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 /* Initial values of fields of the packet that may be changed during
294 * flow processing and needed later. */
295 struct initial_vals {
296 /* This is the value of vlan_tci in the packet as actually received from
297 * dpif. This is the same as the facet's flow.vlan_tci unless the packet
298 * was received via a VLAN splinter. In that case, this value is 0
299 * (because the packet as actually received from the dpif had no 802.1Q
300 * tag) but the facet's flow.vlan_tci is set to the VLAN that the splinter
303 * This member should be removed when the VLAN splinters feature is no
308 static void action_xlate_ctx_init(struct action_xlate_ctx *,
309 struct ofproto_dpif *, const struct flow *,
310 const struct initial_vals *initial_vals,
312 uint8_t tcp_flags, const struct ofpbuf *);
313 static void xlate_actions(struct action_xlate_ctx *,
314 const struct ofpact *ofpacts, size_t ofpacts_len,
315 struct ofpbuf *odp_actions);
316 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
317 const struct ofpact *ofpacts,
319 static void xlate_table_action(struct action_xlate_ctx *, uint16_t in_port,
320 uint8_t table_id, bool may_packet_in);
322 static size_t put_userspace_action(const struct ofproto_dpif *,
323 struct ofpbuf *odp_actions,
325 const union user_action_cookie *);
327 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
328 enum slow_path_reason,
329 uint64_t *stub, size_t stub_size,
330 const struct nlattr **actionsp,
331 size_t *actions_lenp);
333 static void xlate_report(struct action_xlate_ctx *ctx, const char *s);
335 /* A subfacet (see "struct subfacet" below) has three possible installation
338 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
339 * case just after the subfacet is created, just before the subfacet is
340 * destroyed, or if the datapath returns an error when we try to install a
343 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
345 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
346 * ofproto_dpif is installed in the datapath.
349 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
350 SF_FAST_PATH, /* Full actions are installed. */
351 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
354 static const char *subfacet_path_to_string(enum subfacet_path);
356 /* A dpif flow and actions associated with a facet.
358 * See also the large comment on struct facet. */
361 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
362 struct list list_node; /* In struct facet's 'facets' list. */
363 struct facet *facet; /* Owning facet. */
365 enum odp_key_fitness key_fitness;
369 long long int used; /* Time last used; time created if not used. */
371 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
372 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
376 * These should be essentially identical for every subfacet in a facet, but
377 * may differ in trivial ways due to VLAN splinters. */
378 size_t actions_len; /* Number of bytes in actions[]. */
379 struct nlattr *actions; /* Datapath actions. */
381 enum slow_path_reason slow; /* 0 if fast path may be used. */
382 enum subfacet_path path; /* Installed in datapath? */
384 /* Initial values of the packet that may be needed later. */
385 struct initial_vals initial_vals;
387 /* Datapath port the packet arrived on. This is needed to remove
388 * flows for ports that are no longer part of the bridge. Since the
389 * flow definition only has the OpenFlow port number and the port is
390 * no longer part of the bridge, we can't determine the datapath port
391 * number needed to delete the flow from the datapath. */
392 uint32_t odp_in_port;
395 #define SUBFACET_DESTROY_MAX_BATCH 50
397 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
399 static struct subfacet *subfacet_find(struct ofproto_dpif *,
400 const struct nlattr *key, size_t key_len,
402 static void subfacet_destroy(struct subfacet *);
403 static void subfacet_destroy__(struct subfacet *);
404 static void subfacet_destroy_batch(struct ofproto_dpif *,
405 struct subfacet **, int n);
406 static void subfacet_reset_dp_stats(struct subfacet *,
407 struct dpif_flow_stats *);
408 static void subfacet_update_time(struct subfacet *, long long int used);
409 static void subfacet_update_stats(struct subfacet *,
410 const struct dpif_flow_stats *);
411 static void subfacet_make_actions(struct subfacet *,
412 const struct ofpbuf *packet,
413 struct ofpbuf *odp_actions);
414 static int subfacet_install(struct subfacet *,
415 const struct nlattr *actions, size_t actions_len,
416 struct dpif_flow_stats *, enum slow_path_reason);
417 static void subfacet_uninstall(struct subfacet *);
419 static enum subfacet_path subfacet_want_path(enum slow_path_reason);
421 /* An exact-match instantiation of an OpenFlow flow.
423 * A facet associates a "struct flow", which represents the Open vSwitch
424 * userspace idea of an exact-match flow, with one or more subfacets. Each
425 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
426 * the facet. When the kernel module (or other dpif implementation) and Open
427 * vSwitch userspace agree on the definition of a flow key, there is exactly
428 * one subfacet per facet. If the dpif implementation supports more-specific
429 * flow matching than userspace, however, a facet can have more than one
430 * subfacet, each of which corresponds to some distinction in flow that
431 * userspace simply doesn't understand.
433 * Flow expiration works in terms of subfacets, so a facet must have at least
434 * one subfacet or it will never expire, leaking memory. */
437 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
438 struct list list_node; /* In owning rule's 'facets' list. */
439 struct rule_dpif *rule; /* Owning rule. */
442 struct list subfacets;
443 long long int used; /* Time last used; time created if not used. */
450 * - Do include packets and bytes sent "by hand", e.g. with
453 * - Do include packets and bytes that were obtained from the datapath
454 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
455 * DPIF_FP_ZERO_STATS).
457 * - Do not include packets or bytes that can be obtained from the
458 * datapath for any existing subfacet.
460 uint64_t packet_count; /* Number of packets received. */
461 uint64_t byte_count; /* Number of bytes received. */
463 /* Resubmit statistics. */
464 uint64_t prev_packet_count; /* Number of packets from last stats push. */
465 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
466 long long int prev_used; /* Used time from last stats push. */
469 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
470 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
471 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
473 /* Properties of datapath actions.
475 * Every subfacet has its own actions because actions can differ slightly
476 * between splintered and non-splintered subfacets due to the VLAN tag
477 * being initially different (present vs. absent). All of them have these
478 * properties in common so we just store one copy of them here. */
479 bool has_learn; /* Actions include NXAST_LEARN? */
480 bool has_normal; /* Actions output to OFPP_NORMAL? */
481 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
482 tag_type tags; /* Tags that would require revalidation. */
483 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
485 /* Storage for a single subfacet, to reduce malloc() time and space
486 * overhead. (A facet always has at least one subfacet and in the common
487 * case has exactly one subfacet.) */
488 struct subfacet one_subfacet;
491 static struct facet *facet_create(struct rule_dpif *,
492 const struct flow *, uint32_t hash);
493 static void facet_remove(struct facet *);
494 static void facet_free(struct facet *);
496 static struct facet *facet_find(struct ofproto_dpif *,
497 const struct flow *, uint32_t hash);
498 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
499 const struct flow *, uint32_t hash);
500 static void facet_revalidate(struct facet *);
501 static bool facet_check_consistency(struct facet *);
503 static void facet_flush_stats(struct facet *);
505 static void facet_update_time(struct facet *, long long int used);
506 static void facet_reset_counters(struct facet *);
507 static void facet_push_stats(struct facet *);
508 static void facet_learn(struct facet *);
509 static void facet_account(struct facet *);
511 static bool facet_is_controller_flow(struct facet *);
514 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
518 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
519 struct list bundle_node; /* In struct ofbundle's "ports" list. */
520 struct cfm *cfm; /* Connectivity Fault Management, if any. */
521 tag_type tag; /* Tag associated with this port. */
522 bool may_enable; /* May be enabled in bonds. */
523 long long int carrier_seq; /* Carrier status changes. */
524 struct tnl_port *tnl_port; /* Tunnel handle, or null. */
527 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
528 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
529 long long int stp_state_entered;
531 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
533 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
535 * This is deprecated. It is only for compatibility with broken device
536 * drivers in old versions of Linux that do not properly support VLANs when
537 * VLAN devices are not used. When broken device drivers are no longer in
538 * widespread use, we will delete these interfaces. */
539 uint16_t realdev_ofp_port;
543 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
544 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
545 * traffic egressing the 'ofport' with that priority should be marked with. */
546 struct priority_to_dscp {
547 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
548 uint32_t priority; /* Priority of this queue (see struct flow). */
550 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
553 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
555 * This is deprecated. It is only for compatibility with broken device drivers
556 * in old versions of Linux that do not properly support VLANs when VLAN
557 * devices are not used. When broken device drivers are no longer in
558 * widespread use, we will delete these interfaces. */
559 struct vlan_splinter {
560 struct hmap_node realdev_vid_node;
561 struct hmap_node vlandev_node;
562 uint16_t realdev_ofp_port;
563 uint16_t vlandev_ofp_port;
567 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
568 uint32_t realdev, ovs_be16 vlan_tci);
569 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
570 static void vsp_remove(struct ofport_dpif *);
571 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
573 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
575 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
578 static struct ofport_dpif *
579 ofport_dpif_cast(const struct ofport *ofport)
581 ovs_assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
582 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
585 static void port_run(struct ofport_dpif *);
586 static void port_run_fast(struct ofport_dpif *);
587 static void port_wait(struct ofport_dpif *);
588 static int set_cfm(struct ofport *, const struct cfm_settings *);
589 static void ofport_clear_priorities(struct ofport_dpif *);
591 struct dpif_completion {
592 struct list list_node;
593 struct ofoperation *op;
596 /* Extra information about a classifier table.
597 * Currently used just for optimized flow revalidation. */
599 /* If either of these is nonnull, then this table has a form that allows
600 * flows to be tagged to avoid revalidating most flows for the most common
601 * kinds of flow table changes. */
602 struct cls_table *catchall_table; /* Table that wildcards all fields. */
603 struct cls_table *other_table; /* Table with any other wildcard set. */
604 uint32_t basis; /* Keeps each table's tags separate. */
607 /* Reasons that we might need to revalidate every facet, and corresponding
610 * A value of 0 means that there is no need to revalidate.
612 * It would be nice to have some cleaner way to integrate with coverage
613 * counters, but with only a few reasons I guess this is good enough for
615 enum revalidate_reason {
616 REV_RECONFIGURE = 1, /* Switch configuration changed. */
617 REV_STP, /* Spanning tree protocol port status change. */
618 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
619 REV_FLOW_TABLE, /* Flow table changed. */
620 REV_INCONSISTENCY /* Facet self-check failed. */
622 COVERAGE_DEFINE(rev_reconfigure);
623 COVERAGE_DEFINE(rev_stp);
624 COVERAGE_DEFINE(rev_port_toggled);
625 COVERAGE_DEFINE(rev_flow_table);
626 COVERAGE_DEFINE(rev_inconsistency);
628 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
629 * These are datapath flows which have no associated ofproto, if they did we
630 * would use facets. */
632 struct hmap_node hmap_node;
637 /* All datapaths of a given type share a single dpif backer instance. */
642 struct timer next_expiration;
643 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
645 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
647 /* Facet revalidation flags applying to facets which use this backer. */
648 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
649 struct tag_set revalidate_set; /* Revalidate only matching facets. */
651 struct hmap drop_keys; /* Set of dropped odp keys. */
654 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
655 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
657 static void drop_key_clear(struct dpif_backer *);
658 static struct ofport_dpif *
659 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
661 struct ofproto_dpif {
662 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
664 struct dpif_backer *backer;
666 /* Special OpenFlow rules. */
667 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
668 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
674 struct netflow *netflow;
675 struct dpif_sflow *sflow;
676 struct hmap bundles; /* Contains "struct ofbundle"s. */
677 struct mac_learning *ml;
678 struct ofmirror *mirrors[MAX_MIRRORS];
680 bool has_bonded_bundles;
684 struct hmap subfacets;
685 struct governor *governor;
688 struct table_dpif tables[N_TABLES];
690 /* Support for debugging async flow mods. */
691 struct list completions;
693 bool has_bundle_action; /* True when the first bundle action appears. */
694 struct netdev_stats stats; /* To account packets generated and consumed in
699 long long int stp_last_tick;
701 /* VLAN splinters. */
702 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
703 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
706 struct sset ports; /* Set of standard port names. */
707 struct sset ghost_ports; /* Ports with no datapath port. */
708 struct sset port_poll_set; /* Queued names for port_poll() reply. */
709 int port_poll_errno; /* Last errno for port_poll() reply. */
712 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
713 * for debugging the asynchronous flow_mod implementation.) */
716 /* All existing ofproto_dpif instances, indexed by ->up.name. */
717 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
719 static void ofproto_dpif_unixctl_init(void);
721 static struct ofproto_dpif *
722 ofproto_dpif_cast(const struct ofproto *ofproto)
724 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
725 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
728 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
730 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
732 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
733 const struct ofpbuf *,
734 const struct initial_vals *, struct ds *);
736 /* Packet processing. */
737 static void update_learning_table(struct ofproto_dpif *,
738 const struct flow *, int vlan,
741 #define FLOW_MISS_MAX_BATCH 50
742 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
744 /* Flow expiration. */
745 static int expire(struct dpif_backer *);
748 static void send_netflow_active_timeouts(struct ofproto_dpif *);
751 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
752 static size_t compose_sflow_action(const struct ofproto_dpif *,
753 struct ofpbuf *odp_actions,
754 const struct flow *, uint32_t odp_port);
755 static void add_mirror_actions(struct action_xlate_ctx *ctx,
756 const struct flow *flow);
757 /* Global variables. */
758 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
760 /* Initial mappings of port to bridge mappings. */
761 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
763 /* Factory functions. */
766 init(const struct shash *iface_hints)
768 struct shash_node *node;
770 /* Make a local copy, since we don't own 'iface_hints' elements. */
771 SHASH_FOR_EACH(node, iface_hints) {
772 const struct iface_hint *orig_hint = node->data;
773 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
775 new_hint->br_name = xstrdup(orig_hint->br_name);
776 new_hint->br_type = xstrdup(orig_hint->br_type);
777 new_hint->ofp_port = orig_hint->ofp_port;
779 shash_add(&init_ofp_ports, node->name, new_hint);
784 enumerate_types(struct sset *types)
786 dp_enumerate_types(types);
790 enumerate_names(const char *type, struct sset *names)
792 struct ofproto_dpif *ofproto;
795 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
796 if (strcmp(type, ofproto->up.type)) {
799 sset_add(names, ofproto->up.name);
806 del(const char *type, const char *name)
811 error = dpif_open(name, type, &dpif);
813 error = dpif_delete(dpif);
820 port_open_type(const char *datapath_type, const char *port_type)
822 return dpif_port_open_type(datapath_type, port_type);
825 /* Type functions. */
827 static struct ofproto_dpif *
828 lookup_ofproto_dpif_by_port_name(const char *name)
830 struct ofproto_dpif *ofproto;
832 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
833 if (sset_contains(&ofproto->ports, name)) {
842 type_run(const char *type)
844 struct dpif_backer *backer;
848 backer = shash_find_data(&all_dpif_backers, type);
850 /* This is not necessarily a problem, since backers are only
851 * created on demand. */
855 dpif_run(backer->dpif);
857 if (backer->need_revalidate
858 || !tag_set_is_empty(&backer->revalidate_set)) {
859 struct tag_set revalidate_set = backer->revalidate_set;
860 bool need_revalidate = backer->need_revalidate;
861 struct ofproto_dpif *ofproto;
862 struct simap_node *node;
863 struct simap tmp_backers;
865 /* Handle tunnel garbage collection. */
866 simap_init(&tmp_backers);
867 simap_swap(&backer->tnl_backers, &tmp_backers);
869 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
870 struct ofport_dpif *iter;
872 if (backer != ofproto->backer) {
876 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
879 if (!iter->tnl_port) {
883 dp_port = netdev_vport_get_dpif_port(iter->up.netdev);
884 node = simap_find(&tmp_backers, dp_port);
886 simap_put(&backer->tnl_backers, dp_port, node->data);
887 simap_delete(&tmp_backers, node);
888 node = simap_find(&backer->tnl_backers, dp_port);
890 node = simap_find(&backer->tnl_backers, dp_port);
892 uint32_t odp_port = UINT32_MAX;
894 if (!dpif_port_add(backer->dpif, iter->up.netdev,
896 simap_put(&backer->tnl_backers, dp_port, odp_port);
897 node = simap_find(&backer->tnl_backers, dp_port);
902 iter->odp_port = node ? node->data : OVSP_NONE;
903 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
905 backer->need_revalidate = REV_RECONFIGURE;
910 SIMAP_FOR_EACH (node, &tmp_backers) {
911 dpif_port_del(backer->dpif, node->data);
913 simap_destroy(&tmp_backers);
915 switch (backer->need_revalidate) {
916 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
917 case REV_STP: COVERAGE_INC(rev_stp); break;
918 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
919 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
920 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
923 if (backer->need_revalidate) {
924 /* Clear the drop_keys in case we should now be accepting some
925 * formerly dropped flows. */
926 drop_key_clear(backer);
929 /* Clear the revalidation flags. */
930 tag_set_init(&backer->revalidate_set);
931 backer->need_revalidate = 0;
933 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
934 struct facet *facet, *next;
936 if (ofproto->backer != backer) {
940 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
942 || tag_set_intersects(&revalidate_set, facet->tags)) {
943 facet_revalidate(facet);
949 if (timer_expired(&backer->next_expiration)) {
950 int delay = expire(backer);
951 timer_set_duration(&backer->next_expiration, delay);
954 /* Check for port changes in the dpif. */
955 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
956 struct ofproto_dpif *ofproto;
957 struct dpif_port port;
959 /* Don't report on the datapath's device. */
960 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
964 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
965 &all_ofproto_dpifs) {
966 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
971 ofproto = lookup_ofproto_dpif_by_port_name(devname);
972 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
973 /* The port was removed. If we know the datapath,
974 * report it through poll_set(). If we don't, it may be
975 * notifying us of a removal we initiated, so ignore it.
976 * If there's a pending ENOBUFS, let it stand, since
977 * everything will be reevaluated. */
978 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
979 sset_add(&ofproto->port_poll_set, devname);
980 ofproto->port_poll_errno = 0;
982 } else if (!ofproto) {
983 /* The port was added, but we don't know with which
984 * ofproto we should associate it. Delete it. */
985 dpif_port_del(backer->dpif, port.port_no);
987 dpif_port_destroy(&port);
993 if (error != EAGAIN) {
994 struct ofproto_dpif *ofproto;
996 /* There was some sort of error, so propagate it to all
997 * ofprotos that use this backer. */
998 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
999 &all_ofproto_dpifs) {
1000 if (ofproto->backer == backer) {
1001 sset_clear(&ofproto->port_poll_set);
1002 ofproto->port_poll_errno = error;
1011 type_run_fast(const char *type)
1013 struct dpif_backer *backer;
1016 backer = shash_find_data(&all_dpif_backers, type);
1018 /* This is not necessarily a problem, since backers are only
1019 * created on demand. */
1023 /* Handle one or more batches of upcalls, until there's nothing left to do
1024 * or until we do a fixed total amount of work.
