2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-provider.h"
25 #include "byte-order.h"
30 #include "dynamic-string.h"
31 #include "fail-open.h"
35 #include "mac-learning.h"
36 #include "meta-flow.h"
37 #include "multipath.h"
38 #include "netdev-vport.h"
45 #include "ofp-actions.h"
46 #include "ofp-parse.h"
47 #include "ofp-print.h"
48 #include "ofproto-dpif-governor.h"
49 #include "ofproto-dpif-sflow.h"
50 #include "poll-loop.h"
55 #include "unaligned.h"
57 #include "vlan-bitmap.h"
60 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
62 COVERAGE_DEFINE(ofproto_dpif_expired);
63 COVERAGE_DEFINE(ofproto_dpif_xlate);
64 COVERAGE_DEFINE(facet_changed_rule);
65 COVERAGE_DEFINE(facet_revalidate);
66 COVERAGE_DEFINE(facet_unexpected);
67 COVERAGE_DEFINE(facet_suppress);
69 /* Maximum depth of flow table recursion (due to resubmit actions) in a
70 * flow translation. */
71 #define MAX_RESUBMIT_RECURSION 64
73 /* Number of implemented OpenFlow tables. */
74 enum { N_TABLES = 255 };
75 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
76 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
87 * - Do include packets and bytes from facets that have been deleted or
88 * whose own statistics have been folded into the rule.
90 * - Do include packets and bytes sent "by hand" that were accounted to
91 * the rule without any facet being involved (this is a rare corner
92 * case in rule_execute()).
94 * - Do not include packet or bytes that can be obtained from any facet's
95 * packet_count or byte_count member or that can be obtained from the
96 * datapath by, e.g., dpif_flow_get() for any subfacet.
98 uint64_t packet_count; /* Number of packets received. */
99 uint64_t byte_count; /* Number of bytes received. */
101 tag_type tag; /* Caches rule_calculate_tag() result. */
103 struct list facets; /* List of "struct facet"s. */
106 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
108 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
111 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
112 const struct flow *);
113 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
116 static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto,
117 const struct flow *flow);
119 static void rule_credit_stats(struct rule_dpif *,
120 const struct dpif_flow_stats *);
121 static void flow_push_stats(struct rule_dpif *, const struct flow *,
122 const struct dpif_flow_stats *);
123 static tag_type rule_calculate_tag(const struct flow *,
124 const struct minimask *, uint32_t basis);
125 static void rule_invalidate(const struct rule_dpif *);
127 #define MAX_MIRRORS 32
128 typedef uint32_t mirror_mask_t;
129 #define MIRROR_MASK_C(X) UINT32_C(X)
130 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
132 struct ofproto_dpif *ofproto; /* Owning ofproto. */
133 size_t idx; /* In ofproto's "mirrors" array. */
134 void *aux; /* Key supplied by ofproto's client. */
135 char *name; /* Identifier for log messages. */
137 /* Selection criteria. */
138 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
139 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
140 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
142 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
143 struct ofbundle *out; /* Output port or NULL. */
144 int out_vlan; /* Output VLAN or -1. */
145 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
148 int64_t packet_count; /* Number of packets sent. */
149 int64_t byte_count; /* Number of bytes sent. */
152 static void mirror_destroy(struct ofmirror *);
153 static void update_mirror_stats(struct ofproto_dpif *ofproto,
154 mirror_mask_t mirrors,
155 uint64_t packets, uint64_t bytes);
158 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
159 struct ofproto_dpif *ofproto; /* Owning ofproto. */
160 void *aux; /* Key supplied by ofproto's client. */
161 char *name; /* Identifier for log messages. */
164 struct list ports; /* Contains "struct ofport"s. */
165 enum port_vlan_mode vlan_mode; /* VLAN mode */
166 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
167 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
168 * NULL if all VLANs are trunked. */
169 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
170 struct bond *bond; /* Nonnull iff more than one port. */
171 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
174 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
176 /* Port mirroring info. */
177 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
178 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
179 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
182 static void bundle_remove(struct ofport *);
183 static void bundle_update(struct ofbundle *);
184 static void bundle_destroy(struct ofbundle *);
185 static void bundle_del_port(struct ofport_dpif *);
186 static void bundle_run(struct ofbundle *);
187 static void bundle_wait(struct ofbundle *);
188 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
189 uint16_t in_port, bool warn,
190 struct ofport_dpif **in_ofportp);
192 /* A controller may use OFPP_NONE as the ingress port to indicate that
193 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
194 * when an input bundle is needed for validation (e.g., mirroring or
195 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
196 * any 'port' structs, so care must be taken when dealing with it. */
197 static struct ofbundle ofpp_none_bundle = {
199 .vlan_mode = PORT_VLAN_TRUNK
202 static void stp_run(struct ofproto_dpif *ofproto);
203 static void stp_wait(struct ofproto_dpif *ofproto);
204 static int set_stp_port(struct ofport *,
205 const struct ofproto_port_stp_settings *);
207 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
209 struct action_xlate_ctx {
210 /* action_xlate_ctx_init() initializes these members. */
213 struct ofproto_dpif *ofproto;
215 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
216 * this flow when actions change header fields. */
219 /* The packet corresponding to 'flow', or a null pointer if we are
220 * revalidating without a packet to refer to. */
221 const struct ofpbuf *packet;
223 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
224 * actions update the flow table?
