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);
88 * - Do include packets and bytes from facets that have been deleted or
89 * whose own statistics have been folded into the rule.
91 * - Do include packets and bytes sent "by hand" that were accounted to
92 * the rule without any facet being involved (this is a rare corner
93 * case in rule_execute()).
95 * - Do not include packet or bytes that can be obtained from any facet's
96 * packet_count or byte_count member or that can be obtained from the
97 * datapath by, e.g., dpif_flow_get() for any subfacet.
99 uint64_t packet_count; /* Number of packets received. */
100 uint64_t byte_count; /* Number of bytes received. */
102 tag_type tag; /* Caches rule_calculate_tag() result. */
104 struct list facets; /* List of "struct facet"s. */
107 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
109 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
112 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
113 const struct flow *);
114 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
117 static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto,
118 const struct flow *flow);
120 static void rule_credit_stats(struct rule_dpif *,
121 const struct dpif_flow_stats *);
122 static void flow_push_stats(struct facet *, 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 /* stack for the push and pop actions.
220 * Each stack element is of the type "union mf_subvalue". */
222 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
224 /* The packet corresponding to 'flow', or a null pointer if we are
225 * revalidating without a packet to refer to. */
226 const struct ofpbuf *packet;
228 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
229 * actions update the flow table?
231 * We want to update these tables if we are actually processing a packet,
232 * or if we are accounting for packets that the datapath has processed, but
233 * not if we are just revalidating. */
236 /* The rule that we are currently translating, or NULL. */
237 struct rule_dpif *rule;
239 /* Union of the set of TCP flags seen so far in this flow. (Used only by
240 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
244 /* If nonnull, flow translation calls this function just before executing a
245 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
246 * when the recursion depth is exceeded.
248 * 'rule' is the rule being submitted into. It will be null if the
249 * resubmit or OFPP_TABLE action didn't find a matching rule.
251 * This is normally null so the client has to set it manually after
252 * calling action_xlate_ctx_init(). */
253 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule);
255 /* If nonnull, flow translation calls this function to report some
256 * significant decision, e.g. to explain why OFPP_NORMAL translation
257 * dropped a packet. */
258 void (*report_hook)(struct action_xlate_ctx *, const char *s);
260 /* If nonnull, flow translation credits the specified statistics to each
261 * rule reached through a resubmit or OFPP_TABLE action.
263 * This is normally null so the client has to set it manually after
264 * calling action_xlate_ctx_init(). */
265 const struct dpif_flow_stats *resubmit_stats;
267 /* xlate_actions() initializes and uses these members. The client might want
268 * to look at them after it returns. */
270 struct ofpbuf *odp_actions; /* Datapath actions. */
271 tag_type tags; /* Tags associated with actions. */
272 enum slow_path_reason slow; /* 0 if fast path may be used. */
273 bool has_learn; /* Actions include NXAST_LEARN? */
274 bool has_normal; /* Actions output to OFPP_NORMAL? */
275 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
276 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
277 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
279 /* xlate_actions() initializes and uses these members, but the client has no
280 * reason to look at them. */
282 int recurse; /* Recursion level, via xlate_table_action. */
283 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
284 struct flow base_flow; /* Flow at the last commit. */
285 uint32_t orig_skb_priority; /* Priority when packet arrived. */
286 uint8_t table_id; /* OpenFlow table ID where flow was found. */
287 uint32_t sflow_n_outputs; /* Number of output ports. */
288 uint32_t sflow_odp_port; /* Output port for composing sFlow action. */
289 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
290 bool exit; /* No further actions should be processed. */
293 static void action_xlate_ctx_init(struct action_xlate_ctx *,
294 struct ofproto_dpif *, const struct flow *,
295 ovs_be16 initial_tci, struct rule_dpif *,
296 uint8_t tcp_flags, const struct ofpbuf *);
297 static void xlate_actions(struct action_xlate_ctx *,
298 const struct ofpact *ofpacts, size_t ofpacts_len,
299 struct ofpbuf *odp_actions);
300 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
301 const struct ofpact *ofpacts,
303 static void xlate_table_action(struct action_xlate_ctx *, uint16_t in_port,
304 uint8_t table_id, bool may_packet_in);
306 static size_t put_userspace_action(const struct ofproto_dpif *,
307 struct ofpbuf *odp_actions,
309 const union user_action_cookie *);
311 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
312 enum slow_path_reason,
313 uint64_t *stub, size_t stub_size,
314 const struct nlattr **actionsp,
315 size_t *actions_lenp);
317 static void xlate_report(struct action_xlate_ctx *ctx, const char *s);
319 /* A subfacet (see "struct subfacet" below) has three possible installation
322 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
323 * case just after the subfacet is created, just before the subfacet is
324 * destroyed, or if the datapath returns an error when we try to install a
327 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
329 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
330 * ofproto_dpif is installed in the datapath.
333 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
334 SF_FAST_PATH, /* Full actions are installed. */
335 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
338 static const char *subfacet_path_to_string(enum subfacet_path);
340 /* A dpif flow and actions associated with a facet.
342 * See also the large comment on struct facet. */
345 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
346 struct list list_node; /* In struct facet's 'facets' list. */
347 struct facet *facet; /* Owning facet. */
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 static void subfacet_destroy(struct subfacet *);
389 static void subfacet_destroy__(struct subfacet *);
390 static void subfacet_destroy_batch(struct ofproto_dpif *,
391 struct subfacet **, int n);
392 static void subfacet_reset_dp_stats(struct subfacet *,
393 struct dpif_flow_stats *);
394 static void subfacet_update_time(struct subfacet *, long long int used);
395 static void subfacet_update_stats(struct subfacet *,
396 const struct dpif_flow_stats *);
397 static void subfacet_make_actions(struct subfacet *,
398 const struct ofpbuf *packet,
399 struct ofpbuf *odp_actions);
400 static int subfacet_install(struct subfacet *,
401 const struct nlattr *actions, size_t actions_len,
402 struct dpif_flow_stats *, enum slow_path_reason);
403 static void subfacet_uninstall(struct subfacet *);
405 static enum subfacet_path subfacet_want_path(enum slow_path_reason);
407 /* An exact-match instantiation of an OpenFlow flow.
409 * A facet associates a "struct flow", which represents the Open vSwitch
410 * userspace idea of an exact-match flow, with one or more subfacets. Each
411 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
412 * the facet. When the kernel module (or other dpif implementation) and Open
413 * vSwitch userspace agree on the definition of a flow key, there is exactly
414 * one subfacet per facet. If the dpif implementation supports more-specific
415 * flow matching than userspace, however, a facet can have more than one
416 * subfacet, each of which corresponds to some distinction in flow that
417 * userspace simply doesn't understand.
419 * Flow expiration works in terms of subfacets, so a facet must have at least
420 * one subfacet or it will never expire, leaking memory. */
423 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
424 struct list list_node; /* In owning rule's 'facets' list. */
425 struct rule_dpif *rule; /* Owning rule. */
428 struct list subfacets;
429 long long int used; /* Time last used; time created if not used. */
436 * - Do include packets and bytes sent "by hand", e.g. with
439 * - Do include packets and bytes that were obtained from the datapath
440 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
441 * DPIF_FP_ZERO_STATS).
443 * - Do not include packets or bytes that can be obtained from the
444 * datapath for any existing subfacet.
446 uint64_t packet_count; /* Number of packets received. */
447 uint64_t byte_count; /* Number of bytes received. */
449 /* Resubmit statistics. */
450 uint64_t prev_packet_count; /* Number of packets from last stats push. */
451 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
452 long long int prev_used; /* Used time from last stats push. */
455 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
456 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
457 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
459 /* Properties of datapath actions.
461 * Every subfacet has its own actions because actions can differ slightly
462 * between splintered and non-splintered subfacets due to the VLAN tag
463 * being initially different (present vs. absent). All of them have these
464 * properties in common so we just store one copy of them here. */
465 bool has_learn; /* Actions include NXAST_LEARN? */
466 bool has_normal; /* Actions output to OFPP_NORMAL? */
467 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
468 tag_type tags; /* Tags that would require revalidation. */
469 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
471 /* Storage for a single subfacet, to reduce malloc() time and space
472 * overhead. (A facet always has at least one subfacet and in the common
473 * case has exactly one subfacet.) */
474 struct subfacet one_subfacet;
477 static struct facet *facet_create(struct rule_dpif *,
478 const struct flow *, uint32_t hash);
479 static void facet_remove(struct facet *);
480 static void facet_free(struct facet *);
482 static struct facet *facet_find(struct ofproto_dpif *,
483 const struct flow *, uint32_t hash);
484 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
485 const struct flow *, uint32_t hash);
486 static void facet_revalidate(struct facet *);
487 static bool facet_check_consistency(struct facet *);
489 static void facet_flush_stats(struct facet *);
491 static void facet_update_time(struct facet *, long long int used);
492 static void facet_reset_counters(struct facet *);
493 static void facet_push_stats(struct facet *);
494 static void facet_learn(struct facet *);
495 static void facet_account(struct facet *);
497 static bool facet_is_controller_flow(struct facet *);
500 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
504 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
505 struct list bundle_node; /* In struct ofbundle's "ports" list. */
506 struct cfm *cfm; /* Connectivity Fault Management, if any. */
507 tag_type tag; /* Tag associated with this port. */
508 bool may_enable; /* May be enabled in bonds. */
509 long long int carrier_seq; /* Carrier status changes. */
510 struct tnl_port *tnl_port; /* Tunnel handle, or null. */
513 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
514 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
515 long long int stp_state_entered;
517 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
519 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
521 * This is deprecated. It is only for compatibility with broken device
522 * drivers in old versions of Linux that do not properly support VLANs when
523 * VLAN devices are not used. When broken device drivers are no longer in
524 * widespread use, we will delete these interfaces. */
525 uint16_t realdev_ofp_port;
529 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
530 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
531 * traffic egressing the 'ofport' with that priority should be marked with. */
532 struct priority_to_dscp {
533 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
534 uint32_t priority; /* Priority of this queue (see struct flow). */
536 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
539 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
541 * This is deprecated. It is only for compatibility with broken device drivers
542 * in old versions of Linux that do not properly support VLANs when VLAN
543 * devices are not used. When broken device drivers are no longer in
544 * widespread use, we will delete these interfaces. */
545 struct vlan_splinter {
546 struct hmap_node realdev_vid_node;
547 struct hmap_node vlandev_node;
548 uint16_t realdev_ofp_port;
549 uint16_t vlandev_ofp_port;
553 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
554 uint32_t realdev, ovs_be16 vlan_tci);
555 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
556 static void vsp_remove(struct ofport_dpif *);
557 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
559 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
561 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
564 static struct ofport_dpif *
565 ofport_dpif_cast(const struct ofport *ofport)
567 ovs_assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
568 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
571 static void port_run(struct ofport_dpif *);
572 static void port_run_fast(struct ofport_dpif *);
573 static void port_wait(struct ofport_dpif *);
574 static int set_cfm(struct ofport *, const struct cfm_settings *);
575 static void ofport_clear_priorities(struct ofport_dpif *);
577 struct dpif_completion {
578 struct list list_node;
579 struct ofoperation *op;
582 /* Extra information about a classifier table.
583 * Currently used just for optimized flow revalidation. */
585 /* If either of these is nonnull, then this table has a form that allows
586 * flows to be tagged to avoid revalidating most flows for the most common
587 * kinds of flow table changes. */
588 struct cls_table *catchall_table; /* Table that wildcards all fields. */
589 struct cls_table *other_table; /* Table with any other wildcard set. */
590 uint32_t basis; /* Keeps each table's tags separate. */
593 /* Reasons that we might need to revalidate every facet, and corresponding
596 * A value of 0 means that there is no need to revalidate.
598 * It would be nice to have some cleaner way to integrate with coverage
599 * counters, but with only a few reasons I guess this is good enough for
601 enum revalidate_reason {
602 REV_RECONFIGURE = 1, /* Switch configuration changed. */
603 REV_STP, /* Spanning tree protocol port status change. */
604 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
605 REV_FLOW_TABLE, /* Flow table changed. */
606 REV_INCONSISTENCY /* Facet self-check failed. */
608 COVERAGE_DEFINE(rev_reconfigure);
609 COVERAGE_DEFINE(rev_stp);
610 COVERAGE_DEFINE(rev_port_toggled);
611 COVERAGE_DEFINE(rev_flow_table);
612 COVERAGE_DEFINE(rev_inconsistency);
614 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
615 * These are datapath flows which have no associated ofproto, if they did we
616 * would use facets. */
618 struct hmap_node hmap_node;
623 /* All datapaths of a given type share a single dpif backer instance. */
628 struct timer next_expiration;
629 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
631 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
633 /* Facet revalidation flags applying to facets which use this backer. */
634 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
635 struct tag_set revalidate_set; /* Revalidate only matching facets. */
637 struct hmap drop_keys; /* Set of dropped odp keys. */
640 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
641 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
643 static void drop_key_clear(struct dpif_backer *);
644 static struct ofport_dpif *
645 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
647 struct ofproto_dpif {
648 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
650 struct dpif_backer *backer;
652 /* Special OpenFlow rules. */
653 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
654 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
660 struct netflow *netflow;
661 struct dpif_sflow *sflow;
662 struct hmap bundles; /* Contains "struct ofbundle"s. */
663 struct mac_learning *ml;
664 struct ofmirror *mirrors[MAX_MIRRORS];
666 bool has_bonded_bundles;
670 struct hmap subfacets;
671 struct governor *governor;
674 struct table_dpif tables[N_TABLES];
676 /* Support for debugging async flow mods. */
677 struct list completions;
679 bool has_bundle_action; /* True when the first bundle action appears. */
680 struct netdev_stats stats; /* To account packets generated and consumed in
685 long long int stp_last_tick;
687 /* VLAN splinters. */
688 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
689 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
692 struct sset ports; /* Set of standard port names. */
693 struct sset ghost_ports; /* Ports with no datapath port. */
694 struct sset port_poll_set; /* Queued names for port_poll() reply. */
695 int port_poll_errno; /* Last errno for port_poll() reply. */
698 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
699 * for debugging the asynchronous flow_mod implementation.) */
702 /* All existing ofproto_dpif instances, indexed by ->up.name. */
703 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
705 static void ofproto_dpif_unixctl_init(void);
707 static struct ofproto_dpif *
708 ofproto_dpif_cast(const struct ofproto *ofproto)
710 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
711 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
714 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
716 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
718 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
719 const struct ofpbuf *, ovs_be16 initial_tci,
722 /* Packet processing. */
723 static void update_learning_table(struct ofproto_dpif *,
724 const struct flow *, int vlan,
727 #define FLOW_MISS_MAX_BATCH 50
728 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
730 /* Flow expiration. */
731 static int expire(struct dpif_backer *);
734 static void send_netflow_active_timeouts(struct ofproto_dpif *);
737 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
738 static size_t compose_sflow_action(const struct ofproto_dpif *,
739 struct ofpbuf *odp_actions,
740 const struct flow *, uint32_t odp_port);
741 static void add_mirror_actions(struct action_xlate_ctx *ctx,
742 const struct flow *flow);
743 /* Global variables. */
744 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
746 /* Initial mappings of port to bridge mappings. */
747 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
749 /* Factory functions. */
752 init(const struct shash *iface_hints)
754 struct shash_node *node;
756 /* Make a local copy, since we don't own 'iface_hints' elements. */
757 SHASH_FOR_EACH(node, iface_hints) {
758 const struct iface_hint *orig_hint = node->data;
759 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
761 new_hint->br_name = xstrdup(orig_hint->br_name);
762 new_hint->br_type = xstrdup(orig_hint->br_type);
763 new_hint->ofp_port = orig_hint->ofp_port;
765 shash_add(&init_ofp_ports, node->name, new_hint);
770 enumerate_types(struct sset *types)
772 dp_enumerate_types(types);
776 enumerate_names(const char *type, struct sset *names)
778 struct ofproto_dpif *ofproto;
781 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
782 if (strcmp(type, ofproto->up.type)) {
785 sset_add(names, ofproto->up.name);
792 del(const char *type, const char *name)
797 error = dpif_open(name, type, &dpif);
799 error = dpif_delete(dpif);
806 port_open_type(const char *datapath_type, const char *port_type)
808 return dpif_port_open_type(datapath_type, port_type);
811 /* Type functions. */
813 static struct ofproto_dpif *
814 lookup_ofproto_dpif_by_port_name(const char *name)
816 struct ofproto_dpif *ofproto;
818 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
819 if (sset_contains(&ofproto->ports, name)) {
828 type_run(const char *type)
830 struct dpif_backer *backer;
834 backer = shash_find_data(&all_dpif_backers, type);
836 /* This is not necessarily a problem, since backers are only
837 * created on demand. */
841 dpif_run(backer->dpif);
843 if (backer->need_revalidate
844 || !tag_set_is_empty(&backer->revalidate_set)) {
845 struct tag_set revalidate_set = backer->revalidate_set;
846 bool need_revalidate = backer->need_revalidate;
847 struct ofproto_dpif *ofproto;
848 struct simap_node *node;
849 struct simap tmp_backers;
851 /* Handle tunnel garbage collection. */
852 simap_init(&tmp_backers);
853 simap_swap(&backer->tnl_backers, &tmp_backers);
855 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
856 struct ofport_dpif *iter;
858 if (backer != ofproto->backer) {
862 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
865 if (!iter->tnl_port) {
869 dp_port = netdev_vport_get_dpif_port(iter->up.netdev);
870 node = simap_find(&tmp_backers, dp_port);
872 simap_put(&backer->tnl_backers, dp_port, node->data);
873 simap_delete(&tmp_backers, node);
874 node = simap_find(&backer->tnl_backers, dp_port);
876 node = simap_find(&backer->tnl_backers, dp_port);
878 uint32_t odp_port = UINT32_MAX;
880 if (!dpif_port_add(backer->dpif, iter->up.netdev,
882 simap_put(&backer->tnl_backers, dp_port, odp_port);
883 node = simap_find(&backer->tnl_backers, dp_port);
888 iter->odp_port = node ? node->data : OVSP_NONE;
889 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
891 backer->need_revalidate = REV_RECONFIGURE;
896 SIMAP_FOR_EACH (node, &tmp_backers) {
897 dpif_port_del(backer->dpif, node->data);
899 simap_destroy(&tmp_backers);
901 switch (backer->need_revalidate) {
902 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
903 case REV_STP: COVERAGE_INC(rev_stp); break;
904 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
905 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
906 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
909 if (backer->need_revalidate) {
910 /* Clear the drop_keys in case we should now be accepting some
911 * formerly dropped flows. */
912 drop_key_clear(backer);
915 /* Clear the revalidation flags. */
916 tag_set_init(&backer->revalidate_set);
917 backer->need_revalidate = 0;
919 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
920 struct facet *facet, *next;
922 if (ofproto->backer != backer) {
926 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
928 || tag_set_intersects(&revalidate_set, facet->tags)) {
929 facet_revalidate(facet);
935 if (timer_expired(&backer->next_expiration)) {
936 int delay = expire(backer);
937 timer_set_duration(&backer->next_expiration, delay);
940 /* Check for port changes in the dpif. */
941 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
942 struct ofproto_dpif *ofproto;
943 struct dpif_port port;
945 /* Don't report on the datapath's device. */
946 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
950 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
951 &all_ofproto_dpifs) {
952 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
957 ofproto = lookup_ofproto_dpif_by_port_name(devname);
958 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
959 /* The port was removed. If we know the datapath,
960 * report it through poll_set(). If we don't, it may be
961 * notifying us of a removal we initiated, so ignore it.
962 * If there's a pending ENOBUFS, let it stand, since
963 * everything will be reevaluated. */
964 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
965 sset_add(&ofproto->port_poll_set, devname);
966 ofproto->port_poll_errno = 0;
968 } else if (!ofproto) {
969 /* The port was added, but we don't know with which
970 * ofproto we should associate it. Delete it. */
971 dpif_port_del(backer->dpif, port.port_no);
973 dpif_port_destroy(&port);
979 if (error != EAGAIN) {
980 struct ofproto_dpif *ofproto;
982 /* There was some sort of error, so propagate it to all
983 * ofprotos that use this backer. */
984 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
985 &all_ofproto_dpifs) {
986 if (ofproto->backer == backer) {
987 sset_clear(&ofproto->port_poll_set);
988 ofproto->port_poll_errno = error;
997 type_run_fast(const char *type)
999 struct dpif_backer *backer;
1002 backer = shash_find_data(&all_dpif_backers, type);
1004 /* This is not necessarily a problem, since backers are only
1005 * created on demand. */
1009 /* Handle one or more batches of upcalls, until there's nothing left to do
1010 * or until we do a fixed total amount of work.
