproto_receive(ofproto->backer, NULL, odp_key.data,
7645 odp_key.size, &flow, NULL, NULL, NULL,
7647 unixctl_command_reply_error(conn, "Invalid flow");
7653 error_s = parse_ofp_exact_flow(&flow, argv[2]);
7655 unixctl_command_reply_error(conn, error_s);
7660 initial_tci = flow.vlan_tci;
7663 /* Generate a packet, if requested. */
7665 packet = ofpbuf_new(0);
7666 flow_compose(packet, &flow);
7668 } else if (argc == 7) {
7669 /* ofproto/trace dpname priority tun_id in_port mark packet */
7670 const char *priority_s = argv[2];
7671 const char *tun_id_s = argv[3];
7672 const char *in_port_s = argv[4];
7673 const char *mark_s = argv[5];
7674 const char *packet_s = argv[6];
7675 uint32_t in_port = atoi(in_port_s);
7676 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
7677 uint32_t priority = atoi(priority_s);
7678 uint32_t mark = atoi(mark_s);
7681 msg = eth_from_hex(packet_s, &packet);
7683 unixctl_command_reply_error(conn, msg);
7687 ds_put_cstr(&result, "Packet: ");
7688 s = ofp_packet_to_string(packet->data, packet->size);
7689 ds_put_cstr(&result, s);
7692 flow_extract(packet, priority, mark, NULL, in_port, &flow);
7693 flow.tunnel.tun_id = tun_id;
7694 initial_tci = flow.vlan_tci;
7696 unixctl_command_reply_error(conn, "Bad command syntax");
7700 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
7701 unixctl_command_reply(conn, ds_cstr(&result));
7704 ds_destroy(&result);
7705 ofpbuf_delete(packet);
7706 ofpbuf_uninit(&odp_key);
7710 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
7711 const struct ofpbuf *packet, ovs_be16 initial_tci,
7714 struct rule_dpif *rule;
7716 ds_put_cstr(ds, "Flow: ");
7717 flow_format(ds, flow);
7718 ds_put_char(ds, '\n');
7720 rule = rule_dpif_lookup(ofproto, flow);
7722 trace_format_rule(ds, 0, 0, rule);
7723 if (rule == ofproto->miss_rule) {
7724 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
7725 } else if (rule == ofproto->no_packet_in_rule) {
7726 ds_put_cstr(ds, "\nNo match, packets dropped because "
7727 "OFPPC_NO_PACKET_IN is set on in_port.\n");
7731 uint64_t odp_actions_stub[1024 / 8];
7732 struct ofpbuf odp_actions;
7734 struct trace_ctx trace;
7737 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
7740 ofpbuf_use_stub(&odp_actions,
7741 odp_actions_stub, sizeof odp_actions_stub);
7742 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
7743 rule, tcp_flags, packet);
7744 trace.ctx.resubmit_hook = trace_resubmit;
7745 trace.ctx.report_hook = trace_report;
7746 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
7749 ds_put_char(ds, '\n');
7750 trace_format_flow(ds, 0, "Final flow", &trace);
7751 ds_put_cstr(ds, "Datapath actions: ");
7752 format_odp_actions(ds, odp_actions.data, odp_actions.size);
7753 ofpbuf_uninit(&odp_actions);
7755 if (trace.ctx.slow) {
7756 enum slow_path_reason slow;
7758 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
7759 "slow path because it:");
7760 for (slow = trace.ctx.slow; slow; ) {
7761 enum slow_path_reason bit = rightmost_1bit(slow);
7765 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
7768 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
7771 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
7774 ds_put_cstr(ds, "\n\t- Needs in-band special case "
7777 ds_put_cstr(ds, "\n\t (The datapath actions are "
7778 "incomplete--for complete actions, "
7779 "please supply a packet.)");
7782 case SLOW_CONTROLLER:
7783 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
7784 "to the OpenFlow controller.");
7787 ds_put_cstr(ds, "\n\t- Needs more specific matching "
7788 "than the datapath supports.");
7795 if (slow & ~SLOW_MATCH) {
7796 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
7797 "the special slow-path processing.");
7804 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7805 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7808 unixctl_command_reply(conn, NULL);
7812 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7813 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7816 unixctl_command_reply(conn, NULL);
7819 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
7820 * 'reply' describing the results. */
7822 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
7824 struct facet *facet;