226 * We want to update these tables if we are actually processing a packet,
227 * or if we are accounting for packets that the datapath has processed, but
228 * not if we are just revalidating. */
231 /* The rule that we are currently translating, or NULL. */
232 struct rule_dpif *rule;
234 /* Union of the set of TCP flags seen so far in this flow. (Used only by
235 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
239 /* If nonnull, flow translation calls this function just before executing a
240 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
241 * when the recursion depth is exceeded.
243 * 'rule' is the rule being submitted into. It will be null if the
244 * resubmit or OFPP_TABLE action didn't find a matching rule.
246 * This is normally null so the client has to set it manually after
247 * calling action_xlate_ctx_init(). */
248 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule);
250 /* If nonnull, flow translation calls this function to report some
251 * significant decision, e.g. to explain why OFPP_NORMAL translation
252 * dropped a packet. */
253 void (*report_hook)(struct action_xlate_ctx *, const char *s);
255 /* If nonnull, flow translation credits the specified statistics to each
256 * rule reached through a resubmit or OFPP_TABLE action.
258 * This is normally null so the client has to set it manually after
259 * calling action_xlate_ctx_init(). */
260 const struct dpif_flow_stats *resubmit_stats;
262 /* xlate_actions() initializes and uses these members. The client might want
263 * to look at them after it returns. */
265 struct ofpbuf *odp_actions; /* Datapath actions. */
266 tag_type tags; /* Tags associated with actions. */
267 enum slow_path_reason slow; /* 0 if fast path may be used. */
268 bool has_learn; /* Actions include NXAST_LEARN? */
269 bool has_normal; /* Actions output to OFPP_NORMAL? */
270 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
271 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
272 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
274 /* xlate_actions() initializes and uses these members, but the client has no
275 * reason to look at them. */
277 int recurse; /* Recursion level, via xlate_table_action. */
278 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
279 struct flow base_flow; /* Flow at the last commit. */
280 uint32_t orig_skb_priority; /* Priority when packet arrived. */
281 uint8_t table_id; /* OpenFlow table ID where flow was found. */
282 uint32_t sflow_n_outputs; /* Number of output ports. */
283 uint32_t sflow_odp_port; /* Output port for composing sFlow action. */
284 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
285 bool exit; /* No further actions should be processed. */
288 static void action_xlate_ctx_init(struct action_xlate_ctx *,
289 struct ofproto_dpif *, const struct flow *,
290 ovs_be16 initial_tci, struct rule_dpif *,
291 uint8_t tcp_flags, const struct ofpbuf *);
292 static void xlate_actions(struct action_xlate_ctx *,
293 const struct ofpact *ofpacts, size_t ofpacts_len,
294 struct ofpbuf *odp_actions);
295 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
296 const struct ofpact *ofpacts,
298 static void xlate_table_action(struct action_xlate_ctx *, uint16_t in_port,
299 uint8_t table_id, bool may_packet_in);
301 static size_t put_userspace_action(const struct ofproto_dpif *,
302 struct ofpbuf *odp_actions,
304 const union user_action_cookie *);
306 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
307 enum slow_path_reason,
308 uint64_t *stub, size_t stub_size,
309 const struct nlattr **actionsp,
310 size_t *actions_lenp);
312 static void xlate_report(struct action_xlate_ctx *ctx, const char *s);
314 /* A subfacet (see "struct subfacet" below) has three possible installation
317 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
318 * case just after the subfacet is created, just before the subfacet is
319 * destroyed, or if the datapath returns an error when we try to install a
322 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
324 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
325 * ofproto_dpif is installed in the datapath.