1012 * We do work in batches because it can be much cheaper to set up a number
1013 * of flows and fire off their patches all at once. We do multiple batches
1014 * because in some cases handling a packet can cause another packet to be
1015 * queued almost immediately as part of the return flow. Both
1016 * optimizations can make major improvements on some benchmarks and
1017 * presumably for real traffic as well. */
1019 while (work < FLOW_MISS_MAX_BATCH) {
1020 int retval = handle_upcalls(backer, FLOW_MISS_MAX_BATCH - work);
1031 type_wait(const char *type)
1033 struct dpif_backer *backer;
1035 backer = shash_find_data(&all_dpif_backers, type);
1037 /* This is not necessarily a problem, since backers are only
1038 * created on demand. */
1042 timer_wait(&backer->next_expiration);
1045 /* Basic life-cycle. */
1047 static int add_internal_flows(struct ofproto_dpif *);
1049 static struct ofproto *
1052 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1053 return &ofproto->up;
1057 dealloc(struct ofproto *ofproto_)
1059 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1064 close_dpif_backer(struct dpif_backer *backer)
1066 struct shash_node *node;
1068 ovs_assert(backer->refcount > 0);
1070 if (--backer->refcount) {
1074 drop_key_clear(backer);
1075 hmap_destroy(&backer->drop_keys);
1077 simap_destroy(&backer->tnl_backers);
1078 hmap_destroy(&backer->odp_to_ofport_map);
1079 node = shash_find(&all_dpif_backers, backer->type);
1081 shash_delete(&all_dpif_backers, node);
1082 dpif_close(backer->dpif);
1087 /* Datapath port slated for removal from datapath. */
1088 struct odp_garbage {
1089 struct list list_node;
1094 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1096 struct dpif_backer *backer;
1097 struct dpif_port_dump port_dump;
1098 struct dpif_port port;
1099 struct shash_node *node;
1100 struct list garbage_list;
1101 struct odp_garbage *garbage, *next;
1107 backer = shash_find_data(&all_dpif_backers, type);
1114 backer_name = xasprintf("ovs-%s", type);
1116 /* Remove any existing datapaths, since we assume we're the only
1117 * userspace controlling the datapath. */
1119 dp_enumerate_names(type, &names);
1120 SSET_FOR_EACH(name, &names) {
1121 struct dpif *old_dpif;
1123 /* Don't remove our backer if it exists. */
1124 if (!strcmp(name, backer_name)) {
1128 if (dpif_open(name, type, &old_dpif)) {
1129 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1131 dpif_delete(old_dpif);
1132 dpif_close(old_dpif);
1135 sset_destroy(&names);
1137 backer = xmalloc(sizeof *backer);
1139 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1142 VLOG_ERR("failed to open datapath of type %s: %s", type,
1148 backer->type = xstrdup(type);
1149 backer->refcount = 1;
1150 hmap_init(&backer->odp_to_ofport_map);
1151 hmap_init(&backer->drop_keys);
1152 timer_set_duration(&backer->next_expiration, 1000);
1153 backer->need_revalidate = 0;
1154 simap_init(&backer->tnl_backers);
1155 tag_set_init(&backer->revalidate_set);
1158 dpif_flow_flush(backer->dpif);
1160 /* Loop through the ports already on the datapath and remove any
1161 * that we don't need anymore. */
1162 list_init(&garbage_list);
1163 dpif_port_dump_start(&port_dump, backer->dpif);
1164 while (dpif_port_dump_next(&port_dump, &port)) {
1165 node = shash_find(&init_ofp_ports, port.name);
1166 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1167 garbage = xmalloc(sizeof *garbage);
1168 garbage->odp_port = port.port_no;
1169 list_push_front(&garbage_list, &garbage->list_node);
1172 dpif_port_dump_done(&port_dump);
1174 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1175 dpif_port_del(backer->dpif, garbage->odp_port);
1176 list_remove(&garbage->list_node);
1180 shash_add(&all_dpif_backers, type, backer);
1182 error = dpif_recv_set(backer->dpif, true);
1184 VLOG_ERR("failed to listen on datapath of type %s: %s",
1185 type, strerror(error));
1186 close_dpif_backer(backer);
1194 construct(struct ofproto *ofproto_)
1196 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1197 struct shash_node *node, *next;
1202 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1207 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1208 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1210 ofproto->n_matches = 0;
1212 ofproto->netflow = NULL;
1213 ofproto->sflow = NULL;
1214 ofproto->stp = NULL;
1215 hmap_init(&ofproto->bundles);
1216 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1217 for (i = 0; i < MAX_MIRRORS; i++) {
1218 ofproto->mirrors[i] = NULL;
1220 ofproto->has_bonded_bundles = false;
1222 hmap_init(&ofproto->facets);
1223 hmap_init(&ofproto->subfacets);
1224 ofproto->governor = NULL;
1226 for (i = 0; i < N_TABLES; i++) {
1227 struct table_dpif *table = &ofproto->tables[i];
1229 table->catchall_table = NULL;
1230 table->other_table = NULL;
1231 table->basis = random_uint32();
1234 list_init(&ofproto->completions);
1236 ofproto_dpif_unixctl_init();
1238 ofproto->has_mirrors = false;
1239 ofproto->has_bundle_action = false;
1241 hmap_init(&ofproto->vlandev_map);
1242 hmap_init(&ofproto->realdev_vid_map);
1244 sset_init(&ofproto->ports);
1245 sset_init(&ofproto->ghost_ports);
1246 sset_init(&ofproto->port_poll_set);
1247 ofproto->port_poll_errno = 0;
1249 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1250 struct iface_hint *iface_hint = node->data;
1252 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1253 /* Check if the datapath already has this port. */
1254 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1255 sset_add(&ofproto->ports, node->name);
1258 free(iface_hint->br_name);
1259 free(iface_hint->br_type);
1261 shash_delete(&init_ofp_ports, node);
1265 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1266 hash_string(ofproto->up.name, 0));
1267 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1269 ofproto_init_tables(ofproto_, N_TABLES);
1270 error = add_internal_flows(ofproto);
1271 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1277 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1278 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1280 struct ofputil_flow_mod fm;
1283 match_init_catchall(&fm.match);
1285 match_set_reg(&fm.match, 0, id);
1286 fm.new_cookie = htonll(0);
1287 fm.cookie = htonll(0);
1288 fm.cookie_mask = htonll(0);
1289 fm.table_id = TBL_INTERNAL;
1290 fm.command = OFPFC_ADD;
1291 fm.idle_timeout = 0;
1292 fm.hard_timeout = 0;
1296 fm.ofpacts = ofpacts->data;
1297 fm.ofpacts_len = ofpacts->size;
1299 error = ofproto_flow_mod(&ofproto->up, &fm);
1301 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1302 id, ofperr_to_string(error));
1306 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1307 ovs_assert(*rulep != NULL);
1313 add_internal_flows(struct ofproto_dpif *ofproto)
1315 struct ofpact_controller *controller;
1316 uint64_t ofpacts_stub[128 / 8];
1317 struct ofpbuf ofpacts;
1321 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1324 controller = ofpact_put_CONTROLLER(&ofpacts);
1325 controller->max_len = UINT16_MAX;
1326 controller->controller_id = 0;
1327 controller->reason = OFPR_NO_MATCH;
1328 ofpact_pad(&ofpacts);
1330 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1335 ofpbuf_clear(&ofpacts);
1336 error = add_internal_flow(ofproto, id++, &ofpacts,
1337 &ofproto->no_packet_in_rule);
1342 complete_operations(struct ofproto_dpif *ofproto)
1344 struct dpif_completion *c, *next;
1346 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1347 ofoperation_complete(c->op, 0);
1348 list_remove(&c->list_node);
1354 destruct(struct ofproto *ofproto_)
1356 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1357 struct rule_dpif *rule, *next_rule;
1358 struct oftable *table;
1361 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1362 complete_operations(ofproto);
1364 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1365 struct cls_cursor cursor;
1367 cls_cursor_init(&cursor, &table->cls, NULL);
1368 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1369 ofproto_rule_destroy(&rule->up);
1373 for (i = 0; i < MAX_MIRRORS; i++) {
1374 mirror_destroy(ofproto->mirrors[i]);
1377 netflow_destroy(ofproto->netflow);
1378 dpif_sflow_destroy(ofproto->sflow);
1379 hmap_destroy(&ofproto->bundles);
1380 mac_learning_destroy(ofproto->ml);
1382 hmap_destroy(&ofproto->facets);
1383 hmap_destroy(&ofproto->subfacets);
1384 governor_destroy(ofproto->governor);
1386 hmap_destroy(&ofproto->vlandev_map);
1387 hmap_destroy(&ofproto->realdev_vid_map);
1389 sset_destroy(&ofproto->ports);
1390 sset_destroy(&ofproto->ghost_ports);
1391 sset_destroy(&ofproto->port_poll_set);
1393 close_dpif_backer(ofproto->backer);
1397 run_fast(struct ofproto *ofproto_)
1399 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1400 struct ofport_dpif *ofport;
1402 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1403 port_run_fast(ofport);
1410 run(struct ofproto *ofproto_)
1412 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1413 struct ofport_dpif *ofport;
1414 struct ofbundle *bundle;
1418 complete_operations(ofproto);
1421 error = run_fast(ofproto_);
1426 if (ofproto->netflow) {
1427 if (netflow_run(ofproto->netflow)) {
1428 send_netflow_active_timeouts(ofproto);
1431 if (ofproto->sflow) {
1432 dpif_sflow_run(ofproto->sflow);
1435 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1438 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1443 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1445 /* Check the consistency of a random facet, to aid debugging. */
1446 if (!hmap_is_empty(&ofproto->facets)
1447 && !ofproto->backer->need_revalidate) {
1448 struct facet *facet;
1450 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1451 struct facet, hmap_node);
1452 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1454 if (!facet_check_consistency(facet)) {
1455 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1460 if (ofproto->governor) {
1463 governor_run(ofproto->governor);
1465 /* If the governor has shrunk to its minimum size and the number of
1466 * subfacets has dwindled, then drop the governor entirely.
1468 * For hysteresis, the number of subfacets to drop the governor is
1469 * smaller than the number needed to trigger its creation. */
1470 n_subfacets = hmap_count(&ofproto->subfacets);
1471 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1472 && governor_is_idle(ofproto->governor)) {
1473 governor_destroy(ofproto->governor);
1474 ofproto->governor = NULL;
1482 wait(struct ofproto *ofproto_)
1484 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1485 struct ofport_dpif *ofport;
1486 struct ofbundle *bundle;
1488 if (!clogged && !list_is_empty(&ofproto->completions)) {
1489 poll_immediate_wake();
1492 dpif_wait(ofproto->backer->dpif);
1493 dpif_recv_wait(ofproto->backer->dpif);
1494 if (ofproto->sflow) {
1495 dpif_sflow_wait(ofproto->sflow);
1497 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1498 poll_immediate_wake();
1500 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1503 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1504 bundle_wait(bundle);
1506 if (ofproto->netflow) {
1507 netflow_wait(ofproto->netflow);
1509 mac_learning_wait(ofproto->ml);
1511 if (ofproto->backer->need_revalidate) {
1512 /* Shouldn't happen, but if it does just go around again. */
1513 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1514 poll_immediate_wake();
1516 if (ofproto->governor) {
1517 governor_wait(ofproto->governor);
1522 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1524 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1526 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1527 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1531 flush(struct ofproto *ofproto_)
1533 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1534 struct subfacet *subfacet, *next_subfacet;
1535 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1539 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1540 &ofproto->subfacets) {
1541 if (subfacet->path != SF_NOT_INSTALLED) {
1542 batch[n_batch++] = subfacet;
1543 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1544 subfacet_destroy_batch(ofproto, batch, n_batch);
1548 subfacet_destroy(subfacet);
1553 subfacet_destroy_batch(ofproto, batch, n_batch);
1558 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1559 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1561 *arp_match_ip = true;
1562 *actions = (OFPUTIL_A_OUTPUT |
1563 OFPUTIL_A_SET_VLAN_VID |
1564 OFPUTIL_A_SET_VLAN_PCP |
1565 OFPUTIL_A_STRIP_VLAN |
1566 OFPUTIL_A_SET_DL_SRC |
1567 OFPUTIL_A_SET_DL_DST |
1568 OFPUTIL_A_SET_NW_SRC |
1569 OFPUTIL_A_SET_NW_DST |
1570 OFPUTIL_A_SET_NW_TOS |
1571 OFPUTIL_A_SET_TP_SRC |
1572 OFPUTIL_A_SET_TP_DST |
1577 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1579 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1580 struct dpif_dp_stats s;
1582 strcpy(ots->name, "classifier");
1584 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1586 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1587 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1590 static struct ofport *
1593 struct ofport_dpif *port = xmalloc(sizeof *port);
1598 port_dealloc(struct ofport *port_)
1600 struct ofport_dpif *port = ofport_dpif_cast(port_);
1605 port_construct(struct ofport *port_)
1607 struct ofport_dpif *port = ofport_dpif_cast(port_);
1608 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1609 const struct netdev *netdev = port->up.netdev;
1610 struct dpif_port dpif_port;
1613 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1614 port->bundle = NULL;
1616 port->tag = tag_create_random();
1617 port->may_enable = true;
1618 port->stp_port = NULL;
1619 port->stp_state = STP_DISABLED;
1620 port->tnl_port = NULL;
1621 hmap_init(&port->priorities);
1622 port->realdev_ofp_port = 0;
1623 port->vlandev_vid = 0;
1624 port->carrier_seq = netdev_get_carrier_resets(netdev);
1626 if (netdev_vport_is_patch(netdev)) {
1627 /* XXX By bailing out here, we don't do required sFlow work. */
1628 port->odp_port = OVSP_NONE;
1632 error = dpif_port_query_by_name(ofproto->backer->dpif,
1633 netdev_vport_get_dpif_port(netdev),
1639 port->odp_port = dpif_port.port_no;
1641 if (netdev_get_tunnel_config(netdev)) {
1642 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1644 /* Sanity-check that a mapping doesn't already exist. This
1645 * shouldn't happen for non-tunnel ports. */
1646 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1647 VLOG_ERR("port %s already has an OpenFlow port number",
1649 dpif_port_destroy(&dpif_port);
1653 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1654 hash_int(port->odp_port, 0));
1656 dpif_port_destroy(&dpif_port);
1658 if (ofproto->sflow) {
1659 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1666 port_destruct(struct ofport *port_)
1668 struct ofport_dpif *port = ofport_dpif_cast(port_);
1669 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1670 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1671 const char *devname = netdev_get_name(port->up.netdev);
1673 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1674 /* The underlying device is still there, so delete it. This
1675 * happens when the ofproto is being destroyed, since the caller
1676 * assumes that removal of attached ports will happen as part of
1678 if (!port->tnl_port) {
1679 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1681 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1684 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1685 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1688 tnl_port_del(port->tnl_port);
1689 sset_find_and_delete(&ofproto->ports, devname);
1690 sset_find_and_delete(&ofproto->ghost_ports, devname);
1691 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1692 bundle_remove(port_);
1693 set_cfm(port_, NULL);
1694 if (ofproto->sflow) {
1695 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1698 ofport_clear_priorities(port);
1699 hmap_destroy(&port->priorities);
1703 port_modified(struct ofport *port_)
1705 struct ofport_dpif *port = ofport_dpif_cast(port_);
1707 if (port->bundle && port->bundle->bond) {
1708 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1713 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1715 struct ofport_dpif *port = ofport_dpif_cast(port_);
1716 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1717 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1719 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1720 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1721 OFPUTIL_PC_NO_PACKET_IN)) {
1722 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1724 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1725 bundle_update(port->bundle);
1731 set_sflow(struct ofproto *ofproto_,
1732 const struct ofproto_sflow_options *sflow_options)
1734 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1735 struct dpif_sflow *ds = ofproto->sflow;
1737 if (sflow_options) {
1739 struct ofport_dpif *ofport;
1741 ds = ofproto->sflow = dpif_sflow_create();
1742 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1743 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1745 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1747 dpif_sflow_set_options(ds, sflow_options);
1750 dpif_sflow_destroy(ds);
1751 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1752 ofproto->sflow = NULL;
1759 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1761 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1768 struct ofproto_dpif *ofproto;
1770 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1771 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1772 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1775 if (cfm_configure(ofport->cfm, s)) {
1781 cfm_destroy(ofport->cfm);
1787 get_cfm_status(const struct ofport *ofport_,
1788 struct ofproto_cfm_status *status)
1790 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1793 status->faults = cfm_get_fault(ofport->cfm);
1794 status->remote_opstate = cfm_get_opup(ofport->cfm);
1795 status->health = cfm_get_health(ofport->cfm);
1796 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1803 /* Spanning Tree. */
1806 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1808 struct ofproto_dpif *ofproto = ofproto_;
1809 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1810 struct ofport_dpif *ofport;
1812 ofport = stp_port_get_aux(sp);
1814 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1815 ofproto->up.name, port_num);
1817 struct eth_header *eth = pkt->l2;
1819 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1820 if (eth_addr_is_zero(eth->eth_src)) {
1821 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1822 "with unknown MAC", ofproto->up.name, port_num);
1824 send_packet(ofport, pkt);
1830 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1832 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1834 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1836 /* Only revalidate flows if the configuration changed. */
1837 if (!s != !ofproto->stp) {
1838 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1842 if (!ofproto->stp) {
1843 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1844 send_bpdu_cb, ofproto);
1845 ofproto->stp_last_tick = time_msec();
1848 stp_set_bridge_id(ofproto->stp, s->system_id);
1849 stp_set_bridge_priority(ofproto->stp, s->priority);
1850 stp_set_hello_time(ofproto->stp, s->hello_time);
1851 stp_set_max_age(ofproto->stp, s->max_age);
1852 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1854 struct ofport *ofport;
1856 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1857 set_stp_port(ofport, NULL);
1860 stp_destroy(ofproto->stp);
1861 ofproto->stp = NULL;
1868 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1870 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1874 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1875 s->designated_root = stp_get_designated_root(ofproto->stp);
1876 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1885 update_stp_port_state(struct ofport_dpif *ofport)
1887 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1888 enum stp_state state;
1890 /* Figure out new state. */
1891 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1895 if (ofport->stp_state != state) {
1896 enum ofputil_port_state of_state;
1899 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1900 netdev_get_name(ofport->up.netdev),
1901 stp_state_name(ofport->stp_state),
1902 stp_state_name(state));
1903 if (stp_learn_in_state(ofport->stp_state)
1904 != stp_learn_in_state(state)) {
1905 /* xxx Learning action flows should also be flushed. */
1906 mac_learning_flush(ofproto->ml,
1907 &ofproto->backer->revalidate_set);
1909 fwd_change = stp_forward_in_state(ofport->stp_state)
1910 != stp_forward_in_state(state);
1912 ofproto->backer->need_revalidate = REV_STP;
1913 ofport->stp_state = state;
1914 ofport->stp_state_entered = time_msec();
1916 if (fwd_change && ofport->bundle) {
1917 bundle_update(ofport->bundle);
1920 /* Update the STP state bits in the OpenFlow port description. */
1921 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1922 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1923 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1924 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1925 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1927 ofproto_port_set_state(&ofport->up, of_state);
1931 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1932 * caller is responsible for assigning STP port numbers and ensuring
1933 * there are no duplicates. */
1935 set_stp_port(struct ofport *ofport_,
1936 const struct ofproto_port_stp_settings *s)
1938 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1939 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1940 struct stp_port *sp = ofport->stp_port;
1942 if (!s || !s->enable) {
1944 ofport->stp_port = NULL;
1945 stp_port_disable(sp);
1946 update_stp_port_state(ofport);
1949 } else if (sp && stp_port_no(sp) != s->port_num
1950 && ofport == stp_port_get_aux(sp)) {
1951 /* The port-id changed, so disable the old one if it's not
1952 * already in use by another port. */
1953 stp_port_disable(sp);
1956 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1957 stp_port_enable(sp);
1959 stp_port_set_aux(sp, ofport);
1960 stp_port_set_priority(sp, s->priority);
1961 stp_port_set_path_cost(sp, s->path_cost);
1963 update_stp_port_state(ofport);
1969 get_stp_port_status(struct ofport *ofport_,
1970 struct ofproto_port_stp_status *s)
1972 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1973 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1974 struct stp_port *sp = ofport->stp_port;
1976 if (!ofproto->stp || !sp) {
1982 s->port_id = stp_port_get_id(sp);
1983 s->state = stp_port_get_state(sp);
1984 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1985 s->role = stp_port_get_role(sp);
1986 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1992 stp_run(struct ofproto_dpif *ofproto)
1995 long long int now = time_msec();
1996 long long int elapsed = now - ofproto->stp_last_tick;
1997 struct stp_port *sp;
2000 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2001 ofproto->stp_last_tick = now;
2003 while (stp_get_changed_port(ofproto->stp, &sp)) {
2004 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2007 update_stp_port_state(ofport);
2011 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2012 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2018 stp_wait(struct ofproto_dpif *ofproto)
2021 poll_timer_wait(1000);
2025 /* Returns true if STP should process 'flow'. */
2027 stp_should_process_flow(const struct flow *flow)
2029 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2033 stp_process_packet(const struct ofport_dpif *ofport,
2034 const struct ofpbuf *packet)
2036 struct ofpbuf payload = *packet;
2037 struct eth_header *eth = payload.data;
2038 struct stp_port *sp = ofport->stp_port;
2040 /* Sink packets on ports that have STP disabled when the bridge has
2042 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2046 /* Trim off padding on payload. */
2047 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2048 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2051 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2052 stp_received_bpdu(sp, payload.data, payload.size);
2056 static struct priority_to_dscp *
2057 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2059 struct priority_to_dscp *pdscp;
2062 hash = hash_int(priority, 0);
2063 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2064 if (pdscp->priority == priority) {
2072 ofport_clear_priorities(struct ofport_dpif *ofport)
2074 struct priority_to_dscp *pdscp, *next;
2076 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2077 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2083 set_queues(struct ofport *ofport_,
2084 const struct ofproto_port_queue *qdscp_list,
2087 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2088 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2089 struct hmap new = HMAP_INITIALIZER(&new);
2092 for (i = 0; i < n_qdscp; i++) {
2093 struct priority_to_dscp *pdscp;
2097 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2098 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2103 pdscp = get_priority(ofport, priority);
2105 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2107 pdscp = xmalloc(sizeof *pdscp);
2108 pdscp->priority = priority;
2110 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2113 if (pdscp->dscp != dscp) {
2115 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2118 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2121 if (!hmap_is_empty(&ofport->priorities)) {
2122 ofport_clear_priorities(ofport);
2123 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2126 hmap_swap(&new, &ofport->priorities);
2134 /* Expires all MAC learning entries associated with 'bundle' and forces its
2135 * ofproto to revalidate every flow.