1026 * We do work in batches because it can be much cheaper to set up a number
1027 * of flows and fire off their patches all at once. We do multiple batches
1028 * because in some cases handling a packet can cause another packet to be
1029 * queued almost immediately as part of the return flow. Both
1030 * optimizations can make major improvements on some benchmarks and
1031 * presumably for real traffic as well. */
1033 while (work < FLOW_MISS_MAX_BATCH) {
1034 int retval = handle_upcalls(backer, FLOW_MISS_MAX_BATCH - work);
1045 type_wait(const char *type)
1047 struct dpif_backer *backer;
1049 backer = shash_find_data(&all_dpif_backers, type);
1051 /* This is not necessarily a problem, since backers are only
1052 * created on demand. */
1056 timer_wait(&backer->next_expiration);
1059 /* Basic life-cycle. */
1061 static int add_internal_flows(struct ofproto_dpif *);
1063 static struct ofproto *
1066 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1067 return &ofproto->up;
1071 dealloc(struct ofproto *ofproto_)
1073 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1078 close_dpif_backer(struct dpif_backer *backer)
1080 struct shash_node *node;
1082 ovs_assert(backer->refcount > 0);
1084 if (--backer->refcount) {
1088 drop_key_clear(backer);
1089 hmap_destroy(&backer->drop_keys);
1091 simap_destroy(&backer->tnl_backers);
1092 hmap_destroy(&backer->odp_to_ofport_map);
1093 node = shash_find(&all_dpif_backers, backer->type);
1095 shash_delete(&all_dpif_backers, node);
1096 dpif_close(backer->dpif);
1101 /* Datapath port slated for removal from datapath. */
1102 struct odp_garbage {
1103 struct list list_node;
1108 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1110 struct dpif_backer *backer;
1111 struct dpif_port_dump port_dump;
1112 struct dpif_port port;
1113 struct shash_node *node;
1114 struct list garbage_list;
1115 struct odp_garbage *garbage, *next;
1121 backer = shash_find_data(&all_dpif_backers, type);
1128 backer_name = xasprintf("ovs-%s", type);
1130 /* Remove any existing datapaths, since we assume we're the only
1131 * userspace controlling the datapath. */
1133 dp_enumerate_names(type, &names);
1134 SSET_FOR_EACH(name, &names) {
1135 struct dpif *old_dpif;
1137 /* Don't remove our backer if it exists. */
1138 if (!strcmp(name, backer_name)) {
1142 if (dpif_open(name, type, &old_dpif)) {
1143 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1145 dpif_delete(old_dpif);
1146 dpif_close(old_dpif);
1149 sset_destroy(&names);
1151 backer = xmalloc(sizeof *backer);
1153 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1156 VLOG_ERR("failed to open datapath of type %s: %s", type,
1162 backer->type = xstrdup(type);
1163 backer->refcount = 1;
1164 hmap_init(&backer->odp_to_ofport_map);
1165 hmap_init(&backer->drop_keys);
1166 timer_set_duration(&backer->next_expiration, 1000);
1167 backer->need_revalidate = 0;
1168 simap_init(&backer->tnl_backers);
1169 tag_set_init(&backer->revalidate_set);
1172 dpif_flow_flush(backer->dpif);
1174 /* Loop through the ports already on the datapath and remove any
1175 * that we don't need anymore. */
1176 list_init(&garbage_list);
1177 dpif_port_dump_start(&port_dump, backer->dpif);
1178 while (dpif_port_dump_next(&port_dump, &port)) {
1179 node = shash_find(&init_ofp_ports, port.name);
1180 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1181 garbage = xmalloc(sizeof *garbage);
1182 garbage->odp_port = port.port_no;
1183 list_push_front(&garbage_list, &garbage->list_node);
1186 dpif_port_dump_done(&port_dump);
1188 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1189 dpif_port_del(backer->dpif, garbage->odp_port);
1190 list_remove(&garbage->list_node);
1194 shash_add(&all_dpif_backers, type, backer);
1196 error = dpif_recv_set(backer->dpif, true);
1198 VLOG_ERR("failed to listen on datapath of type %s: %s",
1199 type, strerror(error));
1200 close_dpif_backer(backer);
1208 construct(struct ofproto *ofproto_)
1210 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1211 struct shash_node *node, *next;
1216 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1221 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1222 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1224 ofproto->n_matches = 0;
1226 ofproto->netflow = NULL;
1227 ofproto->sflow = NULL;
1228 ofproto->stp = NULL;
1229 hmap_init(&ofproto->bundles);
1230 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1231 for (i = 0; i < MAX_MIRRORS; i++) {
1232 ofproto->mirrors[i] = NULL;
1234 ofproto->has_bonded_bundles = false;
1236 hmap_init(&ofproto->facets);
1237 hmap_init(&ofproto->subfacets);
1238 ofproto->governor = NULL;
1240 for (i = 0; i < N_TABLES; i++) {
1241 struct table_dpif *table = &ofproto->tables[i];
1243 table->catchall_table = NULL;
1244 table->other_table = NULL;
1245 table->basis = random_uint32();
1248 list_init(&ofproto->completions);
1250 ofproto_dpif_unixctl_init();
1252 ofproto->has_mirrors = false;
1253 ofproto->has_bundle_action = false;
1255 hmap_init(&ofproto->vlandev_map);
1256 hmap_init(&ofproto->realdev_vid_map);
1258 sset_init(&ofproto->ports);
1259 sset_init(&ofproto->ghost_ports);
1260 sset_init(&ofproto->port_poll_set);
1261 ofproto->port_poll_errno = 0;
1263 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1264 struct iface_hint *iface_hint = node->data;
1266 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1267 /* Check if the datapath already has this port. */
1268 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1269 sset_add(&ofproto->ports, node->name);
1272 free(iface_hint->br_name);
1273 free(iface_hint->br_type);
1275 shash_delete(&init_ofp_ports, node);
1279 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1280 hash_string(ofproto->up.name, 0));
1281 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1283 ofproto_init_tables(ofproto_, N_TABLES);
1284 error = add_internal_flows(ofproto);
1285 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1291 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1292 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1294 struct ofputil_flow_mod fm;
1297 match_init_catchall(&fm.match);
1299 match_set_reg(&fm.match, 0, id);
1300 fm.new_cookie = htonll(0);
1301 fm.cookie = htonll(0);
1302 fm.cookie_mask = htonll(0);
1303 fm.table_id = TBL_INTERNAL;
1304 fm.command = OFPFC_ADD;
1305 fm.idle_timeout = 0;
1306 fm.hard_timeout = 0;
1310 fm.ofpacts = ofpacts->data;
1311 fm.ofpacts_len = ofpacts->size;
1313 error = ofproto_flow_mod(&ofproto->up, &fm);
1315 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1316 id, ofperr_to_string(error));
1320 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1321 ovs_assert(*rulep != NULL);
1327 add_internal_flows(struct ofproto_dpif *ofproto)
1329 struct ofpact_controller *controller;
1330 uint64_t ofpacts_stub[128 / 8];
1331 struct ofpbuf ofpacts;
1335 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1338 controller = ofpact_put_CONTROLLER(&ofpacts);
1339 controller->max_len = UINT16_MAX;
1340 controller->controller_id = 0;
1341 controller->reason = OFPR_NO_MATCH;
1342 ofpact_pad(&ofpacts);
1344 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1349 ofpbuf_clear(&ofpacts);
1350 error = add_internal_flow(ofproto, id++, &ofpacts,
1351 &ofproto->no_packet_in_rule);
1356 complete_operations(struct ofproto_dpif *ofproto)
1358 struct dpif_completion *c, *next;
1360 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1361 ofoperation_complete(c->op, 0);
1362 list_remove(&c->list_node);
1368 destruct(struct ofproto *ofproto_)
1370 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1371 struct rule_dpif *rule, *next_rule;
1372 struct oftable *table;
1375 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1376 complete_operations(ofproto);
1378 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1379 struct cls_cursor cursor;
1381 cls_cursor_init(&cursor, &table->cls, NULL);
1382 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1383 ofproto_rule_destroy(&rule->up);
1387 for (i = 0; i < MAX_MIRRORS; i++) {
1388 mirror_destroy(ofproto->mirrors[i]);
1391 netflow_destroy(ofproto->netflow);
1392 dpif_sflow_destroy(ofproto->sflow);
1393 hmap_destroy(&ofproto->bundles);
1394 mac_learning_destroy(ofproto->ml);
1396 hmap_destroy(&ofproto->facets);
1397 hmap_destroy(&ofproto->subfacets);
1398 governor_destroy(ofproto->governor);
1400 hmap_destroy(&ofproto->vlandev_map);
1401 hmap_destroy(&ofproto->realdev_vid_map);
1403 sset_destroy(&ofproto->ports);
1404 sset_destroy(&ofproto->ghost_ports);
1405 sset_destroy(&ofproto->port_poll_set);
1407 close_dpif_backer(ofproto->backer);
1411 run_fast(struct ofproto *ofproto_)
1413 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1414 struct ofport_dpif *ofport;
1416 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1417 port_run_fast(ofport);
1424 run(struct ofproto *ofproto_)
1426 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1427 struct ofport_dpif *ofport;
1428 struct ofbundle *bundle;
1432 complete_operations(ofproto);
1435 error = run_fast(ofproto_);
1440 if (ofproto->netflow) {
1441 if (netflow_run(ofproto->netflow)) {
1442 send_netflow_active_timeouts(ofproto);
1445 if (ofproto->sflow) {
1446 dpif_sflow_run(ofproto->sflow);
1449 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1452 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1457 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1459 /* Check the consistency of a random facet, to aid debugging. */
1460 if (!hmap_is_empty(&ofproto->facets)
1461 && !ofproto->backer->need_revalidate) {
1462 struct facet *facet;
1464 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1465 struct facet, hmap_node);
1466 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1468 if (!facet_check_consistency(facet)) {
1469 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1474 if (ofproto->governor) {
1477 governor_run(ofproto->governor);
1479 /* If the governor has shrunk to its minimum size and the number of
1480 * subfacets has dwindled, then drop the governor entirely.
1482 * For hysteresis, the number of subfacets to drop the governor is
1483 * smaller than the number needed to trigger its creation. */
1484 n_subfacets = hmap_count(&ofproto->subfacets);
1485 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1486 && governor_is_idle(ofproto->governor)) {
1487 governor_destroy(ofproto->governor);
1488 ofproto->governor = NULL;
1496 wait(struct ofproto *ofproto_)
1498 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1499 struct ofport_dpif *ofport;
1500 struct ofbundle *bundle;
1502 if (!clogged && !list_is_empty(&ofproto->completions)) {
1503 poll_immediate_wake();
1506 dpif_wait(ofproto->backer->dpif);
1507 dpif_recv_wait(ofproto->backer->dpif);
1508 if (ofproto->sflow) {
1509 dpif_sflow_wait(ofproto->sflow);
1511 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1512 poll_immediate_wake();
1514 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1517 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1518 bundle_wait(bundle);
1520 if (ofproto->netflow) {
1521 netflow_wait(ofproto->netflow);
1523 mac_learning_wait(ofproto->ml);
1525 if (ofproto->backer->need_revalidate) {
1526 /* Shouldn't happen, but if it does just go around again. */
1527 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1528 poll_immediate_wake();
1530 if (ofproto->governor) {
1531 governor_wait(ofproto->governor);
1536 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1538 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1540 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1541 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1545 flush(struct ofproto *ofproto_)
1547 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1548 struct subfacet *subfacet, *next_subfacet;
1549 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1553 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1554 &ofproto->subfacets) {
1555 if (subfacet->path != SF_NOT_INSTALLED) {
1556 batch[n_batch++] = subfacet;
1557 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1558 subfacet_destroy_batch(ofproto, batch, n_batch);
1562 subfacet_destroy(subfacet);
1567 subfacet_destroy_batch(ofproto, batch, n_batch);
1572 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1573 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1575 *arp_match_ip = true;
1576 *actions = (OFPUTIL_A_OUTPUT |
1577 OFPUTIL_A_SET_VLAN_VID |
1578 OFPUTIL_A_SET_VLAN_PCP |
1579 OFPUTIL_A_STRIP_VLAN |
1580 OFPUTIL_A_SET_DL_SRC |
1581 OFPUTIL_A_SET_DL_DST |
1582 OFPUTIL_A_SET_NW_SRC |
1583 OFPUTIL_A_SET_NW_DST |
1584 OFPUTIL_A_SET_NW_TOS |
1585 OFPUTIL_A_SET_TP_SRC |
1586 OFPUTIL_A_SET_TP_DST |
1591 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1593 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1594 struct dpif_dp_stats s;
1596 strcpy(ots->name, "classifier");
1598 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1600 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1601 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1604 static struct ofport *
1607 struct ofport_dpif *port = xmalloc(sizeof *port);
1612 port_dealloc(struct ofport *port_)
1614 struct ofport_dpif *port = ofport_dpif_cast(port_);
1619 port_construct(struct ofport *port_)
1621 struct ofport_dpif *port = ofport_dpif_cast(port_);
1622 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1623 const struct netdev *netdev = port->up.netdev;
1624 struct dpif_port dpif_port;
1627 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1628 port->bundle = NULL;
1630 port->tag = tag_create_random();
1631 port->may_enable = true;
1632 port->stp_port = NULL;
1633 port->stp_state = STP_DISABLED;
1634 port->tnl_port = NULL;
1635 hmap_init(&port->priorities);
1636 port->realdev_ofp_port = 0;
1637 port->vlandev_vid = 0;
1638 port->carrier_seq = netdev_get_carrier_resets(netdev);
1640 if (netdev_vport_is_patch(netdev)) {
1641 /* XXX By bailing out here, we don't do required sFlow work. */
1642 port->odp_port = OVSP_NONE;
1646 error = dpif_port_query_by_name(ofproto->backer->dpif,
1647 netdev_vport_get_dpif_port(netdev),
1653 port->odp_port = dpif_port.port_no;
1655 if (netdev_get_tunnel_config(netdev)) {
1656 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1658 /* Sanity-check that a mapping doesn't already exist. This
1659 * shouldn't happen for non-tunnel ports. */
1660 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1661 VLOG_ERR("port %s already has an OpenFlow port number",
1663 dpif_port_destroy(&dpif_port);
1667 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1668 hash_int(port->odp_port, 0));
1670 dpif_port_destroy(&dpif_port);
1672 if (ofproto->sflow) {
1673 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1680 port_destruct(struct ofport *port_)
1682 struct ofport_dpif *port = ofport_dpif_cast(port_);
1683 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1684 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1685 const char *devname = netdev_get_name(port->up.netdev);
1687 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1688 /* The underlying device is still there, so delete it. This
1689 * happens when the ofproto is being destroyed, since the caller
1690 * assumes that removal of attached ports will happen as part of
1692 if (!port->tnl_port) {
1693 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1695 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1698 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1699 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1702 tnl_port_del(port->tnl_port);
1703 sset_find_and_delete(&ofproto->ports, devname);
1704 sset_find_and_delete(&ofproto->ghost_ports, devname);
1705 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1706 bundle_remove(port_);
1707 set_cfm(port_, NULL);
1708 if (ofproto->sflow) {
1709 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1712 ofport_clear_priorities(port);
1713 hmap_destroy(&port->priorities);
1717 port_modified(struct ofport *port_)
1719 struct ofport_dpif *port = ofport_dpif_cast(port_);
1721 if (port->bundle && port->bundle->bond) {
1722 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1727 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1729 struct ofport_dpif *port = ofport_dpif_cast(port_);
1730 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1731 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1733 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1734 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1735 OFPUTIL_PC_NO_PACKET_IN)) {
1736 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1738 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1739 bundle_update(port->bundle);
1745 set_sflow(struct ofproto *ofproto_,
1746 const struct ofproto_sflow_options *sflow_options)
1748 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1749 struct dpif_sflow *ds = ofproto->sflow;
1751 if (sflow_options) {
1753 struct ofport_dpif *ofport;
1755 ds = ofproto->sflow = dpif_sflow_create();
1756 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1757 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1759 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1761 dpif_sflow_set_options(ds, sflow_options);
1764 dpif_sflow_destroy(ds);
1765 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1766 ofproto->sflow = NULL;
1773 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1775 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1782 struct ofproto_dpif *ofproto;
1784 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1785 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1786 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1789 if (cfm_configure(ofport->cfm, s)) {
1795 cfm_destroy(ofport->cfm);
1801 get_cfm_status(const struct ofport *ofport_,
1802 struct ofproto_cfm_status *status)
1804 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1807 status->faults = cfm_get_fault(ofport->cfm);
1808 status->remote_opstate = cfm_get_opup(ofport->cfm);
1809 status->health = cfm_get_health(ofport->cfm);
1810 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1817 /* Spanning Tree. */
1820 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1822 struct ofproto_dpif *ofproto = ofproto_;
1823 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1824 struct ofport_dpif *ofport;
1826 ofport = stp_port_get_aux(sp);
1828 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1829 ofproto->up.name, port_num);
1831 struct eth_header *eth = pkt->l2;
1833 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1834 if (eth_addr_is_zero(eth->eth_src)) {
1835 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1836 "with unknown MAC", ofproto->up.name, port_num);
1838 send_packet(ofport, pkt);
1844 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1846 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1848 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1850 /* Only revalidate flows if the configuration changed. */
1851 if (!s != !ofproto->stp) {
1852 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1856 if (!ofproto->stp) {
1857 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1858 send_bpdu_cb, ofproto);
1859 ofproto->stp_last_tick = time_msec();
1862 stp_set_bridge_id(ofproto->stp, s->system_id);
1863 stp_set_bridge_priority(ofproto->stp, s->priority);
1864 stp_set_hello_time(ofproto->stp, s->hello_time);
1865 stp_set_max_age(ofproto->stp, s->max_age);
1866 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1868 struct ofport *ofport;
1870 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1871 set_stp_port(ofport, NULL);
1874 stp_destroy(ofproto->stp);
1875 ofproto->stp = NULL;
1882 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1884 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1888 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1889 s->designated_root = stp_get_designated_root(ofproto->stp);
1890 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1899 update_stp_port_state(struct ofport_dpif *ofport)
1901 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1902 enum stp_state state;
1904 /* Figure out new state. */
1905 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1909 if (ofport->stp_state != state) {
1910 enum ofputil_port_state of_state;
1913 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1914 netdev_get_name(ofport->up.netdev),
1915 stp_state_name(ofport->stp_state),
1916 stp_state_name(state));
1917 if (stp_learn_in_state(ofport->stp_state)
1918 != stp_learn_in_state(state)) {
1919 /* xxx Learning action flows should also be flushed. */
1920 mac_learning_flush(ofproto->ml,
1921 &ofproto->backer->revalidate_set);
1923 fwd_change = stp_forward_in_state(ofport->stp_state)
1924 != stp_forward_in_state(state);
1926 ofproto->backer->need_revalidate = REV_STP;
1927 ofport->stp_state = state;
1928 ofport->stp_state_entered = time_msec();
1930 if (fwd_change && ofport->bundle) {
1931 bundle_update(ofport->bundle);
1934 /* Update the STP state bits in the OpenFlow port description. */
1935 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1936 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1937 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1938 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1939 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1941 ofproto_port_set_state(&ofport->up, of_state);
1945 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1946 * caller is responsible for assigning STP port numbers and ensuring
1947 * there are no duplicates. */
1949 set_stp_port(struct ofport *ofport_,
1950 const struct ofproto_port_stp_settings *s)
1952 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1953 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1954 struct stp_port *sp = ofport->stp_port;
1956 if (!s || !s->enable) {
1958 ofport->stp_port = NULL;
1959 stp_port_disable(sp);
1960 update_stp_port_state(ofport);
1963 } else if (sp && stp_port_no(sp) != s->port_num
1964 && ofport == stp_port_get_aux(sp)) {
1965 /* The port-id changed, so disable the old one if it's not
1966 * already in use by another port. */
1967 stp_port_disable(sp);
1970 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1971 stp_port_enable(sp);
1973 stp_port_set_aux(sp, ofport);
1974 stp_port_set_priority(sp, s->priority);
1975 stp_port_set_path_cost(sp, s->path_cost);
1977 update_stp_port_state(ofport);
1983 get_stp_port_status(struct ofport *ofport_,
1984 struct ofproto_port_stp_status *s)
1986 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1987 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1988 struct stp_port *sp = ofport->stp_port;
1990 if (!ofproto->stp || !sp) {
1996 s->port_id = stp_port_get_id(sp);
1997 s->state = stp_port_get_state(sp);
1998 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1999 s->role = stp_port_get_role(sp);
2000 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2006 stp_run(struct ofproto_dpif *ofproto)
2009 long long int now = time_msec();
2010 long long int elapsed = now - ofproto->stp_last_tick;
2011 struct stp_port *sp;
2014 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2015 ofproto->stp_last_tick = now;
2017 while (stp_get_changed_port(ofproto->stp, &sp)) {
2018 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2021 update_stp_port_state(ofport);
2025 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2026 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2032 stp_wait(struct ofproto_dpif *ofproto)
2035 poll_timer_wait(1000);
2039 /* Returns true if STP should process 'flow'. */
2041 stp_should_process_flow(const struct flow *flow)
2043 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2047 stp_process_packet(const struct ofport_dpif *ofport,
2048 const struct ofpbuf *packet)
2050 struct ofpbuf payload = *packet;
2051 struct eth_header *eth = payload.data;
2052 struct stp_port *sp = ofport->stp_port;
2054 /* Sink packets on ports that have STP disabled when the bridge has
2056 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2060 /* Trim off padding on payload. */
2061 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2062 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2065 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2066 stp_received_bpdu(sp, payload.data, payload.size);
2070 static struct priority_to_dscp *
2071 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2073 struct priority_to_dscp *pdscp;
2076 hash = hash_int(priority, 0);
2077 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2078 if (pdscp->priority == priority) {
2086 ofport_clear_priorities(struct ofport_dpif *ofport)
2088 struct priority_to_dscp *pdscp, *next;
2090 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2091 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2097 set_queues(struct ofport *ofport_,
2098 const struct ofproto_port_queue *qdscp_list,
2101 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2102 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2103 struct hmap new = HMAP_INITIALIZER(&new);
2106 for (i = 0; i < n_qdscp; i++) {
2107 struct priority_to_dscp *pdscp;
2111 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2112 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2117 pdscp = get_priority(ofport, priority);
2119 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2121 pdscp = xmalloc(sizeof *pdscp);
2122 pdscp->priority = priority;
2124 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2127 if (pdscp->dscp != dscp) {
2129 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2132 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2135 if (!hmap_is_empty(&ofport->priorities)) {
2136 ofport_clear_priorities(ofport);
2137 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2140 hmap_swap(&new, &ofport->priorities);
2148 /* Expires all MAC learning entries associated with 'bundle' and forces its
2149 * ofproto to revalidate every flow.