328 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
329 SF_FAST_PATH, /* Full actions are installed. */
330 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
333 static const char *subfacet_path_to_string(enum subfacet_path);
335 /* A dpif flow and actions associated with a facet.
337 * See also the large comment on struct facet. */
340 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
341 struct list list_node; /* In struct facet's 'facets' list. */
342 struct facet *facet; /* Owning facet. */
344 enum odp_key_fitness key_fitness;
348 long long int used; /* Time last used; time created if not used. */
350 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
351 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
355 * These should be essentially identical for every subfacet in a facet, but
356 * may differ in trivial ways due to VLAN splinters. */
357 size_t actions_len; /* Number of bytes in actions[]. */
358 struct nlattr *actions; /* Datapath actions. */
360 enum slow_path_reason slow; /* 0 if fast path may be used. */
361 enum subfacet_path path; /* Installed in datapath? */
363 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
364 * splinters can cause it to differ. This value should be removed when
365 * the VLAN splinters feature is no longer needed. */
366 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
368 /* Datapath port the packet arrived on. This is needed to remove
369 * flows for ports that are no longer part of the bridge. Since the
370 * flow definition only has the OpenFlow port number and the port is
371 * no longer part of the bridge, we can't determine the datapath port
372 * number needed to delete the flow from the datapath. */
373 uint32_t odp_in_port;
376 #define SUBFACET_DESTROY_MAX_BATCH 50
378 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
380 static struct subfacet *subfacet_find(struct ofproto_dpif *,
381 const struct nlattr *key, size_t key_len,
383 static void subfacet_destroy(struct subfacet *);
384 static void subfacet_destroy__(struct subfacet *);
385 static void subfacet_destroy_batch(struct ofproto_dpif *,
386 struct subfacet **, int n);
387 static void subfacet_reset_dp_stats(struct subfacet *,
388 struct dpif_flow_stats *);
389 static void subfacet_update_time(struct subfacet *, long long int used);
390 static void subfacet_update_stats(struct subfacet *,
391 const struct dpif_flow_stats *);
392 static void subfacet_make_actions(struct subfacet *,
393 const struct ofpbuf *packet,
394 struct ofpbuf *odp_actions);
395 static int subfacet_install(struct subfacet *,
396 const struct nlattr *actions, size_t actions_len,
397 struct dpif_flow_stats *, enum slow_path_reason);
398 static void subfacet_uninstall(struct subfacet *);
400 static enum subfacet_path subfacet_want_path(enum slow_path_reason);
402 /* An exact-match instantiation of an OpenFlow flow.
404 * A facet associates a "struct flow", which represents the Open vSwitch
405 * userspace idea of an exact-match flow, with one or more subfacets. Each
406 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
407 * the facet. When the kernel module (or other dpif implementation) and Open
408 * vSwitch userspace agree on the definition of a flow key, there is exactly
409 * one subfacet per facet. If the dpif implementation supports more-specific
410 * flow matching than userspace, however, a facet can have more than one
411 * subfacet, each of which corresponds to some distinction in flow that
412 * userspace simply doesn't understand.
414 * Flow expiration works in terms of subfacets, so a facet must have at least
415 * one subfacet or it will never expire, leaking memory. */
418 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
419 struct list list_node; /* In owning rule's 'facets' list. */
420 struct rule_dpif *rule; /* Owning rule. */
423 struct list subfacets;
424 long long int used; /* Time last used; time created if not used. */
431 * - Do include packets and bytes sent "by hand", e.g. with
434 * - Do include packets and bytes that were obtained from the datapath
435 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
436 * DPIF_FP_ZERO_STATS).
438 * - Do not include packets or bytes that can be obtained from the
439 * datapath for any existing subfacet.
441 uint64_t packet_count; /* Number of packets received. */
442 uint64_t byte_count; /* Number of bytes received. */
444 /* Resubmit statistics. */
445 uint64_t prev_packet_count; /* Number of packets from last stats push. */
446 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
447 long long int prev_used; /* Used time from last stats push. */
450 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
451 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
452 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
454 /* Properties of datapath actions.