2137 * Normally MAC learning entries are removed only from the ofproto associated
2138 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2139 * are removed from every ofproto. When patch ports and SLB bonds are in use
2140 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2141 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2142 * with the host from which it migrated. */
2144 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2146 struct ofproto_dpif *ofproto = bundle->ofproto;
2147 struct mac_learning *ml = ofproto->ml;
2148 struct mac_entry *mac, *next_mac;
2150 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2151 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2152 if (mac->port.p == bundle) {
2154 struct ofproto_dpif *o;
2156 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2158 struct mac_entry *e;
2160 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2163 mac_learning_expire(o->ml, e);
2169 mac_learning_expire(ml, mac);
2174 static struct ofbundle *
2175 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2177 struct ofbundle *bundle;
2179 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2180 &ofproto->bundles) {
2181 if (bundle->aux == aux) {
2188 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2189 * ones that are found to 'bundles'. */
2191 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2192 void **auxes, size_t n_auxes,
2193 struct hmapx *bundles)
2197 hmapx_init(bundles);
2198 for (i = 0; i < n_auxes; i++) {
2199 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2201 hmapx_add(bundles, bundle);
2207 bundle_update(struct ofbundle *bundle)
2209 struct ofport_dpif *port;
2211 bundle->floodable = true;
2212 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2213 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2214 || !stp_forward_in_state(port->stp_state)) {
2215 bundle->floodable = false;
2222 bundle_del_port(struct ofport_dpif *port)
2224 struct ofbundle *bundle = port->bundle;
2226 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2228 list_remove(&port->bundle_node);
2229 port->bundle = NULL;
2232 lacp_slave_unregister(bundle->lacp, port);
2235 bond_slave_unregister(bundle->bond, port);
2238 bundle_update(bundle);
2242 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2243 struct lacp_slave_settings *lacp)
2245 struct ofport_dpif *port;
2247 port = get_ofp_port(bundle->ofproto, ofp_port);
2252 if (port->bundle != bundle) {
2253 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2255 bundle_del_port(port);
2258 port->bundle = bundle;
2259 list_push_back(&bundle->ports, &port->bundle_node);
2260 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2261 || !stp_forward_in_state(port->stp_state)) {
2262 bundle->floodable = false;
2266 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2267 lacp_slave_register(bundle->lacp, port, lacp);
2274 bundle_destroy(struct ofbundle *bundle)
2276 struct ofproto_dpif *ofproto;
2277 struct ofport_dpif *port, *next_port;
2284 ofproto = bundle->ofproto;
2285 for (i = 0; i < MAX_MIRRORS; i++) {
2286 struct ofmirror *m = ofproto->mirrors[i];
2288 if (m->out == bundle) {
2290 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2291 || hmapx_find_and_delete(&m->dsts, bundle)) {
2292 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2297 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2298 bundle_del_port(port);
2301 bundle_flush_macs(bundle, true);
2302 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2304 free(bundle->trunks);
2305 lacp_destroy(bundle->lacp);
2306 bond_destroy(bundle->bond);
2311 bundle_set(struct ofproto *ofproto_, void *aux,
2312 const struct ofproto_bundle_settings *s)
2314 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2315 bool need_flush = false;
2316 struct ofport_dpif *port;
2317 struct ofbundle *bundle;
2318 unsigned long *trunks;
2324 bundle_destroy(bundle_lookup(ofproto, aux));
2328 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2329 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2331 bundle = bundle_lookup(ofproto, aux);
2333 bundle = xmalloc(sizeof *bundle);
2335 bundle->ofproto = ofproto;
2336 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2337 hash_pointer(aux, 0));
2339 bundle->name = NULL;
2341 list_init(&bundle->ports);
2342 bundle->vlan_mode = PORT_VLAN_TRUNK;
2344 bundle->trunks = NULL;
2345 bundle->use_priority_tags = s->use_priority_tags;
2346 bundle->lacp = NULL;
2347 bundle->bond = NULL;
2349 bundle->floodable = true;
2351 bundle->src_mirrors = 0;
2352 bundle->dst_mirrors = 0;
2353 bundle->mirror_out = 0;
2356 if (!bundle->name || strcmp(s->name, bundle->name)) {
2358 bundle->name = xstrdup(s->name);
2363 if (!bundle->lacp) {
2364 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2365 bundle->lacp = lacp_create();
2367 lacp_configure(bundle->lacp, s->lacp);
2369 lacp_destroy(bundle->lacp);
2370 bundle->lacp = NULL;
2373 /* Update set of ports. */
2375 for (i = 0; i < s->n_slaves; i++) {
2376 if (!bundle_add_port(bundle, s->slaves[i],
2377 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2381 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2382 struct ofport_dpif *next_port;
2384 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2385 for (i = 0; i < s->n_slaves; i++) {
2386 if (s->slaves[i] == port->up.ofp_port) {
2391 bundle_del_port(port);
2395 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2397 if (list_is_empty(&bundle->ports)) {
2398 bundle_destroy(bundle);
2402 /* Set VLAN tagging mode */
2403 if (s->vlan_mode != bundle->vlan_mode
2404 || s->use_priority_tags != bundle->use_priority_tags) {
2405 bundle->vlan_mode = s->vlan_mode;
2406 bundle->use_priority_tags = s->use_priority_tags;
2411 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2412 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2414 if (vlan != bundle->vlan) {
2415 bundle->vlan = vlan;
2419 /* Get trunked VLANs. */
2420 switch (s->vlan_mode) {
2421 case PORT_VLAN_ACCESS:
2425 case PORT_VLAN_TRUNK:
2426 trunks = CONST_CAST(unsigned long *, s->trunks);
2429 case PORT_VLAN_NATIVE_UNTAGGED:
2430 case PORT_VLAN_NATIVE_TAGGED:
2431 if (vlan != 0 && (!s->trunks
2432 || !bitmap_is_set(s->trunks, vlan)
2433 || bitmap_is_set(s->trunks, 0))) {
2434 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2436 trunks = bitmap_clone(s->trunks, 4096);
2438 trunks = bitmap_allocate1(4096);
2440 bitmap_set1(trunks, vlan);
2441 bitmap_set0(trunks, 0);
2443 trunks = CONST_CAST(unsigned long *, s->trunks);
2450 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2451 free(bundle->trunks);
2452 if (trunks == s->trunks) {
2453 bundle->trunks = vlan_bitmap_clone(trunks);
2455 bundle->trunks = trunks;
2460 if (trunks != s->trunks) {
2465 if (!list_is_short(&bundle->ports)) {
2466 bundle->ofproto->has_bonded_bundles = true;
2468 if (bond_reconfigure(bundle->bond, s->bond)) {
2469 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2472 bundle->bond = bond_create(s->bond);
2473 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2476 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2477 bond_slave_register(bundle->bond, port, port->up.netdev);
2480 bond_destroy(bundle->bond);
2481 bundle->bond = NULL;
2484 /* If we changed something that would affect MAC learning, un-learn
2485 * everything on this port and force flow revalidation. */
2487 bundle_flush_macs(bundle, false);
2494 bundle_remove(struct ofport *port_)
2496 struct ofport_dpif *port = ofport_dpif_cast(port_);
2497 struct ofbundle *bundle = port->bundle;
2500 bundle_del_port(port);
2501 if (list_is_empty(&bundle->ports)) {
2502 bundle_destroy(bundle);
2503 } else if (list_is_short(&bundle->ports)) {
2504 bond_destroy(bundle->bond);
2505 bundle->bond = NULL;
2511 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2513 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2514 struct ofport_dpif *port = port_;
2515 uint8_t ea[ETH_ADDR_LEN];
2518 error = netdev_get_etheraddr(port->up.netdev, ea);
2520 struct ofpbuf packet;
2523 ofpbuf_init(&packet, 0);
2524 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2526 memcpy(packet_pdu, pdu, pdu_size);
2528 send_packet(port, &packet);
2529 ofpbuf_uninit(&packet);
2531 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2532 "%s (%s)", port->bundle->name,
2533 netdev_get_name(port->up.netdev), strerror(error));
2538 bundle_send_learning_packets(struct ofbundle *bundle)
2540 struct ofproto_dpif *ofproto = bundle->ofproto;
2541 int error, n_packets, n_errors;
2542 struct mac_entry *e;
2544 error = n_packets = n_errors = 0;
2545 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2546 if (e->port.p != bundle) {
2547 struct ofpbuf *learning_packet;
2548 struct ofport_dpif *port;
2552 /* The assignment to "port" is unnecessary but makes "grep"ing for
2553 * struct ofport_dpif more effective. */
2554 learning_packet = bond_compose_learning_packet(bundle->bond,
2558 ret = send_packet(port, learning_packet);
2559 ofpbuf_delete(learning_packet);
2569 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2570 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2571 "packets, last error was: %s",
2572 bundle->name, n_errors, n_packets, strerror(error));
2574 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2575 bundle->name, n_packets);
2580 bundle_run(struct ofbundle *bundle)
2583 lacp_run(bundle->lacp, send_pdu_cb);
2586 struct ofport_dpif *port;
2588 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2589 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2592 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2593 lacp_status(bundle->lacp));
2594 if (bond_should_send_learning_packets(bundle->bond)) {
2595 bundle_send_learning_packets(bundle);
2601 bundle_wait(struct ofbundle *bundle)
2604 lacp_wait(bundle->lacp);
2607 bond_wait(bundle->bond);
2614 mirror_scan(struct ofproto_dpif *ofproto)
2618 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2619 if (!ofproto->mirrors[idx]) {
2626 static struct ofmirror *
2627 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2631 for (i = 0; i < MAX_MIRRORS; i++) {
2632 struct ofmirror *mirror = ofproto->mirrors[i];
2633 if (mirror && mirror->aux == aux) {
2641 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2643 mirror_update_dups(struct ofproto_dpif *ofproto)
2647 for (i = 0; i < MAX_MIRRORS; i++) {
2648 struct ofmirror *m = ofproto->mirrors[i];
2651 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2655 for (i = 0; i < MAX_MIRRORS; i++) {
2656 struct ofmirror *m1 = ofproto->mirrors[i];
2663 for (j = i + 1; j < MAX_MIRRORS; j++) {
2664 struct ofmirror *m2 = ofproto->mirrors[j];
2666 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2667 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2668 m2->dup_mirrors |= m1->dup_mirrors;
2675 mirror_set(struct ofproto *ofproto_, void *aux,
2676 const struct ofproto_mirror_settings *s)
2678 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2679 mirror_mask_t mirror_bit;
2680 struct ofbundle *bundle;
2681 struct ofmirror *mirror;
2682 struct ofbundle *out;
2683 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2684 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2687 mirror = mirror_lookup(ofproto, aux);
2689 mirror_destroy(mirror);
2695 idx = mirror_scan(ofproto);
2697 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2699 ofproto->up.name, MAX_MIRRORS, s->name);
2703 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2704 mirror->ofproto = ofproto;
2707 mirror->out_vlan = -1;
2708 mirror->name = NULL;
2711 if (!mirror->name || strcmp(s->name, mirror->name)) {
2713 mirror->name = xstrdup(s->name);
2716 /* Get the new configuration. */
2717 if (s->out_bundle) {
2718 out = bundle_lookup(ofproto, s->out_bundle);
2720 mirror_destroy(mirror);
2726 out_vlan = s->out_vlan;
2728 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2729 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2731 /* If the configuration has not changed, do nothing. */
2732 if (hmapx_equals(&srcs, &mirror->srcs)
2733 && hmapx_equals(&dsts, &mirror->dsts)
2734 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2735 && mirror->out == out
2736 && mirror->out_vlan == out_vlan)
2738 hmapx_destroy(&srcs);
2739 hmapx_destroy(&dsts);
2743 hmapx_swap(&srcs, &mirror->srcs);
2744 hmapx_destroy(&srcs);
2746 hmapx_swap(&dsts, &mirror->dsts);
2747 hmapx_destroy(&dsts);
2749 free(mirror->vlans);
2750 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2753 mirror->out_vlan = out_vlan;
2755 /* Update bundles. */
2756 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2757 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2758 if (hmapx_contains(&mirror->srcs, bundle)) {
2759 bundle->src_mirrors |= mirror_bit;
2761 bundle->src_mirrors &= ~mirror_bit;
2764 if (hmapx_contains(&mirror->dsts, bundle)) {
2765 bundle->dst_mirrors |= mirror_bit;
2767 bundle->dst_mirrors &= ~mirror_bit;
2770 if (mirror->out == bundle) {
2771 bundle->mirror_out |= mirror_bit;
2773 bundle->mirror_out &= ~mirror_bit;
2777 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2778 ofproto->has_mirrors = true;
2779 mac_learning_flush(ofproto->ml,
2780 &ofproto->backer->revalidate_set);
2781 mirror_update_dups(ofproto);
2787 mirror_destroy(struct ofmirror *mirror)
2789 struct ofproto_dpif *ofproto;
2790 mirror_mask_t mirror_bit;
2791 struct ofbundle *bundle;
2798 ofproto = mirror->ofproto;
2799 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2800 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2802 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2803 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2804 bundle->src_mirrors &= ~mirror_bit;
2805 bundle->dst_mirrors &= ~mirror_bit;
2806 bundle->mirror_out &= ~mirror_bit;
2809 hmapx_destroy(&mirror->srcs);
2810 hmapx_destroy(&mirror->dsts);
2811 free(mirror->vlans);
2813 ofproto->mirrors[mirror->idx] = NULL;
2817 mirror_update_dups(ofproto);
2819 ofproto->has_mirrors = false;
2820 for (i = 0; i < MAX_MIRRORS; i++) {
2821 if (ofproto->mirrors[i]) {
2822 ofproto->has_mirrors = true;
2829 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2830 uint64_t *packets, uint64_t *bytes)
2832 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2833 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2836 *packets = *bytes = UINT64_MAX;
2840 *packets = mirror->packet_count;
2841 *bytes = mirror->byte_count;
2847 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2849 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2850 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2851 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2857 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2859 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2860 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2861 return bundle && bundle->mirror_out != 0;
2865 forward_bpdu_changed(struct ofproto *ofproto_)
2867 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2868 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2872 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2875 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2876 mac_learning_set_idle_time(ofproto->ml, idle_time);
2877 mac_learning_set_max_entries(ofproto->ml, max_entries);
2882 static struct ofport_dpif *
2883 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2885 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2886 return ofport ? ofport_dpif_cast(ofport) : NULL;
2889 static struct ofport_dpif *
2890 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2892 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2893 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2897 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2898 struct ofproto_port *ofproto_port,
2899 struct dpif_port *dpif_port)
2901 ofproto_port->name = dpif_port->name;
2902 ofproto_port->type = dpif_port->type;
2903 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2906 static struct ofport_dpif *
2907 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
2909 const struct ofproto_dpif *ofproto;
2912 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
2917 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2918 struct ofport *ofport;
2920 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
2921 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
2922 return ofport_dpif_cast(ofport);
2929 port_run_fast(struct ofport_dpif *ofport)
2931 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2932 struct ofpbuf packet;
2934 ofpbuf_init(&packet, 0);
2935 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2936 send_packet(ofport, &packet);
2937 ofpbuf_uninit(&packet);
2942 port_run(struct ofport_dpif *ofport)
2944 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2945 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2946 bool enable = netdev_get_carrier(ofport->up.netdev);
2948 ofport->carrier_seq = carrier_seq;
2950 port_run_fast(ofport);
2952 if (ofport->tnl_port
2953 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
2954 &ofport->tnl_port)) {
2955 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
2959 int cfm_opup = cfm_get_opup(ofport->cfm);
2961 cfm_run(ofport->cfm);
2962 enable = enable && !cfm_get_fault(ofport->cfm);
2964 if (cfm_opup >= 0) {
2965 enable = enable && cfm_opup;
2969 if (ofport->bundle) {
2970 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2971 if (carrier_changed) {
2972 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2976 if (ofport->may_enable != enable) {
2977 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2979 if (ofproto->has_bundle_action) {
2980 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2984 ofport->may_enable = enable;
2988 port_wait(struct ofport_dpif *ofport)
2991 cfm_wait(ofport->cfm);
2996 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2997 struct ofproto_port *ofproto_port)
2999 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3000 struct dpif_port dpif_port;
3003 if (sset_contains(&ofproto->ghost_ports, devname)) {
3004 const char *type = netdev_get_type_from_name(devname);
3006 /* We may be called before ofproto->up.port_by_name is populated with
3007 * the appropriate ofport. For this reason, we must get the name and
3008 * type from the netdev layer directly. */
3010 const struct ofport *ofport;
3012 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3013 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3014 ofproto_port->name = xstrdup(devname);
3015 ofproto_port->type = xstrdup(type);
3021 if (!sset_contains(&ofproto->ports, devname)) {
3024 error = dpif_port_query_by_name(ofproto->backer->dpif,
3025 devname, &dpif_port);
3027 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3033 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3035 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3036 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3037 const char *devname = netdev_get_name(netdev);
3039 if (netdev_vport_is_patch(netdev)) {
3040 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3044 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3045 uint32_t port_no = UINT32_MAX;
3048 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3052 if (netdev_get_tunnel_config(netdev)) {
3053 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3057 if (netdev_get_tunnel_config(netdev)) {
3058 sset_add(&ofproto->ghost_ports, devname);
3060 sset_add(&ofproto->ports, devname);
3066 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3068 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3069 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3076 sset_find_and_delete(&ofproto->ghost_ports,
3077 netdev_get_name(ofport->up.netdev));
3078 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3079 if (!ofport->tnl_port) {
3080 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3082 /* The caller is going to close ofport->up.netdev. If this is a
3083 * bonded port, then the bond is using that netdev, so remove it
3084 * from the bond. The client will need to reconfigure everything
3085 * after deleting ports, so then the slave will get re-added. */
3086 bundle_remove(&ofport->up);
3093 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3095 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3098 error = netdev_get_stats(ofport->up.netdev, stats);
3100 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3101 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3103 /* ofproto->stats.tx_packets represents packets that we created
3104 * internally and sent to some port (e.g. packets sent with
3105 * send_packet()). Account for them as if they had come from
3106 * OFPP_LOCAL and got forwarded. */
3108 if (stats->rx_packets != UINT64_MAX) {
3109 stats->rx_packets += ofproto->stats.tx_packets;
3112 if (stats->rx_bytes != UINT64_MAX) {
3113 stats->rx_bytes += ofproto->stats.tx_bytes;
3116 /* ofproto->stats.rx_packets represents packets that were received on
3117 * some port and we processed internally and dropped (e.g. STP).
3118 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3120 if (stats->tx_packets != UINT64_MAX) {
3121 stats->tx_packets += ofproto->stats.rx_packets;
3124 if (stats->tx_bytes != UINT64_MAX) {
3125 stats->tx_bytes += ofproto->stats.rx_bytes;
3132 /* Account packets for LOCAL port. */
3134 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
3135 size_t tx_size, size_t rx_size)
3137 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3140 ofproto->stats.rx_packets++;
3141 ofproto->stats.rx_bytes += rx_size;
3144 ofproto->stats.tx_packets++;
3145 ofproto->stats.tx_bytes += tx_size;
3149 struct port_dump_state {
3154 struct ofproto_port port;
3159 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3161 *statep = xzalloc(sizeof(struct port_dump_state));
3166 port_dump_next(const struct ofproto *ofproto_, void *state_,
3167 struct ofproto_port *port)
3169 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3170 struct port_dump_state *state = state_;
3171 const struct sset *sset;
3172 struct sset_node *node;
3174 if (state->has_port) {
3175 ofproto_port_destroy(&state->port);
3176 state->has_port = false;
3178 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3179 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3182 error = port_query_by_name(ofproto_, node->name, &state->port);
3184 *port = state->port;
3185 state->has_port = true;
3187 } else if (error != ENODEV) {
3192 if (!state->ghost) {
3193 state->ghost = true;
3196 return port_dump_next(ofproto_, state_, port);
3203 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3205 struct port_dump_state *state = state_;
3207 if (state->has_port) {
3208 ofproto_port_destroy(&state->port);
3215 port_poll(const struct ofproto *ofproto_, char **devnamep)
3217 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3219 if (ofproto->port_poll_errno) {
3220 int error = ofproto->port_poll_errno;
3221 ofproto->port_poll_errno = 0;
3225 if (sset_is_empty(&ofproto->port_poll_set)) {
3229 *devnamep = sset_pop(&ofproto->port_poll_set);
3234 port_poll_wait(const struct ofproto *ofproto_)
3236 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3237 dpif_port_poll_wait(ofproto->backer->dpif);
3241 port_is_lacp_current(const struct ofport *ofport_)
3243 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3244 return (ofport->bundle && ofport->bundle->lacp
3245 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3249 /* Upcall handling. */
3251 /* Flow miss batching.
3253 * Some dpifs implement operations faster when you hand them off in a batch.
3254 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3255 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3256 * more packets, plus possibly installing the flow in the dpif.
3258 * So far we only batch the operations that affect flow setup time the most.
3259 * It's possible to batch more than that, but the benefit might be minimal. */
3261 struct hmap_node hmap_node;
3262 struct ofproto_dpif *ofproto;
3264 enum odp_key_fitness key_fitness;
3265 const struct nlattr *key;
3267 ovs_be16 initial_tci;
3268 struct list packets;
3269 enum dpif_upcall_type upcall_type;
3270 uint32_t odp_in_port;
3273 struct flow_miss_op {
3274 struct dpif_op dpif_op;
3275 void *garbage; /* Pointer to pass to free(), NULL if none. */
3276 uint64_t stub[1024 / 8]; /* Temporary buffer. */
3279 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3280 * OpenFlow controller as necessary according to their individual
3281 * configurations. */
3283 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3284 const struct flow *flow)
3286 struct ofputil_packet_in pin;
3288 pin.packet = packet->data;
3289 pin.packet_len = packet->size;
3290 pin.reason = OFPR_NO_MATCH;
3291 pin.controller_id = 0;
3296 pin.send_len = 0; /* not used for flow table misses */
3298 flow_get_metadata(flow, &pin.fmd);
3300 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3303 static enum slow_path_reason
3304 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3305 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3309 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3311 cfm_process_heartbeat(ofport->cfm, packet);
3314 } else if (ofport->bundle && ofport->bundle->lacp
3315 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3317 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3320 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3322 stp_process_packet(ofport, packet);
3330 static struct flow_miss *
3331 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3332 const struct flow *flow, uint32_t hash)
3334 struct flow_miss *miss;
3336 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3337 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3345 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3346 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3347 * 'miss' is associated with a subfacet the caller must also initialize the
3348 * returned op->subfacet, and if anything needs to be freed after processing
3349 * the op, the caller must initialize op->garbage also. */
3351 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3352 struct flow_miss_op *op)
3354 if (miss->flow.vlan_tci != miss->initial_tci) {
3355 /* This packet was received on a VLAN splinter port. We
3356 * added a VLAN to the packet to make the packet resemble
3357 * the flow, but the actions were composed assuming that
3358 * the packet contained no VLAN. So, we must remove the
3359 * VLAN header from the packet before trying to execute the
3361 eth_pop_vlan(packet);
3365 op->dpif_op.type = DPIF_OP_EXECUTE;
3366 op->dpif_op.u.execute.key = miss->key;
3367 op->dpif_op.u.execute.key_len = miss->key_len;
3368 op->dpif_op.u.execute.packet = packet;
3371 /* Helper for handle_flow_miss_without_facet() and
3372 * handle_flow_miss_with_facet(). */
3374 handle_flow_miss_common(struct rule_dpif *rule,
3375 struct ofpbuf *packet, const struct flow *flow)
3377 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3379 ofproto->n_matches++;
3381 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3383 * Extra-special case for fail-open mode.
3385 * We are in fail-open mode and the packet matched the fail-open
3386 * rule, but we are connected to a controller too. We should send
3387 * the packet up to the controller in the hope that it will try to
3388 * set up a flow and thereby allow us to exit fail-open.
3390 * See the top-level comment in fail-open.c for more information.