2151 * Normally MAC learning entries are removed only from the ofproto associated
2152 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2153 * are removed from every ofproto. When patch ports and SLB bonds are in use
2154 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2155 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2156 * with the host from which it migrated. */
2158 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2160 struct ofproto_dpif *ofproto = bundle->ofproto;
2161 struct mac_learning *ml = ofproto->ml;
2162 struct mac_entry *mac, *next_mac;
2164 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2165 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2166 if (mac->port.p == bundle) {
2168 struct ofproto_dpif *o;
2170 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2172 struct mac_entry *e;
2174 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2177 mac_learning_expire(o->ml, e);
2183 mac_learning_expire(ml, mac);
2188 static struct ofbundle *
2189 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2191 struct ofbundle *bundle;
2193 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2194 &ofproto->bundles) {
2195 if (bundle->aux == aux) {
2202 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2203 * ones that are found to 'bundles'. */
2205 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2206 void **auxes, size_t n_auxes,
2207 struct hmapx *bundles)
2211 hmapx_init(bundles);
2212 for (i = 0; i < n_auxes; i++) {
2213 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2215 hmapx_add(bundles, bundle);
2221 bundle_update(struct ofbundle *bundle)
2223 struct ofport_dpif *port;
2225 bundle->floodable = true;
2226 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2227 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2228 || !stp_forward_in_state(port->stp_state)) {
2229 bundle->floodable = false;
2236 bundle_del_port(struct ofport_dpif *port)
2238 struct ofbundle *bundle = port->bundle;
2240 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2242 list_remove(&port->bundle_node);
2243 port->bundle = NULL;
2246 lacp_slave_unregister(bundle->lacp, port);
2249 bond_slave_unregister(bundle->bond, port);
2252 bundle_update(bundle);
2256 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2257 struct lacp_slave_settings *lacp)
2259 struct ofport_dpif *port;
2261 port = get_ofp_port(bundle->ofproto, ofp_port);
2266 if (port->bundle != bundle) {
2267 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2269 bundle_del_port(port);
2272 port->bundle = bundle;
2273 list_push_back(&bundle->ports, &port->bundle_node);
2274 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2275 || !stp_forward_in_state(port->stp_state)) {
2276 bundle->floodable = false;
2280 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2281 lacp_slave_register(bundle->lacp, port, lacp);
2288 bundle_destroy(struct ofbundle *bundle)
2290 struct ofproto_dpif *ofproto;
2291 struct ofport_dpif *port, *next_port;
2298 ofproto = bundle->ofproto;
2299 for (i = 0; i < MAX_MIRRORS; i++) {
2300 struct ofmirror *m = ofproto->mirrors[i];
2302 if (m->out == bundle) {
2304 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2305 || hmapx_find_and_delete(&m->dsts, bundle)) {
2306 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2311 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2312 bundle_del_port(port);
2315 bundle_flush_macs(bundle, true);
2316 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2318 free(bundle->trunks);
2319 lacp_destroy(bundle->lacp);
2320 bond_destroy(bundle->bond);
2325 bundle_set(struct ofproto *ofproto_, void *aux,
2326 const struct ofproto_bundle_settings *s)
2328 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2329 bool need_flush = false;
2330 struct ofport_dpif *port;
2331 struct ofbundle *bundle;
2332 unsigned long *trunks;
2338 bundle_destroy(bundle_lookup(ofproto, aux));
2342 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2343 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2345 bundle = bundle_lookup(ofproto, aux);
2347 bundle = xmalloc(sizeof *bundle);
2349 bundle->ofproto = ofproto;
2350 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2351 hash_pointer(aux, 0));
2353 bundle->name = NULL;
2355 list_init(&bundle->ports);
2356 bundle->vlan_mode = PORT_VLAN_TRUNK;
2358 bundle->trunks = NULL;
2359 bundle->use_priority_tags = s->use_priority_tags;
2360 bundle->lacp = NULL;
2361 bundle->bond = NULL;
2363 bundle->floodable = true;
2365 bundle->src_mirrors = 0;
2366 bundle->dst_mirrors = 0;
2367 bundle->mirror_out = 0;
2370 if (!bundle->name || strcmp(s->name, bundle->name)) {
2372 bundle->name = xstrdup(s->name);
2377 if (!bundle->lacp) {
2378 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2379 bundle->lacp = lacp_create();
2381 lacp_configure(bundle->lacp, s->lacp);
2383 lacp_destroy(bundle->lacp);
2384 bundle->lacp = NULL;
2387 /* Update set of ports. */
2389 for (i = 0; i < s->n_slaves; i++) {
2390 if (!bundle_add_port(bundle, s->slaves[i],
2391 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2395 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2396 struct ofport_dpif *next_port;
2398 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2399 for (i = 0; i < s->n_slaves; i++) {
2400 if (s->slaves[i] == port->up.ofp_port) {
2405 bundle_del_port(port);
2409 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2411 if (list_is_empty(&bundle->ports)) {
2412 bundle_destroy(bundle);
2416 /* Set VLAN tagging mode */
2417 if (s->vlan_mode != bundle->vlan_mode
2418 || s->use_priority_tags != bundle->use_priority_tags) {
2419 bundle->vlan_mode = s->vlan_mode;
2420 bundle->use_priority_tags = s->use_priority_tags;
2425 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2426 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2428 if (vlan != bundle->vlan) {
2429 bundle->vlan = vlan;
2433 /* Get trunked VLANs. */
2434 switch (s->vlan_mode) {
2435 case PORT_VLAN_ACCESS:
2439 case PORT_VLAN_TRUNK:
2440 trunks = CONST_CAST(unsigned long *, s->trunks);
2443 case PORT_VLAN_NATIVE_UNTAGGED:
2444 case PORT_VLAN_NATIVE_TAGGED:
2445 if (vlan != 0 && (!s->trunks
2446 || !bitmap_is_set(s->trunks, vlan)
2447 || bitmap_is_set(s->trunks, 0))) {
2448 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2450 trunks = bitmap_clone(s->trunks, 4096);
2452 trunks = bitmap_allocate1(4096);
2454 bitmap_set1(trunks, vlan);
2455 bitmap_set0(trunks, 0);
2457 trunks = CONST_CAST(unsigned long *, s->trunks);
2464 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2465 free(bundle->trunks);
2466 if (trunks == s->trunks) {
2467 bundle->trunks = vlan_bitmap_clone(trunks);
2469 bundle->trunks = trunks;
2474 if (trunks != s->trunks) {
2479 if (!list_is_short(&bundle->ports)) {
2480 bundle->ofproto->has_bonded_bundles = true;
2482 if (bond_reconfigure(bundle->bond, s->bond)) {
2483 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2486 bundle->bond = bond_create(s->bond);
2487 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2490 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2491 bond_slave_register(bundle->bond, port, port->up.netdev);
2494 bond_destroy(bundle->bond);
2495 bundle->bond = NULL;
2498 /* If we changed something that would affect MAC learning, un-learn
2499 * everything on this port and force flow revalidation. */
2501 bundle_flush_macs(bundle, false);
2508 bundle_remove(struct ofport *port_)
2510 struct ofport_dpif *port = ofport_dpif_cast(port_);
2511 struct ofbundle *bundle = port->bundle;
2514 bundle_del_port(port);
2515 if (list_is_empty(&bundle->ports)) {
2516 bundle_destroy(bundle);
2517 } else if (list_is_short(&bundle->ports)) {
2518 bond_destroy(bundle->bond);
2519 bundle->bond = NULL;
2525 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2527 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2528 struct ofport_dpif *port = port_;
2529 uint8_t ea[ETH_ADDR_LEN];
2532 error = netdev_get_etheraddr(port->up.netdev, ea);
2534 struct ofpbuf packet;
2537 ofpbuf_init(&packet, 0);
2538 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2540 memcpy(packet_pdu, pdu, pdu_size);
2542 send_packet(port, &packet);
2543 ofpbuf_uninit(&packet);
2545 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2546 "%s (%s)", port->bundle->name,
2547 netdev_get_name(port->up.netdev), strerror(error));
2552 bundle_send_learning_packets(struct ofbundle *bundle)
2554 struct ofproto_dpif *ofproto = bundle->ofproto;
2555 int error, n_packets, n_errors;
2556 struct mac_entry *e;
2558 error = n_packets = n_errors = 0;
2559 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2560 if (e->port.p != bundle) {
2561 struct ofpbuf *learning_packet;
2562 struct ofport_dpif *port;
2566 /* The assignment to "port" is unnecessary but makes "grep"ing for
2567 * struct ofport_dpif more effective. */
2568 learning_packet = bond_compose_learning_packet(bundle->bond,
2572 ret = send_packet(port, learning_packet);
2573 ofpbuf_delete(learning_packet);
2583 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2584 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2585 "packets, last error was: %s",
2586 bundle->name, n_errors, n_packets, strerror(error));
2588 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2589 bundle->name, n_packets);
2594 bundle_run(struct ofbundle *bundle)
2597 lacp_run(bundle->lacp, send_pdu_cb);
2600 struct ofport_dpif *port;
2602 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2603 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2606 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2607 lacp_status(bundle->lacp));
2608 if (bond_should_send_learning_packets(bundle->bond)) {
2609 bundle_send_learning_packets(bundle);
2615 bundle_wait(struct ofbundle *bundle)
2618 lacp_wait(bundle->lacp);
2621 bond_wait(bundle->bond);
2628 mirror_scan(struct ofproto_dpif *ofproto)
2632 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2633 if (!ofproto->mirrors[idx]) {
2640 static struct ofmirror *
2641 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2645 for (i = 0; i < MAX_MIRRORS; i++) {
2646 struct ofmirror *mirror = ofproto->mirrors[i];
2647 if (mirror && mirror->aux == aux) {
2655 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2657 mirror_update_dups(struct ofproto_dpif *ofproto)
2661 for (i = 0; i < MAX_MIRRORS; i++) {
2662 struct ofmirror *m = ofproto->mirrors[i];
2665 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2669 for (i = 0; i < MAX_MIRRORS; i++) {
2670 struct ofmirror *m1 = ofproto->mirrors[i];
2677 for (j = i + 1; j < MAX_MIRRORS; j++) {
2678 struct ofmirror *m2 = ofproto->mirrors[j];
2680 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2681 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2682 m2->dup_mirrors |= m1->dup_mirrors;
2689 mirror_set(struct ofproto *ofproto_, void *aux,
2690 const struct ofproto_mirror_settings *s)
2692 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2693 mirror_mask_t mirror_bit;
2694 struct ofbundle *bundle;
2695 struct ofmirror *mirror;
2696 struct ofbundle *out;
2697 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2698 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2701 mirror = mirror_lookup(ofproto, aux);
2703 mirror_destroy(mirror);
2709 idx = mirror_scan(ofproto);
2711 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2713 ofproto->up.name, MAX_MIRRORS, s->name);
2717 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2718 mirror->ofproto = ofproto;
2721 mirror->out_vlan = -1;
2722 mirror->name = NULL;
2725 if (!mirror->name || strcmp(s->name, mirror->name)) {
2727 mirror->name = xstrdup(s->name);
2730 /* Get the new configuration. */
2731 if (s->out_bundle) {
2732 out = bundle_lookup(ofproto, s->out_bundle);
2734 mirror_destroy(mirror);
2740 out_vlan = s->out_vlan;
2742 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2743 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2745 /* If the configuration has not changed, do nothing. */
2746 if (hmapx_equals(&srcs, &mirror->srcs)
2747 && hmapx_equals(&dsts, &mirror->dsts)
2748 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2749 && mirror->out == out
2750 && mirror->out_vlan == out_vlan)
2752 hmapx_destroy(&srcs);
2753 hmapx_destroy(&dsts);
2757 hmapx_swap(&srcs, &mirror->srcs);
2758 hmapx_destroy(&srcs);
2760 hmapx_swap(&dsts, &mirror->dsts);
2761 hmapx_destroy(&dsts);
2763 free(mirror->vlans);
2764 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2767 mirror->out_vlan = out_vlan;
2769 /* Update bundles. */
2770 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2771 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2772 if (hmapx_contains(&mirror->srcs, bundle)) {
2773 bundle->src_mirrors |= mirror_bit;
2775 bundle->src_mirrors &= ~mirror_bit;
2778 if (hmapx_contains(&mirror->dsts, bundle)) {
2779 bundle->dst_mirrors |= mirror_bit;
2781 bundle->dst_mirrors &= ~mirror_bit;
2784 if (mirror->out == bundle) {
2785 bundle->mirror_out |= mirror_bit;
2787 bundle->mirror_out &= ~mirror_bit;
2791 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2792 ofproto->has_mirrors = true;
2793 mac_learning_flush(ofproto->ml,
2794 &ofproto->backer->revalidate_set);
2795 mirror_update_dups(ofproto);
2801 mirror_destroy(struct ofmirror *mirror)
2803 struct ofproto_dpif *ofproto;
2804 mirror_mask_t mirror_bit;
2805 struct ofbundle *bundle;
2812 ofproto = mirror->ofproto;
2813 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2814 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2816 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2817 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2818 bundle->src_mirrors &= ~mirror_bit;
2819 bundle->dst_mirrors &= ~mirror_bit;
2820 bundle->mirror_out &= ~mirror_bit;
2823 hmapx_destroy(&mirror->srcs);
2824 hmapx_destroy(&mirror->dsts);
2825 free(mirror->vlans);
2827 ofproto->mirrors[mirror->idx] = NULL;
2831 mirror_update_dups(ofproto);
2833 ofproto->has_mirrors = false;
2834 for (i = 0; i < MAX_MIRRORS; i++) {
2835 if (ofproto->mirrors[i]) {
2836 ofproto->has_mirrors = true;
2843 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2844 uint64_t *packets, uint64_t *bytes)
2846 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2847 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2850 *packets = *bytes = UINT64_MAX;
2854 *packets = mirror->packet_count;
2855 *bytes = mirror->byte_count;
2861 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2863 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2864 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2865 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2871 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2873 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2874 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2875 return bundle && bundle->mirror_out != 0;
2879 forward_bpdu_changed(struct ofproto *ofproto_)
2881 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2882 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2886 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2889 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2890 mac_learning_set_idle_time(ofproto->ml, idle_time);
2891 mac_learning_set_max_entries(ofproto->ml, max_entries);
2896 static struct ofport_dpif *
2897 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2899 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2900 return ofport ? ofport_dpif_cast(ofport) : NULL;
2903 static struct ofport_dpif *
2904 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2906 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2907 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2911 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2912 struct ofproto_port *ofproto_port,
2913 struct dpif_port *dpif_port)
2915 ofproto_port->name = dpif_port->name;
2916 ofproto_port->type = dpif_port->type;
2917 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2920 static struct ofport_dpif *
2921 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
2923 const struct ofproto_dpif *ofproto;
2926 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
2931 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2932 struct ofport *ofport;
2934 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
2935 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
2936 return ofport_dpif_cast(ofport);
2943 port_run_fast(struct ofport_dpif *ofport)
2945 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2946 struct ofpbuf packet;
2948 ofpbuf_init(&packet, 0);
2949 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2950 send_packet(ofport, &packet);
2951 ofpbuf_uninit(&packet);
2956 port_run(struct ofport_dpif *ofport)
2958 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2959 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2960 bool enable = netdev_get_carrier(ofport->up.netdev);
2962 ofport->carrier_seq = carrier_seq;
2964 port_run_fast(ofport);
2966 if (ofport->tnl_port
2967 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
2968 &ofport->tnl_port)) {
2969 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
2973 int cfm_opup = cfm_get_opup(ofport->cfm);
2975 cfm_run(ofport->cfm);
2976 enable = enable && !cfm_get_fault(ofport->cfm);
2978 if (cfm_opup >= 0) {
2979 enable = enable && cfm_opup;
2983 if (ofport->bundle) {
2984 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2985 if (carrier_changed) {
2986 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2990 if (ofport->may_enable != enable) {
2991 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2993 if (ofproto->has_bundle_action) {
2994 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2998 ofport->may_enable = enable;
3002 port_wait(struct ofport_dpif *ofport)
3005 cfm_wait(ofport->cfm);
3010 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3011 struct ofproto_port *ofproto_port)
3013 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3014 struct dpif_port dpif_port;
3017 if (sset_contains(&ofproto->ghost_ports, devname)) {
3018 const char *type = netdev_get_type_from_name(devname);
3020 /* We may be called before ofproto->up.port_by_name is populated with
3021 * the appropriate ofport. For this reason, we must get the name and
3022 * type from the netdev layer directly. */
3024 const struct ofport *ofport;
3026 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3027 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3028 ofproto_port->name = xstrdup(devname);
3029 ofproto_port->type = xstrdup(type);
3035 if (!sset_contains(&ofproto->ports, devname)) {
3038 error = dpif_port_query_by_name(ofproto->backer->dpif,
3039 devname, &dpif_port);
3041 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3047 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3049 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3050 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3051 const char *devname = netdev_get_name(netdev);
3053 if (netdev_vport_is_patch(netdev)) {
3054 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3058 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3059 uint32_t port_no = UINT32_MAX;
3062 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3066 if (netdev_get_tunnel_config(netdev)) {
3067 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3071 if (netdev_get_tunnel_config(netdev)) {
3072 sset_add(&ofproto->ghost_ports, devname);
3074 sset_add(&ofproto->ports, devname);
3080 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3082 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3083 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3090 sset_find_and_delete(&ofproto->ghost_ports,
3091 netdev_get_name(ofport->up.netdev));
3092 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3093 if (!ofport->tnl_port) {
3094 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3096 /* The caller is going to close ofport->up.netdev. If this is a
3097 * bonded port, then the bond is using that netdev, so remove it
3098 * from the bond. The client will need to reconfigure everything
3099 * after deleting ports, so then the slave will get re-added. */
3100 bundle_remove(&ofport->up);
3107 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3109 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3112 error = netdev_get_stats(ofport->up.netdev, stats);
3114 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3115 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3117 /* ofproto->stats.tx_packets represents packets that we created
3118 * internally and sent to some port (e.g. packets sent with
3119 * send_packet()). Account for them as if they had come from
3120 * OFPP_LOCAL and got forwarded. */
3122 if (stats->rx_packets != UINT64_MAX) {
3123 stats->rx_packets += ofproto->stats.tx_packets;
3126 if (stats->rx_bytes != UINT64_MAX) {
3127 stats->rx_bytes += ofproto->stats.tx_bytes;
3130 /* ofproto->stats.rx_packets represents packets that were received on
3131 * some port and we processed internally and dropped (e.g. STP).
3132 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3134 if (stats->tx_packets != UINT64_MAX) {
3135 stats->tx_packets += ofproto->stats.rx_packets;
3138 if (stats->tx_bytes != UINT64_MAX) {
3139 stats->tx_bytes += ofproto->stats.rx_bytes;
3146 /* Account packets for LOCAL port. */
3148 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
3149 size_t tx_size, size_t rx_size)
3151 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3154 ofproto->stats.rx_packets++;
3155 ofproto->stats.rx_bytes += rx_size;
3158 ofproto->stats.tx_packets++;
3159 ofproto->stats.tx_bytes += tx_size;
3163 struct port_dump_state {
3168 struct ofproto_port port;
3173 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3175 *statep = xzalloc(sizeof(struct port_dump_state));
3180 port_dump_next(const struct ofproto *ofproto_, void *state_,
3181 struct ofproto_port *port)
3183 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3184 struct port_dump_state *state = state_;
3185 const struct sset *sset;
3186 struct sset_node *node;
3188 if (state->has_port) {
3189 ofproto_port_destroy(&state->port);
3190 state->has_port = false;
3192 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3193 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3196 error = port_query_by_name(ofproto_, node->name, &state->port);
3198 *port = state->port;
3199 state->has_port = true;
3201 } else if (error != ENODEV) {
3206 if (!state->ghost) {
3207 state->ghost = true;
3210 return port_dump_next(ofproto_, state_, port);
3217 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3219 struct port_dump_state *state = state_;
3221 if (state->has_port) {
3222 ofproto_port_destroy(&state->port);
3229 port_poll(const struct ofproto *ofproto_, char **devnamep)
3231 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3233 if (ofproto->port_poll_errno) {
3234 int error = ofproto->port_poll_errno;
3235 ofproto->port_poll_errno = 0;
3239 if (sset_is_empty(&ofproto->port_poll_set)) {
3243 *devnamep = sset_pop(&ofproto->port_poll_set);
3248 port_poll_wait(const struct ofproto *ofproto_)
3250 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3251 dpif_port_poll_wait(ofproto->backer->dpif);
3255 port_is_lacp_current(const struct ofport *ofport_)
3257 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3258 return (ofport->bundle && ofport->bundle->lacp
3259 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3263 /* Upcall handling. */
3265 /* Flow miss batching.
3267 * Some dpifs implement operations faster when you hand them off in a batch.
3268 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3269 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3270 * more packets, plus possibly installing the flow in the dpif.
3272 * So far we only batch the operations that affect flow setup time the most.
3273 * It's possible to batch more than that, but the benefit might be minimal. */
3275 struct hmap_node hmap_node;
3276 struct ofproto_dpif *ofproto;
3278 enum odp_key_fitness key_fitness;
3279 const struct nlattr *key;
3281 struct initial_vals initial_vals;
3282 struct list packets;
3283 enum dpif_upcall_type upcall_type;
3284 uint32_t odp_in_port;
3287 struct flow_miss_op {
3288 struct dpif_op dpif_op;
3289 void *garbage; /* Pointer to pass to free(), NULL if none. */
3290 uint64_t stub[1024 / 8]; /* Temporary buffer. */
3293 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3294 * OpenFlow controller as necessary according to their individual
3295 * configurations. */
3297 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3298 const struct flow *flow)
3300 struct ofputil_packet_in pin;
3302 pin.packet = packet->data;
3303 pin.packet_len = packet->size;
3304 pin.reason = OFPR_NO_MATCH;
3305 pin.controller_id = 0;
3310 pin.send_len = 0; /* not used for flow table misses */
3312 flow_get_metadata(flow, &pin.fmd);
3314 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3317 static enum slow_path_reason
3318 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3319 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3323 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3325 cfm_process_heartbeat(ofport->cfm, packet);
3328 } else if (ofport->bundle && ofport->bundle->lacp
3329 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3331 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3334 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3336 stp_process_packet(ofport, packet);
3344 static struct flow_miss *
3345 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3346 const struct flow *flow, uint32_t hash)
3348 struct flow_miss *miss;
3350 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3351 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3359 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3360 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3361 * 'miss' is associated with a subfacet the caller must also initialize the
3362 * returned op->subfacet, and if anything needs to be freed after processing
3363 * the op, the caller must initialize op->garbage also. */
3365 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3366 struct flow_miss_op *op)
3368 if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) {
3369 /* This packet was received on a VLAN splinter port. We
3370 * added a VLAN to the packet to make the packet resemble
3371 * the flow, but the actions were composed assuming that
3372 * the packet contained no VLAN. So, we must remove the
3373 * VLAN header from the packet before trying to execute the
3375 eth_pop_vlan(packet);
3379 op->dpif_op.type = DPIF_OP_EXECUTE;
3380 op->dpif_op.u.execute.key = miss->key;
3381 op->dpif_op.u.execute.key_len = miss->key_len;
3382 op->dpif_op.u.execute.packet = packet;
3385 /* Helper for handle_flow_miss_without_facet() and
3386 * handle_flow_miss_with_facet(). */
3388 handle_flow_miss_common(struct rule_dpif *rule,
3389 struct ofpbuf *packet, const struct flow *flow)
3391 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3393 ofproto->n_matches++;
3395 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3397 * Extra-special case for fail-open mode.
3399 * We are in fail-open mode and the packet matched the fail-open
3400 * rule, but we are connected to a controller too. We should send
3401 * the packet up to the controller in the hope that it will try to
3402 * set up a flow and thereby allow us to exit fail-open.
3404 * See the top-level comment in fail-open.c for more information.
3406 send_packet_in_miss(ofproto, packet, flow);
3410 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3411 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3412 * installing a datapath flow. The answer is usually "yes" (a return value of
3413 * true). However, for short flows the cost of bookkeeping is much higher than
3414 * the benefits, so when the datapath holds a large number of flows we impose
3415 * some heuristics to decide which flows are likely to be worth tracking. */
3417 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3418 struct flow_miss *miss, uint32_t hash)
3420 if (!ofproto->governor) {
3423 n_subfacets = hmap_count(&ofproto->subfacets);
3424 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3428 ofproto->governor = governor_create(ofproto->up.name);
3431 return governor_should_install_flow(ofproto->governor, hash,
3432 list_size(&miss->packets));
3435 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3436 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3437 * increment '*n_ops'. */
3439 handle_flow_miss_without_facet(struct flow_miss *miss,
3440 struct rule_dpif *rule,
3441 struct flow_miss_op *ops, size_t *n_ops)
3443 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3444 long long int now = time_msec();
3445 struct action_xlate_ctx ctx;
3446 struct ofpbuf *packet;
3448 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3449 struct flow_miss_op *op = &ops[*n_ops];
3450 struct dpif_flow_stats stats;
3451 struct ofpbuf odp_actions;
3453 COVERAGE_INC(facet_suppress);
3455 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3457 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3458 rule_credit_stats(rule, &stats);
3460 action_xlate_ctx_init(&ctx, ofproto, &miss->flow,
3461 &miss->initial_vals, rule, 0, packet);
3462 ctx.resubmit_stats = &stats;
3463 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3466 if (odp_actions.size) {
3467 struct dpif_execute *execute = &op->dpif_op.u.execute;
3469 init_flow_miss_execute_op(miss, packet, op);
3470 execute->actions = odp_actions.data;
3471 execute->actions_len = odp_actions.size;
3472 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3476 ofpbuf_uninit(&odp_actions);
3481 /* Handles 'miss', which matches 'facet'. May add any required datapath
3482 * operations to 'ops', incrementing '*n_ops' for each new op.