456 * Every subfacet has its own actions because actions can differ slightly
457 * between splintered and non-splintered subfacets due to the VLAN tag
458 * being initially different (present vs. absent). All of them have these
459 * properties in common so we just store one copy of them here. */
460 bool has_learn; /* Actions include NXAST_LEARN? */
461 bool has_normal; /* Actions output to OFPP_NORMAL? */
462 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
463 tag_type tags; /* Tags that would require revalidation. */
464 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
466 /* Storage for a single subfacet, to reduce malloc() time and space
467 * overhead. (A facet always has at least one subfacet and in the common
468 * case has exactly one subfacet.) */
469 struct subfacet one_subfacet;
472 static struct facet *facet_create(struct rule_dpif *,
473 const struct flow *, uint32_t hash);
474 static void facet_remove(struct facet *);
475 static void facet_free(struct facet *);
477 static struct facet *facet_find(struct ofproto_dpif *,
478 const struct flow *, uint32_t hash);
479 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
480 const struct flow *, uint32_t hash);
481 static void facet_revalidate(struct facet *);
482 static bool facet_check_consistency(struct facet *);
484 static void facet_flush_stats(struct facet *);
486 static void facet_update_time(struct facet *, long long int used);
487 static void facet_reset_counters(struct facet *);
488 static void facet_push_stats(struct facet *);
489 static void facet_learn(struct facet *);
490 static void facet_account(struct facet *);
492 static bool facet_is_controller_flow(struct facet *);
495 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
499 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
500 struct list bundle_node; /* In struct ofbundle's "ports" list. */
501 struct cfm *cfm; /* Connectivity Fault Management, if any. */
502 tag_type tag; /* Tag associated with this port. */
503 bool may_enable; /* May be enabled in bonds. */
504 long long int carrier_seq; /* Carrier status changes. */
505 struct tnl_port *tnl_port; /* Tunnel handle, or null. */
508 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
509 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
510 long long int stp_state_entered;
512 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
514 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
516 * This is deprecated. It is only for compatibility with broken device
517 * drivers in old versions of Linux that do not properly support VLANs when
518 * VLAN devices are not used. When broken device drivers are no longer in
519 * widespread use, we will delete these interfaces. */
520 uint16_t realdev_ofp_port;
524 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
525 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
526 * traffic egressing the 'ofport' with that priority should be marked with. */
527 struct priority_to_dscp {
528 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
529 uint32_t priority; /* Priority of this queue (see struct flow). */
531 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
534 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
536 * This is deprecated. It is only for compatibility with broken device drivers
537 * in old versions of Linux that do not properly support VLANs when VLAN
538 * devices are not used. When broken device drivers are no longer in
539 * widespread use, we will delete these interfaces. */
540 struct vlan_splinter {
541 struct hmap_node realdev_vid_node;
542 struct hmap_node vlandev_node;
543 uint16_t realdev_ofp_port;
544 uint16_t vlandev_ofp_port;
548 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
549 uint32_t realdev, ovs_be16 vlan_tci);
550 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
551 static void vsp_remove(struct ofport_dpif *);
552 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
554 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
556 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
559 static struct ofport_dpif *
560 ofport_dpif_cast(const struct ofport *ofport)
562 ovs_assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
563 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
566 static void port_run(struct ofport_dpif *);
567 static void port_run_fast(struct ofport_dpif *);
568 static void port_wait(struct ofport_dpif *);
569 static int set_cfm(struct ofport *, const struct cfm_settings *);
570 static void ofport_clear_priorities(struct ofport_dpif *);
572 struct dpif_completion {
573 struct list list_node;
574 struct ofoperation *op;
577 /* Extra information about a classifier table.
578 * Currently used just for optimized flow revalidation. */
580 /* If either of these is nonnull, then this table has a form that allows
581 * flows to be tagged to avoid revalidating most flows for the most common
582 * kinds of flow table changes. */
583 struct cls_table *catchall_table; /* Table that wildcards all fields. */
584 struct cls_table *other_table; /* Table with any other wildcard set. */
585 uint32_t basis; /* Keeps each table's tags separate. */
588 /* Reasons that we might need to revalidate every facet, and corresponding
591 * A value of 0 means that there is no need to revalidate.