3392 send_packet_in_miss(ofproto, packet, flow);
3396 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3397 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3398 * installing a datapath flow. The answer is usually "yes" (a return value of
3399 * true). However, for short flows the cost of bookkeeping is much higher than
3400 * the benefits, so when the datapath holds a large number of flows we impose
3401 * some heuristics to decide which flows are likely to be worth tracking. */
3403 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3404 struct flow_miss *miss, uint32_t hash)
3406 if (!ofproto->governor) {
3409 n_subfacets = hmap_count(&ofproto->subfacets);
3410 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3414 ofproto->governor = governor_create(ofproto->up.name);
3417 return governor_should_install_flow(ofproto->governor, hash,
3418 list_size(&miss->packets));
3421 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3422 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3423 * increment '*n_ops'. */
3425 handle_flow_miss_without_facet(struct flow_miss *miss,
3426 struct rule_dpif *rule,
3427 struct flow_miss_op *ops, size_t *n_ops)
3429 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3430 long long int now = time_msec();
3431 struct action_xlate_ctx ctx;
3432 struct ofpbuf *packet;
3434 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3435 struct flow_miss_op *op = &ops[*n_ops];
3436 struct dpif_flow_stats stats;
3437 struct ofpbuf odp_actions;
3439 COVERAGE_INC(facet_suppress);
3441 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3443 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3444 rule_credit_stats(rule, &stats);
3446 action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci,
3448 ctx.resubmit_stats = &stats;
3449 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3452 if (odp_actions.size) {
3453 struct dpif_execute *execute = &op->dpif_op.u.execute;
3455 init_flow_miss_execute_op(miss, packet, op);
3456 execute->actions = odp_actions.data;
3457 execute->actions_len = odp_actions.size;
3458 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3462 ofpbuf_uninit(&odp_actions);
3467 /* Handles 'miss', which matches 'facet'. May add any required datapath
3468 * operations to 'ops', incrementing '*n_ops' for each new op.
3470 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3471 * This is really important only for new facets: if we just called time_msec()
3472 * here, then the new subfacet or its packets could look (occasionally) as
3473 * though it was used some time after the facet was used. That can make a
3474 * one-packet flow look like it has a nonzero duration, which looks odd in
3475 * e.g. NetFlow statistics. */
3477 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3479 struct flow_miss_op *ops, size_t *n_ops)
3481 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3482 enum subfacet_path want_path;
3483 struct subfacet *subfacet;
3484 struct ofpbuf *packet;
3486 subfacet = subfacet_create(facet, miss, now);
3488 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3489 struct flow_miss_op *op = &ops[*n_ops];
3490 struct dpif_flow_stats stats;
3491 struct ofpbuf odp_actions;
3493 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3495 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3496 if (!subfacet->actions || subfacet->slow) {
3497 subfacet_make_actions(subfacet, packet, &odp_actions);
3500 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3501 subfacet_update_stats(subfacet, &stats);
3503 if (subfacet->actions_len) {
3504 struct dpif_execute *execute = &op->dpif_op.u.execute;
3506 init_flow_miss_execute_op(miss, packet, op);
3507 if (!subfacet->slow) {
3508 execute->actions = subfacet->actions;
3509 execute->actions_len = subfacet->actions_len;
3510 ofpbuf_uninit(&odp_actions);
3512 execute->actions = odp_actions.data;
3513 execute->actions_len = odp_actions.size;
3514 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3519 ofpbuf_uninit(&odp_actions);
3523 want_path = subfacet_want_path(subfacet->slow);
3524 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3525 struct flow_miss_op *op = &ops[(*n_ops)++];
3526 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3528 subfacet->path = want_path;
3531 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3532 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3533 put->key = miss->key;
3534 put->key_len = miss->key_len;
3535 if (want_path == SF_FAST_PATH) {
3536 put->actions = subfacet->actions;
3537 put->actions_len = subfacet->actions_len;
3539 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
3540 op->stub, sizeof op->stub,
3541 &put->actions, &put->actions_len);
3547 /* Handles flow miss 'miss'. May add any required datapath operations
3548 * to 'ops', incrementing '*n_ops' for each new op. */
3550 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3553 struct ofproto_dpif *ofproto = miss->ofproto;
3554 struct facet *facet;
3558 /* The caller must ensure that miss->hmap_node.hash contains
3559 * flow_hash(miss->flow, 0). */
3560 hash = miss->hmap_node.hash;
3562 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3564 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
3566 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
3567 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
3571 facet = facet_create(rule, &miss->flow, hash);
3576 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3579 static struct drop_key *
3580 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3583 struct drop_key *drop_key;
3585 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3586 &backer->drop_keys) {
3587 if (drop_key->key_len == key_len
3588 && !memcmp(drop_key->key, key, key_len)) {
3596 drop_key_clear(struct dpif_backer *backer)
3598 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3599 struct drop_key *drop_key, *next;
3601 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3604 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3606 if (error && !VLOG_DROP_WARN(&rl)) {
3607 struct ds ds = DS_EMPTY_INITIALIZER;
3608 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3609 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3614 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3615 free(drop_key->key);
3620 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3621 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3622 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3623 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3624 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3625 * 'packet' ingressed.
3627 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3628 * 'flow''s in_port to OFPP_NONE.
3630 * This function does post-processing on data returned from
3631 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3632 * of the upcall processing logic. In particular, if the extracted in_port is
3633 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3634 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3635 * a VLAN header onto 'packet' (if it is nonnull).
3637 * Optionally, if nonnull, sets '*initial_tci' to the VLAN TCI with which the
3638 * packet was really received, that is, the actual VLAN TCI extracted by
3639 * odp_flow_key_to_flow(). (This differs from the value returned in
3640 * flow->vlan_tci only for packets received on VLAN splinters.)
3642 * Similarly, this function also includes some logic to help with tunnels. It
3643 * may modify 'flow' as necessary to make the tunneling implementation
3644 * transparent to the upcall processing logic.
3646 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3647 * or some other positive errno if there are other problems. */
3649 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3650 const struct nlattr *key, size_t key_len,
3651 struct flow *flow, enum odp_key_fitness *fitnessp,
3652 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3653 ovs_be16 *initial_tci)
3655 const struct ofport_dpif *port;
3656 enum odp_key_fitness fitness;
3659 fitness = odp_flow_key_to_flow(key, key_len, flow);
3660 if (fitness == ODP_FIT_ERROR) {
3666 *initial_tci = flow->vlan_tci;
3670 *odp_in_port = flow->in_port;
3673 if (tnl_port_should_receive(flow)) {
3674 const struct ofport *ofport = tnl_port_receive(flow);
3676 flow->in_port = OFPP_NONE;
3679 port = ofport_dpif_cast(ofport);
3681 /* We can't reproduce 'key' from 'flow'. */
3682 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3684 /* XXX: Since the tunnel module is not scoped per backer, it's
3685 * theoretically possible that we'll receive an ofport belonging to an
3686 * entirely different datapath. In practice, this can't happen because
3687 * no platforms has two separate datapaths which each support
3689 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3691 port = odp_port_to_ofport(backer, flow->in_port);
3693 flow->in_port = OFPP_NONE;
3697 flow->in_port = port->up.ofp_port;
3698 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3700 /* Make the packet resemble the flow, so that it gets sent to
3701 * an OpenFlow controller properly, so that it looks correct
3702 * for sFlow, and so that flow_extract() will get the correct
3703 * vlan_tci if it is called on 'packet'.
3705 * The allocated space inside 'packet' probably also contains
3706 * 'key', that is, both 'packet' and 'key' are probably part of
3707 * a struct dpif_upcall (see the large comment on that
3708 * structure definition), so pushing data on 'packet' is in
3709 * general not a good idea since it could overwrite 'key' or
3710 * free it as a side effect. However, it's OK in this special
3711 * case because we know that 'packet' is inside a Netlink
3712 * attribute: pushing 4 bytes will just overwrite the 4-byte
3713 * "struct nlattr", which is fine since we don't need that
3714 * header anymore. */
3715 eth_push_vlan(packet, flow->vlan_tci);
3717 /* We can't reproduce 'key' from 'flow'. */
3718 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3724 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3729 *fitnessp = fitness;
3735 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3738 struct dpif_upcall *upcall;
3739 struct flow_miss *miss;
3740 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3741 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3742 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3752 /* Construct the to-do list.
3754 * This just amounts to extracting the flow from each packet and sticking
3755 * the packets that have the same flow in the same "flow_miss" structure so
3756 * that we can process them together. */
3759 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3760 struct flow_miss *miss = &misses[n_misses];
3761 struct flow_miss *existing_miss;
3762 struct ofproto_dpif *ofproto;
3763 uint32_t odp_in_port;
3768 error = ofproto_receive(backer, upcall->packet, upcall->key,
3769 upcall->key_len, &flow, &miss->key_fitness,
3770 &ofproto, &odp_in_port, &miss->initial_tci);
3771 if (error == ENODEV) {
3772 struct drop_key *drop_key;
3774 /* Received packet on port for which we couldn't associate
3775 * an ofproto. This can happen if a port is removed while
3776 * traffic is being received. Print a rate-limited message
3777 * in case it happens frequently. Install a drop flow so
3778 * that future packets of the flow are inexpensively dropped
3780 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
3783 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3785 drop_key = xmalloc(sizeof *drop_key);
3786 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3787 drop_key->key_len = upcall->key_len;
3789 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3790 hash_bytes(drop_key->key, drop_key->key_len, 0));
3791 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3792 drop_key->key, drop_key->key_len, NULL, 0, NULL);
3799 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3800 &flow.tunnel, flow.in_port, &miss->flow);
3802 /* Add other packets to a to-do list. */
3803 hash = flow_hash(&miss->flow, 0);
3804 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
3805 if (!existing_miss) {
3806 hmap_insert(&todo, &miss->hmap_node, hash);
3807 miss->ofproto = ofproto;
3808 miss->key = upcall->key;
3809 miss->key_len = upcall->key_len;
3810 miss->upcall_type = upcall->type;
3811 miss->odp_in_port = odp_in_port;
3812 list_init(&miss->packets);
3816 miss = existing_miss;
3818 list_push_back(&miss->packets, &upcall->packet->list_node);
3821 /* Process each element in the to-do list, constructing the set of
3822 * operations to batch. */
3824 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3825 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3827 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3829 /* Execute batch. */
3830 for (i = 0; i < n_ops; i++) {
3831 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3833 dpif_operate(backer->dpif, dpif_ops, n_ops);
3836 for (i = 0; i < n_ops; i++) {
3837 free(flow_miss_ops[i].garbage);
3839 hmap_destroy(&todo);
3842 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
3843 classify_upcall(const struct dpif_upcall *upcall)
3845 union user_action_cookie cookie;
3847 /* First look at the upcall type. */
3848 switch (upcall->type) {
3849 case DPIF_UC_ACTION:
3855 case DPIF_N_UC_TYPES:
3857 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3861 /* "action" upcalls need a closer look. */
3862 if (!upcall->userdata) {
3863 VLOG_WARN_RL(&rl, "action upcall missing cookie");
3866 if (nl_attr_get_size(upcall->userdata) != sizeof(cookie)) {
3867 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
3868 nl_attr_get_size(upcall->userdata));
3871 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof(cookie));
3872 switch (cookie.type) {
3873 case USER_ACTION_COOKIE_SFLOW:
3874 return SFLOW_UPCALL;
3876 case USER_ACTION_COOKIE_SLOW_PATH:
3879 case USER_ACTION_COOKIE_UNSPEC:
3881 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64,
3882 nl_attr_get_u64(upcall->userdata));
3888 handle_sflow_upcall(struct dpif_backer *backer,
3889 const struct dpif_upcall *upcall)
3891 struct ofproto_dpif *ofproto;
3892 union user_action_cookie cookie;
3894 uint32_t odp_in_port;
3896 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3897 &flow, NULL, &ofproto, &odp_in_port, NULL)
3898 || !ofproto->sflow) {
3902 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof(cookie));
3903 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3904 odp_in_port, &cookie);
3908 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3910 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3911 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3912 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3917 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
3920 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3921 struct dpif_upcall *upcall = &misses[n_misses];
3922 struct ofpbuf *buf = &miss_bufs[n_misses];
3925 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3926 sizeof miss_buf_stubs[n_misses]);
3927 error = dpif_recv(backer->dpif, upcall, buf);
3933 switch (classify_upcall(upcall)) {
3935 /* Handle it later. */
3940 handle_sflow_upcall(backer, upcall);
3950 /* Handle deferred MISS_UPCALL processing. */
3951 handle_miss_upcalls(backer, misses, n_misses);
3952 for (i = 0; i < n_misses; i++) {
3953 ofpbuf_uninit(&miss_bufs[i]);
3959 /* Flow expiration. */
3961 static int subfacet_max_idle(const struct ofproto_dpif *);
3962 static void update_stats(struct dpif_backer *);
3963 static void rule_expire(struct rule_dpif *);
3964 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
3966 /* This function is called periodically by run(). Its job is to collect
3967 * updates for the flows that have been installed into the datapath, most
3968 * importantly when they last were used, and then use that information to
3969 * expire flows that have not been used recently.
3971 * Returns the number of milliseconds after which it should be called again. */
3973 expire(struct dpif_backer *backer)
3975 struct ofproto_dpif *ofproto;
3976 int max_idle = INT32_MAX;
3978 /* Periodically clear out the drop keys in an effort to keep them
3979 * relatively few. */
3980 drop_key_clear(backer);
3982 /* Update stats for each flow in the backer. */
3983 update_stats(backer);
3985 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3986 struct rule *rule, *next_rule;
3989 if (ofproto->backer != backer) {
3993 /* Expire subfacets that have been idle too long. */
3994 dp_max_idle = subfacet_max_idle(ofproto);
3995 expire_subfacets(ofproto, dp_max_idle);
3997 max_idle = MIN(max_idle, dp_max_idle);
3999 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4001 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4002 &ofproto->up.expirable) {
4003 rule_expire(rule_dpif_cast(rule));
4006 /* All outstanding data in existing flows has been accounted, so it's a
4007 * good time to do bond rebalancing. */
4008 if (ofproto->has_bonded_bundles) {
4009 struct ofbundle *bundle;
4011 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4013 bond_rebalance(bundle->bond, &backer->revalidate_set);
4019 return MIN(max_idle, 1000);
4022 /* Updates flow table statistics given that the datapath just reported 'stats'
4023 * as 'subfacet''s statistics. */
4025 update_subfacet_stats(struct subfacet *subfacet,
4026 const struct dpif_flow_stats *stats)
4028 struct facet *facet = subfacet->facet;
4030 if (stats->n_packets >= subfacet->dp_packet_count) {
4031 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
4032 facet->packet_count += extra;
4034 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4037 if (stats->n_bytes >= subfacet->dp_byte_count) {
4038 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
4040 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4043 subfacet->dp_packet_count = stats->n_packets;
4044 subfacet->dp_byte_count = stats->n_bytes;
4046 facet->tcp_flags |= stats->tcp_flags;
4048 subfacet_update_time(subfacet, stats->used);
4049 if (facet->accounted_bytes < facet->byte_count) {
4051 facet_account(facet);
4052 facet->accounted_bytes = facet->byte_count;
4054 facet_push_stats(facet);
4057 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4058 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4060 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4061 const struct nlattr *key, size_t key_len)
4063 if (!VLOG_DROP_WARN(&rl)) {
4067 odp_flow_key_format(key, key_len, &s);
4068 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4072 COVERAGE_INC(facet_unexpected);
4073 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4076 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4078 * This function also pushes statistics updates to rules which each facet
4079 * resubmits into. Generally these statistics will be accurate. However, if a
4080 * facet changes the rule it resubmits into at some time in between
4081 * update_stats() runs, it is possible that statistics accrued to the
4082 * old rule will be incorrectly attributed to the new rule. This could be
4083 * avoided by calling update_stats() whenever rules are created or
4084 * deleted. However, the performance impact of making so many calls to the
4085 * datapath do not justify the benefit of having perfectly accurate statistics.
4088 update_stats(struct dpif_backer *backer)
4090 const struct dpif_flow_stats *stats;
4091 struct dpif_flow_dump dump;
4092 const struct nlattr *key;
4095 dpif_flow_dump_start(&dump, backer->dpif);
4096 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4098 struct subfacet *subfacet;
4099 struct ofproto_dpif *ofproto;
4100 struct ofport_dpif *ofport;
4103 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4108 ofport = get_ofp_port(ofproto, flow.in_port);
4109 if (ofport && ofport->tnl_port) {
4110 netdev_vport_inc_rx(ofport->up.netdev, stats);
4113 key_hash = odp_flow_key_hash(key, key_len);
4114 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4115 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4117 update_subfacet_stats(subfacet, stats);
4121 /* Stats are updated per-packet. */
4124 case SF_NOT_INSTALLED:
4126 delete_unexpected_flow(ofproto, key, key_len);
4130 dpif_flow_dump_done(&dump);
4133 /* Calculates and returns the number of milliseconds of idle time after which
4134 * subfacets should expire from the datapath. When a subfacet expires, we fold
4135 * its statistics into its facet, and when a facet's last subfacet expires, we
4136 * fold its statistic into its rule. */
4138 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4141 * Idle time histogram.
4143 * Most of the time a switch has a relatively small number of subfacets.
4144 * When this is the case we might as well keep statistics for all of them
4145 * in userspace and to cache them in the kernel datapath for performance as
4148 * As the number of subfacets increases, the memory required to maintain
4149 * statistics about them in userspace and in the kernel becomes
4150 * significant. However, with a large number of subfacets it is likely
4151 * that only a few of them are "heavy hitters" that consume a large amount
4152 * of bandwidth. At this point, only heavy hitters are worth caching in
4153 * the kernel and maintaining in userspaces; other subfacets we can
4156 * The technique used to compute the idle time is to build a histogram with
4157 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4158 * that is installed in the kernel gets dropped in the appropriate bucket.
4159 * After the histogram has been built, we compute the cutoff so that only
4160 * the most-recently-used 1% of subfacets (but at least
4161 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4162 * the most-recently-used bucket of subfacets is kept, so actually an
4163 * arbitrary number of subfacets can be kept in any given expiration run
4164 * (though the next run will delete most of those unless they receive
4167 * This requires a second pass through the subfacets, in addition to the
4168 * pass made by update_stats(), because the former function never looks at
4169 * uninstallable subfacets.
4171 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4172 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4173 int buckets[N_BUCKETS] = { 0 };
4174 int total, subtotal, bucket;
4175 struct subfacet *subfacet;
4179 total = hmap_count(&ofproto->subfacets);
4180 if (total <= ofproto->up.flow_eviction_threshold) {
4181 return N_BUCKETS * BUCKET_WIDTH;
4184 /* Build histogram. */
4186 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4187 long long int idle = now - subfacet->used;
4188 int bucket = (idle <= 0 ? 0
4189 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4190 : (unsigned int) idle / BUCKET_WIDTH);
4194 /* Find the first bucket whose flows should be expired. */
4195 subtotal = bucket = 0;
4197 subtotal += buckets[bucket++];
4198 } while (bucket < N_BUCKETS &&
4199 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4201 if (VLOG_IS_DBG_ENABLED()) {
4205 ds_put_cstr(&s, "keep");
4206 for (i = 0; i < N_BUCKETS; i++) {
4208 ds_put_cstr(&s, ", drop");
4211 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4214 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4218 return bucket * BUCKET_WIDTH;
4222 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4224 /* Cutoff time for most flows. */
4225 long long int normal_cutoff = time_msec() - dp_max_idle;
4227 /* We really want to keep flows for special protocols around, so use a more
4228 * conservative cutoff. */
4229 long long int special_cutoff = time_msec() - 10000;
4231 struct subfacet *subfacet, *next_subfacet;
4232 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4236 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4237 &ofproto->subfacets) {
4238 long long int cutoff;
4240 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
4243 if (subfacet->used < cutoff) {
4244 if (subfacet->path != SF_NOT_INSTALLED) {
4245 batch[n_batch++] = subfacet;
4246 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4247 subfacet_destroy_batch(ofproto, batch, n_batch);
4251 subfacet_destroy(subfacet);
4257 subfacet_destroy_batch(ofproto, batch, n_batch);
4261 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4262 * then delete it entirely. */
4264 rule_expire(struct rule_dpif *rule)
4266 struct facet *facet, *next_facet;
4270 if (rule->up.pending) {
4271 /* We'll have to expire it later. */
4275 /* Has 'rule' expired? */
4277 if (rule->up.hard_timeout
4278 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4279 reason = OFPRR_HARD_TIMEOUT;
4280 } else if (rule->up.idle_timeout
4281 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4282 reason = OFPRR_IDLE_TIMEOUT;
4287 COVERAGE_INC(ofproto_dpif_expired);
4289 /* Update stats. (This is a no-op if the rule expired due to an idle
4290 * timeout, because that only happens when the rule has no facets left.) */
4291 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4292 facet_remove(facet);
4295 /* Get rid of the rule. */
4296 ofproto_rule_expire(&rule->up, reason);
4301 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
4303 * The caller must already have determined that no facet with an identical
4304 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
4305 * the ofproto's classifier table.
4307 * 'hash' must be the return value of flow_hash(flow, 0).
4309 * The facet will initially have no subfacets. The caller should create (at
4310 * least) one subfacet with subfacet_create(). */
4311 static struct facet *
4312 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
4314 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4315 struct facet *facet;
4317 facet = xzalloc(sizeof *facet);
4318 facet->used = time_msec();
4319 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4320 list_push_back(&rule->facets, &facet->list_node);
4322 facet->flow = *flow;
4323 list_init(&facet->subfacets);
4324 netflow_flow_init(&facet->nf_flow);
4325 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4331 facet_free(struct facet *facet)
4336 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4337 * 'packet', which arrived on 'in_port'. */
4339 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4340 const struct nlattr *odp_actions, size_t actions_len,
4341 struct ofpbuf *packet)
4343 struct odputil_keybuf keybuf;
4347 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4348 odp_flow_key_from_flow(&key, flow,
4349 ofp_port_to_odp_port(ofproto, flow->in_port));
4351 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4352 odp_actions, actions_len, packet);
4356 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4358 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4359 * rule's statistics, via subfacet_uninstall().
4361 * - Removes 'facet' from its rule and from ofproto->facets.
4364 facet_remove(struct facet *facet)
4366 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4367 struct subfacet *subfacet, *next_subfacet;
4369 ovs_assert(!list_is_empty(&facet->subfacets));
4371 /* First uninstall all of the subfacets to get final statistics. */
4372 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4373 subfacet_uninstall(subfacet);
4376 /* Flush the final stats to the rule.