3484 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3485 * This is really important only for new facets: if we just called time_msec()
3486 * here, then the new subfacet or its packets could look (occasionally) as
3487 * though it was used some time after the facet was used. That can make a
3488 * one-packet flow look like it has a nonzero duration, which looks odd in
3489 * e.g. NetFlow statistics. */
3491 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3493 struct flow_miss_op *ops, size_t *n_ops)
3495 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3496 enum subfacet_path want_path;
3497 struct subfacet *subfacet;
3498 struct ofpbuf *packet;
3500 subfacet = subfacet_create(facet, miss, now);
3502 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3503 struct flow_miss_op *op = &ops[*n_ops];
3504 struct dpif_flow_stats stats;
3505 struct ofpbuf odp_actions;
3507 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3509 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3510 if (!subfacet->actions || subfacet->slow) {
3511 subfacet_make_actions(subfacet, packet, &odp_actions);
3514 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3515 subfacet_update_stats(subfacet, &stats);
3517 if (subfacet->actions_len) {
3518 struct dpif_execute *execute = &op->dpif_op.u.execute;
3520 init_flow_miss_execute_op(miss, packet, op);
3521 if (!subfacet->slow) {
3522 execute->actions = subfacet->actions;
3523 execute->actions_len = subfacet->actions_len;
3524 ofpbuf_uninit(&odp_actions);
3526 execute->actions = odp_actions.data;
3527 execute->actions_len = odp_actions.size;
3528 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3533 ofpbuf_uninit(&odp_actions);
3537 want_path = subfacet_want_path(subfacet->slow);
3538 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3539 struct flow_miss_op *op = &ops[(*n_ops)++];
3540 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3542 subfacet->path = want_path;
3545 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3546 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3547 put->key = miss->key;
3548 put->key_len = miss->key_len;
3549 if (want_path == SF_FAST_PATH) {
3550 put->actions = subfacet->actions;
3551 put->actions_len = subfacet->actions_len;
3553 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
3554 op->stub, sizeof op->stub,
3555 &put->actions, &put->actions_len);
3561 /* Handles flow miss 'miss'. May add any required datapath operations
3562 * to 'ops', incrementing '*n_ops' for each new op. */
3564 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3567 struct ofproto_dpif *ofproto = miss->ofproto;
3568 struct facet *facet;
3572 /* The caller must ensure that miss->hmap_node.hash contains
3573 * flow_hash(miss->flow, 0). */
3574 hash = miss->hmap_node.hash;
3576 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3578 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
3580 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
3581 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
3585 facet = facet_create(rule, &miss->flow, hash);
3590 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3593 static struct drop_key *
3594 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3597 struct drop_key *drop_key;
3599 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3600 &backer->drop_keys) {
3601 if (drop_key->key_len == key_len
3602 && !memcmp(drop_key->key, key, key_len)) {
3610 drop_key_clear(struct dpif_backer *backer)
3612 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3613 struct drop_key *drop_key, *next;
3615 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3618 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3620 if (error && !VLOG_DROP_WARN(&rl)) {
3621 struct ds ds = DS_EMPTY_INITIALIZER;
3622 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3623 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3628 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3629 free(drop_key->key);
3634 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3635 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3636 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3637 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3638 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3639 * 'packet' ingressed.
3641 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3642 * 'flow''s in_port to OFPP_NONE.
3644 * This function does post-processing on data returned from
3645 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3646 * of the upcall processing logic. In particular, if the extracted in_port is
3647 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3648 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3649 * a VLAN header onto 'packet' (if it is nonnull).
3651 * Optionally, if nonnull, sets 'initial_vals->vlan_tci' to the VLAN TCI
3652 * with which the packet was really received, that is, the actual VLAN
3653 * TCI extracted by odp_flow_key_to_flow(). (This differs from the
3654 * value returned in flow->vlan_tci only for packets received on VLAN
3657 * Similarly, this function also includes some logic to help with tunnels. It
3658 * may modify 'flow' as necessary to make the tunneling implementation
3659 * transparent to the upcall processing logic.
3661 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3662 * or some other positive errno if there are other problems. */
3664 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3665 const struct nlattr *key, size_t key_len,
3666 struct flow *flow, enum odp_key_fitness *fitnessp,
3667 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3668 struct initial_vals *initial_vals)
3670 const struct ofport_dpif *port;
3671 enum odp_key_fitness fitness;
3674 fitness = odp_flow_key_to_flow(key, key_len, flow);
3675 if (fitness == ODP_FIT_ERROR) {
3681 initial_vals->vlan_tci = flow->vlan_tci;
3685 *odp_in_port = flow->in_port;
3688 if (tnl_port_should_receive(flow)) {
3689 const struct ofport *ofport = tnl_port_receive(flow);
3691 flow->in_port = OFPP_NONE;
3694 port = ofport_dpif_cast(ofport);
3696 /* We can't reproduce 'key' from 'flow'. */
3697 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3699 /* XXX: Since the tunnel module is not scoped per backer, it's
3700 * theoretically possible that we'll receive an ofport belonging to an
3701 * entirely different datapath. In practice, this can't happen because
3702 * no platforms has two separate datapaths which each support
3704 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3706 port = odp_port_to_ofport(backer, flow->in_port);
3708 flow->in_port = OFPP_NONE;
3712 flow->in_port = port->up.ofp_port;
3713 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3715 /* Make the packet resemble the flow, so that it gets sent to
3716 * an OpenFlow controller properly, so that it looks correct
3717 * for sFlow, and so that flow_extract() will get the correct
3718 * vlan_tci if it is called on 'packet'.
3720 * The allocated space inside 'packet' probably also contains
3721 * 'key', that is, both 'packet' and 'key' are probably part of
3722 * a struct dpif_upcall (see the large comment on that
3723 * structure definition), so pushing data on 'packet' is in
3724 * general not a good idea since it could overwrite 'key' or
3725 * free it as a side effect. However, it's OK in this special
3726 * case because we know that 'packet' is inside a Netlink
3727 * attribute: pushing 4 bytes will just overwrite the 4-byte
3728 * "struct nlattr", which is fine since we don't need that
3729 * header anymore. */
3730 eth_push_vlan(packet, flow->vlan_tci);
3732 /* We can't reproduce 'key' from 'flow'. */
3733 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3739 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3744 *fitnessp = fitness;
3750 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3753 struct dpif_upcall *upcall;
3754 struct flow_miss *miss;
3755 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3756 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3757 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3767 /* Construct the to-do list.
3769 * This just amounts to extracting the flow from each packet and sticking
3770 * the packets that have the same flow in the same "flow_miss" structure so
3771 * that we can process them together. */
3774 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3775 struct flow_miss *miss = &misses[n_misses];
3776 struct flow_miss *existing_miss;
3777 struct ofproto_dpif *ofproto;
3778 uint32_t odp_in_port;
3783 error = ofproto_receive(backer, upcall->packet, upcall->key,
3784 upcall->key_len, &flow, &miss->key_fitness,
3785 &ofproto, &odp_in_port, &miss->initial_vals);
3786 if (error == ENODEV) {
3787 struct drop_key *drop_key;
3789 /* Received packet on port for which we couldn't associate
3790 * an ofproto. This can happen if a port is removed while
3791 * traffic is being received. Print a rate-limited message
3792 * in case it happens frequently. Install a drop flow so
3793 * that future packets of the flow are inexpensively dropped
3795 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
3798 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3800 drop_key = xmalloc(sizeof *drop_key);
3801 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3802 drop_key->key_len = upcall->key_len;
3804 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3805 hash_bytes(drop_key->key, drop_key->key_len, 0));
3806 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3807 drop_key->key, drop_key->key_len, NULL, 0, NULL);
3814 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3815 &flow.tunnel, flow.in_port, &miss->flow);
3817 /* Add other packets to a to-do list. */
3818 hash = flow_hash(&miss->flow, 0);
3819 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
3820 if (!existing_miss) {
3821 hmap_insert(&todo, &miss->hmap_node, hash);
3822 miss->ofproto = ofproto;
3823 miss->key = upcall->key;
3824 miss->key_len = upcall->key_len;
3825 miss->upcall_type = upcall->type;
3826 miss->odp_in_port = odp_in_port;
3827 list_init(&miss->packets);
3831 miss = existing_miss;
3833 list_push_back(&miss->packets, &upcall->packet->list_node);
3836 /* Process each element in the to-do list, constructing the set of
3837 * operations to batch. */
3839 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3840 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3842 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3844 /* Execute batch. */
3845 for (i = 0; i < n_ops; i++) {
3846 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3848 dpif_operate(backer->dpif, dpif_ops, n_ops);
3851 for (i = 0; i < n_ops; i++) {
3852 free(flow_miss_ops[i].garbage);
3854 hmap_destroy(&todo);
3857 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
3858 classify_upcall(const struct dpif_upcall *upcall)
3860 union user_action_cookie cookie;
3862 /* First look at the upcall type. */
3863 switch (upcall->type) {
3864 case DPIF_UC_ACTION:
3870 case DPIF_N_UC_TYPES:
3872 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3876 /* "action" upcalls need a closer look. */
3877 if (!upcall->userdata) {
3878 VLOG_WARN_RL(&rl, "action upcall missing cookie");
3881 if (nl_attr_get_size(upcall->userdata) != sizeof(cookie)) {
3882 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
3883 nl_attr_get_size(upcall->userdata));
3886 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof(cookie));
3887 switch (cookie.type) {
3888 case USER_ACTION_COOKIE_SFLOW:
3889 return SFLOW_UPCALL;
3891 case USER_ACTION_COOKIE_SLOW_PATH:
3894 case USER_ACTION_COOKIE_UNSPEC:
3896 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64,
3897 nl_attr_get_u64(upcall->userdata));
3903 handle_sflow_upcall(struct dpif_backer *backer,
3904 const struct dpif_upcall *upcall)
3906 struct ofproto_dpif *ofproto;
3907 union user_action_cookie cookie;
3909 uint32_t odp_in_port;
3911 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3912 &flow, NULL, &ofproto, &odp_in_port, NULL)
3913 || !ofproto->sflow) {
3917 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof(cookie));
3918 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3919 odp_in_port, &cookie);
3923 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3925 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3926 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3927 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3932 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
3935 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3936 struct dpif_upcall *upcall = &misses[n_misses];
3937 struct ofpbuf *buf = &miss_bufs[n_misses];
3940 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3941 sizeof miss_buf_stubs[n_misses]);
3942 error = dpif_recv(backer->dpif, upcall, buf);
3948 switch (classify_upcall(upcall)) {
3950 /* Handle it later. */
3955 handle_sflow_upcall(backer, upcall);
3965 /* Handle deferred MISS_UPCALL processing. */
3966 handle_miss_upcalls(backer, misses, n_misses);
3967 for (i = 0; i < n_misses; i++) {
3968 ofpbuf_uninit(&miss_bufs[i]);
3974 /* Flow expiration. */
3976 static int subfacet_max_idle(const struct ofproto_dpif *);
3977 static void update_stats(struct dpif_backer *);
3978 static void rule_expire(struct rule_dpif *);
3979 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
3981 /* This function is called periodically by run(). Its job is to collect
3982 * updates for the flows that have been installed into the datapath, most
3983 * importantly when they last were used, and then use that information to
3984 * expire flows that have not been used recently.
3986 * Returns the number of milliseconds after which it should be called again. */
3988 expire(struct dpif_backer *backer)
3990 struct ofproto_dpif *ofproto;
3991 int max_idle = INT32_MAX;
3993 /* Periodically clear out the drop keys in an effort to keep them
3994 * relatively few. */
3995 drop_key_clear(backer);
3997 /* Update stats for each flow in the backer. */
3998 update_stats(backer);
4000 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4001 struct rule *rule, *next_rule;
4004 if (ofproto->backer != backer) {
4008 /* Expire subfacets that have been idle too long. */
4009 dp_max_idle = subfacet_max_idle(ofproto);
4010 expire_subfacets(ofproto, dp_max_idle);
4012 max_idle = MIN(max_idle, dp_max_idle);
4014 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4016 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4017 &ofproto->up.expirable) {
4018 rule_expire(rule_dpif_cast(rule));
4021 /* All outstanding data in existing flows has been accounted, so it's a
4022 * good time to do bond rebalancing. */
4023 if (ofproto->has_bonded_bundles) {
4024 struct ofbundle *bundle;
4026 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4028 bond_rebalance(bundle->bond, &backer->revalidate_set);
4034 return MIN(max_idle, 1000);
4037 /* Updates flow table statistics given that the datapath just reported 'stats'
4038 * as 'subfacet''s statistics. */
4040 update_subfacet_stats(struct subfacet *subfacet,
4041 const struct dpif_flow_stats *stats)
4043 struct facet *facet = subfacet->facet;
4045 if (stats->n_packets >= subfacet->dp_packet_count) {
4046 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
4047 facet->packet_count += extra;
4049 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4052 if (stats->n_bytes >= subfacet->dp_byte_count) {
4053 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
4055 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4058 subfacet->dp_packet_count = stats->n_packets;
4059 subfacet->dp_byte_count = stats->n_bytes;
4061 facet->tcp_flags |= stats->tcp_flags;
4063 subfacet_update_time(subfacet, stats->used);
4064 if (facet->accounted_bytes < facet->byte_count) {
4066 facet_account(facet);
4067 facet->accounted_bytes = facet->byte_count;
4069 facet_push_stats(facet);
4072 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4073 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4075 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4076 const struct nlattr *key, size_t key_len)
4078 if (!VLOG_DROP_WARN(&rl)) {
4082 odp_flow_key_format(key, key_len, &s);
4083 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4087 COVERAGE_INC(facet_unexpected);
4088 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4091 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4093 * This function also pushes statistics updates to rules which each facet
4094 * resubmits into. Generally these statistics will be accurate. However, if a
4095 * facet changes the rule it resubmits into at some time in between
4096 * update_stats() runs, it is possible that statistics accrued to the
4097 * old rule will be incorrectly attributed to the new rule. This could be
4098 * avoided by calling update_stats() whenever rules are created or
4099 * deleted. However, the performance impact of making so many calls to the
4100 * datapath do not justify the benefit of having perfectly accurate statistics.
4103 update_stats(struct dpif_backer *backer)
4105 const struct dpif_flow_stats *stats;
4106 struct dpif_flow_dump dump;
4107 const struct nlattr *key;
4110 dpif_flow_dump_start(&dump, backer->dpif);
4111 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4113 struct subfacet *subfacet;
4114 struct ofproto_dpif *ofproto;
4115 struct ofport_dpif *ofport;
4118 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4123 ofport = get_ofp_port(ofproto, flow.in_port);
4124 if (ofport && ofport->tnl_port) {
4125 netdev_vport_inc_rx(ofport->up.netdev, stats);
4128 key_hash = odp_flow_key_hash(key, key_len);
4129 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4130 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4132 update_subfacet_stats(subfacet, stats);
4136 /* Stats are updated per-packet. */
4139 case SF_NOT_INSTALLED:
4141 delete_unexpected_flow(ofproto, key, key_len);
4145 dpif_flow_dump_done(&dump);
4148 /* Calculates and returns the number of milliseconds of idle time after which
4149 * subfacets should expire from the datapath. When a subfacet expires, we fold
4150 * its statistics into its facet, and when a facet's last subfacet expires, we
4151 * fold its statistic into its rule. */
4153 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4156 * Idle time histogram.
4158 * Most of the time a switch has a relatively small number of subfacets.
4159 * When this is the case we might as well keep statistics for all of them
4160 * in userspace and to cache them in the kernel datapath for performance as
4163 * As the number of subfacets increases, the memory required to maintain
4164 * statistics about them in userspace and in the kernel becomes
4165 * significant. However, with a large number of subfacets it is likely
4166 * that only a few of them are "heavy hitters" that consume a large amount
4167 * of bandwidth. At this point, only heavy hitters are worth caching in
4168 * the kernel and maintaining in userspaces; other subfacets we can
4171 * The technique used to compute the idle time is to build a histogram with
4172 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4173 * that is installed in the kernel gets dropped in the appropriate bucket.
4174 * After the histogram has been built, we compute the cutoff so that only
4175 * the most-recently-used 1% of subfacets (but at least
4176 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4177 * the most-recently-used bucket of subfacets is kept, so actually an
4178 * arbitrary number of subfacets can be kept in any given expiration run
4179 * (though the next run will delete most of those unless they receive
4182 * This requires a second pass through the subfacets, in addition to the
4183 * pass made by update_stats(), because the former function never looks at
4184 * uninstallable subfacets.
4186 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4187 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4188 int buckets[N_BUCKETS] = { 0 };
4189 int total, subtotal, bucket;
4190 struct subfacet *subfacet;
4194 total = hmap_count(&ofproto->subfacets);
4195 if (total <= ofproto->up.flow_eviction_threshold) {
4196 return N_BUCKETS * BUCKET_WIDTH;
4199 /* Build histogram. */
4201 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4202 long long int idle = now - subfacet->used;
4203 int bucket = (idle <= 0 ? 0
4204 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4205 : (unsigned int) idle / BUCKET_WIDTH);
4209 /* Find the first bucket whose flows should be expired. */
4210 subtotal = bucket = 0;
4212 subtotal += buckets[bucket++];
4213 } while (bucket < N_BUCKETS &&
4214 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4216 if (VLOG_IS_DBG_ENABLED()) {
4220 ds_put_cstr(&s, "keep");
4221 for (i = 0; i < N_BUCKETS; i++) {
4223 ds_put_cstr(&s, ", drop");
4226 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4229 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4233 return bucket * BUCKET_WIDTH;
4237 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4239 /* Cutoff time for most flows. */
4240 long long int normal_cutoff = time_msec() - dp_max_idle;
4242 /* We really want to keep flows for special protocols around, so use a more
4243 * conservative cutoff. */
4244 long long int special_cutoff = time_msec() - 10000;
4246 struct subfacet *subfacet, *next_subfacet;
4247 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4251 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4252 &ofproto->subfacets) {
4253 long long int cutoff;
4255 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
4258 if (subfacet->used < cutoff) {
4259 if (subfacet->path != SF_NOT_INSTALLED) {
4260 batch[n_batch++] = subfacet;
4261 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4262 subfacet_destroy_batch(ofproto, batch, n_batch);
4266 subfacet_destroy(subfacet);
4272 subfacet_destroy_batch(ofproto, batch, n_batch);
4276 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4277 * then delete it entirely. */
4279 rule_expire(struct rule_dpif *rule)
4281 struct facet *facet, *next_facet;
4285 if (rule->up.pending) {
4286 /* We'll have to expire it later. */
4290 /* Has 'rule' expired? */
4292 if (rule->up.hard_timeout
4293 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4294 reason = OFPRR_HARD_TIMEOUT;
4295 } else if (rule->up.idle_timeout
4296 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4297 reason = OFPRR_IDLE_TIMEOUT;
4302 COVERAGE_INC(ofproto_dpif_expired);
4304 /* Update stats. (This is a no-op if the rule expired due to an idle
4305 * timeout, because that only happens when the rule has no facets left.) */
4306 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4307 facet_remove(facet);
4310 /* Get rid of the rule. */
4311 ofproto_rule_expire(&rule->up, reason);
4316 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
4318 * The caller must already have determined that no facet with an identical
4319 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
4320 * the ofproto's classifier table.
4322 * 'hash' must be the return value of flow_hash(flow, 0).
4324 * The facet will initially have no subfacets. The caller should create (at
4325 * least) one subfacet with subfacet_create(). */
4326 static struct facet *
4327 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
4329 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4330 struct facet *facet;
4332 facet = xzalloc(sizeof *facet);
4333 facet->used = time_msec();
4334 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4335 list_push_back(&rule->facets, &facet->list_node);
4337 facet->flow = *flow;
4338 list_init(&facet->subfacets);
4339 netflow_flow_init(&facet->nf_flow);
4340 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4346 facet_free(struct facet *facet)
4351 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4352 * 'packet', which arrived on 'in_port'. */
4354 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4355 const struct nlattr *odp_actions, size_t actions_len,
4356 struct ofpbuf *packet)
4358 struct odputil_keybuf keybuf;
4362 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4363 odp_flow_key_from_flow(&key, flow,
4364 ofp_port_to_odp_port(ofproto, flow->in_port));
4366 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4367 odp_actions, actions_len, packet);
4371 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4373 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4374 * rule's statistics, via subfacet_uninstall().
4376 * - Removes 'facet' from its rule and from ofproto->facets.
4379 facet_remove(struct facet *facet)
4381 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4382 struct subfacet *subfacet, *next_subfacet;
4384 ovs_assert(!list_is_empty(&facet->subfacets));
4386 /* First uninstall all of the subfacets to get final statistics. */
4387 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4388 subfacet_uninstall(subfacet);
4391 /* Flush the final stats to the rule.
4393 * This might require us to have at least one subfacet around so that we
4394 * can use its actions for accounting in facet_account(), which is why we
4395 * have uninstalled but not yet destroyed the subfacets. */
4396 facet_flush_stats(facet);
4398 /* Now we're really all done so destroy everything. */
4399 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4400 &facet->subfacets) {
4401 subfacet_destroy__(subfacet);
4403 hmap_remove(&ofproto->facets, &facet->hmap_node);
4404 list_remove(&facet->list_node);
4408 /* Feed information from 'facet' back into the learning table to keep it in
4409 * sync with what is actually flowing through the datapath. */
4411 facet_learn(struct facet *facet)
4413 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4414 struct subfacet *subfacet= CONTAINER_OF(list_front(&facet->subfacets),
4415 struct subfacet, list_node);
4416 struct action_xlate_ctx ctx;
4418 if (!facet->has_learn
4419 && !facet->has_normal
4420 && (!facet->has_fin_timeout
4421 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4425 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4426 &subfacet->initial_vals,
4427 facet->rule, facet->tcp_flags, NULL);
4428 ctx.may_learn = true;
4429 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
4430 facet->rule->up.ofpacts_len);
4434 facet_account(struct facet *facet)
4436 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4437 struct subfacet *subfacet;
4438 const struct nlattr *a;
4443 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
4446 n_bytes = facet->byte_count - facet->accounted_bytes;
4448 /* This loop feeds byte counters to bond_account() for rebalancing to use
4449 * as a basis. We also need to track the actual VLAN on which the packet
4450 * is going to be sent to ensure that it matches the one passed to
4451 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4454 * We use the actions from an arbitrary subfacet because they should all
4455 * be equally valid for our purpose. */
4456 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
4457 struct subfacet, list_node);
4458 vlan_tci = facet->flow.vlan_tci;
4459 NL_ATTR_FOR_EACH_UNSAFE (a, left,
4460 subfacet->actions, subfacet->actions_len) {
4461 const struct ovs_action_push_vlan *vlan;
4462 struct ofport_dpif *port;
4464 switch (nl_attr_type(a)) {
4465 case OVS_ACTION_ATTR_OUTPUT:
4466 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4467 if (port && port->bundle && port->bundle->bond) {
4468 bond_account(port->bundle->bond, &facet->flow,
4469 vlan_tci_to_vid(vlan_tci), n_bytes);
4473 case OVS_ACTION_ATTR_POP_VLAN:
4474 vlan_tci = htons(0);
4477 case OVS_ACTION_ATTR_PUSH_VLAN:
4478 vlan = nl_attr_get(a);
4479 vlan_tci = vlan->vlan_tci;
4485 /* Returns true if the only action for 'facet' is to send to the controller.