7828 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7829 if (!facet_check_consistency(facet)) {
7834 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
7838 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
7839 ofproto->up.name, errors);
7841 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
7846 ofproto_dpif_self_check(struct unixctl_conn *conn,
7847 int argc, const char *argv[], void *aux OVS_UNUSED)
7849 struct ds reply = DS_EMPTY_INITIALIZER;
7850 struct ofproto_dpif *ofproto;
7853 ofproto = ofproto_dpif_lookup(argv[1]);
7855 unixctl_command_reply_error(conn, "Unknown ofproto (use "
7856 "ofproto/list for help)");
7859 ofproto_dpif_self_check__(ofproto, &reply);
7861 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7862 ofproto_dpif_self_check__(ofproto, &reply);
7866 unixctl_command_reply(conn, ds_cstr(&reply));
7870 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
7871 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
7872 * to destroy 'ofproto_shash' and free the returned value. */
7873 static const struct shash_node **
7874 get_ofprotos(struct shash *ofproto_shash)
7876 const struct ofproto_dpif *ofproto;
7878 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7879 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
7880 shash_add_nocopy(ofproto_shash, name, ofproto);
7883 return shash_sort(ofproto_shash);
7887 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
7888 const char *argv[] OVS_UNUSED,
7889 void *aux OVS_UNUSED)
7891 struct ds ds = DS_EMPTY_INITIALIZER;
7892 struct shash ofproto_shash;
7893 const struct shash_node **sorted_ofprotos;
7896 shash_init(&ofproto_shash);
7897 sorted_ofprotos = get_ofprotos(&ofproto_shash);
7898 for (i = 0; i < shash_count(&ofproto_shash); i++) {
7899 const struct shash_node *node = sorted_ofprotos[i];
7900 ds_put_format(&ds, "%s\n", node->name);
7903 shash_destroy(&ofproto_shash);
7904 free(sorted_ofprotos);
7906 unixctl_command_reply(conn, ds_cstr(&ds));
7911 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
7913 struct dpif_dp_stats s;
7914 const struct shash_node **ports;
7917 dpif_get_dp_stats(ofproto->backer->dpif, &s);
7919 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
7920 dpif_name(ofproto->backer->dpif));
7921 /* xxx It would be better to show bridge-specific stats instead
7922 * xxx of dp ones. */
7924 "\tlookups: hit:%"PRIu64" missed:%"PRIu64" lost:%"PRIu64"\n",
7925 s.n_hit, s.n_missed, s.n_lost);
7926 ds_put_format(ds, "\tflows: %zu\n",
7927 hmap_count(&ofproto->subfacets));
7929 ports = shash_sort(&ofproto->up.port_by_name);
7930 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
7931 const struct shash_node *node = ports[i];
7932 struct ofport *ofport = node->data;
7933 const char *name = netdev_get_name(ofport->netdev);
7934 const char *type = netdev_get_type(ofport->netdev);
7937 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
7939 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
7940 if (odp_port != OVSP_NONE) {
7941 ds_put_format(ds, "%"PRIu32":", odp_port);
7943 ds_put_cstr(ds, "none:");
7946 if (strcmp(type, "system")) {
7947 struct netdev *netdev;
7950 ds_put_format(ds, " (%s", type);
7952 error = netdev_open(name, type, &netdev);
7957 error = netdev_get_config(netdev, &config);
7959 const struct smap_node **nodes;
7962 nodes = smap_sort(&config);
7963 for (i = 0; i < smap_count(&config); i++) {
7964 const struct smap_node *node = nodes[i];
7965 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
7966 node->key, node->value);
7970 smap_destroy(&config);
7972 netdev_close(netdev);
7974 ds_put_char(ds, ')');
7976 ds_put_char(ds, '\n');
7982 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
7983 const char *argv[], void *aux OVS_UNUSED)
7985 struct ds ds = DS_EMPTY_INITIALIZER;
7986 const struct ofproto_dpif *ofproto;
7990 for (i = 1; i < argc; i++) {
7991 ofproto = ofproto_dpif_lookup(argv[i]);
7993 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
7994 "for help)", argv[i]);
7995 unixctl_command_reply_error(conn, ds_cstr(&ds));
7998 show_dp_format(ofproto, &ds);
8001 struct shash ofproto_shash;
8002 const struct shash_node **sorted_ofprotos;
8005 shash_init(&ofproto_shash);
8006 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8007 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8008 const struct shash_node *node = sorted_ofprotos[i];