593 * It would be nice to have some cleaner way to integrate with coverage
594 * counters, but with only a few reasons I guess this is good enough for
596 enum revalidate_reason {
597 REV_RECONFIGURE = 1, /* Switch configuration changed. */
598 REV_STP, /* Spanning tree protocol port status change. */
599 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
600 REV_FLOW_TABLE, /* Flow table changed. */
601 REV_INCONSISTENCY /* Facet self-check failed. */
603 COVERAGE_DEFINE(rev_reconfigure);
604 COVERAGE_DEFINE(rev_stp);
605 COVERAGE_DEFINE(rev_port_toggled);
606 COVERAGE_DEFINE(rev_flow_table);
607 COVERAGE_DEFINE(rev_inconsistency);
609 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
610 * These are datapath flows which have no associated ofproto, if they did we
611 * would use facets. */
613 struct hmap_node hmap_node;
618 /* All datapaths of a given type share a single dpif backer instance. */
623 struct timer next_expiration;
624 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
626 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
628 /* Facet revalidation flags applying to facets which use this backer. */
629 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
630 struct tag_set revalidate_set; /* Revalidate only matching facets. */
632 struct hmap drop_keys; /* Set of dropped odp keys. */
635 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
636 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
638 static void drop_key_clear(struct dpif_backer *);
639 static struct ofport_dpif *
640 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
642 struct ofproto_dpif {
643 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
645 struct dpif_backer *backer;
647 /* Special OpenFlow rules. */
648 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
649 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
655 struct netflow *netflow;
656 struct dpif_sflow *sflow;
657 struct hmap bundles; /* Contains "struct ofbundle"s. */
658 struct mac_learning *ml;
659 struct ofmirror *mirrors[MAX_MIRRORS];
661 bool has_bonded_bundles;
665 struct hmap subfacets;
666 struct governor *governor;
669 struct table_dpif tables[N_TABLES];
671 /* Support for debugging async flow mods. */
672 struct list completions;
674 bool has_bundle_action; /* True when the first bundle action appears. */
675 struct netdev_stats stats; /* To account packets generated and consumed in
680 long long int stp_last_tick;
682 /* VLAN splinters. */
683 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
684 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
687 struct sset ports; /* Set of standard port names. */
688 struct sset ghost_ports; /* Ports with no datapath port. */
689 struct sset port_poll_set; /* Queued names for port_poll() reply. */
690 int port_poll_errno; /* Last errno for port_poll() reply. */
693 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
694 * for debugging the asynchronous flow_mod implementation.) */
697 /* All existing ofproto_dpif instances, indexed by ->up.name. */
698 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
700 static void ofproto_dpif_unixctl_init(void);
702 static struct ofproto_dpif *
703 ofproto_dpif_cast(const struct ofproto *ofproto)
705 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
706 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
709 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
711 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
713 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
714 const struct ofpbuf *, ovs_be16 initial_tci,
717 /* Packet processing. */
718 static void update_learning_table(struct ofproto_dpif *,
719 const struct flow *, int vlan,
722 #define FLOW_MISS_MAX_BATCH 50
723 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
725 /* Flow expiration. */
726 static int expire(struct dpif_backer *);
729 static void send_netflow_active_timeouts(struct ofproto_dpif *);
732 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
733 static size_t compose_sflow_action(const struct ofproto_dpif *,
734 struct ofpbuf *odp_actions,
735 const struct flow *, uint32_t odp_port);
736 static void add_mirror_actions(struct action_xlate_ctx *ctx,
737 const struct flow *flow);
738 /* Global variables. */
739 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
741 /* Initial mappings of port to bridge mappings. */
742 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
744 /* Factory functions. */
747 init(const struct shash *iface_hints)
749 struct shash_node *node;