4378 * This might require us to have at least one subfacet around so that we
4379 * can use its actions for accounting in facet_account(), which is why we
4380 * have uninstalled but not yet destroyed the subfacets. */
4381 facet_flush_stats(facet);
4383 /* Now we're really all done so destroy everything. */
4384 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4385 &facet->subfacets) {
4386 subfacet_destroy__(subfacet);
4388 hmap_remove(&ofproto->facets, &facet->hmap_node);
4389 list_remove(&facet->list_node);
4393 /* Feed information from 'facet' back into the learning table to keep it in
4394 * sync with what is actually flowing through the datapath. */
4396 facet_learn(struct facet *facet)
4398 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4399 struct action_xlate_ctx ctx;
4401 if (!facet->has_learn
4402 && !facet->has_normal
4403 && (!facet->has_fin_timeout
4404 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4408 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4409 facet->flow.vlan_tci,
4410 facet->rule, facet->tcp_flags, NULL);
4411 ctx.may_learn = true;
4412 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
4413 facet->rule->up.ofpacts_len);
4417 facet_account(struct facet *facet)
4419 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4420 struct subfacet *subfacet;
4421 const struct nlattr *a;
4426 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
4429 n_bytes = facet->byte_count - facet->accounted_bytes;
4431 /* This loop feeds byte counters to bond_account() for rebalancing to use
4432 * as a basis. We also need to track the actual VLAN on which the packet
4433 * is going to be sent to ensure that it matches the one passed to
4434 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4437 * We use the actions from an arbitrary subfacet because they should all
4438 * be equally valid for our purpose. */
4439 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
4440 struct subfacet, list_node);
4441 vlan_tci = facet->flow.vlan_tci;
4442 NL_ATTR_FOR_EACH_UNSAFE (a, left,
4443 subfacet->actions, subfacet->actions_len) {
4444 const struct ovs_action_push_vlan *vlan;
4445 struct ofport_dpif *port;
4447 switch (nl_attr_type(a)) {
4448 case OVS_ACTION_ATTR_OUTPUT:
4449 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4450 if (port && port->bundle && port->bundle->bond) {
4451 bond_account(port->bundle->bond, &facet->flow,
4452 vlan_tci_to_vid(vlan_tci), n_bytes);
4456 case OVS_ACTION_ATTR_POP_VLAN:
4457 vlan_tci = htons(0);
4460 case OVS_ACTION_ATTR_PUSH_VLAN:
4461 vlan = nl_attr_get(a);
4462 vlan_tci = vlan->vlan_tci;
4468 /* Returns true if the only action for 'facet' is to send to the controller.
4469 * (We don't report NetFlow expiration messages for such facets because they
4470 * are just part of the control logic for the network, not real traffic). */
4472 facet_is_controller_flow(struct facet *facet)
4475 const struct rule *rule = &facet->rule->up;
4476 const struct ofpact *ofpacts = rule->ofpacts;
4477 size_t ofpacts_len = rule->ofpacts_len;
4479 if (ofpacts_len > 0 &&
4480 ofpacts->type == OFPACT_CONTROLLER &&
4481 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4488 /* Folds all of 'facet''s statistics into its rule. Also updates the
4489 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4490 * 'facet''s statistics in the datapath should have been zeroed and folded into
4491 * its packet and byte counts before this function is called. */
4493 facet_flush_stats(struct facet *facet)
4495 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4496 struct subfacet *subfacet;
4498 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4499 ovs_assert(!subfacet->dp_byte_count);
4500 ovs_assert(!subfacet->dp_packet_count);
4503 facet_push_stats(facet);
4504 if (facet->accounted_bytes < facet->byte_count) {
4505 facet_account(facet);
4506 facet->accounted_bytes = facet->byte_count;
4509 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4510 struct ofexpired expired;
4511 expired.flow = facet->flow;
4512 expired.packet_count = facet->packet_count;
4513 expired.byte_count = facet->byte_count;
4514 expired.used = facet->used;
4515 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4518 facet->rule->packet_count += facet->packet_count;
4519 facet->rule->byte_count += facet->byte_count;
4521 /* Reset counters to prevent double counting if 'facet' ever gets
4523 facet_reset_counters(facet);
4525 netflow_flow_clear(&facet->nf_flow);
4526 facet->tcp_flags = 0;
4529 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4530 * Returns it if found, otherwise a null pointer.
4532 * 'hash' must be the return value of flow_hash(flow, 0).
4534 * The returned facet might need revalidation; use facet_lookup_valid()
4535 * instead if that is important. */
4536 static struct facet *
4537 facet_find(struct ofproto_dpif *ofproto,
4538 const struct flow *flow, uint32_t hash)
4540 struct facet *facet;
4542 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4543 if (flow_equal(flow, &facet->flow)) {
4551 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4552 * Returns it if found, otherwise a null pointer.
4554 * 'hash' must be the return value of flow_hash(flow, 0).
4556 * The returned facet is guaranteed to be valid. */
4557 static struct facet *
4558 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4561 struct facet *facet;
4563 facet = facet_find(ofproto, flow, hash);
4565 && (ofproto->backer->need_revalidate
4566 || tag_set_intersects(&ofproto->backer->revalidate_set,
4568 facet_revalidate(facet);
4570 /* facet_revalidate() may have destroyed 'facet'. */
4571 facet = facet_find(ofproto, flow, hash);
4578 subfacet_path_to_string(enum subfacet_path path)
4581 case SF_NOT_INSTALLED:
4582 return "not installed";
4584 return "in fast path";
4586 return "in slow path";
4592 /* Returns the path in which a subfacet should be installed if its 'slow'
4593 * member has the specified value. */
4594 static enum subfacet_path
4595 subfacet_want_path(enum slow_path_reason slow)
4597 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
4600 /* Returns true if 'subfacet' needs to have its datapath flow updated,
4601 * supposing that its actions have been recalculated as 'want_actions' and that
4602 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
4604 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
4605 const struct ofpbuf *want_actions)
4607 enum subfacet_path want_path = subfacet_want_path(slow);
4608 return (want_path != subfacet->path
4609 || (want_path == SF_FAST_PATH
4610 && (subfacet->actions_len != want_actions->size
4611 || memcmp(subfacet->actions, want_actions->data,
4612 subfacet->actions_len))));
4616 facet_check_consistency(struct facet *facet)
4618 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4620 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4622 uint64_t odp_actions_stub[1024 / 8];
4623 struct ofpbuf odp_actions;
4625 struct rule_dpif *rule;
4626 struct subfacet *subfacet;
4627 bool may_log = false;
4630 /* Check the rule for consistency. */
4631 rule = rule_dpif_lookup(ofproto, &facet->flow);
4632 ok = rule == facet->rule;
4634 may_log = !VLOG_DROP_WARN(&rl);
4639 flow_format(&s, &facet->flow);
4640 ds_put_format(&s, ": facet associated with wrong rule (was "
4641 "table=%"PRIu8",", facet->rule->up.table_id);
4642 cls_rule_format(&facet->rule->up.cr, &s);
4643 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4645 cls_rule_format(&rule->up.cr, &s);
4646 ds_put_char(&s, ')');
4648 VLOG_WARN("%s", ds_cstr(&s));
4653 /* Check the datapath actions for consistency. */
4654 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4655 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4656 enum subfacet_path want_path;
4657 struct action_xlate_ctx ctx;
4660 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4661 subfacet->initial_tci, rule, 0, NULL);
4662 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
4665 if (subfacet->path == SF_NOT_INSTALLED) {
4666 /* This only happens if the datapath reported an error when we
4667 * tried to install the flow. Don't flag another error here. */
4671 want_path = subfacet_want_path(subfacet->slow);
4672 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
4673 /* The actions for slow-path flows may legitimately vary from one
4674 * packet to the next. We're done. */
4678 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
4682 /* Inconsistency! */
4684 may_log = !VLOG_DROP_WARN(&rl);
4688 /* Rate-limited, skip reporting. */
4693 odp_flow_key_format(subfacet->key, subfacet->key_len, &s);
4695 ds_put_cstr(&s, ": inconsistency in subfacet");
4696 if (want_path != subfacet->path) {
4697 enum odp_key_fitness fitness = subfacet->key_fitness;
4699 ds_put_format(&s, " (%s, fitness=%s)",
4700 subfacet_path_to_string(subfacet->path),
4701 odp_key_fitness_to_string(fitness));
4702 ds_put_format(&s, " (should have been %s)",
4703 subfacet_path_to_string(want_path));
4704 } else if (want_path == SF_FAST_PATH) {
4705 ds_put_cstr(&s, " (actions were: ");
4706 format_odp_actions(&s, subfacet->actions,
4707 subfacet->actions_len);
4708 ds_put_cstr(&s, ") (correct actions: ");
4709 format_odp_actions(&s, odp_actions.data, odp_actions.size);
4710 ds_put_char(&s, ')');
4712 ds_put_cstr(&s, " (actions: ");
4713 format_odp_actions(&s, subfacet->actions,
4714 subfacet->actions_len);
4715 ds_put_char(&s, ')');
4717 VLOG_WARN("%s", ds_cstr(&s));
4720 ofpbuf_uninit(&odp_actions);
4725 /* Re-searches the classifier for 'facet':
4727 * - If the rule found is different from 'facet''s current rule, moves
4728 * 'facet' to the new rule and recompiles its actions.
4730 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4731 * where it is and recompiles its actions anyway.
4733 * - If any of 'facet''s subfacets correspond to a new flow according to
4734 * ofproto_receive(), 'facet' is removed. */
4736 facet_revalidate(struct facet *facet)
4738 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4740 struct nlattr *odp_actions;
4743 struct actions *new_actions;
4745 struct action_xlate_ctx ctx;
4746 uint64_t odp_actions_stub[1024 / 8];
4747 struct ofpbuf odp_actions;
4749 struct rule_dpif *new_rule;
4750 struct subfacet *subfacet;
4753 COVERAGE_INC(facet_revalidate);
4755 /* Check that child subfacets still correspond to this facet. Tunnel
4756 * configuration changes could cause a subfacet's OpenFlow in_port to
4758 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4759 struct ofproto_dpif *recv_ofproto;
4760 struct flow recv_flow;
4763 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
4764 subfacet->key_len, &recv_flow, NULL,
4765 &recv_ofproto, NULL, NULL);
4767 || recv_ofproto != ofproto
4768 || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) {
4769 facet_remove(facet);
4774 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
4776 /* Calculate new datapath actions.
4778 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4779 * emit a NetFlow expiration and, if so, we need to have the old state
4780 * around to properly compose it. */
4782 /* If the datapath actions changed or the installability changed,
4783 * then we need to talk to the datapath. */
4786 memset(&ctx, 0, sizeof ctx);
4787 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4788 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4789 enum slow_path_reason slow;
4791 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4792 subfacet->initial_tci, new_rule, 0, NULL);
4793 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
4796 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4797 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
4798 struct dpif_flow_stats stats;
4800 subfacet_install(subfacet,
4801 odp_actions.data, odp_actions.size, &stats, slow);
4802 subfacet_update_stats(subfacet, &stats);
4805 new_actions = xcalloc(list_size(&facet->subfacets),
4806 sizeof *new_actions);
4808 new_actions[i].odp_actions = xmemdup(odp_actions.data,
4810 new_actions[i].actions_len = odp_actions.size;
4815 ofpbuf_uninit(&odp_actions);
4818 facet_flush_stats(facet);
4821 /* Update 'facet' now that we've taken care of all the old state. */
4822 facet->tags = ctx.tags;
4823 facet->nf_flow.output_iface = ctx.nf_output_iface;
4824 facet->has_learn = ctx.has_learn;
4825 facet->has_normal = ctx.has_normal;
4826 facet->has_fin_timeout = ctx.has_fin_timeout;
4827 facet->mirrors = ctx.mirrors;
4830 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4831 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4833 if (new_actions && new_actions[i].odp_actions) {
4834 free(subfacet->actions);
4835 subfacet->actions = new_actions[i].odp_actions;
4836 subfacet->actions_len = new_actions[i].actions_len;
4842 if (facet->rule != new_rule) {
4843 COVERAGE_INC(facet_changed_rule);
4844 list_remove(&facet->list_node);
4845 list_push_back(&new_rule->facets, &facet->list_node);
4846 facet->rule = new_rule;
4847 facet->used = new_rule->up.created;
4848 facet->prev_used = facet->used;
4852 /* Updates 'facet''s used time. Caller is responsible for calling
4853 * facet_push_stats() to update the flows which 'facet' resubmits into. */
4855 facet_update_time(struct facet *facet, long long int used)
4857 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4858 if (used > facet->used) {
4860 ofproto_rule_update_used(&facet->rule->up, used);
4861 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
4866 facet_reset_counters(struct facet *facet)
4868 facet->packet_count = 0;
4869 facet->byte_count = 0;
4870 facet->prev_packet_count = 0;
4871 facet->prev_byte_count = 0;
4872 facet->accounted_bytes = 0;
4876 facet_push_stats(struct facet *facet)
4878 struct dpif_flow_stats stats;
4880 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4881 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4882 ovs_assert(facet->used >= facet->prev_used);
4884 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4885 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4886 stats.used = facet->used;
4887 stats.tcp_flags = 0;
4889 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
4890 facet->prev_packet_count = facet->packet_count;
4891 facet->prev_byte_count = facet->byte_count;
4892 facet->prev_used = facet->used;
4894 flow_push_stats(facet, &stats);
4896 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
4897 facet->mirrors, stats.n_packets, stats.n_bytes);
4902 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4904 rule->packet_count += stats->n_packets;
4905 rule->byte_count += stats->n_bytes;
4906 ofproto_rule_update_used(&rule->up, stats->used);
4909 /* Pushes flow statistics to the rules which 'facet->flow' resubmits
4910 * into given 'facet->rule''s actions and mirrors. */
4912 flow_push_stats(struct facet *facet, const struct dpif_flow_stats *stats)
4914 struct rule_dpif *rule = facet->rule;
4915 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4916 struct action_xlate_ctx ctx;
4918 ofproto_rule_update_used(&rule->up, stats->used);
4920 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, facet->flow.vlan_tci,
4922 ctx.resubmit_stats = stats;
4923 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
4924 rule->up.ofpacts_len);
4929 static struct subfacet *
4930 subfacet_find(struct ofproto_dpif *ofproto,
4931 const struct nlattr *key, size_t key_len, uint32_t key_hash)
4933 struct subfacet *subfacet;
4935 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4936 &ofproto->subfacets) {
4937 if (subfacet->key_len == key_len
4938 && !memcmp(key, subfacet->key, key_len)) {
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);
4972 if (subfacet->facet == facet) {
4976 /* This shouldn't happen. */
4977 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4978 subfacet_destroy(subfacet);
4981 subfacet = xmalloc(sizeof *subfacet);
4984 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
4985 list_push_back(&facet->subfacets, &subfacet->list_node);
4986 subfacet->facet = facet;
4987 subfacet->key_fitness = key_fitness;
4988 subfacet->key = xmemdup(key, key_len);
4989 subfacet->key_len = key_len;
4990 subfacet->used = now;
4991 subfacet->dp_packet_count = 0;
4992 subfacet->dp_byte_count = 0;
4993 subfacet->actions_len = 0;
4994 subfacet->actions = NULL;
4995 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
4998 subfacet->path = SF_NOT_INSTALLED;
4999 subfacet->initial_tci = miss->initial_tci;
5000 subfacet->odp_in_port = miss->odp_in_port;
5005 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5006 * its facet within 'ofproto', and frees it. */
5008 subfacet_destroy__(struct subfacet *subfacet)
5010 struct facet *facet = subfacet->facet;
5011 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5013 subfacet_uninstall(subfacet);
5014 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5015 list_remove(&subfacet->list_node);
5016 free(subfacet->key);
5017 free(subfacet->actions);
5018 if (subfacet != &facet->one_subfacet) {
5023 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5024 * last remaining subfacet in its facet destroys the facet too. */
5026 subfacet_destroy(struct subfacet *subfacet)
5028 struct facet *facet = subfacet->facet;
5030 if (list_is_singleton(&facet->subfacets)) {
5031 /* facet_remove() needs at least one subfacet (it will remove it). */
5032 facet_remove(facet);
5034 subfacet_destroy__(subfacet);
5039 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5040 struct subfacet **subfacets, int n)
5042 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5043 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5044 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5047 for (i = 0; i < n; i++) {
5048 ops[i].type = DPIF_OP_FLOW_DEL;
5049 ops[i].u.flow_del.key = subfacets[i]->key;
5050 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5051 ops[i].u.flow_del.stats = &stats[i];
5055 dpif_operate(ofproto->backer->dpif, opsp, n);
5056 for (i = 0; i < n; i++) {
5057 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5058 subfacets[i]->path = SF_NOT_INSTALLED;
5059 subfacet_destroy(subfacets[i]);
5063 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
5064 * Translates the actions into 'odp_actions', which the caller must have
5065 * initialized and is responsible for uninitializing. */
5067 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
5068 struct ofpbuf *odp_actions)
5070 struct facet *facet = subfacet->facet;
5071 struct rule_dpif *rule = facet->rule;
5072 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5074 struct action_xlate_ctx ctx;
5076 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
5078 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
5079 facet->tags = ctx.tags;
5080 facet->has_learn = ctx.has_learn;
5081 facet->has_normal = ctx.has_normal;
5082 facet->has_fin_timeout = ctx.has_fin_timeout;
5083 facet->nf_flow.output_iface = ctx.nf_output_iface;
5084 facet->mirrors = ctx.mirrors;
5086 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
5087 if (subfacet->actions_len != odp_actions->size
5088 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
5089 free(subfacet->actions);
5090 subfacet->actions_len = odp_actions->size;
5091 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
5095 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5096 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5097 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5098 * since 'subfacet' was last updated.
5100 * Returns 0 if successful, otherwise a positive errno value. */
5102 subfacet_install(struct subfacet *subfacet,
5103 const struct nlattr *actions, size_t actions_len,
5104 struct dpif_flow_stats *stats,
5105 enum slow_path_reason slow)
5107 struct facet *facet = subfacet->facet;
5108 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5109 enum subfacet_path path = subfacet_want_path(slow);
5110 uint64_t slow_path_stub[128 / 8];
5111 enum dpif_flow_put_flags flags;
5114 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5116 flags |= DPIF_FP_ZERO_STATS;
5119 if (path == SF_SLOW_PATH) {
5120 compose_slow_path(ofproto, &facet->flow, slow,
5121 slow_path_stub, sizeof slow_path_stub,
5122 &actions, &actions_len);
5125 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5126 subfacet->key_len, actions, actions_len, stats);
5129 subfacet_reset_dp_stats(subfacet, stats);
5133 subfacet->path = path;
5139 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
5141 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
5142 stats, subfacet->slow);
5145 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5147 subfacet_uninstall(struct subfacet *subfacet)
5149 if (subfacet->path != SF_NOT_INSTALLED) {
5150 struct rule_dpif *rule = subfacet->facet->rule;
5151 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5152 struct dpif_flow_stats stats;
5155 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5156 subfacet->key_len, &stats);
5157 subfacet_reset_dp_stats(subfacet, &stats);
5159 subfacet_update_stats(subfacet, &stats);
5161 subfacet->path = SF_NOT_INSTALLED;
5163 ovs_assert(subfacet->dp_packet_count == 0);
5164 ovs_assert(subfacet->dp_byte_count == 0);
5168 /* Resets 'subfacet''s datapath statistics counters. This should be called
5169 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5170 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5171 * was reset in the datapath. 'stats' will be modified to include only
5172 * statistics new since 'subfacet' was last updated. */
5174 subfacet_reset_dp_stats(struct subfacet *subfacet,
5175 struct dpif_flow_stats *stats)
5178 && subfacet->dp_packet_count <= stats->n_packets
5179 && subfacet->dp_byte_count <= stats->n_bytes) {
5180 stats->n_packets -= subfacet->dp_packet_count;
5181 stats->n_bytes -= subfacet->dp_byte_count;
5184 subfacet->dp_packet_count = 0;
5185 subfacet->dp_byte_count = 0;
5188 /* Updates 'subfacet''s used time. The caller is responsible for calling
5189 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
5191 subfacet_update_time(struct subfacet *subfacet, long long int used)
5193 if (used > subfacet->used) {
5194 subfacet->used = used;
5195 facet_update_time(subfacet->facet, used);
5199 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5201 * Because of the meaning of a subfacet's counters, it only makes sense to do
5202 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5203 * represents a packet that was sent by hand or if it represents statistics
5204 * that have been cleared out of the datapath. */
5206 subfacet_update_stats(struct subfacet *subfacet,
5207 const struct dpif_flow_stats *stats)
5209 if (stats->n_packets || stats->used > subfacet->used) {
5210 struct facet *facet = subfacet->facet;
5212 subfacet_update_time(subfacet, stats->used);
5213 facet->packet_count += stats->n_packets;
5214 facet->byte_count += stats->n_bytes;
5215 facet->tcp_flags |= stats->tcp_flags;
5216 facet_push_stats(facet);
5217 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
5223 static struct rule_dpif *
5224 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5226 struct rule_dpif *rule;
5228 rule = rule_dpif_lookup__(ofproto, flow, 0);
5233 return rule_dpif_miss_rule(ofproto, flow);
5236 static struct rule_dpif *
5237 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5240 struct cls_rule *cls_rule;
5241 struct classifier *cls;
5243 if (table_id >= N_TABLES) {
5247 cls = &ofproto->up.tables[table_id].cls;
5248 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5249 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5250 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5251 * are unavailable. */
5252 struct flow ofpc_normal_flow = *flow;
5253 ofpc_normal_flow.tp_src = htons(0);
5254 ofpc_normal_flow.tp_dst = htons(0);
5255 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5257 cls_rule = classifier_lookup(cls, flow);
5259 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5262 static struct rule_dpif *
5263 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5265 struct ofport_dpif *port;
5267 port = get_ofp_port(ofproto, flow->in_port);
5269 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5270 return ofproto->miss_rule;
5273 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5274 return ofproto->no_packet_in_rule;
5276 return ofproto->miss_rule;
5280 complete_operation(struct rule_dpif *rule)
5282 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5284 rule_invalidate(rule);
5286 struct dpif_completion *c = xmalloc(sizeof *c);
5287 c->op = rule->up.pending;
5288 list_push_back(&ofproto->completions, &c->list_node);
5290 ofoperation_complete(rule->up.pending, 0);
5294 static struct rule *
5297 struct rule_dpif *rule = xmalloc(sizeof *rule);
5302 rule_dealloc(struct rule *rule_)
5304 struct rule_dpif *rule = rule_dpif_cast(rule_);
5309 rule_construct(struct rule *rule_)
5311 struct rule_dpif *rule = rule_dpif_cast(rule_);
5312 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5313 struct rule_dpif *victim;
5316 rule->packet_count = 0;
5317 rule->byte_count = 0;
5319 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5320 if (victim && !list_is_empty(&victim->facets)) {
5321 struct facet *facet;
5323 rule->facets = victim->facets;
5324 list_moved(&rule->facets);
5325 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5326 /* XXX: We're only clearing our local counters here. It's possible
5327 * that quite a few packets are unaccounted for in the datapath
5328 * statistics. These will be accounted to the new rule instead of
5329 * cleared as required. This could be fixed by clearing out the
5330 * datapath statistics for this facet, but currently it doesn't
5332 facet_reset_counters(facet);
5336 /* Must avoid list_moved() in this case. */
5337 list_init(&rule->facets);
5340 table_id = rule->up.table_id;
5342 rule->tag = victim->tag;
5343 } else if (table_id == 0) {
5348 miniflow_expand(&rule->up.cr.match.flow, &flow);
5349 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5350 ofproto->tables[table_id].basis);
5353 complete_operation(rule);
5358 rule_destruct(struct rule *rule_)
5360 struct rule_dpif *rule = rule_dpif_cast(rule_);
5361 struct facet *facet, *next_facet;
5363 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5364 facet_revalidate(facet);
5367 complete_operation(rule);
5371 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5373 struct rule_dpif *rule = rule_dpif_cast(rule_);
5374 struct facet *facet;
5376 /* Start from historical data for 'rule' itself that are no longer tracked
5377 * in facets. This counts, for example, facets that have expired. */
5378 *packets = rule->packet_count;
5379 *bytes = rule->byte_count;
5381 /* Add any statistics that are tracked by facets. This includes
5382 * statistical data recently updated by ofproto_update_stats() as well as
5383 * stats for packets that were executed "by hand" via dpif_execute(). */
5384 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5385 *packets += facet->packet_count;
5386 *bytes += facet->byte_count;
5391 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5392 struct ofpbuf *packet)
5394 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5396 struct dpif_flow_stats stats;
5398 struct action_xlate_ctx ctx;
5399 uint64_t odp_actions_stub[1024 / 8];
5400 struct ofpbuf odp_actions;
5402 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5403 rule_credit_stats(rule, &stats);
5405 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5406 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
5407 rule, stats.tcp_flags, packet);
5408 ctx.resubmit_stats = &stats;
5409 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
5411 execute_odp_actions(ofproto, flow, odp_actions.data,
5412 odp_actions.size, packet);
5414 ofpbuf_uninit(&odp_actions);
5418 rule_execute(struct rule *rule, const struct flow *flow,
5419 struct ofpbuf *packet)
5421 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5422 ofpbuf_delete(packet);
5427 rule_modify_actions(struct rule *rule_)
5429 struct rule_dpif *rule = rule_dpif_cast(rule_);
5431 complete_operation(rule);
5434 /* Sends 'packet' out 'ofport'.