4486 * (We don't report NetFlow expiration messages for such facets because they
4487 * are just part of the control logic for the network, not real traffic). */
4489 facet_is_controller_flow(struct facet *facet)
4492 const struct rule *rule = &facet->rule->up;
4493 const struct ofpact *ofpacts = rule->ofpacts;
4494 size_t ofpacts_len = rule->ofpacts_len;
4496 if (ofpacts_len > 0 &&
4497 ofpacts->type == OFPACT_CONTROLLER &&
4498 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4505 /* Folds all of 'facet''s statistics into its rule. Also updates the
4506 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4507 * 'facet''s statistics in the datapath should have been zeroed and folded into
4508 * its packet and byte counts before this function is called. */
4510 facet_flush_stats(struct facet *facet)
4512 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4513 struct subfacet *subfacet;
4515 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4516 ovs_assert(!subfacet->dp_byte_count);
4517 ovs_assert(!subfacet->dp_packet_count);
4520 facet_push_stats(facet);
4521 if (facet->accounted_bytes < facet->byte_count) {
4522 facet_account(facet);
4523 facet->accounted_bytes = facet->byte_count;
4526 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4527 struct ofexpired expired;
4528 expired.flow = facet->flow;
4529 expired.packet_count = facet->packet_count;
4530 expired.byte_count = facet->byte_count;
4531 expired.used = facet->used;
4532 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4535 facet->rule->packet_count += facet->packet_count;
4536 facet->rule->byte_count += facet->byte_count;
4538 /* Reset counters to prevent double counting if 'facet' ever gets
4540 facet_reset_counters(facet);
4542 netflow_flow_clear(&facet->nf_flow);
4543 facet->tcp_flags = 0;
4546 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4547 * Returns it if found, otherwise a null pointer.
4549 * 'hash' must be the return value of flow_hash(flow, 0).
4551 * The returned facet might need revalidation; use facet_lookup_valid()
4552 * instead if that is important. */
4553 static struct facet *
4554 facet_find(struct ofproto_dpif *ofproto,
4555 const struct flow *flow, uint32_t hash)
4557 struct facet *facet;
4559 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4560 if (flow_equal(flow, &facet->flow)) {
4568 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4569 * Returns it if found, otherwise a null pointer.
4571 * 'hash' must be the return value of flow_hash(flow, 0).
4573 * The returned facet is guaranteed to be valid. */
4574 static struct facet *
4575 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4578 struct facet *facet;
4580 facet = facet_find(ofproto, flow, hash);
4582 && (ofproto->backer->need_revalidate
4583 || tag_set_intersects(&ofproto->backer->revalidate_set,
4585 facet_revalidate(facet);
4587 /* facet_revalidate() may have destroyed 'facet'. */
4588 facet = facet_find(ofproto, flow, hash);
4595 subfacet_path_to_string(enum subfacet_path path)
4598 case SF_NOT_INSTALLED:
4599 return "not installed";
4601 return "in fast path";
4603 return "in slow path";
4609 /* Returns the path in which a subfacet should be installed if its 'slow'
4610 * member has the specified value. */
4611 static enum subfacet_path
4612 subfacet_want_path(enum slow_path_reason slow)
4614 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
4617 /* Returns true if 'subfacet' needs to have its datapath flow updated,
4618 * supposing that its actions have been recalculated as 'want_actions' and that
4619 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
4621 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
4622 const struct ofpbuf *want_actions)
4624 enum subfacet_path want_path = subfacet_want_path(slow);
4625 return (want_path != subfacet->path
4626 || (want_path == SF_FAST_PATH
4627 && (subfacet->actions_len != want_actions->size
4628 || memcmp(subfacet->actions, want_actions->data,
4629 subfacet->actions_len))));
4633 facet_check_consistency(struct facet *facet)
4635 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4637 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4639 uint64_t odp_actions_stub[1024 / 8];
4640 struct ofpbuf odp_actions;
4642 struct rule_dpif *rule;
4643 struct subfacet *subfacet;
4644 bool may_log = false;
4647 /* Check the rule for consistency. */
4648 rule = rule_dpif_lookup(ofproto, &facet->flow);
4649 ok = rule == facet->rule;
4651 may_log = !VLOG_DROP_WARN(&rl);
4656 flow_format(&s, &facet->flow);
4657 ds_put_format(&s, ": facet associated with wrong rule (was "
4658 "table=%"PRIu8",", facet->rule->up.table_id);
4659 cls_rule_format(&facet->rule->up.cr, &s);
4660 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4662 cls_rule_format(&rule->up.cr, &s);
4663 ds_put_char(&s, ')');
4665 VLOG_WARN("%s", ds_cstr(&s));
4670 /* Check the datapath actions for consistency. */
4671 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4672 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4673 enum subfacet_path want_path;
4674 struct action_xlate_ctx ctx;
4677 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4678 &subfacet->initial_vals, rule, 0, NULL);
4679 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
4682 if (subfacet->path == SF_NOT_INSTALLED) {
4683 /* This only happens if the datapath reported an error when we
4684 * tried to install the flow. Don't flag another error here. */
4688 want_path = subfacet_want_path(subfacet->slow);
4689 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
4690 /* The actions for slow-path flows may legitimately vary from one
4691 * packet to the next. We're done. */
4695 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
4699 /* Inconsistency! */
4701 may_log = !VLOG_DROP_WARN(&rl);
4705 /* Rate-limited, skip reporting. */
4710 odp_flow_key_format(subfacet->key, subfacet->key_len, &s);
4712 ds_put_cstr(&s, ": inconsistency in subfacet");
4713 if (want_path != subfacet->path) {
4714 enum odp_key_fitness fitness = subfacet->key_fitness;
4716 ds_put_format(&s, " (%s, fitness=%s)",
4717 subfacet_path_to_string(subfacet->path),
4718 odp_key_fitness_to_string(fitness));
4719 ds_put_format(&s, " (should have been %s)",
4720 subfacet_path_to_string(want_path));
4721 } else if (want_path == SF_FAST_PATH) {
4722 ds_put_cstr(&s, " (actions were: ");
4723 format_odp_actions(&s, subfacet->actions,
4724 subfacet->actions_len);
4725 ds_put_cstr(&s, ") (correct actions: ");
4726 format_odp_actions(&s, odp_actions.data, odp_actions.size);
4727 ds_put_char(&s, ')');
4729 ds_put_cstr(&s, " (actions: ");
4730 format_odp_actions(&s, subfacet->actions,
4731 subfacet->actions_len);
4732 ds_put_char(&s, ')');
4734 VLOG_WARN("%s", ds_cstr(&s));
4737 ofpbuf_uninit(&odp_actions);
4742 /* Re-searches the classifier for 'facet':
4744 * - If the rule found is different from 'facet''s current rule, moves
4745 * 'facet' to the new rule and recompiles its actions.
4747 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4748 * where it is and recompiles its actions anyway.
4750 * - If any of 'facet''s subfacets correspond to a new flow according to
4751 * ofproto_receive(), 'facet' is removed. */
4753 facet_revalidate(struct facet *facet)
4755 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4757 struct nlattr *odp_actions;
4760 struct actions *new_actions;
4762 struct action_xlate_ctx ctx;
4763 uint64_t odp_actions_stub[1024 / 8];
4764 struct ofpbuf odp_actions;
4766 struct rule_dpif *new_rule;
4767 struct subfacet *subfacet;
4770 COVERAGE_INC(facet_revalidate);
4772 /* Check that child subfacets still correspond to this facet. Tunnel
4773 * configuration changes could cause a subfacet's OpenFlow in_port to
4775 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4776 struct ofproto_dpif *recv_ofproto;
4777 struct flow recv_flow;
4780 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
4781 subfacet->key_len, &recv_flow, NULL,
4782 &recv_ofproto, NULL, NULL);
4784 || recv_ofproto != ofproto
4785 || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) {
4786 facet_remove(facet);
4791 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
4793 /* Calculate new datapath actions.
4795 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4796 * emit a NetFlow expiration and, if so, we need to have the old state
4797 * around to properly compose it. */
4799 /* If the datapath actions changed or the installability changed,
4800 * then we need to talk to the datapath. */
4803 memset(&ctx, 0, sizeof ctx);
4804 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4805 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4806 enum slow_path_reason slow;
4808 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4809 &subfacet->initial_vals, new_rule, 0, NULL);
4810 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
4813 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4814 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
4815 struct dpif_flow_stats stats;
4817 subfacet_install(subfacet,
4818 odp_actions.data, odp_actions.size, &stats, slow);
4819 subfacet_update_stats(subfacet, &stats);
4822 new_actions = xcalloc(list_size(&facet->subfacets),
4823 sizeof *new_actions);
4825 new_actions[i].odp_actions = xmemdup(odp_actions.data,
4827 new_actions[i].actions_len = odp_actions.size;
4832 ofpbuf_uninit(&odp_actions);
4835 facet_flush_stats(facet);
4838 /* Update 'facet' now that we've taken care of all the old state. */
4839 facet->tags = ctx.tags;
4840 facet->nf_flow.output_iface = ctx.nf_output_iface;
4841 facet->has_learn = ctx.has_learn;
4842 facet->has_normal = ctx.has_normal;
4843 facet->has_fin_timeout = ctx.has_fin_timeout;
4844 facet->mirrors = ctx.mirrors;
4847 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4848 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4850 if (new_actions && new_actions[i].odp_actions) {
4851 free(subfacet->actions);
4852 subfacet->actions = new_actions[i].odp_actions;
4853 subfacet->actions_len = new_actions[i].actions_len;
4859 if (facet->rule != new_rule) {
4860 COVERAGE_INC(facet_changed_rule);
4861 list_remove(&facet->list_node);
4862 list_push_back(&new_rule->facets, &facet->list_node);
4863 facet->rule = new_rule;
4864 facet->used = new_rule->up.created;
4865 facet->prev_used = facet->used;
4869 /* Updates 'facet''s used time. Caller is responsible for calling
4870 * facet_push_stats() to update the flows which 'facet' resubmits into. */
4872 facet_update_time(struct facet *facet, long long int used)
4874 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4875 if (used > facet->used) {
4877 ofproto_rule_update_used(&facet->rule->up, used);
4878 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
4883 facet_reset_counters(struct facet *facet)
4885 facet->packet_count = 0;
4886 facet->byte_count = 0;
4887 facet->prev_packet_count = 0;
4888 facet->prev_byte_count = 0;
4889 facet->accounted_bytes = 0;
4893 facet_push_stats(struct facet *facet)
4895 struct dpif_flow_stats stats;
4897 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4898 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4899 ovs_assert(facet->used >= facet->prev_used);
4901 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4902 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4903 stats.used = facet->used;
4904 stats.tcp_flags = 0;
4906 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
4907 facet->prev_packet_count = facet->packet_count;
4908 facet->prev_byte_count = facet->byte_count;
4909 facet->prev_used = facet->used;
4911 flow_push_stats(facet, &stats);
4913 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
4914 facet->mirrors, stats.n_packets, stats.n_bytes);
4919 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4921 rule->packet_count += stats->n_packets;
4922 rule->byte_count += stats->n_bytes;
4923 ofproto_rule_update_used(&rule->up, stats->used);
4926 /* Pushes flow statistics to the rules which 'facet->flow' resubmits
4927 * into given 'facet->rule''s actions and mirrors. */
4929 flow_push_stats(struct facet *facet, const struct dpif_flow_stats *stats)
4931 struct rule_dpif *rule = facet->rule;
4932 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4933 struct subfacet *subfacet = CONTAINER_OF(list_front(&facet->subfacets),
4934 struct subfacet, list_node);
4935 struct action_xlate_ctx ctx;
4937 ofproto_rule_update_used(&rule->up, stats->used);
4939 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4940 &subfacet->initial_vals, rule, 0, NULL);
4941 ctx.resubmit_stats = stats;
4942 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
4943 rule->up.ofpacts_len);
4948 static struct subfacet *
4949 subfacet_find(struct ofproto_dpif *ofproto,
4950 const struct nlattr *key, size_t key_len, uint32_t key_hash)
4952 struct subfacet *subfacet;
4954 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4955 &ofproto->subfacets) {
4956 if (subfacet->key_len == key_len
4957 && !memcmp(key, subfacet->key, key_len)) {
4965 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4966 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4967 * existing subfacet if there is one, otherwise creates and returns a
4970 * If the returned subfacet is new, then subfacet->actions will be NULL, in
4971 * which case the caller must populate the actions with
4972 * subfacet_make_actions(). */
4973 static struct subfacet *
4974 subfacet_create(struct facet *facet, struct flow_miss *miss,
4977 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4978 enum odp_key_fitness key_fitness = miss->key_fitness;
4979 const struct nlattr *key = miss->key;
4980 size_t key_len = miss->key_len;
4982 struct subfacet *subfacet;
4984 key_hash = odp_flow_key_hash(key, key_len);
4986 if (list_is_empty(&facet->subfacets)) {
4987 subfacet = &facet->one_subfacet;
4989 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4991 if (subfacet->facet == facet) {
4995 /* This shouldn't happen. */
4996 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4997 subfacet_destroy(subfacet);
5000 subfacet = xmalloc(sizeof *subfacet);
5003 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
5004 list_push_back(&facet->subfacets, &subfacet->list_node);
5005 subfacet->facet = facet;
5006 subfacet->key_fitness = key_fitness;
5007 subfacet->key = xmemdup(key, key_len);
5008 subfacet->key_len = key_len;
5009 subfacet->used = now;
5010 subfacet->dp_packet_count = 0;
5011 subfacet->dp_byte_count = 0;
5012 subfacet->actions_len = 0;
5013 subfacet->actions = NULL;
5014 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
5017 subfacet->path = SF_NOT_INSTALLED;
5018 subfacet->initial_vals = miss->initial_vals;
5019 subfacet->odp_in_port = miss->odp_in_port;
5024 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5025 * its facet within 'ofproto', and frees it. */
5027 subfacet_destroy__(struct subfacet *subfacet)
5029 struct facet *facet = subfacet->facet;
5030 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5032 subfacet_uninstall(subfacet);
5033 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5034 list_remove(&subfacet->list_node);
5035 free(subfacet->key);
5036 free(subfacet->actions);
5037 if (subfacet != &facet->one_subfacet) {
5042 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5043 * last remaining subfacet in its facet destroys the facet too. */
5045 subfacet_destroy(struct subfacet *subfacet)
5047 struct facet *facet = subfacet->facet;
5049 if (list_is_singleton(&facet->subfacets)) {
5050 /* facet_remove() needs at least one subfacet (it will remove it). */
5051 facet_remove(facet);
5053 subfacet_destroy__(subfacet);
5058 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5059 struct subfacet **subfacets, int n)
5061 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5062 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5063 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5066 for (i = 0; i < n; i++) {
5067 ops[i].type = DPIF_OP_FLOW_DEL;
5068 ops[i].u.flow_del.key = subfacets[i]->key;
5069 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5070 ops[i].u.flow_del.stats = &stats[i];
5074 dpif_operate(ofproto->backer->dpif, opsp, n);
5075 for (i = 0; i < n; i++) {
5076 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5077 subfacets[i]->path = SF_NOT_INSTALLED;
5078 subfacet_destroy(subfacets[i]);
5082 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
5083 * Translates the actions into 'odp_actions', which the caller must have
5084 * initialized and is responsible for uninitializing. */
5086 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
5087 struct ofpbuf *odp_actions)
5089 struct facet *facet = subfacet->facet;
5090 struct rule_dpif *rule = facet->rule;
5091 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5093 struct action_xlate_ctx ctx;
5095 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5096 &subfacet->initial_vals, rule, 0, packet);
5097 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
5098 facet->tags = ctx.tags;
5099 facet->has_learn = ctx.has_learn;
5100 facet->has_normal = ctx.has_normal;
5101 facet->has_fin_timeout = ctx.has_fin_timeout;
5102 facet->nf_flow.output_iface = ctx.nf_output_iface;
5103 facet->mirrors = ctx.mirrors;
5105 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
5106 if (subfacet->actions_len != odp_actions->size
5107 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
5108 free(subfacet->actions);
5109 subfacet->actions_len = odp_actions->size;
5110 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
5114 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5115 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5116 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5117 * since 'subfacet' was last updated.
5119 * Returns 0 if successful, otherwise a positive errno value. */
5121 subfacet_install(struct subfacet *subfacet,
5122 const struct nlattr *actions, size_t actions_len,
5123 struct dpif_flow_stats *stats,
5124 enum slow_path_reason slow)
5126 struct facet *facet = subfacet->facet;
5127 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5128 enum subfacet_path path = subfacet_want_path(slow);
5129 uint64_t slow_path_stub[128 / 8];
5130 enum dpif_flow_put_flags flags;
5133 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5135 flags |= DPIF_FP_ZERO_STATS;
5138 if (path == SF_SLOW_PATH) {
5139 compose_slow_path(ofproto, &facet->flow, slow,
5140 slow_path_stub, sizeof slow_path_stub,
5141 &actions, &actions_len);
5144 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5145 subfacet->key_len, actions, actions_len, stats);
5148 subfacet_reset_dp_stats(subfacet, stats);
5152 subfacet->path = path;
5158 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
5160 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
5161 stats, subfacet->slow);
5164 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5166 subfacet_uninstall(struct subfacet *subfacet)
5168 if (subfacet->path != SF_NOT_INSTALLED) {
5169 struct rule_dpif *rule = subfacet->facet->rule;
5170 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5171 struct dpif_flow_stats stats;
5174 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5175 subfacet->key_len, &stats);
5176 subfacet_reset_dp_stats(subfacet, &stats);
5178 subfacet_update_stats(subfacet, &stats);
5180 subfacet->path = SF_NOT_INSTALLED;
5182 ovs_assert(subfacet->dp_packet_count == 0);
5183 ovs_assert(subfacet->dp_byte_count == 0);
5187 /* Resets 'subfacet''s datapath statistics counters. This should be called
5188 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5189 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5190 * was reset in the datapath. 'stats' will be modified to include only
5191 * statistics new since 'subfacet' was last updated. */
5193 subfacet_reset_dp_stats(struct subfacet *subfacet,
5194 struct dpif_flow_stats *stats)
5197 && subfacet->dp_packet_count <= stats->n_packets
5198 && subfacet->dp_byte_count <= stats->n_bytes) {
5199 stats->n_packets -= subfacet->dp_packet_count;
5200 stats->n_bytes -= subfacet->dp_byte_count;
5203 subfacet->dp_packet_count = 0;
5204 subfacet->dp_byte_count = 0;
5207 /* Updates 'subfacet''s used time. The caller is responsible for calling
5208 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
5210 subfacet_update_time(struct subfacet *subfacet, long long int used)
5212 if (used > subfacet->used) {
5213 subfacet->used = used;
5214 facet_update_time(subfacet->facet, used);
5218 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5220 * Because of the meaning of a subfacet's counters, it only makes sense to do
5221 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5222 * represents a packet that was sent by hand or if it represents statistics
5223 * that have been cleared out of the datapath. */
5225 subfacet_update_stats(struct subfacet *subfacet,
5226 const struct dpif_flow_stats *stats)
5228 if (stats->n_packets || stats->used > subfacet->used) {
5229 struct facet *facet = subfacet->facet;
5231 subfacet_update_time(subfacet, stats->used);
5232 facet->packet_count += stats->n_packets;
5233 facet->byte_count += stats->n_bytes;
5234 facet->tcp_flags |= stats->tcp_flags;
5235 facet_push_stats(facet);
5236 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
5242 static struct rule_dpif *
5243 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5245 struct rule_dpif *rule;
5247 rule = rule_dpif_lookup__(ofproto, flow, 0);
5252 return rule_dpif_miss_rule(ofproto, flow);
5255 static struct rule_dpif *
5256 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5259 struct cls_rule *cls_rule;
5260 struct classifier *cls;
5262 if (table_id >= N_TABLES) {
5266 cls = &ofproto->up.tables[table_id].cls;
5267 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5268 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5269 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5270 * are unavailable. */
5271 struct flow ofpc_normal_flow = *flow;
5272 ofpc_normal_flow.tp_src = htons(0);
5273 ofpc_normal_flow.tp_dst = htons(0);
5274 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5276 cls_rule = classifier_lookup(cls, flow);
5278 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5281 static struct rule_dpif *
5282 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5284 struct ofport_dpif *port;
5286 port = get_ofp_port(ofproto, flow->in_port);
5288 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5289 return ofproto->miss_rule;
5292 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5293 return ofproto->no_packet_in_rule;
5295 return ofproto->miss_rule;
5299 complete_operation(struct rule_dpif *rule)
5301 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5303 rule_invalidate(rule);
5305 struct dpif_completion *c = xmalloc(sizeof *c);
5306 c->op = rule->up.pending;
5307 list_push_back(&ofproto->completions, &c->list_node);
5309 ofoperation_complete(rule->up.pending, 0);
5313 static struct rule *
5316 struct rule_dpif *rule = xmalloc(sizeof *rule);
5321 rule_dealloc(struct rule *rule_)
5323 struct rule_dpif *rule = rule_dpif_cast(rule_);
5328 rule_construct(struct rule *rule_)
5330 struct rule_dpif *rule = rule_dpif_cast(rule_);
5331 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5332 struct rule_dpif *victim;
5335 rule->packet_count = 0;
5336 rule->byte_count = 0;
5338 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5339 if (victim && !list_is_empty(&victim->facets)) {
5340 struct facet *facet;
5342 rule->facets = victim->facets;
5343 list_moved(&rule->facets);
5344 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5345 /* XXX: We're only clearing our local counters here. It's possible
5346 * that quite a few packets are unaccounted for in the datapath
5347 * statistics. These will be accounted to the new rule instead of
5348 * cleared as required. This could be fixed by clearing out the
5349 * datapath statistics for this facet, but currently it doesn't
5351 facet_reset_counters(facet);
5355 /* Must avoid list_moved() in this case. */
5356 list_init(&rule->facets);
5359 table_id = rule->up.table_id;
5361 rule->tag = victim->tag;
5362 } else if (table_id == 0) {
5367 miniflow_expand(&rule->up.cr.match.flow, &flow);
5368 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5369 ofproto->tables[table_id].basis);
5372 complete_operation(rule);
5377 rule_destruct(struct rule *rule_)
5379 struct rule_dpif *rule = rule_dpif_cast(rule_);
5380 struct facet *facet, *next_facet;
5382 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5383 facet_revalidate(facet);
5386 complete_operation(rule);
5390 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5392 struct rule_dpif *rule = rule_dpif_cast(rule_);
5393 struct facet *facet;
5395 /* Start from historical data for 'rule' itself that are no longer tracked
5396 * in facets. This counts, for example, facets that have expired. */
5397 *packets = rule->packet_count;
5398 *bytes = rule->byte_count;
5400 /* Add any statistics that are tracked by facets. This includes
5401 * statistical data recently updated by ofproto_update_stats() as well as
5402 * stats for packets that were executed "by hand" via dpif_execute(). */
5403 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5404 *packets += facet->packet_count;
5405 *bytes += facet->byte_count;
5410 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5411 struct ofpbuf *packet)
5413 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5414 struct initial_vals initial_vals;
5415 struct dpif_flow_stats stats;
5416 struct action_xlate_ctx ctx;
5417 uint64_t odp_actions_stub[1024 / 8];
5418 struct ofpbuf odp_actions;
5420 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5421 rule_credit_stats(rule, &stats);
5423 initial_vals.vlan_tci = flow->vlan_tci;
5424 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5425 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals,
5426 rule, stats.tcp_flags, packet);
5427 ctx.resubmit_stats = &stats;
5428 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
5430 execute_odp_actions(ofproto, flow, odp_actions.data,
5431 odp_actions.size, packet);
5433 ofpbuf_uninit(&odp_actions);
5437 rule_execute(struct rule *rule, const struct flow *flow,
5438 struct ofpbuf *packet)
5440 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5441 ofpbuf_delete(packet);
5446 rule_modify_actions(struct rule *rule_)
5448 struct rule_dpif *rule = rule_dpif_cast(rule_);
5450 complete_operation(rule);
5453 /* Sends 'packet' out 'ofport'.