751 /* Make a local copy, since we don't own 'iface_hints' elements. */
752 SHASH_FOR_EACH(node, iface_hints) {
753 const struct iface_hint *orig_hint = node->data;
754 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
756 new_hint->br_name = xstrdup(orig_hint->br_name);
757 new_hint->br_type = xstrdup(orig_hint->br_type);
758 new_hint->ofp_port = orig_hint->ofp_port;
760 shash_add(&init_ofp_ports, node->name, new_hint);
765 enumerate_types(struct sset *types)
767 dp_enumerate_types(types);
771 enumerate_names(const char *type, struct sset *names)
773 struct ofproto_dpif *ofproto;
776 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
777 if (strcmp(type, ofproto->up.type)) {
780 sset_add(names, ofproto->up.name);
787 del(const char *type, const char *name)
792 error = dpif_open(name, type, &dpif);
794 error = dpif_delete(dpif);
801 port_open_type(const char *datapath_type, const char *port_type)
803 return dpif_port_open_type(datapath_type, port_type);
806 /* Type functions. */
808 static struct ofproto_dpif *
809 lookup_ofproto_dpif_by_port_name(const char *name)
811 struct ofproto_dpif *ofproto;
813 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
814 if (sset_contains(&ofproto->ports, name)) {
823 type_run(const char *type)
825 struct dpif_backer *backer;
829 backer = shash_find_data(&all_dpif_backers, type);
831 /* This is not necessarily a problem, since backers are only
832 * created on demand. */
836 dpif_run(backer->dpif);
838 if (backer->need_revalidate
839 || !tag_set_is_empty(&backer->revalidate_set)) {
840 struct tag_set revalidate_set = backer->revalidate_set;
841 bool need_revalidate = backer->need_revalidate;
842 struct ofproto_dpif *ofproto;
843 struct simap_node *node;
844 struct simap tmp_backers;
846 /* Handle tunnel garbage collection. */
847 simap_init(&tmp_backers);
848 simap_swap(&backer->tnl_backers, &tmp_backers);
850 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
851 struct ofport_dpif *iter;
853 if (backer != ofproto->backer) {
857 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
860 if (!iter->tnl_port) {
864 dp_port = netdev_vport_get_dpif_port(iter->up.netdev);
865 node = simap_find(&tmp_backers, dp_port);
867 simap_put(&backer->tnl_backers, dp_port, node->data);
868 simap_delete(&tmp_backers, node);
869 node = simap_find(&backer->tnl_backers, dp_port);
871 node = simap_find(&backer->tnl_backers, dp_port);
873 uint32_t odp_port = UINT32_MAX;
875 if (!dpif_port_add(backer->dpif, iter->up.netdev,
877 simap_put(&backer->tnl_backers, dp_port, odp_port);
878 node = simap_find(&backer->tnl_backers, dp_port);
883 iter->odp_port = node ? node->data : OVSP_NONE;
884 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
886 backer->need_revalidate = REV_RECONFIGURE;
891 SIMAP_FOR_EACH (node, &tmp_backers) {
892 dpif_port_del(backer->dpif, node->data);
894 simap_destroy(&tmp_backers);
896 switch (backer->need_revalidate) {
897 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
898 case REV_STP: COVERAGE_INC(rev_stp); break;
899 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
900 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
901 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
904 if (backer->need_revalidate) {
905 /* Clear the drop_keys in case we should now be accepting some
906 * formerly dropped flows. */
907 drop_key_clear(backer);
910 /* Clear the revalidation flags. */
911 tag_set_init(&backer->revalidate_set);
912 backer->need_revalidate = 0;
914 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
915 struct facet *facet, *next;
917 if (ofproto->backer != backer) {
921 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
923 || tag_set_intersects(&revalidate_set, facet->tags)) {
924 facet_revalidate(facet);
930 if (timer_expired(&backer->next_expiration)) {
931 int delay = expire(backer);
932 timer_set_duration(&backer->next_expiration, delay);
935 /* Check for port changes in the dpif. */
936 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
937 struct ofproto_dpif *ofproto;
938 struct dpif_port port;
940 /* Don't report on the datapath's device. */
941 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
945 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
946 &all_ofproto_dpifs) {
947 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
952 ofproto = lookup_ofproto_dpif_by_port_name(devname);
953 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
954 /* The port was removed. If we know the datapath,
955 * report it through poll_set(). If we don't, it may be
956 * notifying us of a removal we initiated, so ignore it.