8009 show_dp_format(node->data, &ds);
8012 shash_destroy(&ofproto_shash);
8013 free(sorted_ofprotos);
8016 unixctl_command_reply(conn, ds_cstr(&ds));
8021 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8022 int argc OVS_UNUSED, const char *argv[],
8023 void *aux OVS_UNUSED)
8025 struct ds ds = DS_EMPTY_INITIALIZER;
8026 const struct ofproto_dpif *ofproto;
8027 struct subfacet *subfacet;
8029 ofproto = ofproto_dpif_lookup(argv[1]);
8031 unixctl_command_reply_error(conn, "no such bridge");
8035 update_stats(ofproto->backer);
8037 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8038 struct odputil_keybuf keybuf;
8041 subfacet_get_key(subfacet, &keybuf, &key);
8042 odp_flow_key_format(key.data, key.size, &ds);
8044 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8045 subfacet->dp_packet_count, subfacet->dp_byte_count);
8046 if (subfacet->used) {
8047 ds_put_format(&ds, "%.3fs",
8048 (time_msec() - subfacet->used) / 1000.0);
8050 ds_put_format(&ds, "never");
8052 if (subfacet->facet->tcp_flags) {
8053 ds_put_cstr(&ds, ", flags:");
8054 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8057 ds_put_cstr(&ds, ", actions:");
8058 format_odp_actions(&ds, subfacet->actions, subfacet->actions_len);
8059 ds_put_char(&ds, '\n');
8062 unixctl_command_reply(conn, ds_cstr(&ds));
8067 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8068 int argc OVS_UNUSED, const char *argv[],
8069 void *aux OVS_UNUSED)
8071 struct ds ds = DS_EMPTY_INITIALIZER;
8072 struct ofproto_dpif *ofproto;
8074 ofproto = ofproto_dpif_lookup(argv[1]);
8076 unixctl_command_reply_error(conn, "no such bridge");
8080 flush(&ofproto->up);
8082 unixctl_command_reply(conn, ds_cstr(&ds));
8087 ofproto_dpif_unixctl_init(void)
8089 static bool registered;
8095 unixctl_command_register(
8097 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
8098 2, 6, ofproto_unixctl_trace, NULL);
8099 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8100 ofproto_unixctl_fdb_flush, NULL);
8101 unixctl_command_register("fdb/show", "bridge", 1, 1,
8102 ofproto_unixctl_fdb_show, NULL);
8103 unixctl_command_register("ofproto/clog", "", 0, 0,
8104 ofproto_dpif_clog, NULL);
8105 unixctl_command_register("ofproto/unclog", "", 0, 0,
8106 ofproto_dpif_unclog, NULL);
8107 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8108 ofproto_dpif_self_check, NULL);
8109 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8110 ofproto_unixctl_dpif_dump_dps, NULL);
8111 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8112 ofproto_unixctl_dpif_show, NULL);
8113 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8114 ofproto_unixctl_dpif_dump_flows, NULL);
8115 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8116 ofproto_unixctl_dpif_del_flows, NULL);
8119 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8121 * This is deprecated. It is only for compatibility with broken device drivers
8122 * in old versions of Linux that do not properly support VLANs when VLAN
8123 * devices are not used. When broken device drivers are no longer in
8124 * widespread use, we will delete these interfaces. */
8127 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8129 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8130 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8132 if (realdev_ofp_port == ofport->realdev_ofp_port
8133 && vid == ofport->vlandev_vid) {
8137 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8139 if (ofport->realdev_ofp_port) {
8142 if (realdev_ofp_port && ofport->bundle) {
8143 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8144 * themselves be part of a bundle. */
8145 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8148 ofport->realdev_ofp_port = realdev_ofp_port;
8149 ofport->vlandev_vid = vid;
8151 if (realdev_ofp_port) {
8152 vsp_add(ofport, realdev_ofp_port, vid);
8159 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8161 return hash_2words(realdev_ofp_port, vid);
8164 /* Returns the ODP port number of the Linux VLAN device that corresponds to
8165 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
8166 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
8167 * it would return the port number of eth0.9.