5454 * May modify 'packet'.
5455 * Returns 0 if successful, otherwise a positive errno value. */
5457 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5459 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5460 uint64_t odp_actions_stub[1024 / 8];
5461 struct ofpbuf key, odp_actions;
5462 struct odputil_keybuf keybuf;
5467 flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow);
5468 if (netdev_vport_is_patch(ofport->up.netdev)) {
5469 struct ofproto_dpif *peer_ofproto;
5470 struct dpif_flow_stats stats;
5471 struct ofport_dpif *peer;
5472 struct rule_dpif *rule;
5474 peer = ofport_get_peer(ofport);
5479 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5480 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5481 netdev_vport_inc_rx(peer->up.netdev, &stats);
5483 flow.in_port = peer->up.ofp_port;
5484 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5485 rule = rule_dpif_lookup(peer_ofproto, &flow);
5486 rule_dpif_execute(rule, &flow, packet);
5491 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5493 if (ofport->tnl_port) {
5494 struct dpif_flow_stats stats;
5496 odp_port = tnl_port_send(ofport->tnl_port, &flow);
5497 if (odp_port == OVSP_NONE) {
5501 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5502 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5503 odp_put_tunnel_action(&flow.tunnel, &odp_actions);
5504 odp_put_skb_mark_action(flow.skb_mark, &odp_actions);
5506 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5508 if (odp_port != ofport->odp_port) {
5509 eth_pop_vlan(packet);
5510 flow.vlan_tci = htons(0);
5514 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5515 odp_flow_key_from_flow(&key, &flow,
5516 ofp_port_to_odp_port(ofproto, flow.in_port));
5518 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5520 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5521 error = dpif_execute(ofproto->backer->dpif,
5523 odp_actions.data, odp_actions.size,
5525 ofpbuf_uninit(&odp_actions);
5528 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5529 ofproto->up.name, odp_port, strerror(error));
5531 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5535 /* OpenFlow to datapath action translation. */
5537 static bool may_receive(const struct ofport_dpif *, struct action_xlate_ctx *);
5538 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5539 struct action_xlate_ctx *);
5540 static void xlate_normal(struct action_xlate_ctx *);
5542 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5543 * The action will state 'slow' as the reason that the action is in the slow
5544 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5545 * dump-flows" output to see why a flow is in the slow path.)
5547 * The 'stub_size' bytes in 'stub' will be used to store the action.
5548 * 'stub_size' must be large enough for the action.
5550 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5553 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5554 enum slow_path_reason slow,
5555 uint64_t *stub, size_t stub_size,
5556 const struct nlattr **actionsp, size_t *actions_lenp)
5558 union user_action_cookie cookie;
5561 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5562 cookie.slow_path.unused = 0;
5563 cookie.slow_path.reason = slow;
5565 ofpbuf_use_stack(&buf, stub, stub_size);
5566 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
5567 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5568 odp_put_userspace_action(pid, &cookie, sizeof cookie, &buf);
5570 put_userspace_action(ofproto, &buf, flow, &cookie);
5572 *actionsp = buf.data;
5573 *actions_lenp = buf.size;
5577 put_userspace_action(const struct ofproto_dpif *ofproto,
5578 struct ofpbuf *odp_actions,
5579 const struct flow *flow,
5580 const union user_action_cookie *cookie)
5584 pid = dpif_port_get_pid(ofproto->backer->dpif,
5585 ofp_port_to_odp_port(ofproto, flow->in_port));
5587 return odp_put_userspace_action(pid, cookie, sizeof *cookie, odp_actions);
5591 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5592 ovs_be16 vlan_tci, uint32_t odp_port,
5593 unsigned int n_outputs, union user_action_cookie *cookie)
5597 cookie->type = USER_ACTION_COOKIE_SFLOW;
5598 cookie->sflow.vlan_tci = vlan_tci;
5600 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5601 * port information") for the interpretation of cookie->output. */
5602 switch (n_outputs) {
5604 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5605 cookie->sflow.output = 0x40000000 | 256;
5609 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5611 cookie->sflow.output = ifindex;
5616 /* 0x80000000 means "multiple output ports. */
5617 cookie->sflow.output = 0x80000000 | n_outputs;
5622 /* Compose SAMPLE action for sFlow. */
5624 compose_sflow_action(const struct ofproto_dpif *ofproto,
5625 struct ofpbuf *odp_actions,
5626 const struct flow *flow,
5629 uint32_t probability;
5630 union user_action_cookie cookie;
5631 size_t sample_offset, actions_offset;
5634 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5638 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5640 /* Number of packets out of UINT_MAX to sample. */
5641 probability = dpif_sflow_get_probability(ofproto->sflow);
5642 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5644 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5645 compose_sflow_cookie(ofproto, htons(0), odp_port,
5646 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5647 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
5649 nl_msg_end_nested(odp_actions, actions_offset);
5650 nl_msg_end_nested(odp_actions, sample_offset);
5651 return cookie_offset;
5654 /* SAMPLE action must be first action in any given list of actions.
5655 * At this point we do not have all information required to build it. So try to
5656 * build sample action as complete as possible. */
5658 add_sflow_action(struct action_xlate_ctx *ctx)
5660 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5662 &ctx->flow, OVSP_NONE);
5663 ctx->sflow_odp_port = 0;
5664 ctx->sflow_n_outputs = 0;
5667 /* Fix SAMPLE action according to data collected while composing ODP actions.
5668 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5669 * USERSPACE action's user-cookie which is required for sflow. */
5671 fix_sflow_action(struct action_xlate_ctx *ctx)
5673 const struct flow *base = &ctx->base_flow;
5674 union user_action_cookie *cookie;
5676 if (!ctx->user_cookie_offset) {
5680 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
5682 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5684 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5685 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5689 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
5692 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5693 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
5694 ovs_be64 flow_tun_id = ctx->flow.tunnel.tun_id;
5695 uint8_t flow_nw_tos = ctx->flow.nw_tos;
5696 struct priority_to_dscp *pdscp;
5697 uint32_t out_port, odp_port;
5699 /* If 'struct flow' gets additional metadata, we'll need to zero it out
5700 * before traversing a patch port. */
5701 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 19);
5704 xlate_report(ctx, "Nonexistent output port");
5706 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5707 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5709 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5710 xlate_report(ctx, "STP not in forwarding state, skipping output");
5714 if (netdev_vport_is_patch(ofport->up.netdev)) {
5715 struct ofport_dpif *peer = ofport_get_peer(ofport);
5716 struct flow old_flow = ctx->flow;
5717 const struct ofproto_dpif *peer_ofproto;
5718 enum slow_path_reason special;
5719 struct ofport_dpif *in_port;
5722 xlate_report(ctx, "Nonexistent patch port peer");
5726 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5727 if (peer_ofproto->backer != ctx->ofproto->backer) {
5728 xlate_report(ctx, "Patch port peer on a different datapath");
5732 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
5733 ctx->flow.in_port = peer->up.ofp_port;
5734 ctx->flow.metadata = htonll(0);
5735 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
5736 memset(ctx->flow.regs, 0, sizeof ctx->flow.regs);
5738 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5739 special = process_special(ctx->ofproto, &ctx->flow, in_port,
5742 ctx->slow |= special;
5743 } else if (!in_port || may_receive(in_port, ctx)) {
5744 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
5745 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5747 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
5748 * learning action look at the packet, then drop it. */
5749 struct flow old_base_flow = ctx->base_flow;
5750 size_t old_size = ctx->odp_actions->size;
5751 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5752 ctx->base_flow = old_base_flow;
5753 ctx->odp_actions->size = old_size;
5757 ctx->flow = old_flow;
5758 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5760 if (ctx->resubmit_stats) {
5761 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
5762 netdev_vport_inc_rx(peer->up.netdev, ctx->resubmit_stats);
5768 pdscp = get_priority(ofport, ctx->flow.skb_priority);
5770 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5771 ctx->flow.nw_tos |= pdscp->dscp;
5774 odp_port = ofp_port_to_odp_port(ctx->ofproto, ofp_port);
5775 if (ofport->tnl_port) {
5776 odp_port = tnl_port_send(ofport->tnl_port, &ctx->flow);
5777 if (odp_port == OVSP_NONE) {
5778 xlate_report(ctx, "Tunneling decided against output");
5782 if (ctx->resubmit_stats) {
5783 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
5785 out_port = odp_port;
5786 commit_odp_tunnel_action(&ctx->flow, &ctx->base_flow,
5789 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
5790 ctx->flow.vlan_tci);
5791 if (out_port != odp_port) {
5792 ctx->flow.vlan_tci = htons(0);
5795 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
5796 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
5798 ctx->sflow_odp_port = odp_port;
5799 ctx->sflow_n_outputs++;
5800 ctx->nf_output_iface = ofp_port;
5801 ctx->flow.tunnel.tun_id = flow_tun_id;
5802 ctx->flow.vlan_tci = flow_vlan_tci;
5803 ctx->flow.nw_tos = flow_nw_tos;
5807 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
5809 compose_output_action__(ctx, ofp_port, true);
5813 xlate_table_action(struct action_xlate_ctx *ctx,
5814 uint16_t in_port, uint8_t table_id, bool may_packet_in)
5816 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
5817 struct ofproto_dpif *ofproto = ctx->ofproto;
5818 struct rule_dpif *rule;
5819 uint16_t old_in_port;
5820 uint8_t old_table_id;
5822 old_table_id = ctx->table_id;
5823 ctx->table_id = table_id;
5825 /* Look up a flow with 'in_port' as the input port. */
5826 old_in_port = ctx->flow.in_port;
5827 ctx->flow.in_port = in_port;
5828 rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id);
5831 if (table_id > 0 && table_id < N_TABLES) {
5832 struct table_dpif *table = &ofproto->tables[table_id];
5833 if (table->other_table) {
5834 ctx->tags |= (rule && rule->tag
5836 : rule_calculate_tag(&ctx->flow,
5837 &table->other_table->mask,
5842 /* Restore the original input port. Otherwise OFPP_NORMAL and
5843 * OFPP_IN_PORT will have surprising behavior. */
5844 ctx->flow.in_port = old_in_port;
5846 if (ctx->resubmit_hook) {
5847 ctx->resubmit_hook(ctx, rule);
5850 if (rule == NULL && may_packet_in) {
5852 * check if table configuration flags
5853 * OFPTC_TABLE_MISS_CONTROLLER, default.
5854 * OFPTC_TABLE_MISS_CONTINUE,
5855 * OFPTC_TABLE_MISS_DROP
5856 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
5858 rule = rule_dpif_miss_rule(ofproto, &ctx->flow);
5862 struct rule_dpif *old_rule = ctx->rule;
5864 if (ctx->resubmit_stats) {
5865 rule_credit_stats(rule, ctx->resubmit_stats);
5870 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
5871 ctx->rule = old_rule;
5875 ctx->table_id = old_table_id;
5877 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5879 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
5880 MAX_RESUBMIT_RECURSION);
5881 ctx->max_resubmit_trigger = true;
5886 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
5887 const struct ofpact_resubmit *resubmit)
5892 in_port = resubmit->in_port;
5893 if (in_port == OFPP_IN_PORT) {
5894 in_port = ctx->flow.in_port;
5897 table_id = resubmit->table_id;
5898 if (table_id == 255) {
5899 table_id = ctx->table_id;
5902 xlate_table_action(ctx, in_port, table_id, false);
5906 flood_packets(struct action_xlate_ctx *ctx, bool all)
5908 struct ofport_dpif *ofport;
5910 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
5911 uint16_t ofp_port = ofport->up.ofp_port;
5913 if (ofp_port == ctx->flow.in_port) {
5918 compose_output_action__(ctx, ofp_port, false);
5919 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
5920 compose_output_action(ctx, ofp_port);
5924 ctx->nf_output_iface = NF_OUT_FLOOD;
5928 execute_controller_action(struct action_xlate_ctx *ctx, int len,
5929 enum ofp_packet_in_reason reason,
5930 uint16_t controller_id)
5932 struct ofputil_packet_in pin;
5933 struct ofpbuf *packet;
5935 ctx->slow |= SLOW_CONTROLLER;
5940 packet = ofpbuf_clone(ctx->packet);
5942 if (packet->l2 && packet->l3) {
5943 struct eth_header *eh;
5944 uint16_t mpls_depth;
5946 eth_pop_vlan(packet);
5949 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
5950 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
5952 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
5953 eth_push_vlan(packet, ctx->flow.vlan_tci);
5956 mpls_depth = eth_mpls_depth(packet);
5958 if (mpls_depth < ctx->flow.mpls_depth) {
5959 push_mpls(packet, ctx->flow.dl_type, ctx->flow.mpls_lse);
5960 } else if (mpls_depth > ctx->flow.mpls_depth) {
5961 pop_mpls(packet, ctx->flow.dl_type);
5962 } else if (mpls_depth) {
5963 set_mpls_lse(packet, ctx->flow.mpls_lse);
5967 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5968 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
5969 ctx->flow.nw_tos, ctx->flow.nw_ttl);
5973 if (ctx->flow.nw_proto == IPPROTO_TCP) {
5974 packet_set_tcp_port(packet, ctx->flow.tp_src,
5976 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
5977 packet_set_udp_port(packet, ctx->flow.tp_src,
5984 pin.packet = packet->data;
5985 pin.packet_len = packet->size;
5986 pin.reason = reason;
5987 pin.controller_id = controller_id;
5988 pin.table_id = ctx->table_id;
5989 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
5992 flow_get_metadata(&ctx->flow, &pin.fmd);
5994 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
5995 ofpbuf_delete(packet);
5999 execute_mpls_push_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
6001 ovs_assert(eth_type_mpls(eth_type));
6003 if (ctx->base_flow.mpls_depth) {
6004 ctx->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK);
6005 ctx->flow.mpls_depth++;
6010 if (ctx->flow.dl_type == htons(ETH_TYPE_IPV6)) {
6011 label = htonl(0x2); /* IPV6 Explicit Null. */
6013 label = htonl(0x0); /* IPV4 Explicit Null. */
6015 tc = (ctx->flow.nw_tos & IP_DSCP_MASK) >> 2;
6016 ttl = ctx->flow.nw_ttl ? ctx->flow.nw_ttl : 0x40;
6017 ctx->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
6018 ctx->flow.encap_dl_type = ctx->flow.dl_type;
6019 ctx->flow.mpls_depth = 1;
6021 ctx->flow.dl_type = eth_type;
6025 execute_mpls_pop_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
6027 ovs_assert(eth_type_mpls(ctx->flow.dl_type));
6028 ovs_assert(!eth_type_mpls(eth_type));
6030 if (ctx->flow.mpls_depth) {
6031 ctx->flow.mpls_depth--;
6032 ctx->flow.mpls_lse = htonl(0);
6033 if (!ctx->flow.mpls_depth) {
6034 ctx->flow.dl_type = eth_type;
6035 ctx->flow.encap_dl_type = htons(0);
6041 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6043 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
6044 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6048 if (ctx->flow.nw_ttl > 1) {
6054 for (i = 0; i < ids->n_controllers; i++) {
6055 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6059 /* Stop processing for current table. */
6065 execute_set_mpls_ttl_action(struct action_xlate_ctx *ctx, uint8_t ttl)
6067 if (!eth_type_mpls(ctx->flow.dl_type)) {
6071 set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl);
6076 execute_dec_mpls_ttl_action(struct action_xlate_ctx *ctx)
6078 uint8_t ttl = mpls_lse_to_ttl(ctx->flow.mpls_lse);
6080 if (!eth_type_mpls(ctx->flow.dl_type)) {
6086 set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl);
6089 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
6091 /* Stop processing for current table. */
6097 xlate_output_action(struct action_xlate_ctx *ctx,
6098 uint16_t port, uint16_t max_len, bool may_packet_in)
6100 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
6102 ctx->nf_output_iface = NF_OUT_DROP;
6106 compose_output_action(ctx, ctx->flow.in_port);
6109 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
6115 flood_packets(ctx, false);
6118 flood_packets(ctx, true);
6120 case OFPP_CONTROLLER:
6121 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6127 if (port != ctx->flow.in_port) {
6128 compose_output_action(ctx, port);
6130 xlate_report(ctx, "skipping output to input port");
6135 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6136 ctx->nf_output_iface = NF_OUT_FLOOD;
6137 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
6138 ctx->nf_output_iface = prev_nf_output_iface;
6139 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6140 ctx->nf_output_iface != NF_OUT_FLOOD) {
6141 ctx->nf_output_iface = NF_OUT_MULTI;
6146 xlate_output_reg_action(struct action_xlate_ctx *ctx,
6147 const struct ofpact_output_reg *or)
6149 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
6150 if (port <= UINT16_MAX) {
6151 xlate_output_action(ctx, port, or->max_len, false);
6156 xlate_enqueue_action(struct action_xlate_ctx *ctx,
6157 const struct ofpact_enqueue *enqueue)
6159 uint16_t ofp_port = enqueue->port;
6160 uint32_t queue_id = enqueue->queue;
6161 uint32_t flow_priority, priority;
6164 /* Translate queue to priority. */
6165 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6166 queue_id, &priority);
6168 /* Fall back to ordinary output action. */
6169 xlate_output_action(ctx, enqueue->port, 0, false);
6173 /* Check output port. */
6174 if (ofp_port == OFPP_IN_PORT) {
6175 ofp_port = ctx->flow.in_port;
6176 } else if (ofp_port == ctx->flow.in_port) {
6180 /* Add datapath actions. */
6181 flow_priority = ctx->flow.skb_priority;
6182 ctx->flow.skb_priority = priority;
6183 compose_output_action(ctx, ofp_port);
6184 ctx->flow.skb_priority = flow_priority;
6186 /* Update NetFlow output port. */
6187 if (ctx->nf_output_iface == NF_OUT_DROP) {
6188 ctx->nf_output_iface = ofp_port;
6189 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
6190 ctx->nf_output_iface = NF_OUT_MULTI;
6195 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
6197 uint32_t skb_priority;
6199 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6200 queue_id, &skb_priority)) {
6201 ctx->flow.skb_priority = skb_priority;
6203 /* Couldn't translate queue to a priority. Nothing to do. A warning
6204 * has already been logged. */
6208 struct xlate_reg_state {
6214 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6216 struct ofproto_dpif *ofproto = ofproto_;
6217 struct ofport_dpif *port;
6227 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6230 port = get_ofp_port(ofproto, ofp_port);
6231 return port ? port->may_enable : false;
6236 xlate_bundle_action(struct action_xlate_ctx *ctx,
6237 const struct ofpact_bundle *bundle)
6241 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
6242 if (bundle->dst.field) {
6243 nxm_reg_load(&bundle->dst, port, &ctx->flow);
6245 xlate_output_action(ctx, port, 0, false);
6250 xlate_learn_action(struct action_xlate_ctx *ctx,
6251 const struct ofpact_learn *learn)
6253 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6254 struct ofputil_flow_mod fm;
6255 uint64_t ofpacts_stub[1024 / 8];
6256 struct ofpbuf ofpacts;
6259 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6260 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
6262 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6263 if (error && !VLOG_DROP_WARN(&rl)) {
6264 VLOG_WARN("learning action failed to modify flow table (%s)",
6265 ofperr_get_name(error));
6268 ofpbuf_uninit(&ofpacts);
6271 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6272 * means "infinite". */
6274 reduce_timeout(uint16_t max, uint16_t *timeout)
6276 if (max && (!*timeout || *timeout > max)) {
6282 xlate_fin_timeout(struct action_xlate_ctx *ctx,
6283 const struct ofpact_fin_timeout *oft)
6285 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6286 struct rule_dpif *rule = ctx->rule;
6288 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6289 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6294 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
6296 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
6297 ? OFPUTIL_PC_NO_RECV_STP
6298 : OFPUTIL_PC_NO_RECV)) {
6302 /* Only drop packets here if both forwarding and learning are
6303 * disabled. If just learning is enabled, we need to have
6304 * OFPP_NORMAL and the learning action have a look at the packet
6305 * before we can drop it. */
6306 if (!stp_forward_in_state(port->stp_state)
6307 && !stp_learn_in_state(port->stp_state)) {
6315 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6316 struct action_xlate_ctx *ctx)
6318 bool was_evictable = true;
6319 const struct ofpact *a;
6322 /* Don't let the rule we're working on get evicted underneath us. */
6323 was_evictable = ctx->rule->up.evictable;
6324 ctx->rule->up.evictable = false;
6326 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6327 struct ofpact_controller *controller;
6328 const struct ofpact_metadata *metadata;
6336 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6337 ofpact_get_OUTPUT(a)->max_len, true);
6340 case OFPACT_CONTROLLER:
6341 controller = ofpact_get_CONTROLLER(a);
6342 execute_controller_action(ctx, controller->max_len,
6344 controller->controller_id);
6347 case OFPACT_ENQUEUE:
6348 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6351 case OFPACT_SET_VLAN_VID:
6352 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6353 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6357 case OFPACT_SET_VLAN_PCP:
6358 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6359 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
6364 case OFPACT_STRIP_VLAN:
6365 ctx->flow.vlan_tci = htons(0);
6368 case OFPACT_PUSH_VLAN:
6369 /* XXX 802.1AD(QinQ) */
6370 ctx->flow.vlan_tci = htons(VLAN_CFI);
6373 case OFPACT_SET_ETH_SRC:
6374 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6378 case OFPACT_SET_ETH_DST:
6379 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6383 case OFPACT_SET_IPV4_SRC:
6384 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6387 case OFPACT_SET_IPV4_DST:
6388 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6391 case OFPACT_SET_IPV4_DSCP:
6392 /* OpenFlow 1.0 only supports IPv4. */
6393 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6394 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
6395 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6399 case OFPACT_SET_L4_SRC_PORT:
6400 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6403 case OFPACT_SET_L4_DST_PORT:
6404 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6407 case OFPACT_RESUBMIT:
6408 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6411 case OFPACT_SET_TUNNEL:
6412 ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6415 case OFPACT_SET_QUEUE:
6416 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6419 case OFPACT_POP_QUEUE:
6420 ctx->flow.skb_priority = ctx->orig_skb_priority;
6423 case OFPACT_REG_MOVE:
6424 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
6427 case OFPACT_REG_LOAD:
6428 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
6431 case OFPACT_STACK_PUSH:
6432 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->flow,
6436 case OFPACT_STACK_POP:
6437 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->flow,
6441 case OFPACT_PUSH_MPLS:
6442 execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
6445 case OFPACT_POP_MPLS:
6446 execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
6449 case OFPACT_SET_MPLS_TTL:
6450 if (execute_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl)) {
6455 case OFPACT_DEC_MPLS_TTL:
6456 if (execute_dec_mpls_ttl_action(ctx)) {
6461 case OFPACT_DEC_TTL:
6462 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6468 /* Nothing to do. */
6471 case OFPACT_MULTIPATH:
6472 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
6476 ctx->ofproto->has_bundle_action = true;
6477 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6480 case OFPACT_OUTPUT_REG:
6481 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6485 ctx->has_learn = true;
6486 if (ctx->may_learn) {
6487 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6495 case OFPACT_FIN_TIMEOUT:
6496 ctx->has_fin_timeout = true;
6497 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6500 case OFPACT_CLEAR_ACTIONS:
6502 * Nothing to do because writa-actions is not supported for now.