5435 * May modify 'packet'.
5436 * Returns 0 if successful, otherwise a positive errno value. */
5438 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5440 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5441 uint64_t odp_actions_stub[1024 / 8];
5442 struct ofpbuf key, odp_actions;
5443 struct odputil_keybuf keybuf;
5448 flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow);
5449 if (netdev_vport_is_patch(ofport->up.netdev)) {
5450 struct ofproto_dpif *peer_ofproto;
5451 struct dpif_flow_stats stats;
5452 struct ofport_dpif *peer;
5453 struct rule_dpif *rule;
5455 peer = ofport_get_peer(ofport);
5460 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5461 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5462 netdev_vport_inc_rx(peer->up.netdev, &stats);
5464 flow.in_port = peer->up.ofp_port;
5465 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5466 rule = rule_dpif_lookup(peer_ofproto, &flow);
5467 rule_dpif_execute(rule, &flow, packet);
5472 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5474 if (ofport->tnl_port) {
5475 struct dpif_flow_stats stats;
5477 odp_port = tnl_port_send(ofport->tnl_port, &flow);
5478 if (odp_port == OVSP_NONE) {
5482 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5483 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5484 odp_put_tunnel_action(&flow.tunnel, &odp_actions);
5485 odp_put_skb_mark_action(flow.skb_mark, &odp_actions);
5487 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5489 if (odp_port != ofport->odp_port) {
5490 eth_pop_vlan(packet);
5491 flow.vlan_tci = htons(0);
5495 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5496 odp_flow_key_from_flow(&key, &flow,
5497 ofp_port_to_odp_port(ofproto, flow.in_port));
5499 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5501 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5502 error = dpif_execute(ofproto->backer->dpif,
5504 odp_actions.data, odp_actions.size,
5506 ofpbuf_uninit(&odp_actions);
5509 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5510 ofproto->up.name, odp_port, strerror(error));
5512 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5516 /* OpenFlow to datapath action translation. */
5518 static bool may_receive(const struct ofport_dpif *, struct action_xlate_ctx *);
5519 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5520 struct action_xlate_ctx *);
5521 static void xlate_normal(struct action_xlate_ctx *);
5523 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5524 * The action will state 'slow' as the reason that the action is in the slow
5525 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5526 * dump-flows" output to see why a flow is in the slow path.)
5528 * The 'stub_size' bytes in 'stub' will be used to store the action.
5529 * 'stub_size' must be large enough for the action.
5531 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5534 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5535 enum slow_path_reason slow,
5536 uint64_t *stub, size_t stub_size,
5537 const struct nlattr **actionsp, size_t *actions_lenp)
5539 union user_action_cookie cookie;
5542 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5543 cookie.slow_path.unused = 0;
5544 cookie.slow_path.reason = slow;
5546 ofpbuf_use_stack(&buf, stub, stub_size);
5547 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
5548 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5549 odp_put_userspace_action(pid, &cookie, sizeof cookie, &buf);
5551 put_userspace_action(ofproto, &buf, flow, &cookie);
5553 *actionsp = buf.data;
5554 *actions_lenp = buf.size;
5558 put_userspace_action(const struct ofproto_dpif *ofproto,
5559 struct ofpbuf *odp_actions,
5560 const struct flow *flow,
5561 const union user_action_cookie *cookie)
5565 pid = dpif_port_get_pid(ofproto->backer->dpif,
5566 ofp_port_to_odp_port(ofproto, flow->in_port));
5568 return odp_put_userspace_action(pid, cookie, sizeof *cookie, odp_actions);
5572 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5573 ovs_be16 vlan_tci, uint32_t odp_port,
5574 unsigned int n_outputs, union user_action_cookie *cookie)
5578 cookie->type = USER_ACTION_COOKIE_SFLOW;
5579 cookie->sflow.vlan_tci = vlan_tci;
5581 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5582 * port information") for the interpretation of cookie->output. */
5583 switch (n_outputs) {
5585 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5586 cookie->sflow.output = 0x40000000 | 256;
5590 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5592 cookie->sflow.output = ifindex;
5597 /* 0x80000000 means "multiple output ports. */
5598 cookie->sflow.output = 0x80000000 | n_outputs;
5603 /* Compose SAMPLE action for sFlow. */
5605 compose_sflow_action(const struct ofproto_dpif *ofproto,
5606 struct ofpbuf *odp_actions,
5607 const struct flow *flow,
5610 uint32_t probability;
5611 union user_action_cookie cookie;
5612 size_t sample_offset, actions_offset;
5615 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5619 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5621 /* Number of packets out of UINT_MAX to sample. */
5622 probability = dpif_sflow_get_probability(ofproto->sflow);
5623 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5625 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5626 compose_sflow_cookie(ofproto, htons(0), odp_port,
5627 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5628 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
5630 nl_msg_end_nested(odp_actions, actions_offset);
5631 nl_msg_end_nested(odp_actions, sample_offset);
5632 return cookie_offset;
5635 /* SAMPLE action must be first action in any given list of actions.
5636 * At this point we do not have all information required to build it. So try to
5637 * build sample action as complete as possible. */
5639 add_sflow_action(struct action_xlate_ctx *ctx)
5641 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5643 &ctx->flow, OVSP_NONE);
5644 ctx->sflow_odp_port = 0;
5645 ctx->sflow_n_outputs = 0;
5648 /* Fix SAMPLE action according to data collected while composing ODP actions.
5649 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5650 * USERSPACE action's user-cookie which is required for sflow. */
5652 fix_sflow_action(struct action_xlate_ctx *ctx)
5654 const struct flow *base = &ctx->base_flow;
5655 union user_action_cookie *cookie;
5657 if (!ctx->user_cookie_offset) {
5661 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
5663 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5665 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5666 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5670 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
5673 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5674 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
5675 ovs_be64 flow_tun_id = ctx->flow.tunnel.tun_id;
5676 uint8_t flow_nw_tos = ctx->flow.nw_tos;
5677 struct priority_to_dscp *pdscp;
5678 uint32_t out_port, odp_port;
5680 /* If 'struct flow' gets additional metadata, we'll need to zero it out
5681 * before traversing a patch port. */
5682 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 19);
5685 xlate_report(ctx, "Nonexistent output port");
5687 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5688 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5690 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5691 xlate_report(ctx, "STP not in forwarding state, skipping output");
5695 if (netdev_vport_is_patch(ofport->up.netdev)) {
5696 struct ofport_dpif *peer = ofport_get_peer(ofport);
5697 struct flow old_flow = ctx->flow;
5698 const struct ofproto_dpif *peer_ofproto;
5699 enum slow_path_reason special;
5700 struct ofport_dpif *in_port;
5703 xlate_report(ctx, "Nonexistent patch port peer");
5707 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5708 if (peer_ofproto->backer != ctx->ofproto->backer) {
5709 xlate_report(ctx, "Patch port peer on a different datapath");
5713 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
5714 ctx->flow.in_port = peer->up.ofp_port;
5715 ctx->flow.metadata = htonll(0);
5716 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
5717 memset(ctx->flow.regs, 0, sizeof ctx->flow.regs);
5719 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5720 special = process_special(ctx->ofproto, &ctx->flow, in_port,
5723 ctx->slow |= special;
5724 } else if (!in_port || may_receive(in_port, ctx)) {
5725 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
5726 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5728 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
5729 * learning action look at the packet, then drop it. */
5730 struct flow old_base_flow = ctx->base_flow;
5731 size_t old_size = ctx->odp_actions->size;
5732 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5733 ctx->base_flow = old_base_flow;
5734 ctx->odp_actions->size = old_size;
5738 ctx->flow = old_flow;
5739 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5741 if (ctx->resubmit_stats) {
5742 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
5743 netdev_vport_inc_rx(peer->up.netdev, ctx->resubmit_stats);
5749 pdscp = get_priority(ofport, ctx->flow.skb_priority);
5751 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5752 ctx->flow.nw_tos |= pdscp->dscp;
5755 odp_port = ofp_port_to_odp_port(ctx->ofproto, ofp_port);
5756 if (ofport->tnl_port) {
5757 odp_port = tnl_port_send(ofport->tnl_port, &ctx->flow);
5758 if (odp_port == OVSP_NONE) {
5759 xlate_report(ctx, "Tunneling decided against output");
5763 if (ctx->resubmit_stats) {
5764 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
5766 out_port = odp_port;
5767 commit_odp_tunnel_action(&ctx->flow, &ctx->base_flow,
5770 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
5771 ctx->flow.vlan_tci);
5772 if (out_port != odp_port) {
5773 ctx->flow.vlan_tci = htons(0);
5776 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
5777 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
5779 ctx->sflow_odp_port = odp_port;
5780 ctx->sflow_n_outputs++;
5781 ctx->nf_output_iface = ofp_port;
5782 ctx->flow.tunnel.tun_id = flow_tun_id;
5783 ctx->flow.vlan_tci = flow_vlan_tci;
5784 ctx->flow.nw_tos = flow_nw_tos;
5788 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
5790 compose_output_action__(ctx, ofp_port, true);
5794 xlate_table_action(struct action_xlate_ctx *ctx,
5795 uint16_t in_port, uint8_t table_id, bool may_packet_in)
5797 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
5798 struct ofproto_dpif *ofproto = ctx->ofproto;
5799 struct rule_dpif *rule;
5800 uint16_t old_in_port;
5801 uint8_t old_table_id;
5803 old_table_id = ctx->table_id;
5804 ctx->table_id = table_id;
5806 /* Look up a flow with 'in_port' as the input port. */
5807 old_in_port = ctx->flow.in_port;
5808 ctx->flow.in_port = in_port;
5809 rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id);
5812 if (table_id > 0 && table_id < N_TABLES) {
5813 struct table_dpif *table = &ofproto->tables[table_id];
5814 if (table->other_table) {
5815 ctx->tags |= (rule && rule->tag
5817 : rule_calculate_tag(&ctx->flow,
5818 &table->other_table->mask,
5823 /* Restore the original input port. Otherwise OFPP_NORMAL and
5824 * OFPP_IN_PORT will have surprising behavior. */
5825 ctx->flow.in_port = old_in_port;
5827 if (ctx->resubmit_hook) {
5828 ctx->resubmit_hook(ctx, rule);
5831 if (rule == NULL && may_packet_in) {
5833 * check if table configuration flags
5834 * OFPTC_TABLE_MISS_CONTROLLER, default.
5835 * OFPTC_TABLE_MISS_CONTINUE,
5836 * OFPTC_TABLE_MISS_DROP
5837 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
5839 rule = rule_dpif_miss_rule(ofproto, &ctx->flow);
5843 struct rule_dpif *old_rule = ctx->rule;
5845 if (ctx->resubmit_stats) {
5846 rule_credit_stats(rule, ctx->resubmit_stats);
5851 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
5852 ctx->rule = old_rule;
5856 ctx->table_id = old_table_id;
5858 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5860 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
5861 MAX_RESUBMIT_RECURSION);
5862 ctx->max_resubmit_trigger = true;
5867 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
5868 const struct ofpact_resubmit *resubmit)
5873 in_port = resubmit->in_port;
5874 if (in_port == OFPP_IN_PORT) {
5875 in_port = ctx->flow.in_port;
5878 table_id = resubmit->table_id;
5879 if (table_id == 255) {
5880 table_id = ctx->table_id;
5883 xlate_table_action(ctx, in_port, table_id, false);
5887 flood_packets(struct action_xlate_ctx *ctx, bool all)
5889 struct ofport_dpif *ofport;
5891 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
5892 uint16_t ofp_port = ofport->up.ofp_port;
5894 if (ofp_port == ctx->flow.in_port) {
5899 compose_output_action__(ctx, ofp_port, false);
5900 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
5901 compose_output_action(ctx, ofp_port);
5905 ctx->nf_output_iface = NF_OUT_FLOOD;
5909 execute_controller_action(struct action_xlate_ctx *ctx, int len,
5910 enum ofp_packet_in_reason reason,
5911 uint16_t controller_id)
5913 struct ofputil_packet_in pin;
5914 struct ofpbuf *packet;
5916 ctx->slow |= SLOW_CONTROLLER;
5921 packet = ofpbuf_clone(ctx->packet);
5923 if (packet->l2 && packet->l3) {
5924 struct eth_header *eh;
5925 uint16_t mpls_depth;
5927 eth_pop_vlan(packet);
5930 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
5931 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
5933 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
5934 eth_push_vlan(packet, ctx->flow.vlan_tci);
5937 mpls_depth = eth_mpls_depth(packet);
5939 if (mpls_depth < ctx->flow.mpls_depth) {
5940 push_mpls(packet, ctx->flow.dl_type, ctx->flow.mpls_lse);
5941 } else if (mpls_depth > ctx->flow.mpls_depth) {
5942 pop_mpls(packet, ctx->flow.dl_type);
5943 } else if (mpls_depth) {
5944 set_mpls_lse(packet, ctx->flow.mpls_lse);
5948 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5949 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
5950 ctx->flow.nw_tos, ctx->flow.nw_ttl);
5954 if (ctx->flow.nw_proto == IPPROTO_TCP) {
5955 packet_set_tcp_port(packet, ctx->flow.tp_src,
5957 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
5958 packet_set_udp_port(packet, ctx->flow.tp_src,
5965 pin.packet = packet->data;
5966 pin.packet_len = packet->size;
5967 pin.reason = reason;
5968 pin.controller_id = controller_id;
5969 pin.table_id = ctx->table_id;
5970 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
5973 flow_get_metadata(&ctx->flow, &pin.fmd);
5975 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
5976 ofpbuf_delete(packet);
5980 execute_mpls_push_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
5982 ovs_assert(eth_type_mpls(eth_type));
5984 if (ctx->base_flow.mpls_depth) {
5985 ctx->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK);
5986 ctx->flow.mpls_depth++;
5991 if (ctx->flow.dl_type == htons(ETH_TYPE_IPV6)) {
5992 label = htonl(0x2); /* IPV6 Explicit Null. */
5994 label = htonl(0x0); /* IPV4 Explicit Null. */
5996 tc = (ctx->flow.nw_tos & IP_DSCP_MASK) >> 2;
5997 ttl = ctx->flow.nw_ttl ? ctx->flow.nw_ttl : 0x40;
5998 ctx->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
5999 ctx->flow.encap_dl_type = ctx->flow.dl_type;
6000 ctx->flow.mpls_depth = 1;
6002 ctx->flow.dl_type = eth_type;
6006 execute_mpls_pop_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
6008 ovs_assert(eth_type_mpls(ctx->flow.dl_type));
6009 ovs_assert(!eth_type_mpls(eth_type));
6011 if (ctx->flow.mpls_depth) {
6012 ctx->flow.mpls_depth--;
6013 ctx->flow.mpls_lse = htonl(0);
6014 if (!ctx->flow.mpls_depth) {
6015 ctx->flow.dl_type = eth_type;
6016 ctx->flow.encap_dl_type = htons(0);
6022 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6024 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
6025 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6029 if (ctx->flow.nw_ttl > 1) {
6035 for (i = 0; i < ids->n_controllers; i++) {
6036 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6040 /* Stop processing for current table. */
6046 execute_set_mpls_ttl_action(struct action_xlate_ctx *ctx, uint8_t ttl)
6048 if (!eth_type_mpls(ctx->flow.dl_type)) {
6052 set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl);
6057 execute_dec_mpls_ttl_action(struct action_xlate_ctx *ctx)
6059 uint8_t ttl = mpls_lse_to_ttl(ctx->flow.mpls_lse);
6061 if (!eth_type_mpls(ctx->flow.dl_type)) {
6067 set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl);
6070 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
6072 /* Stop processing for current table. */
6078 xlate_output_action(struct action_xlate_ctx *ctx,
6079 uint16_t port, uint16_t max_len, bool may_packet_in)
6081 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
6083 ctx->nf_output_iface = NF_OUT_DROP;
6087 compose_output_action(ctx, ctx->flow.in_port);
6090 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
6096 flood_packets(ctx, false);
6099 flood_packets(ctx, true);
6101 case OFPP_CONTROLLER:
6102 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6108 if (port != ctx->flow.in_port) {
6109 compose_output_action(ctx, port);
6111 xlate_report(ctx, "skipping output to input port");
6116 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6117 ctx->nf_output_iface = NF_OUT_FLOOD;
6118 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
6119 ctx->nf_output_iface = prev_nf_output_iface;
6120 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6121 ctx->nf_output_iface != NF_OUT_FLOOD) {
6122 ctx->nf_output_iface = NF_OUT_MULTI;
6127 xlate_output_reg_action(struct action_xlate_ctx *ctx,
6128 const struct ofpact_output_reg *or)
6130 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
6131 if (port <= UINT16_MAX) {
6132 xlate_output_action(ctx, port, or->max_len, false);
6137 xlate_enqueue_action(struct action_xlate_ctx *ctx,
6138 const struct ofpact_enqueue *enqueue)
6140 uint16_t ofp_port = enqueue->port;
6141 uint32_t queue_id = enqueue->queue;
6142 uint32_t flow_priority, priority;
6145 /* Translate queue to priority. */
6146 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6147 queue_id, &priority);
6149 /* Fall back to ordinary output action. */
6150 xlate_output_action(ctx, enqueue->port, 0, false);
6154 /* Check output port. */
6155 if (ofp_port == OFPP_IN_PORT) {
6156 ofp_port = ctx->flow.in_port;
6157 } else if (ofp_port == ctx->flow.in_port) {
6161 /* Add datapath actions. */
6162 flow_priority = ctx->flow.skb_priority;
6163 ctx->flow.skb_priority = priority;
6164 compose_output_action(ctx, ofp_port);
6165 ctx->flow.skb_priority = flow_priority;
6167 /* Update NetFlow output port. */
6168 if (ctx->nf_output_iface == NF_OUT_DROP) {
6169 ctx->nf_output_iface = ofp_port;
6170 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
6171 ctx->nf_output_iface = NF_OUT_MULTI;
6176 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
6178 uint32_t skb_priority;
6180 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6181 queue_id, &skb_priority)) {
6182 ctx->flow.skb_priority = skb_priority;
6184 /* Couldn't translate queue to a priority. Nothing to do. A warning
6185 * has already been logged. */
6189 struct xlate_reg_state {
6195 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6197 struct ofproto_dpif *ofproto = ofproto_;
6198 struct ofport_dpif *port;
6208 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6211 port = get_ofp_port(ofproto, ofp_port);
6212 return port ? port->may_enable : false;
6217 xlate_bundle_action(struct action_xlate_ctx *ctx,
6218 const struct ofpact_bundle *bundle)
6222 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
6223 if (bundle->dst.field) {
6224 nxm_reg_load(&bundle->dst, port, &ctx->flow);
6226 xlate_output_action(ctx, port, 0, false);
6231 xlate_learn_action(struct action_xlate_ctx *ctx,
6232 const struct ofpact_learn *learn)
6234 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6235 struct ofputil_flow_mod fm;
6236 uint64_t ofpacts_stub[1024 / 8];
6237 struct ofpbuf ofpacts;
6240 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6241 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
6243 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6244 if (error && !VLOG_DROP_WARN(&rl)) {
6245 VLOG_WARN("learning action failed to modify flow table (%s)",
6246 ofperr_get_name(error));
6249 ofpbuf_uninit(&ofpacts);
6252 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6253 * means "infinite". */
6255 reduce_timeout(uint16_t max, uint16_t *timeout)
6257 if (max && (!*timeout || *timeout > max)) {
6263 xlate_fin_timeout(struct action_xlate_ctx *ctx,
6264 const struct ofpact_fin_timeout *oft)
6266 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6267 struct rule_dpif *rule = ctx->rule;
6269 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6270 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6275 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
6277 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
6278 ? OFPUTIL_PC_NO_RECV_STP
6279 : OFPUTIL_PC_NO_RECV)) {
6283 /* Only drop packets here if both forwarding and learning are
6284 * disabled. If just learning is enabled, we need to have
6285 * OFPP_NORMAL and the learning action have a look at the packet
6286 * before we can drop it. */
6287 if (!stp_forward_in_state(port->stp_state)
6288 && !stp_learn_in_state(port->stp_state)) {
6296 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6297 struct action_xlate_ctx *ctx)
6299 bool was_evictable = true;
6300 const struct ofpact *a;
6303 /* Don't let the rule we're working on get evicted underneath us. */
6304 was_evictable = ctx->rule->up.evictable;
6305 ctx->rule->up.evictable = false;
6307 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6308 struct ofpact_controller *controller;
6309 const struct ofpact_metadata *metadata;
6317 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6318 ofpact_get_OUTPUT(a)->max_len, true);
6321 case OFPACT_CONTROLLER:
6322 controller = ofpact_get_CONTROLLER(a);
6323 execute_controller_action(ctx, controller->max_len,
6325 controller->controller_id);
6328 case OFPACT_ENQUEUE:
6329 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6332 case OFPACT_SET_VLAN_VID:
6333 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6334 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6338 case OFPACT_SET_VLAN_PCP:
6339 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6340 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
6345 case OFPACT_STRIP_VLAN:
6346 ctx->flow.vlan_tci = htons(0);
6349 case OFPACT_PUSH_VLAN:
6350 /* XXX 802.1AD(QinQ) */
6351 ctx->flow.vlan_tci = htons(VLAN_CFI);
6354 case OFPACT_SET_ETH_SRC:
6355 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6359 case OFPACT_SET_ETH_DST:
6360 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6364 case OFPACT_SET_IPV4_SRC:
6365 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6368 case OFPACT_SET_IPV4_DST:
6369 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6372 case OFPACT_SET_IPV4_DSCP:
6373 /* OpenFlow 1.0 only supports IPv4. */
6374 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6375 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
6376 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6380 case OFPACT_SET_L4_SRC_PORT:
6381 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6384 case OFPACT_SET_L4_DST_PORT:
6385 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6388 case OFPACT_RESUBMIT:
6389 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6392 case OFPACT_SET_TUNNEL:
6393 ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6396 case OFPACT_SET_QUEUE:
6397 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6400 case OFPACT_POP_QUEUE:
6401 ctx->flow.skb_priority = ctx->orig_skb_priority;
6404 case OFPACT_REG_MOVE:
6405 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
6408 case OFPACT_REG_LOAD:
6409 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
6412 case OFPACT_STACK_PUSH:
6413 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->flow,
6417 case OFPACT_STACK_POP:
6418 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->flow,
6422 case OFPACT_PUSH_MPLS:
6423 execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
6426 case OFPACT_POP_MPLS:
6427 execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
6430 case OFPACT_SET_MPLS_TTL:
6431 if (execute_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl)) {
6436 case OFPACT_DEC_MPLS_TTL:
6437 if (execute_dec_mpls_ttl_action(ctx)) {
6442 case OFPACT_DEC_TTL:
6443 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6449 /* Nothing to do. */
6452 case OFPACT_MULTIPATH:
6453 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
6457 ctx->ofproto->has_bundle_action = true;
6458 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6461 case OFPACT_OUTPUT_REG:
6462 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6466 ctx->has_learn = true;
6467 if (ctx->may_learn) {
6468 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6476 case OFPACT_FIN_TIMEOUT:
6477 ctx->has_fin_timeout = true;
6478 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6481 case OFPACT_CLEAR_ACTIONS:
6483 * Nothing to do because writa-actions is not supported for now.