8169 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
8170 * function just returns its 'realdev_odp_port' argument. */
8172 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8173 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
8175 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8176 uint16_t realdev_ofp_port;
8177 int vid = vlan_tci_to_vid(vlan_tci);
8178 const struct vlan_splinter *vsp;
8180 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
8181 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8182 hash_realdev_vid(realdev_ofp_port, vid),
8183 &ofproto->realdev_vid_map) {
8184 if (vsp->realdev_ofp_port == realdev_ofp_port
8185 && vsp->vid == vid) {
8186 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
8190 return realdev_odp_port;
8193 static struct vlan_splinter *
8194 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8196 struct vlan_splinter *vsp;
8198 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8199 &ofproto->vlandev_map) {
8200 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8208 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8209 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8210 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8211 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8212 * eth0 and store 9 in '*vid'.
8214 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8215 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8218 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8219 uint16_t vlandev_ofp_port, int *vid)
8221 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8222 const struct vlan_splinter *vsp;
8224 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8229 return vsp->realdev_ofp_port;
8235 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8236 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8237 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8238 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8239 * always the case unless VLAN splinters are enabled), returns false without
8240 * making any changes. */
8242 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8247 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8252 /* Cause the flow to be processed as if it came in on the real device with
8253 * the VLAN device's VLAN ID. */
8254 flow->in_port = realdev;
8255 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8260 vsp_remove(struct ofport_dpif *port)
8262 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8263 struct vlan_splinter *vsp;
8265 vsp = vlandev_find(ofproto, port->up.ofp_port);
8267 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8268 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8271 port->realdev_ofp_port = 0;
8273 VLOG_ERR("missing vlan device record");
8278 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8280 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8282 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8283 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8284 == realdev_ofp_port)) {
8285 struct vlan_splinter *vsp;
8287 vsp = xmalloc(sizeof *vsp);
8288 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8289 hash_int(port->up.ofp_port, 0));
8290 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8291 hash_realdev_vid(realdev_ofp_port, vid));
8292 vsp->realdev_ofp_port = realdev_ofp_port;
8293 vsp->vlandev_ofp_port = port->up.ofp_port;
8296 port->realdev_ofp_port = realdev_ofp_port;
8298 VLOG_ERR("duplicate vlan device record");
8303 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8305 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8306 return ofport ? ofport->odp_port : OVSP_NONE;
8309 static struct ofport_dpif *
8310 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8312 struct ofport_dpif *port;
8314 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8315 hash_int(odp_port, 0),
8316 &backer->odp_to_ofport_map) {
8317 if (port->odp_port == odp_port) {
8326 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8328 struct ofport_dpif *port;
8330 port = odp_port_to_ofport(ofproto->backer, odp_port);
8331 if (port && &ofproto->up == port->up.ofproto) {
8332 return port->up.ofp_port;
8338 const struct ofproto_class ofproto_dpif_class = {
8373 port_is_lacp_current,
8374 NULL, /* rule_choose_table */
8381 rule_modify_actions,
8390 get_cfm_remote_mpids,
8395 get_stp_port_status,
8402 is_mirror_output_bundle,
8403 forward_bpdu_changed,
8404 set_mac_table_config,