6503 * When writa-actions is supported, clear-actions also must
6504 * be supported at the same time.
6508 case OFPACT_WRITE_METADATA:
6509 metadata = ofpact_get_WRITE_METADATA(a);
6510 ctx->flow.metadata &= ~metadata->mask;
6511 ctx->flow.metadata |= metadata->metadata & metadata->mask;
6514 case OFPACT_GOTO_TABLE: {
6515 /* XXX remove recursion */
6516 /* It is assumed that goto-table is last action */
6517 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6518 ovs_assert(ctx->table_id < ogt->table_id);
6519 xlate_table_action(ctx, ctx->flow.in_port, ogt->table_id, true);
6527 ctx->rule->up.evictable = was_evictable;
6532 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
6533 struct ofproto_dpif *ofproto, const struct flow *flow,
6534 const struct initial_vals *initial_vals,
6535 struct rule_dpif *rule,
6536 uint8_t tcp_flags, const struct ofpbuf *packet)
6538 ovs_be64 initial_tun_id = flow->tunnel.tun_id;
6540 /* Flow initialization rules:
6541 * - 'base_flow' must match the kernel's view of the packet at the
6542 * time that action processing starts. 'flow' represents any
6543 * transformations we wish to make through actions.
6544 * - By default 'base_flow' and 'flow' are the same since the input
6545 * packet matches the output before any actions are applied.
6546 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
6547 * of the received packet as seen by the kernel. If we later output
6548 * to another device without any modifications this will cause us to
6549 * insert a new tag since the original one was stripped off by the
6551 * - Tunnel 'flow' is largely cleared when transitioning between
6552 * the input and output stages since it does not make sense to output
6553 * a packet with the exact headers that it was received with (i.e.
6554 * the destination IP is us). The one exception is the tun_id, which
6555 * is preserved to allow use in later resubmit lookups and loads into
6557 * - Tunnel 'base_flow' is completely cleared since that is what the
6558 * kernel does. If we wish to maintain the original values an action
6559 * needs to be generated. */
6561 ctx->ofproto = ofproto;
6563 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
6564 ctx->base_flow = ctx->flow;
6565 ctx->base_flow.vlan_tci = initial_vals->vlan_tci;
6566 ctx->flow.tunnel.tun_id = initial_tun_id;
6568 ctx->packet = packet;
6569 ctx->may_learn = packet != NULL;
6570 ctx->tcp_flags = tcp_flags;
6571 ctx->resubmit_hook = NULL;
6572 ctx->report_hook = NULL;
6573 ctx->resubmit_stats = NULL;
6576 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6577 * into datapath actions in 'odp_actions', using 'ctx'. */
6579 xlate_actions(struct action_xlate_ctx *ctx,
6580 const struct ofpact *ofpacts, size_t ofpacts_len,
6581 struct ofpbuf *odp_actions)
6583 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6584 * that in the future we always keep a copy of the original flow for
6585 * tracing purposes. */
6586 static bool hit_resubmit_limit;
6588 enum slow_path_reason special;
6589 struct ofport_dpif *in_port;
6590 struct flow orig_flow;
6592 COVERAGE_INC(ofproto_dpif_xlate);
6594 ofpbuf_clear(odp_actions);
6595 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
6597 ctx->odp_actions = odp_actions;
6600 ctx->has_learn = false;
6601 ctx->has_normal = false;
6602 ctx->has_fin_timeout = false;
6603 ctx->nf_output_iface = NF_OUT_DROP;
6606 ctx->max_resubmit_trigger = false;
6607 ctx->orig_skb_priority = ctx->flow.skb_priority;
6611 ofpbuf_use_stub(&ctx->stack, ctx->init_stack, sizeof ctx->init_stack);
6613 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
6614 /* Do this conditionally because the copy is expensive enough that it
6615 * shows up in profiles. */
6616 orig_flow = ctx->flow;
6619 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6620 switch (ctx->ofproto->up.frag_handling) {
6621 case OFPC_FRAG_NORMAL:
6622 /* We must pretend that transport ports are unavailable. */
6623 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
6624 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
6627 case OFPC_FRAG_DROP:
6630 case OFPC_FRAG_REASM:
6633 case OFPC_FRAG_NX_MATCH:
6634 /* Nothing to do. */
6637 case OFPC_INVALID_TTL_TO_CONTROLLER:
6642 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
6643 special = process_special(ctx->ofproto, &ctx->flow, in_port, ctx->packet);
6645 ctx->slow |= special;
6647 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6648 struct initial_vals initial_vals;
6649 uint32_t local_odp_port;
6651 initial_vals.vlan_tci = ctx->base_flow.vlan_tci;
6653 add_sflow_action(ctx);
6655 if (!in_port || may_receive(in_port, ctx)) {
6656 do_xlate_actions(ofpacts, ofpacts_len, ctx);
6658 /* We've let OFPP_NORMAL and the learning action look at the
6659 * packet, so drop it now if forwarding is disabled. */
6660 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
6661 ofpbuf_clear(ctx->odp_actions);
6662 add_sflow_action(ctx);
6666 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
6667 if (!hit_resubmit_limit) {
6668 /* We didn't record the original flow. Make sure we do from
6670 hit_resubmit_limit = true;
6671 } else if (!VLOG_DROP_ERR(&trace_rl)) {
6672 struct ds ds = DS_EMPTY_INITIALIZER;
6674 ofproto_trace(ctx->ofproto, &orig_flow, ctx->packet,
6675 &initial_vals, &ds);
6676 VLOG_ERR("Trace triggered by excessive resubmit "
6677 "recursion:\n%s", ds_cstr(&ds));
6682 local_odp_port = ofp_port_to_odp_port(ctx->ofproto, OFPP_LOCAL);
6683 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
6685 ctx->odp_actions->data,
6686 ctx->odp_actions->size)) {
6687 ctx->slow |= SLOW_IN_BAND;
6689 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
6691 compose_output_action(ctx, OFPP_LOCAL);
6694 if (ctx->ofproto->has_mirrors) {
6695 add_mirror_actions(ctx, &orig_flow);
6697 fix_sflow_action(ctx);
6700 ofpbuf_uninit(&ctx->stack);
6703 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
6704 * into datapath actions, using 'ctx', and discards the datapath actions. */
6706 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
6707 const struct ofpact *ofpacts,
6710 uint64_t odp_actions_stub[1024 / 8];
6711 struct ofpbuf odp_actions;
6713 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
6714 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
6715 ofpbuf_uninit(&odp_actions);
6719 xlate_report(struct action_xlate_ctx *ctx, const char *s)
6721 if (ctx->report_hook) {
6722 ctx->report_hook(ctx, s);
6726 /* OFPP_NORMAL implementation. */
6728 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
6730 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
6731 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
6732 * the bundle on which the packet was received, returns the VLAN to which the
6735 * Both 'vid' and the return value are in the range 0...4095. */
6737 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
6739 switch (in_bundle->vlan_mode) {
6740 case PORT_VLAN_ACCESS:
6741 return in_bundle->vlan;
6744 case PORT_VLAN_TRUNK:
6747 case PORT_VLAN_NATIVE_UNTAGGED:
6748 case PORT_VLAN_NATIVE_TAGGED:
6749 return vid ? vid : in_bundle->vlan;
6756 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
6757 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
6760 * 'vid' should be the VID obtained from the 802.1Q header that was received as
6761 * part of a packet (specify 0 if there was no 802.1Q header), in the range
6764 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
6766 /* Allow any VID on the OFPP_NONE port. */
6767 if (in_bundle == &ofpp_none_bundle) {
6771 switch (in_bundle->vlan_mode) {
6772 case PORT_VLAN_ACCESS:
6775 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6776 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
6777 "packet received on port %s configured as VLAN "
6778 "%"PRIu16" access port",
6779 in_bundle->ofproto->up.name, vid,
6780 in_bundle->name, in_bundle->vlan);
6786 case PORT_VLAN_NATIVE_UNTAGGED:
6787 case PORT_VLAN_NATIVE_TAGGED:
6789 /* Port must always carry its native VLAN. */
6793 case PORT_VLAN_TRUNK:
6794 if (!ofbundle_includes_vlan(in_bundle, vid)) {
6796 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6797 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
6798 "received on port %s not configured for trunking "
6800 in_bundle->ofproto->up.name, vid,
6801 in_bundle->name, vid);
6813 /* Given 'vlan', the VLAN that a packet belongs to, and
6814 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
6815 * that should be included in the 802.1Q header. (If the return value is 0,
6816 * then the 802.1Q header should only be included in the packet if there is a
6819 * Both 'vlan' and the return value are in the range 0...4095. */
6821 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
6823 switch (out_bundle->vlan_mode) {
6824 case PORT_VLAN_ACCESS:
6827 case PORT_VLAN_TRUNK:
6828 case PORT_VLAN_NATIVE_TAGGED:
6831 case PORT_VLAN_NATIVE_UNTAGGED:
6832 return vlan == out_bundle->vlan ? 0 : vlan;
6840 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
6843 struct ofport_dpif *port;
6845 ovs_be16 tci, old_tci;
6847 vid = output_vlan_to_vid(out_bundle, vlan);
6848 if (!out_bundle->bond) {
6849 port = ofbundle_get_a_port(out_bundle);
6851 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
6854 /* No slaves enabled, so drop packet. */
6859 old_tci = ctx->flow.vlan_tci;
6861 if (tci || out_bundle->use_priority_tags) {
6862 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
6864 tci |= htons(VLAN_CFI);
6867 ctx->flow.vlan_tci = tci;
6869 compose_output_action(ctx, port->up.ofp_port);
6870 ctx->flow.vlan_tci = old_tci;
6874 mirror_mask_ffs(mirror_mask_t mask)
6876 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
6881 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
6883 return (bundle->vlan_mode != PORT_VLAN_ACCESS
6884 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
6888 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
6890 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
6893 /* Returns an arbitrary interface within 'bundle'. */
6894 static struct ofport_dpif *
6895 ofbundle_get_a_port(const struct ofbundle *bundle)
6897 return CONTAINER_OF(list_front(&bundle->ports),
6898 struct ofport_dpif, bundle_node);
6902 vlan_is_mirrored(const struct ofmirror *m, int vlan)
6904 return !m->vlans || bitmap_is_set(m->vlans, vlan);
6908 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
6910 struct ofproto_dpif *ofproto = ctx->ofproto;
6911 mirror_mask_t mirrors;
6912 struct ofbundle *in_bundle;
6915 const struct nlattr *a;
6918 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
6919 ctx->packet != NULL, NULL);
6923 mirrors = in_bundle->src_mirrors;
6925 /* Drop frames on bundles reserved for mirroring. */
6926 if (in_bundle->mirror_out) {
6927 if (ctx->packet != NULL) {
6928 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6929 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6930 "%s, which is reserved exclusively for mirroring",
6931 ctx->ofproto->up.name, in_bundle->name);
6937 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
6938 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6941 vlan = input_vid_to_vlan(in_bundle, vid);
6943 /* Look at the output ports to check for destination selections. */
6945 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
6946 ctx->odp_actions->size) {
6947 enum ovs_action_attr type = nl_attr_type(a);
6948 struct ofport_dpif *ofport;
6950 if (type != OVS_ACTION_ATTR_OUTPUT) {
6954 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
6955 if (ofport && ofport->bundle) {
6956 mirrors |= ofport->bundle->dst_mirrors;
6964 /* Restore the original packet before adding the mirror actions. */
6965 ctx->flow = *orig_flow;
6970 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6972 if (!vlan_is_mirrored(m, vlan)) {
6973 mirrors = zero_rightmost_1bit(mirrors);
6977 mirrors &= ~m->dup_mirrors;
6978 ctx->mirrors |= m->dup_mirrors;
6980 output_normal(ctx, m->out, vlan);
6981 } else if (vlan != m->out_vlan
6982 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
6983 struct ofbundle *bundle;
6985 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
6986 if (ofbundle_includes_vlan(bundle, m->out_vlan)
6987 && !bundle->mirror_out) {
6988 output_normal(ctx, bundle, m->out_vlan);
6996 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
6997 uint64_t packets, uint64_t bytes)
7003 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
7006 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7009 /* In normal circumstances 'm' will not be NULL. However,
7010 * if mirrors are reconfigured, we can temporarily get out
7011 * of sync in facet_revalidate(). We could "correct" the
7012 * mirror list before reaching here, but doing that would
7013 * not properly account the traffic stats we've currently
7014 * accumulated for previous mirror configuration. */
7018 m->packet_count += packets;
7019 m->byte_count += bytes;
7023 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
7024 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
7025 * indicate this; newer upstream kernels use gratuitous ARP requests. */
7027 is_gratuitous_arp(const struct flow *flow)
7029 return (flow->dl_type == htons(ETH_TYPE_ARP)
7030 && eth_addr_is_broadcast(flow->dl_dst)
7031 && (flow->nw_proto == ARP_OP_REPLY
7032 || (flow->nw_proto == ARP_OP_REQUEST
7033 && flow->nw_src == flow->nw_dst)));
7037 update_learning_table(struct ofproto_dpif *ofproto,
7038 const struct flow *flow, int vlan,
7039 struct ofbundle *in_bundle)
7041 struct mac_entry *mac;
7043 /* Don't learn the OFPP_NONE port. */
7044 if (in_bundle == &ofpp_none_bundle) {
7048 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7052 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7053 if (is_gratuitous_arp(flow)) {
7054 /* We don't want to learn from gratuitous ARP packets that are
7055 * reflected back over bond slaves so we lock the learning table. */
7056 if (!in_bundle->bond) {
7057 mac_entry_set_grat_arp_lock(mac);
7058 } else if (mac_entry_is_grat_arp_locked(mac)) {
7063 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7064 /* The log messages here could actually be useful in debugging,
7065 * so keep the rate limit relatively high. */
7066 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7067 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7068 "on port %s in VLAN %d",
7069 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7070 in_bundle->name, vlan);
7072 mac->port.p = in_bundle;
7073 tag_set_add(&ofproto->backer->revalidate_set,
7074 mac_learning_changed(ofproto->ml, mac));
7078 static struct ofbundle *
7079 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7080 bool warn, struct ofport_dpif **in_ofportp)
7082 struct ofport_dpif *ofport;
7084 /* Find the port and bundle for the received packet. */
7085 ofport = get_ofp_port(ofproto, in_port);
7087 *in_ofportp = ofport;
7089 if (ofport && ofport->bundle) {
7090 return ofport->bundle;
7093 /* Special-case OFPP_NONE, which a controller may use as the ingress
7094 * port for traffic that it is sourcing. */
7095 if (in_port == OFPP_NONE) {
7096 return &ofpp_none_bundle;
7099 /* Odd. A few possible reasons here:
7101 * - We deleted a port but there are still a few packets queued up
7104 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7105 * we don't know about.
7107 * - The ofproto client didn't configure the port as part of a bundle.
7108 * This is particularly likely to happen if a packet was received on the
7109 * port after it was created, but before the client had a chance to
7110 * configure its bundle.
7113 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7115 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7116 "port %"PRIu16, ofproto->up.name, in_port);
7121 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7122 * dropped. Returns true if they may be forwarded, false if they should be
7125 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7126 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7128 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7129 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7130 * checked by input_vid_is_valid().
7132 * May also add tags to '*tags', although the current implementation only does
7133 * so in one special case.
7136 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
7139 struct ofproto_dpif *ofproto = ctx->ofproto;
7140 struct flow *flow = &ctx->flow;
7141 struct ofbundle *in_bundle = in_port->bundle;
7143 /* Drop frames for reserved multicast addresses
7144 * only if forward_bpdu option is absent. */
7145 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7146 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7150 if (in_bundle->bond) {
7151 struct mac_entry *mac;
7153 switch (bond_check_admissibility(in_bundle->bond, in_port,
7154 flow->dl_dst, &ctx->tags)) {
7159 xlate_report(ctx, "bonding refused admissibility, dropping");
7162 case BV_DROP_IF_MOVED:
7163 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7164 if (mac && mac->port.p != in_bundle &&
7165 (!is_gratuitous_arp(flow)
7166 || mac_entry_is_grat_arp_locked(mac))) {
7167 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7179 xlate_normal(struct action_xlate_ctx *ctx)
7181 struct ofport_dpif *in_port;
7182 struct ofbundle *in_bundle;
7183 struct mac_entry *mac;
7187 ctx->has_normal = true;
7189 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
7190 ctx->packet != NULL, &in_port);
7192 xlate_report(ctx, "no input bundle, dropping");
7196 /* Drop malformed frames. */
7197 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7198 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
7199 if (ctx->packet != NULL) {
7200 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7201 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7202 "VLAN tag received on port %s",
7203 ctx->ofproto->up.name, in_bundle->name);
7205 xlate_report(ctx, "partial VLAN tag, dropping");
7209 /* Drop frames on bundles reserved for mirroring. */
7210 if (in_bundle->mirror_out) {
7211 if (ctx->packet != NULL) {
7212 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7213 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7214 "%s, which is reserved exclusively for mirroring",
7215 ctx->ofproto->up.name, in_bundle->name);
7217 xlate_report(ctx, "input port is mirror output port, dropping");
7222 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
7223 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7224 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7227 vlan = input_vid_to_vlan(in_bundle, vid);
7229 /* Check other admissibility requirements. */
7230 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7234 /* Learn source MAC. */
7235 if (ctx->may_learn) {
7236 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
7239 /* Determine output bundle. */
7240 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
7243 if (mac->port.p != in_bundle) {
7244 xlate_report(ctx, "forwarding to learned port");
7245 output_normal(ctx, mac->port.p, vlan);
7247 xlate_report(ctx, "learned port is input port, dropping");
7250 struct ofbundle *bundle;
7252 xlate_report(ctx, "no learned MAC for destination, flooding");
7253 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7254 if (bundle != in_bundle
7255 && ofbundle_includes_vlan(bundle, vlan)
7256 && bundle->floodable
7257 && !bundle->mirror_out) {
7258 output_normal(ctx, bundle, vlan);
7261 ctx->nf_output_iface = NF_OUT_FLOOD;
7265 /* Optimized flow revalidation.
7267 * It's a difficult problem, in general, to tell which facets need to have
7268 * their actions recalculated whenever the OpenFlow flow table changes. We
7269 * don't try to solve that general problem: for most kinds of OpenFlow flow
7270 * table changes, we recalculate the actions for every facet. This is
7271 * relatively expensive, but it's good enough if the OpenFlow flow table
7272 * doesn't change very often.
7274 * However, we can expect one particular kind of OpenFlow flow table change to
7275 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7276 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7277 * table, we add a special case that applies to flow tables in which every rule
7278 * has the same form (that is, the same wildcards), except that the table is
7279 * also allowed to have a single "catch-all" flow that matches all packets. We
7280 * optimize this case by tagging all of the facets that resubmit into the table
7281 * and invalidating the same tag whenever a flow changes in that table. The
7282 * end result is that we revalidate just the facets that need it (and sometimes
7283 * a few more, but not all of the facets or even all of the facets that
7284 * resubmit to the table modified by MAC learning). */
7286 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7287 * into an OpenFlow table with the given 'basis'. */
7289 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7292 if (minimask_is_catchall(mask)) {
7295 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7296 return tag_create_deterministic(hash);
7300 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7301 * taggability of that table.
7303 * This function must be called after *each* change to a flow table. If you
7304 * skip calling it on some changes then the pointer comparisons at the end can
7305 * be invalid if you get unlucky. For example, if a flow removal causes a
7306 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7307 * different wildcards to be created with the same address, then this function
7308 * will incorrectly skip revalidation. */
7310 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7312 struct table_dpif *table = &ofproto->tables[table_id];
7313 const struct oftable *oftable = &ofproto->up.tables[table_id];
7314 struct cls_table *catchall, *other;
7315 struct cls_table *t;
7317 catchall = other = NULL;
7319 switch (hmap_count(&oftable->cls.tables)) {
7321 /* We could tag this OpenFlow table but it would make the logic a
7322 * little harder and it's a corner case that doesn't seem worth it
7328 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7329 if (cls_table_is_catchall(t)) {
7331 } else if (!other) {
7334 /* Indicate that we can't tag this by setting both tables to
7335 * NULL. (We know that 'catchall' is already NULL.) */
7342 /* Can't tag this table. */
7346 if (table->catchall_table != catchall || table->other_table != other) {
7347 table->catchall_table = catchall;
7348 table->other_table = other;
7349 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7353 /* Given 'rule' that has changed in some way (either it is a rule being
7354 * inserted, a rule being deleted, or a rule whose actions are being
7355 * modified), marks facets for revalidation to ensure that packets will be
7356 * forwarded correctly according to the new state of the flow table.