6484 * When writa-actions is supported, clear-actions also must
6485 * be supported at the same time.
6489 case OFPACT_WRITE_METADATA:
6490 metadata = ofpact_get_WRITE_METADATA(a);
6491 ctx->flow.metadata &= ~metadata->mask;
6492 ctx->flow.metadata |= metadata->metadata & metadata->mask;
6495 case OFPACT_GOTO_TABLE: {
6496 /* XXX remove recursion */
6497 /* It is assumed that goto-table is last action */
6498 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6499 ovs_assert(ctx->table_id < ogt->table_id);
6500 xlate_table_action(ctx, ctx->flow.in_port, ogt->table_id, true);
6508 ctx->rule->up.evictable = was_evictable;
6513 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
6514 struct ofproto_dpif *ofproto, const struct flow *flow,
6515 ovs_be16 initial_tci, struct rule_dpif *rule,
6516 uint8_t tcp_flags, const struct ofpbuf *packet)
6518 ovs_be64 initial_tun_id = flow->tunnel.tun_id;
6520 /* Flow initialization rules:
6521 * - 'base_flow' must match the kernel's view of the packet at the
6522 * time that action processing starts. 'flow' represents any
6523 * transformations we wish to make through actions.
6524 * - By default 'base_flow' and 'flow' are the same since the input
6525 * packet matches the output before any actions are applied.
6526 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
6527 * of the received packet as seen by the kernel. If we later output
6528 * to another device without any modifications this will cause us to
6529 * insert a new tag since the original one was stripped off by the
6531 * - Tunnel 'flow' is largely cleared when transitioning between
6532 * the input and output stages since it does not make sense to output
6533 * a packet with the exact headers that it was received with (i.e.
6534 * the destination IP is us). The one exception is the tun_id, which
6535 * is preserved to allow use in later resubmit lookups and loads into
6537 * - Tunnel 'base_flow' is completely cleared since that is what the
6538 * kernel does. If we wish to maintain the original values an action
6539 * needs to be generated. */
6541 ctx->ofproto = ofproto;
6543 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
6544 ctx->base_flow = ctx->flow;
6545 ctx->base_flow.vlan_tci = initial_tci;
6546 ctx->flow.tunnel.tun_id = initial_tun_id;
6548 ctx->packet = packet;
6549 ctx->may_learn = packet != NULL;
6550 ctx->tcp_flags = tcp_flags;
6551 ctx->resubmit_hook = NULL;
6552 ctx->report_hook = NULL;
6553 ctx->resubmit_stats = NULL;
6556 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6557 * into datapath actions in 'odp_actions', using 'ctx'. */
6559 xlate_actions(struct action_xlate_ctx *ctx,
6560 const struct ofpact *ofpacts, size_t ofpacts_len,
6561 struct ofpbuf *odp_actions)
6563 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6564 * that in the future we always keep a copy of the original flow for
6565 * tracing purposes. */
6566 static bool hit_resubmit_limit;
6568 enum slow_path_reason special;
6569 struct ofport_dpif *in_port;
6570 struct flow orig_flow;
6572 COVERAGE_INC(ofproto_dpif_xlate);
6574 ofpbuf_clear(odp_actions);
6575 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
6577 ctx->odp_actions = odp_actions;
6580 ctx->has_learn = false;
6581 ctx->has_normal = false;
6582 ctx->has_fin_timeout = false;
6583 ctx->nf_output_iface = NF_OUT_DROP;
6586 ctx->max_resubmit_trigger = false;
6587 ctx->orig_skb_priority = ctx->flow.skb_priority;
6591 ofpbuf_use_stub(&ctx->stack, ctx->init_stack, sizeof ctx->init_stack);
6593 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
6594 /* Do this conditionally because the copy is expensive enough that it
6595 * shows up in profiles. */
6596 orig_flow = ctx->flow;
6599 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6600 switch (ctx->ofproto->up.frag_handling) {
6601 case OFPC_FRAG_NORMAL:
6602 /* We must pretend that transport ports are unavailable. */
6603 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
6604 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
6607 case OFPC_FRAG_DROP:
6610 case OFPC_FRAG_REASM:
6613 case OFPC_FRAG_NX_MATCH:
6614 /* Nothing to do. */
6617 case OFPC_INVALID_TTL_TO_CONTROLLER:
6622 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
6623 special = process_special(ctx->ofproto, &ctx->flow, in_port, ctx->packet);
6625 ctx->slow |= special;
6627 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6628 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
6629 uint32_t local_odp_port;
6631 add_sflow_action(ctx);
6633 if (!in_port || may_receive(in_port, ctx)) {
6634 do_xlate_actions(ofpacts, ofpacts_len, ctx);
6636 /* We've let OFPP_NORMAL and the learning action look at the
6637 * packet, so drop it now if forwarding is disabled. */
6638 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
6639 ofpbuf_clear(ctx->odp_actions);
6640 add_sflow_action(ctx);
6644 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
6645 if (!hit_resubmit_limit) {
6646 /* We didn't record the original flow. Make sure we do from
6648 hit_resubmit_limit = true;
6649 } else if (!VLOG_DROP_ERR(&trace_rl)) {
6650 struct ds ds = DS_EMPTY_INITIALIZER;
6652 ofproto_trace(ctx->ofproto, &orig_flow, ctx->packet,
6654 VLOG_ERR("Trace triggered by excessive resubmit "
6655 "recursion:\n%s", ds_cstr(&ds));
6660 local_odp_port = ofp_port_to_odp_port(ctx->ofproto, OFPP_LOCAL);
6661 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
6663 ctx->odp_actions->data,
6664 ctx->odp_actions->size)) {
6665 ctx->slow |= SLOW_IN_BAND;
6667 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
6669 compose_output_action(ctx, OFPP_LOCAL);
6672 if (ctx->ofproto->has_mirrors) {
6673 add_mirror_actions(ctx, &orig_flow);
6675 fix_sflow_action(ctx);
6678 ofpbuf_uninit(&ctx->stack);
6681 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
6682 * into datapath actions, using 'ctx', and discards the datapath actions. */
6684 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
6685 const struct ofpact *ofpacts,
6688 uint64_t odp_actions_stub[1024 / 8];
6689 struct ofpbuf odp_actions;
6691 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
6692 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
6693 ofpbuf_uninit(&odp_actions);
6697 xlate_report(struct action_xlate_ctx *ctx, const char *s)
6699 if (ctx->report_hook) {
6700 ctx->report_hook(ctx, s);
6704 /* OFPP_NORMAL implementation. */
6706 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
6708 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
6709 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
6710 * the bundle on which the packet was received, returns the VLAN to which the
6713 * Both 'vid' and the return value are in the range 0...4095. */
6715 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
6717 switch (in_bundle->vlan_mode) {
6718 case PORT_VLAN_ACCESS:
6719 return in_bundle->vlan;
6722 case PORT_VLAN_TRUNK:
6725 case PORT_VLAN_NATIVE_UNTAGGED:
6726 case PORT_VLAN_NATIVE_TAGGED:
6727 return vid ? vid : in_bundle->vlan;
6734 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
6735 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
6738 * 'vid' should be the VID obtained from the 802.1Q header that was received as
6739 * part of a packet (specify 0 if there was no 802.1Q header), in the range
6742 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
6744 /* Allow any VID on the OFPP_NONE port. */
6745 if (in_bundle == &ofpp_none_bundle) {
6749 switch (in_bundle->vlan_mode) {
6750 case PORT_VLAN_ACCESS:
6753 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6754 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
6755 "packet received on port %s configured as VLAN "
6756 "%"PRIu16" access port",
6757 in_bundle->ofproto->up.name, vid,
6758 in_bundle->name, in_bundle->vlan);
6764 case PORT_VLAN_NATIVE_UNTAGGED:
6765 case PORT_VLAN_NATIVE_TAGGED:
6767 /* Port must always carry its native VLAN. */
6771 case PORT_VLAN_TRUNK:
6772 if (!ofbundle_includes_vlan(in_bundle, vid)) {
6774 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6775 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
6776 "received on port %s not configured for trunking "
6778 in_bundle->ofproto->up.name, vid,
6779 in_bundle->name, vid);
6791 /* Given 'vlan', the VLAN that a packet belongs to, and
6792 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
6793 * that should be included in the 802.1Q header. (If the return value is 0,
6794 * then the 802.1Q header should only be included in the packet if there is a
6797 * Both 'vlan' and the return value are in the range 0...4095. */
6799 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
6801 switch (out_bundle->vlan_mode) {
6802 case PORT_VLAN_ACCESS:
6805 case PORT_VLAN_TRUNK:
6806 case PORT_VLAN_NATIVE_TAGGED:
6809 case PORT_VLAN_NATIVE_UNTAGGED:
6810 return vlan == out_bundle->vlan ? 0 : vlan;
6818 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
6821 struct ofport_dpif *port;
6823 ovs_be16 tci, old_tci;
6825 vid = output_vlan_to_vid(out_bundle, vlan);
6826 if (!out_bundle->bond) {
6827 port = ofbundle_get_a_port(out_bundle);
6829 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
6832 /* No slaves enabled, so drop packet. */
6837 old_tci = ctx->flow.vlan_tci;
6839 if (tci || out_bundle->use_priority_tags) {
6840 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
6842 tci |= htons(VLAN_CFI);
6845 ctx->flow.vlan_tci = tci;
6847 compose_output_action(ctx, port->up.ofp_port);
6848 ctx->flow.vlan_tci = old_tci;
6852 mirror_mask_ffs(mirror_mask_t mask)
6854 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
6859 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
6861 return (bundle->vlan_mode != PORT_VLAN_ACCESS
6862 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
6866 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
6868 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
6871 /* Returns an arbitrary interface within 'bundle'. */
6872 static struct ofport_dpif *
6873 ofbundle_get_a_port(const struct ofbundle *bundle)
6875 return CONTAINER_OF(list_front(&bundle->ports),
6876 struct ofport_dpif, bundle_node);
6880 vlan_is_mirrored(const struct ofmirror *m, int vlan)
6882 return !m->vlans || bitmap_is_set(m->vlans, vlan);
6886 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
6888 struct ofproto_dpif *ofproto = ctx->ofproto;
6889 mirror_mask_t mirrors;
6890 struct ofbundle *in_bundle;
6893 const struct nlattr *a;
6896 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
6897 ctx->packet != NULL, NULL);
6901 mirrors = in_bundle->src_mirrors;
6903 /* Drop frames on bundles reserved for mirroring. */
6904 if (in_bundle->mirror_out) {
6905 if (ctx->packet != NULL) {
6906 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6907 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6908 "%s, which is reserved exclusively for mirroring",
6909 ctx->ofproto->up.name, in_bundle->name);
6915 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
6916 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6919 vlan = input_vid_to_vlan(in_bundle, vid);
6921 /* Look at the output ports to check for destination selections. */
6923 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
6924 ctx->odp_actions->size) {
6925 enum ovs_action_attr type = nl_attr_type(a);
6926 struct ofport_dpif *ofport;
6928 if (type != OVS_ACTION_ATTR_OUTPUT) {
6932 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
6933 if (ofport && ofport->bundle) {
6934 mirrors |= ofport->bundle->dst_mirrors;
6942 /* Restore the original packet before adding the mirror actions. */
6943 ctx->flow = *orig_flow;
6948 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6950 if (!vlan_is_mirrored(m, vlan)) {
6951 mirrors = zero_rightmost_1bit(mirrors);
6955 mirrors &= ~m->dup_mirrors;
6956 ctx->mirrors |= m->dup_mirrors;
6958 output_normal(ctx, m->out, vlan);
6959 } else if (vlan != m->out_vlan
6960 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
6961 struct ofbundle *bundle;
6963 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
6964 if (ofbundle_includes_vlan(bundle, m->out_vlan)
6965 && !bundle->mirror_out) {
6966 output_normal(ctx, bundle, m->out_vlan);
6974 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
6975 uint64_t packets, uint64_t bytes)
6981 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
6984 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6987 /* In normal circumstances 'm' will not be NULL. However,
6988 * if mirrors are reconfigured, we can temporarily get out
6989 * of sync in facet_revalidate(). We could "correct" the
6990 * mirror list before reaching here, but doing that would
6991 * not properly account the traffic stats we've currently
6992 * accumulated for previous mirror configuration. */
6996 m->packet_count += packets;
6997 m->byte_count += bytes;
7001 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
7002 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
7003 * indicate this; newer upstream kernels use gratuitous ARP requests. */
7005 is_gratuitous_arp(const struct flow *flow)
7007 return (flow->dl_type == htons(ETH_TYPE_ARP)
7008 && eth_addr_is_broadcast(flow->dl_dst)
7009 && (flow->nw_proto == ARP_OP_REPLY
7010 || (flow->nw_proto == ARP_OP_REQUEST
7011 && flow->nw_src == flow->nw_dst)));
7015 update_learning_table(struct ofproto_dpif *ofproto,
7016 const struct flow *flow, int vlan,
7017 struct ofbundle *in_bundle)
7019 struct mac_entry *mac;
7021 /* Don't learn the OFPP_NONE port. */
7022 if (in_bundle == &ofpp_none_bundle) {
7026 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7030 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7031 if (is_gratuitous_arp(flow)) {
7032 /* We don't want to learn from gratuitous ARP packets that are
7033 * reflected back over bond slaves so we lock the learning table. */
7034 if (!in_bundle->bond) {
7035 mac_entry_set_grat_arp_lock(mac);
7036 } else if (mac_entry_is_grat_arp_locked(mac)) {
7041 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7042 /* The log messages here could actually be useful in debugging,
7043 * so keep the rate limit relatively high. */
7044 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7045 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7046 "on port %s in VLAN %d",
7047 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7048 in_bundle->name, vlan);
7050 mac->port.p = in_bundle;
7051 tag_set_add(&ofproto->backer->revalidate_set,
7052 mac_learning_changed(ofproto->ml, mac));
7056 static struct ofbundle *
7057 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7058 bool warn, struct ofport_dpif **in_ofportp)
7060 struct ofport_dpif *ofport;
7062 /* Find the port and bundle for the received packet. */
7063 ofport = get_ofp_port(ofproto, in_port);
7065 *in_ofportp = ofport;
7067 if (ofport && ofport->bundle) {
7068 return ofport->bundle;
7071 /* Special-case OFPP_NONE, which a controller may use as the ingress
7072 * port for traffic that it is sourcing. */
7073 if (in_port == OFPP_NONE) {
7074 return &ofpp_none_bundle;
7077 /* Odd. A few possible reasons here:
7079 * - We deleted a port but there are still a few packets queued up
7082 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7083 * we don't know about.
7085 * - The ofproto client didn't configure the port as part of a bundle.
7086 * This is particularly likely to happen if a packet was received on the
7087 * port after it was created, but before the client had a chance to
7088 * configure its bundle.
7091 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7093 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7094 "port %"PRIu16, ofproto->up.name, in_port);
7099 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7100 * dropped. Returns true if they may be forwarded, false if they should be
7103 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7104 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7106 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7107 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7108 * checked by input_vid_is_valid().
7110 * May also add tags to '*tags', although the current implementation only does
7111 * so in one special case.
7114 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
7117 struct ofproto_dpif *ofproto = ctx->ofproto;
7118 struct flow *flow = &ctx->flow;
7119 struct ofbundle *in_bundle = in_port->bundle;
7121 /* Drop frames for reserved multicast addresses
7122 * only if forward_bpdu option is absent. */
7123 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7124 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7128 if (in_bundle->bond) {
7129 struct mac_entry *mac;
7131 switch (bond_check_admissibility(in_bundle->bond, in_port,
7132 flow->dl_dst, &ctx->tags)) {
7137 xlate_report(ctx, "bonding refused admissibility, dropping");
7140 case BV_DROP_IF_MOVED:
7141 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7142 if (mac && mac->port.p != in_bundle &&
7143 (!is_gratuitous_arp(flow)
7144 || mac_entry_is_grat_arp_locked(mac))) {
7145 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7157 xlate_normal(struct action_xlate_ctx *ctx)
7159 struct ofport_dpif *in_port;
7160 struct ofbundle *in_bundle;
7161 struct mac_entry *mac;
7165 ctx->has_normal = true;
7167 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
7168 ctx->packet != NULL, &in_port);
7170 xlate_report(ctx, "no input bundle, dropping");
7174 /* Drop malformed frames. */
7175 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7176 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
7177 if (ctx->packet != NULL) {
7178 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7179 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7180 "VLAN tag received on port %s",
7181 ctx->ofproto->up.name, in_bundle->name);
7183 xlate_report(ctx, "partial VLAN tag, dropping");
7187 /* Drop frames on bundles reserved for mirroring. */
7188 if (in_bundle->mirror_out) {
7189 if (ctx->packet != NULL) {
7190 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7191 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7192 "%s, which is reserved exclusively for mirroring",
7193 ctx->ofproto->up.name, in_bundle->name);
7195 xlate_report(ctx, "input port is mirror output port, dropping");
7200 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
7201 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7202 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7205 vlan = input_vid_to_vlan(in_bundle, vid);
7207 /* Check other admissibility requirements. */
7208 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7212 /* Learn source MAC. */
7213 if (ctx->may_learn) {
7214 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
7217 /* Determine output bundle. */
7218 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
7221 if (mac->port.p != in_bundle) {
7222 xlate_report(ctx, "forwarding to learned port");
7223 output_normal(ctx, mac->port.p, vlan);
7225 xlate_report(ctx, "learned port is input port, dropping");
7228 struct ofbundle *bundle;
7230 xlate_report(ctx, "no learned MAC for destination, flooding");
7231 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7232 if (bundle != in_bundle
7233 && ofbundle_includes_vlan(bundle, vlan)
7234 && bundle->floodable
7235 && !bundle->mirror_out) {
7236 output_normal(ctx, bundle, vlan);
7239 ctx->nf_output_iface = NF_OUT_FLOOD;
7243 /* Optimized flow revalidation.
7245 * It's a difficult problem, in general, to tell which facets need to have
7246 * their actions recalculated whenever the OpenFlow flow table changes. We
7247 * don't try to solve that general problem: for most kinds of OpenFlow flow
7248 * table changes, we recalculate the actions for every facet. This is
7249 * relatively expensive, but it's good enough if the OpenFlow flow table
7250 * doesn't change very often.
7252 * However, we can expect one particular kind of OpenFlow flow table change to
7253 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7254 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7255 * table, we add a special case that applies to flow tables in which every rule
7256 * has the same form (that is, the same wildcards), except that the table is
7257 * also allowed to have a single "catch-all" flow that matches all packets. We
7258 * optimize this case by tagging all of the facets that resubmit into the table
7259 * and invalidating the same tag whenever a flow changes in that table. The
7260 * end result is that we revalidate just the facets that need it (and sometimes
7261 * a few more, but not all of the facets or even all of the facets that
7262 * resubmit to the table modified by MAC learning). */
7264 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7265 * into an OpenFlow table with the given 'basis'. */
7267 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7270 if (minimask_is_catchall(mask)) {
7273 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7274 return tag_create_deterministic(hash);
7278 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7279 * taggability of that table.
7281 * This function must be called after *each* change to a flow table. If you
7282 * skip calling it on some changes then the pointer comparisons at the end can
7283 * be invalid if you get unlucky. For example, if a flow removal causes a
7284 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7285 * different wildcards to be created with the same address, then this function
7286 * will incorrectly skip revalidation. */
7288 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7290 struct table_dpif *table = &ofproto->tables[table_id];
7291 const struct oftable *oftable = &ofproto->up.tables[table_id];
7292 struct cls_table *catchall, *other;
7293 struct cls_table *t;
7295 catchall = other = NULL;
7297 switch (hmap_count(&oftable->cls.tables)) {
7299 /* We could tag this OpenFlow table but it would make the logic a
7300 * little harder and it's a corner case that doesn't seem worth it
7306 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7307 if (cls_table_is_catchall(t)) {
7309 } else if (!other) {
7312 /* Indicate that we can't tag this by setting both tables to
7313 * NULL. (We know that 'catchall' is already NULL.) */
7320 /* Can't tag this table. */
7324 if (table->catchall_table != catchall || table->other_table != other) {
7325 table->catchall_table = catchall;
7326 table->other_table = other;
7327 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7331 /* Given 'rule' that has changed in some way (either it is a rule being
7332 * inserted, a rule being deleted, or a rule whose actions are being
7333 * modified), marks facets for revalidation to ensure that packets will be
7334 * forwarded correctly according to the new state of the flow table.
7336 * This function must be called after *each* change to a flow table. See
7337 * the comment on table_update_taggable() for more information. */
7339 rule_invalidate(const struct rule_dpif *rule)
7341 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7343 table_update_taggable(ofproto, rule->up.table_id);
7345 if (!ofproto->backer->need_revalidate) {
7346 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7348 if (table->other_table && rule->tag) {
7349 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7351 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7357 set_frag_handling(struct ofproto *ofproto_,
7358 enum ofp_config_flags frag_handling)
7360 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7361 if (frag_handling != OFPC_FRAG_REASM) {
7362 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7370 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7371 const struct flow *flow,
7372 const struct ofpact *ofpacts, size_t ofpacts_len)
7374 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7375 struct odputil_keybuf keybuf;
7376 struct dpif_flow_stats stats;
7380 struct action_xlate_ctx ctx;
7381 uint64_t odp_actions_stub[1024 / 8];
7382 struct ofpbuf odp_actions;
7384 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7385 odp_flow_key_from_flow(&key, flow,
7386 ofp_port_to_odp_port(ofproto, flow->in_port));
7388 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7390 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL,
7391 packet_get_tcp_flags(packet, flow), packet);
7392 ctx.resubmit_stats = &stats;
7394 ofpbuf_use_stub(&odp_actions,
7395 odp_actions_stub, sizeof odp_actions_stub);
7396 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
7397 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7398 odp_actions.data, odp_actions.size, packet);
7399 ofpbuf_uninit(&odp_actions);
7407 set_netflow(struct ofproto *ofproto_,
7408 const struct netflow_options *netflow_options)
7410 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7412 if (netflow_options) {
7413 if (!ofproto->netflow) {
7414 ofproto->netflow = netflow_create();
7416 return netflow_set_options(ofproto->netflow, netflow_options);
7418 netflow_destroy(ofproto->netflow);
7419 ofproto->netflow = NULL;
7425 get_netflow_ids(const struct ofproto *ofproto_,
7426 uint8_t *engine_type, uint8_t *engine_id)
7428 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7430 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7434 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7436 if (!facet_is_controller_flow(facet) &&
7437 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7438 struct subfacet *subfacet;
7439 struct ofexpired expired;
7441 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7442 if (subfacet->path == SF_FAST_PATH) {
7443 struct dpif_flow_stats stats;
7445 subfacet_reinstall(subfacet, &stats);
7446 subfacet_update_stats(subfacet, &stats);
7450 expired.flow = facet->flow;
7451 expired.packet_count = facet->packet_count;
7452 expired.byte_count = facet->byte_count;
7453 expired.used = facet->used;
7454 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
7459 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7461 struct facet *facet;
7463 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7464 send_active_timeout(ofproto, facet);
7468 static struct ofproto_dpif *
7469 ofproto_dpif_lookup(const char *name)
7471 struct ofproto_dpif *ofproto;
7473 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7474 hash_string(name, 0), &all_ofproto_dpifs) {
7475 if (!strcmp(ofproto->up.name, name)) {
7483 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7484 const char *argv[], void *aux OVS_UNUSED)
7486 struct ofproto_dpif *ofproto;
7489 ofproto = ofproto_dpif_lookup(argv[1]);
7491 unixctl_command_reply_error(conn, "no such bridge");
7494 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7496 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7497 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7501 unixctl_command_reply(conn, "table successfully flushed");
7505 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7506 const char *argv[], void *aux OVS_UNUSED)
7508 struct ds ds = DS_EMPTY_INITIALIZER;
7509 const struct ofproto_dpif *ofproto;
7510 const struct mac_entry *e;
7512 ofproto = ofproto_dpif_lookup(argv[1]);
7514 unixctl_command_reply_error(conn, "no such bridge");
7518 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7519 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7520 struct ofbundle *bundle = e->port.p;
7521 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7522 ofbundle_get_a_port(bundle)->odp_port,
7523 e->vlan, ETH_ADDR_ARGS(e->mac),
7524 mac_entry_age(ofproto->ml, e));
7526 unixctl_command_reply(conn, ds_cstr(&ds));
7531 struct action_xlate_ctx ctx;
7537 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7538 const struct rule_dpif *rule)
7540 ds_put_char_multiple(result, '\t', level);
7542 ds_put_cstr(result, "No match\n");
7546 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7547 table_id, ntohll(rule->up.flow_cookie));
7548 cls_rule_format(&rule->up.cr, result);
7549 ds_put_char(result, '\n');
7551 ds_put_char_multiple(result, '\t', level);
7552 ds_put_cstr(result, "OpenFlow ");
7553 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7554 ds_put_char(result, '\n');
7558 trace_format_flow(struct ds *result, int level, const char *title,
7559 struct trace_ctx *trace)
7561 ds_put_char_multiple(result, '\t', level);
7562 ds_put_format(result, "%s: ", title);
7563 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
7564 ds_put_cstr(result, "unchanged");
7566 flow_format(result, &trace->ctx.flow);
7567 trace->flow = trace->ctx.flow;
7569 ds_put_char(result, '\n');
7573 trace_format_regs(struct ds *result, int level, const char *title,
7574 struct trace_ctx *trace)
7578 ds_put_char_multiple(result, '\t', level);
7579 ds_put_format(result, "%s:", title);
7580 for (i = 0; i < FLOW_N_REGS; i++) {
7581 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7583 ds_put_char(result, '\n');
7587 trace_format_odp(struct ds *result, int level, const char *title,
7588 struct trace_ctx *trace)
7590 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
7592 ds_put_char_multiple(result, '\t', level);
7593 ds_put_format(result, "%s: ", title);
7594 format_odp_actions(result, odp_actions->data, odp_actions->size);
7595 ds_put_char(result, '\n');
7599 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
7601 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7602 struct ds *result = trace->result;
7604 ds_put_char(result, '\n');
7605 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7606 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
7607 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
7608 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
7612 trace_report(struct action_xlate_ctx *ctx, const char *s)
7614 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7615 struct ds *result = trace->result;
7617 ds_put_char_multiple(result, '\t', ctx->recurse);
7618 ds_put_cstr(result, s);
7619 ds_put_char(result, '\n');
7623 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
7624 void *aux OVS_UNUSED)
7626 const char *dpname = argv[1];
7627 struct ofproto_dpif *ofproto;
7628 struct ofpbuf odp_key;
7629 struct ofpbuf *packet;
7630 ovs_be16 initial_tci;
7636 ofpbuf_init(&odp_key, 0);
7639 ofproto = ofproto_dpif_lookup(dpname);
7641 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
7645 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
7646 /* ofproto/trace dpname flow [-generate] */
7647 const char *flow_s = argv[2];
7648 const char *generate_s = argv[3];
7650 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
7651 * flow. We guess which type it is based on whether 'flow_s' contains
7652 * an '(', since a datapath flow always contains '(') but an
7653 * OpenFlow-like flow should not (in fact it's allowed but I believe
7654 * that's not documented anywhere).
7656 * An alternative would be to try to parse 'flow_s' both ways, but then
7657 * it would be tricky giving a sensible error message. After all, do
7658 * you just say "syntax error" or do you present both error messages?
7659 * Both choices seem lousy. */
7660 if (strchr(flow_s, '(')) {
7663 /* Convert string to datapath key. */
7664 ofpbuf_init(&odp_key, 0);
7665 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
7667 unixctl_command_reply_error(conn, "Bad flow syntax");
7671 /* XXX: Since we allow the user to specify an ofproto, it's
7672 * possible they will specify a different ofproto than the one the
7673 * port actually belongs too. Ideally we should simply remove the
7674 * ability to specify the ofproto. */
7675 if (ofproto_receive(ofproto->backer, NULL, odp_key.data,
7676 odp_key.size, &flow, NULL, NULL, NULL,
7678 unixctl_command_reply_error(conn, "Invalid flow");
7684 error_s = parse_ofp_exact_flow(&flow, argv[2]);
7686 unixctl_command_reply_error(conn, error_s);
7691 initial_tci = flow.vlan_tci;
7694 /* Generate a packet, if requested. */
7696 packet = ofpbuf_new(0);
7697 flow_compose(packet, &flow);
7699 } else if (argc == 7) {
7700 /* ofproto/trace dpname priority tun_id in_port mark packet */
7701 const char *priority_s = argv[2];
7702 const char *tun_id_s = argv[3];
7703 const char *in_port_s = argv[4];
7704 const char *mark_s = argv[5];
7705 const char *packet_s = argv[6];
7706 uint32_t in_port = atoi(in_port_s);
7707 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
7708 uint32_t priority = atoi(priority_s);
7709 uint32_t mark = atoi(mark_s);
7712 msg = eth_from_hex(packet_s, &packet);
7714 unixctl_command_reply_error(conn, msg);
7718 ds_put_cstr(&result, "Packet: ");
7719 s = ofp_packet_to_string(packet->data, packet->size);
7720 ds_put_cstr(&result, s);
7723 flow_extract(packet, priority, mark, NULL, in_port, &flow);
7724 flow.tunnel.tun_id = tun_id;
7725 initial_tci = flow.vlan_tci;
7727 unixctl_command_reply_error(conn, "Bad command syntax");
7731 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
7732 unixctl_command_reply(conn, ds_cstr(&result));
7735 ds_destroy(&result);
7736 ofpbuf_delete(packet);
7737 ofpbuf_uninit(&odp_key);
7741 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
7742 const struct ofpbuf *packet, ovs_be16 initial_tci,
7745 struct rule_dpif *rule;
7747 ds_put_cstr(ds, "Flow: ");
7748 flow_format(ds, flow);
7749 ds_put_char(ds, '\n');
7751 rule = rule_dpif_lookup(ofproto, flow);
7753 trace_format_rule(ds, 0, 0, rule);
7754 if (rule == ofproto->miss_rule) {
7755 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
7756 } else if (rule == ofproto->no_packet_in_rule) {
7757 ds_put_cstr(ds, "\nNo match, packets dropped because "
7758 "OFPPC_NO_PACKET_IN is set on in_port.\n");
7762 uint64_t odp_actions_stub[1024 / 8];
7763 struct ofpbuf odp_actions;
7765 struct trace_ctx trace;
7768 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
7771 ofpbuf_use_stub(&odp_actions,
7772 odp_actions_stub, sizeof odp_actions_stub);
7773 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
7774 rule, tcp_flags, packet);
7775 trace.ctx.resubmit_hook = trace_resubmit;
7776 trace.ctx.report_hook = trace_report;
7777 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
7780 ds_put_char(ds, '\n');
7781 trace_format_flow(ds, 0, "Final flow", &trace);
7782 ds_put_cstr(ds, "Datapath actions: ");
7783 format_odp_actions(ds, odp_actions.data, odp_actions.size);
7784 ofpbuf_uninit(&odp_actions);
7786 if (trace.ctx.slow) {
7787 enum slow_path_reason slow;
7789 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
7790 "slow path because it:");
7791 for (slow = trace.ctx.slow; slow; ) {
7792 enum slow_path_reason bit = rightmost_1bit(slow);
7796 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
7799 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
7802 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
7805 ds_put_cstr(ds, "\n\t- Needs in-band special case "
7808 ds_put_cstr(ds, "\n\t (The datapath actions are "
7809 "incomplete--for complete actions, "
7810 "please supply a packet.)");
7813 case SLOW_CONTROLLER:
7814 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
7815 "to the OpenFlow controller.");
7818 ds_put_cstr(ds, "\n\t- Needs more specific matching "
7819 "than the datapath supports.");
7826 if (slow & ~SLOW_MATCH) {
7827 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
7828 "the special slow-path processing.");
7835 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7836 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7839 unixctl_command_reply(conn, NULL);
7843 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7844 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7847 unixctl_command_reply(conn, NULL);
7850 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
7851 * 'reply' describing the results. */
7853 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
7855 struct facet *facet;
7859 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7860 if (!facet_check_consistency(facet)) {
7865 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
7869 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
7870 ofproto->up.name, errors);
7872 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
7877 ofproto_dpif_self_check(struct unixctl_conn *conn,
7878 int argc, const char *argv[], void *aux OVS_UNUSED)
7880 struct ds reply = DS_EMPTY_INITIALIZER;
7881 struct ofproto_dpif *ofproto;
7884 ofproto = ofproto_dpif_lookup(argv[1]);
7886 unixctl_command_reply_error(conn, "Unknown ofproto (use "
7887 "ofproto/list for help)");
7890 ofproto_dpif_self_check__(ofproto, &reply);
7892 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7893 ofproto_dpif_self_check__(ofproto, &reply);
7897 unixctl_command_reply(conn, ds_cstr(&reply));
7901 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
7902 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
7903 * to destroy 'ofproto_shash' and free the returned value. */
7904 static const struct shash_node **
7905 get_ofprotos(struct shash *ofproto_shash)
7907 const struct ofproto_dpif *ofproto;
7909 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7910 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
7911 shash_add_nocopy(ofproto_shash, name, ofproto);
7914 return shash_sort(ofproto_shash);
7918 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
7919 const char *argv[] OVS_UNUSED,
7920 void *aux OVS_UNUSED)
7922 struct ds ds = DS_EMPTY_INITIALIZER;
7923 struct shash ofproto_shash;
7924 const struct shash_node **sorted_ofprotos;
7927 shash_init(&ofproto_shash);
7928 sorted_ofprotos = get_ofprotos(&ofproto_shash);
7929 for (i = 0; i < shash_count(&ofproto_shash); i++) {
7930 const struct shash_node *node = sorted_ofprotos[i];
7931 ds_put_format(&ds, "%s\n", node->name);
7934 shash_destroy(&ofproto_shash);
7935 free(sorted_ofprotos);
7937 unixctl_command_reply(conn, ds_cstr(&ds));
7942 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
7944 struct dpif_dp_stats s;
7945 const struct shash_node **ports;
7948 dpif_get_dp_stats(ofproto->backer->dpif, &s);
7950 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
7951 dpif_name(ofproto->backer->dpif));
7952 /* xxx It would be better to show bridge-specific stats instead
7953 * xxx of dp ones. */
7955 "\tlookups: hit:%"PRIu64" missed:%"PRIu64" lost:%"PRIu64"\n",
7956 s.n_hit, s.n_missed, s.n_lost);
7957 ds_put_format(ds, "\tflows: %zu\n",
7958 hmap_count(&ofproto->subfacets));
7960 ports = shash_sort(&ofproto->up.port_by_name);
7961 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
7962 const struct shash_node *node = ports[i];
7963 struct ofport *ofport = node->data;
7964 const char *name = netdev_get_name(ofport->netdev);
7965 const char *type = netdev_get_type(ofport->netdev);
7968 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
7970 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
7971 if (odp_port != OVSP_NONE) {
7972 ds_put_format(ds, "%"PRIu32":", odp_port);
7974 ds_put_cstr(ds, "none:");
7977 if (strcmp(type, "system")) {
7978 struct netdev *netdev;
7981 ds_put_format(ds, " (%s", type);
7983 error = netdev_open(name, type, &netdev);
7988 error = netdev_get_config(netdev, &config);
7990 const struct smap_node **nodes;
7993 nodes = smap_sort(&config);
7994 for (i = 0; i < smap_count(&config); i++) {
7995 const struct smap_node *node = nodes[i];
7996 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
7997 node->key, node->value);
8001 smap_destroy(&config);
8003 netdev_close(netdev);
8005 ds_put_char(ds, ')');
8007 ds_put_char(ds, '\n');
8013 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
8014 const char *argv[], void *aux OVS_UNUSED)
8016 struct ds ds = DS_EMPTY_INITIALIZER;
8017 const struct ofproto_dpif *ofproto;
8021 for (i = 1; i < argc; i++) {
8022 ofproto = ofproto_dpif_lookup(argv[i]);
8024 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
8025 "for help)", argv[i]);
8026 unixctl_command_reply_error(conn, ds_cstr(&ds));
8029 show_dp_format(ofproto, &ds);
8032 struct shash ofproto_shash;
8033 const struct shash_node **sorted_ofprotos;
8036 shash_init(&ofproto_shash);
8037 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8038 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8039 const struct shash_node *node = sorted_ofprotos[i];
8040 show_dp_format(node->data, &ds);
8043 shash_destroy(&ofproto_shash);
8044 free(sorted_ofprotos);
8047 unixctl_command_reply(conn, ds_cstr(&ds));
8052 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8053 int argc OVS_UNUSED, const char *argv[],
8054 void *aux OVS_UNUSED)
8056 struct ds ds = DS_EMPTY_INITIALIZER;
8057 const struct ofproto_dpif *ofproto;
8058 struct subfacet *subfacet;
8060 ofproto = ofproto_dpif_lookup(argv[1]);
8062 unixctl_command_reply_error(conn, "no such bridge");
8066 update_stats(ofproto->backer);
8068 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8069 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
8071 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8072 subfacet->dp_packet_count, subfacet->dp_byte_count);
8073 if (subfacet->used) {
8074 ds_put_format(&ds, "%.3fs",
8075 (time_msec() - subfacet->used) / 1000.0);
8077 ds_put_format(&ds, "never");
8079 if (subfacet->facet->tcp_flags) {
8080 ds_put_cstr(&ds, ", flags:");
8081 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8084 ds_put_cstr(&ds, ", actions:");
8085 format_odp_actions(&ds, subfacet->actions, subfacet->actions_len);
8086 ds_put_char(&ds, '\n');
8089 unixctl_command_reply(conn, ds_cstr(&ds));
8094 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8095 int argc OVS_UNUSED, const char *argv[],
8096 void *aux OVS_UNUSED)
8098 struct ds ds = DS_EMPTY_INITIALIZER;
8099 struct ofproto_dpif *ofproto;
8101 ofproto = ofproto_dpif_lookup(argv[1]);
8103 unixctl_command_reply_error(conn, "no such bridge");
8107 flush(&ofproto->up);
8109 unixctl_command_reply(conn, ds_cstr(&ds));
8114 ofproto_dpif_unixctl_init(void)
8116 static bool registered;
8122 unixctl_command_register(
8124 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
8125 2, 6, ofproto_unixctl_trace, NULL);
8126 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8127 ofproto_unixctl_fdb_flush, NULL);
8128 unixctl_command_register("fdb/show", "bridge", 1, 1,
8129 ofproto_unixctl_fdb_show, NULL);
8130 unixctl_command_register("ofproto/clog", "", 0, 0,
8131 ofproto_dpif_clog, NULL);
8132 unixctl_command_register("ofproto/unclog", "", 0, 0,
8133 ofproto_dpif_unclog, NULL);
8134 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8135 ofproto_dpif_self_check, NULL);
8136 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8137 ofproto_unixctl_dpif_dump_dps, NULL);
8138 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8139 ofproto_unixctl_dpif_show, NULL);
8140 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8141 ofproto_unixctl_dpif_dump_flows, NULL);
8142 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8143 ofproto_unixctl_dpif_del_flows, NULL);
8146 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8148 * This is deprecated. It is only for compatibility with broken device drivers
8149 * in old versions of Linux that do not properly support VLANs when VLAN
8150 * devices are not used. When broken device drivers are no longer in
8151 * widespread use, we will delete these interfaces. */
8154 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8156 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8157 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8159 if (realdev_ofp_port == ofport->realdev_ofp_port
8160 && vid == ofport->vlandev_vid) {
8164 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8166 if (ofport->realdev_ofp_port) {
8169 if (realdev_ofp_port && ofport->bundle) {
8170 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8171 * themselves be part of a bundle. */
8172 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8175 ofport->realdev_ofp_port = realdev_ofp_port;
8176 ofport->vlandev_vid = vid;
8178 if (realdev_ofp_port) {
8179 vsp_add(ofport, realdev_ofp_port, vid);
8186 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8188 return hash_2words(realdev_ofp_port, vid);
8191 /* Returns the ODP port number of the Linux VLAN device that corresponds to
8192 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
8193 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
8194 * it would return the port number of eth0.9.
8196 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
8197 * function just returns its 'realdev_odp_port' argument. */
8199 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8200 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
8202 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8203 uint16_t realdev_ofp_port;
8204 int vid = vlan_tci_to_vid(vlan_tci);
8205 const struct vlan_splinter *vsp;
8207 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
8208 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8209 hash_realdev_vid(realdev_ofp_port, vid),
8210 &ofproto->realdev_vid_map) {
8211 if (vsp->realdev_ofp_port == realdev_ofp_port
8212 && vsp->vid == vid) {
8213 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
8217 return realdev_odp_port;
8220 static struct vlan_splinter *
8221 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8223 struct vlan_splinter *vsp;
8225 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8226 &ofproto->vlandev_map) {
8227 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8235 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8236 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8237 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8238 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8239 * eth0 and store 9 in '*vid'.
8241 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8242 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8245 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8246 uint16_t vlandev_ofp_port, int *vid)
8248 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8249 const struct vlan_splinter *vsp;
8251 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8256 return vsp->realdev_ofp_port;
8262 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8263 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8264 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8265 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8266 * always the case unless VLAN splinters are enabled), returns false without
8267 * making any changes. */
8269 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8274 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8279 /* Cause the flow to be processed as if it came in on the real device with
8280 * the VLAN device's VLAN ID. */
8281 flow->in_port = realdev;
8282 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8287 vsp_remove(struct ofport_dpif *port)
8289 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8290 struct vlan_splinter *vsp;
8292 vsp = vlandev_find(ofproto, port->up.ofp_port);
8294 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8295 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8298 port->realdev_ofp_port = 0;
8300 VLOG_ERR("missing vlan device record");
8305 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8307 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8309 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8310 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8311 == realdev_ofp_port)) {
8312 struct vlan_splinter *vsp;
8314 vsp = xmalloc(sizeof *vsp);
8315 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8316 hash_int(port->up.ofp_port, 0));
8317 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8318 hash_realdev_vid(realdev_ofp_port, vid));
8319 vsp->realdev_ofp_port = realdev_ofp_port;
8320 vsp->vlandev_ofp_port = port->up.ofp_port;
8323 port->realdev_ofp_port = realdev_ofp_port;
8325 VLOG_ERR("duplicate vlan device record");
8330 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8332 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8333 return ofport ? ofport->odp_port : OVSP_NONE;
8336 static struct ofport_dpif *
8337 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8339 struct ofport_dpif *port;
8341 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8342 hash_int(odp_port, 0),
8343 &backer->odp_to_ofport_map) {
8344 if (port->odp_port == odp_port) {
8353 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8355 struct ofport_dpif *port;
8357 port = odp_port_to_ofport(ofproto->backer, odp_port);
8358 if (port && &ofproto->up == port->up.ofproto) {
8359 return port->up.ofp_port;
8365 const struct ofproto_class ofproto_dpif_class = {
8400 port_is_lacp_current,
8401 NULL, /* rule_choose_table */
8408 rule_modify_actions,
8419 get_stp_port_status,
8426 is_mirror_output_bundle,
8427 forward_bpdu_changed,
8428 set_mac_table_config,