7358 * This function must be called after *each* change to a flow table. See
7359 * the comment on table_update_taggable() for more information. */
7361 rule_invalidate(const struct rule_dpif *rule)
7363 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7365 table_update_taggable(ofproto, rule->up.table_id);
7367 if (!ofproto->backer->need_revalidate) {
7368 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7370 if (table->other_table && rule->tag) {
7371 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7373 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7379 set_frag_handling(struct ofproto *ofproto_,
7380 enum ofp_config_flags frag_handling)
7382 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7383 if (frag_handling != OFPC_FRAG_REASM) {
7384 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7392 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7393 const struct flow *flow,
7394 const struct ofpact *ofpacts, size_t ofpacts_len)
7396 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7397 struct initial_vals initial_vals;
7398 struct odputil_keybuf keybuf;
7399 struct dpif_flow_stats stats;
7403 struct action_xlate_ctx ctx;
7404 uint64_t odp_actions_stub[1024 / 8];
7405 struct ofpbuf odp_actions;
7407 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7408 odp_flow_key_from_flow(&key, flow,
7409 ofp_port_to_odp_port(ofproto, flow->in_port));
7411 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7413 initial_vals.vlan_tci = flow->vlan_tci;
7414 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals, NULL,
7415 packet_get_tcp_flags(packet, flow), packet);
7416 ctx.resubmit_stats = &stats;
7418 ofpbuf_use_stub(&odp_actions,
7419 odp_actions_stub, sizeof odp_actions_stub);
7420 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
7421 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7422 odp_actions.data, odp_actions.size, packet);
7423 ofpbuf_uninit(&odp_actions);
7431 set_netflow(struct ofproto *ofproto_,
7432 const struct netflow_options *netflow_options)
7434 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7436 if (netflow_options) {
7437 if (!ofproto->netflow) {
7438 ofproto->netflow = netflow_create();
7440 return netflow_set_options(ofproto->netflow, netflow_options);
7442 netflow_destroy(ofproto->netflow);
7443 ofproto->netflow = NULL;
7449 get_netflow_ids(const struct ofproto *ofproto_,
7450 uint8_t *engine_type, uint8_t *engine_id)
7452 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7454 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7458 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7460 if (!facet_is_controller_flow(facet) &&
7461 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7462 struct subfacet *subfacet;
7463 struct ofexpired expired;
7465 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7466 if (subfacet->path == SF_FAST_PATH) {
7467 struct dpif_flow_stats stats;
7469 subfacet_reinstall(subfacet, &stats);
7470 subfacet_update_stats(subfacet, &stats);
7474 expired.flow = facet->flow;
7475 expired.packet_count = facet->packet_count;
7476 expired.byte_count = facet->byte_count;
7477 expired.used = facet->used;
7478 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
7483 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7485 struct facet *facet;
7487 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7488 send_active_timeout(ofproto, facet);
7492 static struct ofproto_dpif *
7493 ofproto_dpif_lookup(const char *name)
7495 struct ofproto_dpif *ofproto;
7497 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7498 hash_string(name, 0), &all_ofproto_dpifs) {
7499 if (!strcmp(ofproto->up.name, name)) {
7507 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7508 const char *argv[], void *aux OVS_UNUSED)
7510 struct ofproto_dpif *ofproto;
7513 ofproto = ofproto_dpif_lookup(argv[1]);
7515 unixctl_command_reply_error(conn, "no such bridge");
7518 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7520 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7521 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7525 unixctl_command_reply(conn, "table successfully flushed");
7529 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7530 const char *argv[], void *aux OVS_UNUSED)
7532 struct ds ds = DS_EMPTY_INITIALIZER;
7533 const struct ofproto_dpif *ofproto;
7534 const struct mac_entry *e;
7536 ofproto = ofproto_dpif_lookup(argv[1]);
7538 unixctl_command_reply_error(conn, "no such bridge");
7542 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7543 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7544 struct ofbundle *bundle = e->port.p;
7545 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7546 ofbundle_get_a_port(bundle)->odp_port,
7547 e->vlan, ETH_ADDR_ARGS(e->mac),
7548 mac_entry_age(ofproto->ml, e));
7550 unixctl_command_reply(conn, ds_cstr(&ds));
7555 struct action_xlate_ctx ctx;
7561 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7562 const struct rule_dpif *rule)
7564 ds_put_char_multiple(result, '\t', level);
7566 ds_put_cstr(result, "No match\n");
7570 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7571 table_id, ntohll(rule->up.flow_cookie));
7572 cls_rule_format(&rule->up.cr, result);
7573 ds_put_char(result, '\n');
7575 ds_put_char_multiple(result, '\t', level);
7576 ds_put_cstr(result, "OpenFlow ");
7577 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7578 ds_put_char(result, '\n');
7582 trace_format_flow(struct ds *result, int level, const char *title,
7583 struct trace_ctx *trace)
7585 ds_put_char_multiple(result, '\t', level);
7586 ds_put_format(result, "%s: ", title);
7587 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
7588 ds_put_cstr(result, "unchanged");
7590 flow_format(result, &trace->ctx.flow);
7591 trace->flow = trace->ctx.flow;
7593 ds_put_char(result, '\n');
7597 trace_format_regs(struct ds *result, int level, const char *title,
7598 struct trace_ctx *trace)
7602 ds_put_char_multiple(result, '\t', level);
7603 ds_put_format(result, "%s:", title);
7604 for (i = 0; i < FLOW_N_REGS; i++) {
7605 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7607 ds_put_char(result, '\n');
7611 trace_format_odp(struct ds *result, int level, const char *title,
7612 struct trace_ctx *trace)
7614 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
7616 ds_put_char_multiple(result, '\t', level);
7617 ds_put_format(result, "%s: ", title);
7618 format_odp_actions(result, odp_actions->data, odp_actions->size);
7619 ds_put_char(result, '\n');
7623 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
7625 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7626 struct ds *result = trace->result;
7628 ds_put_char(result, '\n');
7629 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7630 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
7631 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
7632 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
7636 trace_report(struct action_xlate_ctx *ctx, const char *s)
7638 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7639 struct ds *result = trace->result;
7641 ds_put_char_multiple(result, '\t', ctx->recurse);
7642 ds_put_cstr(result, s);
7643 ds_put_char(result, '\n');
7647 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
7648 void *aux OVS_UNUSED)
7650 const char *dpname = argv[1];
7651 struct ofproto_dpif *ofproto;
7652 struct ofpbuf odp_key;
7653 struct ofpbuf *packet;
7654 struct initial_vals initial_vals;
7660 ofpbuf_init(&odp_key, 0);
7663 ofproto = ofproto_dpif_lookup(dpname);
7665 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
7669 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
7670 /* ofproto/trace dpname flow [-generate] */
7671 const char *flow_s = argv[2];
7672 const char *generate_s = argv[3];
7674 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
7675 * flow. We guess which type it is based on whether 'flow_s' contains
7676 * an '(', since a datapath flow always contains '(') but an
7677 * OpenFlow-like flow should not (in fact it's allowed but I believe
7678 * that's not documented anywhere).
7680 * An alternative would be to try to parse 'flow_s' both ways, but then
7681 * it would be tricky giving a sensible error message. After all, do
7682 * you just say "syntax error" or do you present both error messages?
7683 * Both choices seem lousy. */
7684 if (strchr(flow_s, '(')) {
7687 /* Convert string to datapath key. */
7688 ofpbuf_init(&odp_key, 0);
7689 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
7691 unixctl_command_reply_error(conn, "Bad flow syntax");
7695 /* XXX: Since we allow the user to specify an ofproto, it's
7696 * possible they will specify a different ofproto than the one the
7697 * port actually belongs too. Ideally we should simply remove the
7698 * ability to specify the ofproto. */
7699 if (ofproto_receive(ofproto->backer, NULL, odp_key.data,
7700 odp_key.size, &flow, NULL, NULL, NULL,
7702 unixctl_command_reply_error(conn, "Invalid flow");
7708 error_s = parse_ofp_exact_flow(&flow, argv[2]);
7710 unixctl_command_reply_error(conn, error_s);
7715 initial_vals.vlan_tci = flow.vlan_tci;
7718 /* Generate a packet, if requested. */
7720 packet = ofpbuf_new(0);
7721 flow_compose(packet, &flow);
7723 } else if (argc == 7) {
7724 /* ofproto/trace dpname priority tun_id in_port mark packet */
7725 const char *priority_s = argv[2];
7726 const char *tun_id_s = argv[3];
7727 const char *in_port_s = argv[4];
7728 const char *mark_s = argv[5];
7729 const char *packet_s = argv[6];
7730 uint32_t in_port = atoi(in_port_s);
7731 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
7732 uint32_t priority = atoi(priority_s);
7733 uint32_t mark = atoi(mark_s);
7736 msg = eth_from_hex(packet_s, &packet);
7738 unixctl_command_reply_error(conn, msg);
7742 ds_put_cstr(&result, "Packet: ");
7743 s = ofp_packet_to_string(packet->data, packet->size);
7744 ds_put_cstr(&result, s);
7747 flow_extract(packet, priority, mark, NULL, in_port, &flow);
7748 flow.tunnel.tun_id = tun_id;
7749 initial_vals.vlan_tci = flow.vlan_tci;
7751 unixctl_command_reply_error(conn, "Bad command syntax");
7755 ofproto_trace(ofproto, &flow, packet, &initial_vals, &result);
7756 unixctl_command_reply(conn, ds_cstr(&result));
7759 ds_destroy(&result);
7760 ofpbuf_delete(packet);
7761 ofpbuf_uninit(&odp_key);
7765 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
7766 const struct ofpbuf *packet,
7767 const struct initial_vals *initial_vals, struct ds *ds)
7769 struct rule_dpif *rule;
7771 ds_put_cstr(ds, "Flow: ");
7772 flow_format(ds, flow);
7773 ds_put_char(ds, '\n');
7775 rule = rule_dpif_lookup(ofproto, flow);
7777 trace_format_rule(ds, 0, 0, rule);
7778 if (rule == ofproto->miss_rule) {
7779 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
7780 } else if (rule == ofproto->no_packet_in_rule) {
7781 ds_put_cstr(ds, "\nNo match, packets dropped because "
7782 "OFPPC_NO_PACKET_IN is set on in_port.\n");
7786 uint64_t odp_actions_stub[1024 / 8];
7787 struct ofpbuf odp_actions;
7789 struct trace_ctx trace;
7792 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
7795 ofpbuf_use_stub(&odp_actions,
7796 odp_actions_stub, sizeof odp_actions_stub);
7797 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_vals,
7798 rule, tcp_flags, packet);
7799 trace.ctx.resubmit_hook = trace_resubmit;
7800 trace.ctx.report_hook = trace_report;
7801 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
7804 ds_put_char(ds, '\n');
7805 trace_format_flow(ds, 0, "Final flow", &trace);
7806 ds_put_cstr(ds, "Datapath actions: ");
7807 format_odp_actions(ds, odp_actions.data, odp_actions.size);
7808 ofpbuf_uninit(&odp_actions);
7810 if (trace.ctx.slow) {
7811 enum slow_path_reason slow;
7813 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
7814 "slow path because it:");
7815 for (slow = trace.ctx.slow; slow; ) {
7816 enum slow_path_reason bit = rightmost_1bit(slow);
7820 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
7823 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
7826 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
7829 ds_put_cstr(ds, "\n\t- Needs in-band special case "
7832 ds_put_cstr(ds, "\n\t (The datapath actions are "
7833 "incomplete--for complete actions, "
7834 "please supply a packet.)");
7837 case SLOW_CONTROLLER:
7838 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
7839 "to the OpenFlow controller.");
7842 ds_put_cstr(ds, "\n\t- Needs more specific matching "
7843 "than the datapath supports.");
7850 if (slow & ~SLOW_MATCH) {
7851 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
7852 "the special slow-path processing.");
7859 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7860 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7863 unixctl_command_reply(conn, NULL);
7867 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7868 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7871 unixctl_command_reply(conn, NULL);
7874 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
7875 * 'reply' describing the results. */
7877 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
7879 struct facet *facet;
7883 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7884 if (!facet_check_consistency(facet)) {
7889 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
7893 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
7894 ofproto->up.name, errors);
7896 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
7901 ofproto_dpif_self_check(struct unixctl_conn *conn,
7902 int argc, const char *argv[], void *aux OVS_UNUSED)
7904 struct ds reply = DS_EMPTY_INITIALIZER;
7905 struct ofproto_dpif *ofproto;
7908 ofproto = ofproto_dpif_lookup(argv[1]);
7910 unixctl_command_reply_error(conn, "Unknown ofproto (use "
7911 "ofproto/list for help)");
7914 ofproto_dpif_self_check__(ofproto, &reply);
7916 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7917 ofproto_dpif_self_check__(ofproto, &reply);
7921 unixctl_command_reply(conn, ds_cstr(&reply));
7925 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
7926 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
7927 * to destroy 'ofproto_shash' and free the returned value. */
7928 static const struct shash_node **
7929 get_ofprotos(struct shash *ofproto_shash)
7931 const struct ofproto_dpif *ofproto;
7933 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7934 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
7935 shash_add_nocopy(ofproto_shash, name, ofproto);
7938 return shash_sort(ofproto_shash);
7942 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
7943 const char *argv[] OVS_UNUSED,
7944 void *aux OVS_UNUSED)
7946 struct ds ds = DS_EMPTY_INITIALIZER;
7947 struct shash ofproto_shash;
7948 const struct shash_node **sorted_ofprotos;
7951 shash_init(&ofproto_shash);
7952 sorted_ofprotos = get_ofprotos(&ofproto_shash);
7953 for (i = 0; i < shash_count(&ofproto_shash); i++) {
7954 const struct shash_node *node = sorted_ofprotos[i];
7955 ds_put_format(&ds, "%s\n", node->name);
7958 shash_destroy(&ofproto_shash);
7959 free(sorted_ofprotos);
7961 unixctl_command_reply(conn, ds_cstr(&ds));
7966 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
7968 struct dpif_dp_stats s;
7969 const struct shash_node **ports;
7972 dpif_get_dp_stats(ofproto->backer->dpif, &s);
7974 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
7975 dpif_name(ofproto->backer->dpif));
7976 /* xxx It would be better to show bridge-specific stats instead
7977 * xxx of dp ones. */
7979 "\tlookups: hit:%"PRIu64" missed:%"PRIu64" lost:%"PRIu64"\n",
7980 s.n_hit, s.n_missed, s.n_lost);
7981 ds_put_format(ds, "\tflows: %zu\n",
7982 hmap_count(&ofproto->subfacets));
7984 ports = shash_sort(&ofproto->up.port_by_name);
7985 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
7986 const struct shash_node *node = ports[i];
7987 struct ofport *ofport = node->data;
7988 const char *name = netdev_get_name(ofport->netdev);
7989 const char *type = netdev_get_type(ofport->netdev);
7992 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
7994 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
7995 if (odp_port != OVSP_NONE) {
7996 ds_put_format(ds, "%"PRIu32":", odp_port);
7998 ds_put_cstr(ds, "none:");
8001 if (strcmp(type, "system")) {
8002 struct netdev *netdev;
8005 ds_put_format(ds, " (%s", type);
8007 error = netdev_open(name, type, &netdev);
8012 error = netdev_get_config(netdev, &config);
8014 const struct smap_node **nodes;
8017 nodes = smap_sort(&config);
8018 for (i = 0; i < smap_count(&config); i++) {
8019 const struct smap_node *node = nodes[i];
8020 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
8021 node->key, node->value);
8025 smap_destroy(&config);
8027 netdev_close(netdev);
8029 ds_put_char(ds, ')');
8031 ds_put_char(ds, '\n');
8037 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
8038 const char *argv[], void *aux OVS_UNUSED)
8040 struct ds ds = DS_EMPTY_INITIALIZER;
8041 const struct ofproto_dpif *ofproto;
8045 for (i = 1; i < argc; i++) {
8046 ofproto = ofproto_dpif_lookup(argv[i]);
8048 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
8049 "for help)", argv[i]);
8050 unixctl_command_reply_error(conn, ds_cstr(&ds));
8053 show_dp_format(ofproto, &ds);
8056 struct shash ofproto_shash;
8057 const struct shash_node **sorted_ofprotos;
8060 shash_init(&ofproto_shash);
8061 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8062 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8063 const struct shash_node *node = sorted_ofprotos[i];
8064 show_dp_format(node->data, &ds);
8067 shash_destroy(&ofproto_shash);
8068 free(sorted_ofprotos);
8071 unixctl_command_reply(conn, ds_cstr(&ds));
8076 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8077 int argc OVS_UNUSED, const char *argv[],
8078 void *aux OVS_UNUSED)
8080 struct ds ds = DS_EMPTY_INITIALIZER;
8081 const struct ofproto_dpif *ofproto;
8082 struct subfacet *subfacet;
8084 ofproto = ofproto_dpif_lookup(argv[1]);
8086 unixctl_command_reply_error(conn, "no such bridge");
8090 update_stats(ofproto->backer);
8092 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8093 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
8095 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8096 subfacet->dp_packet_count, subfacet->dp_byte_count);
8097 if (subfacet->used) {
8098 ds_put_format(&ds, "%.3fs",
8099 (time_msec() - subfacet->used) / 1000.0);
8101 ds_put_format(&ds, "never");
8103 if (subfacet->facet->tcp_flags) {
8104 ds_put_cstr(&ds, ", flags:");
8105 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8108 ds_put_cstr(&ds, ", actions:");
8109 format_odp_actions(&ds, subfacet->actions, subfacet->actions_len);
8110 ds_put_char(&ds, '\n');
8113 unixctl_command_reply(conn, ds_cstr(&ds));
8118 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8119 int argc OVS_UNUSED, const char *argv[],
8120 void *aux OVS_UNUSED)
8122 struct ds ds = DS_EMPTY_INITIALIZER;
8123 struct ofproto_dpif *ofproto;
8125 ofproto = ofproto_dpif_lookup(argv[1]);
8127 unixctl_command_reply_error(conn, "no such bridge");
8131 flush(&ofproto->up);
8133 unixctl_command_reply(conn, ds_cstr(&ds));
8138 ofproto_dpif_unixctl_init(void)
8140 static bool registered;
8146 unixctl_command_register(
8148 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
8149 2, 6, ofproto_unixctl_trace, NULL);
8150 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8151 ofproto_unixctl_fdb_flush, NULL);
8152 unixctl_command_register("fdb/show", "bridge", 1, 1,
8153 ofproto_unixctl_fdb_show, NULL);
8154 unixctl_command_register("ofproto/clog", "", 0, 0,
8155 ofproto_dpif_clog, NULL);
8156 unixctl_command_register("ofproto/unclog", "", 0, 0,
8157 ofproto_dpif_unclog, NULL);
8158 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8159 ofproto_dpif_self_check, NULL);
8160 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8161 ofproto_unixctl_dpif_dump_dps, NULL);
8162 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8163 ofproto_unixctl_dpif_show, NULL);
8164 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8165 ofproto_unixctl_dpif_dump_flows, NULL);
8166 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8167 ofproto_unixctl_dpif_del_flows, NULL);
8170 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8172 * This is deprecated. It is only for compatibility with broken device drivers
8173 * in old versions of Linux that do not properly support VLANs when VLAN
8174 * devices are not used. When broken device drivers are no longer in
8175 * widespread use, we will delete these interfaces. */
8178 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8180 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8181 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8183 if (realdev_ofp_port == ofport->realdev_ofp_port
8184 && vid == ofport->vlandev_vid) {
8188 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8190 if (ofport->realdev_ofp_port) {
8193 if (realdev_ofp_port && ofport->bundle) {
8194 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8195 * themselves be part of a bundle. */
8196 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8199 ofport->realdev_ofp_port = realdev_ofp_port;
8200 ofport->vlandev_vid = vid;
8202 if (realdev_ofp_port) {
8203 vsp_add(ofport, realdev_ofp_port, vid);
8210 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8212 return hash_2words(realdev_ofp_port, vid);
8215 /* Returns the ODP port number of the Linux VLAN device that corresponds to
8216 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
8217 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
8218 * it would return the port number of eth0.9.
8220 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
8221 * function just returns its 'realdev_odp_port' argument. */
8223 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8224 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
8226 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8227 uint16_t realdev_ofp_port;
8228 int vid = vlan_tci_to_vid(vlan_tci);
8229 const struct vlan_splinter *vsp;
8231 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
8232 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8233 hash_realdev_vid(realdev_ofp_port, vid),
8234 &ofproto->realdev_vid_map) {
8235 if (vsp->realdev_ofp_port == realdev_ofp_port
8236 && vsp->vid == vid) {
8237 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
8241 return realdev_odp_port;
8244 static struct vlan_splinter *
8245 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8247 struct vlan_splinter *vsp;
8249 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8250 &ofproto->vlandev_map) {
8251 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8259 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8260 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8261 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8262 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8263 * eth0 and store 9 in '*vid'.
8265 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8266 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8269 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8270 uint16_t vlandev_ofp_port, int *vid)
8272 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8273 const struct vlan_splinter *vsp;
8275 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8280 return vsp->realdev_ofp_port;
8286 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8287 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8288 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8289 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8290 * always the case unless VLAN splinters are enabled), returns false without
8291 * making any changes. */
8293 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8298 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8303 /* Cause the flow to be processed as if it came in on the real device with
8304 * the VLAN device's VLAN ID. */
8305 flow->in_port = realdev;
8306 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8311 vsp_remove(struct ofport_dpif *port)
8313 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8314 struct vlan_splinter *vsp;
8316 vsp = vlandev_find(ofproto, port->up.ofp_port);
8318 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8319 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8322 port->realdev_ofp_port = 0;
8324 VLOG_ERR("missing vlan device record");
8329 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8331 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8333 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8334 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8335 == realdev_ofp_port)) {
8336 struct vlan_splinter *vsp;
8338 vsp = xmalloc(sizeof *vsp);
8339 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8340 hash_int(port->up.ofp_port, 0));
8341 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8342 hash_realdev_vid(realdev_ofp_port, vid));
8343 vsp->realdev_ofp_port = realdev_ofp_port;
8344 vsp->vlandev_ofp_port = port->up.ofp_port;
8347 port->realdev_ofp_port = realdev_ofp_port;
8349 VLOG_ERR("duplicate vlan device record");
8354 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8356 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8357 return ofport ? ofport->odp_port : OVSP_NONE;
8360 static struct ofport_dpif *
8361 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8363 struct ofport_dpif *port;
8365 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8366 hash_int(odp_port, 0),
8367 &backer->odp_to_ofport_map) {
8368 if (port->odp_port == odp_port) {
8377 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8379 struct ofport_dpif *port;
8381 port = odp_port_to_ofport(ofproto->backer, odp_port);
8382 if (port && &ofproto->up == port->up.ofproto) {
8383 return port->up.ofp_port;
8389 const struct ofproto_class ofproto_dpif_class = {
8424 port_is_lacp_current,
8425 NULL, /* rule_choose_table */
8432 rule_modify_actions,
8443 get_stp_port_status,
8450 is_mirror_output_bundle,
8451 forward_bpdu_changed,
8452 set_mac_table_config,