957 * If there's a pending ENOBUFS, let it stand, since
958 * everything will be reevaluated. */
959 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
960 sset_add(&ofproto->port_poll_set, devname);
961 ofproto->port_poll_errno = 0;
963 } else if (!ofproto) {
964 /* The port was added, but we don't know with which
965 * ofproto we should associate it. Delete it. */
966 dpif_port_del(backer->dpif, port.port_no);
968 dpif_port_destroy(&port);
974 if (error != EAGAIN) {
975 struct ofproto_dpif *ofproto;
977 /* There was some sort of error, so propagate it to all
978 * ofprotos that use this backer. */
979 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
980 &all_ofproto_dpifs) {
981 if (ofproto->backer == backer) {
982 sset_clear(&ofproto->port_poll_set);
983 ofproto->port_poll_errno = error;
992 type_run_fast(const char *type)
994 struct dpif_backer *backer;
997 backer = shash_find_data(&all_dpif_backers, type);
999 /* This is not necessarily a problem, since backers are only
1000 * created on demand. */
1004 /* Handle one or more batches of upcalls, until there's nothing left to do
1005 * or until we do a fixed total amount of work.
1007 * We do work in batches because it can be much cheaper to set up a number
1008 * of flows and fire off their patches all at once. We do multiple batches
1009 * because in some cases handling a packet can cause another packet to be
1010 * queued almost immediately as part of the return flow. Both
1011 * optimizations can make major improvements on some benchmarks and
1012 * presumably for real traffic as well. */
1014 while (work < FLOW_MISS_MAX_BATCH) {
1015 int retval = handle_upcalls(backer, FLOW_MISS_MAX_BATCH - work);
1026 type_wait(const char *type)
1028 struct dpif_backer *backer;
1030 backer = shash_find_data(&all_dpif_backers, type);
1032 /* This is not necessarily a problem, since backers are only
1033 * created on demand. */
1037 timer_wait(&backer->next_expiration);
1040 /* Basic life-cycle. */
1042 static int add_internal_flows(struct ofproto_dpif *);
1044 static struct ofproto *
1047 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1048 return &ofproto->up;
1052 dealloc(struct ofproto *ofproto_)
1054 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1059 close_dpif_backer(struct dpif_backer *backer)
1061 struct shash_node *node;
1063 ovs_assert(backer->refcount > 0);
1065 if (--backer->refcount) {
1069 drop_key_clear(backer);
1070 hmap_destroy(&backer->drop_keys);
1072 simap_destroy(&backer->tnl_backers);
1073 hmap_destroy(&backer->odp_to_ofport_map);
1074 node = shash_find(&all_dpif_backers, backer->type);
1076 shash_delete(&all_dpif_backers, node);
1077 dpif_close(backer->dpif);
1082 /* Datapath port slated for removal from datapath. */
1083 struct odp_garbage {
1084 struct list list_node;
1089 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1091 struct dpif_backer *backer;
1092 struct dpif_port_dump port_dump;
1093 struct dpif_port port;
1094 struct shash_node *node;
1095 struct list garbage_list;
1096 struct odp_garbage *garbage, *next;
1102 backer = shash_find_data(&all_dpif_backers, type);
1109 backer_name = xasprintf("ovs-%s", type);
1111 /* Remove any existing datapaths, since we assume we're the only
1112 * userspace controlling the datapath. */
1114 dp_enumerate_names(type, &names);
1115 SSET_FOR_EACH(name, &names) {
1116 struct dpif *old_dpif;
1118 /* Don't remove our backer if it exists. */
1119 if (!strcmp(name, backer_name)) {
1123 if (dpif_open(name, type, &old_dpif)) {
1124 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1126 dpif_delete(old_dpif);
1127 dpif_close(old_dpif);
1130 sset_destroy(&names);
1132 backer = xmalloc(sizeof *backer);
1134 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1137 VLOG_ERR("failed to open datapath of type %s: %s", type,
1143 backer->type = xstrdup(type);
1144 backer->refcount = 1;
1145 hmap_init(&backer->odp_to_ofport_map);
1146 hmap_init(&backer->drop_keys);
1147 timer_set_duration(&backer->next_expiration, 1000);
1148 backer->need_revalidate = 0;
1149 simap_init(&backer->tnl_backers);
1150 tag_set_init(&backer->revalidate_set);
1153 dpif_flow_flush(backer->dpif);
1155 /* Loop through the ports already on the datapath and remove any
1156 * that we don't need anymore. */
1157 list_init(&garbage_list);
1158 dpif_port_dump_start(&port_dump, backer->dpif);
1159 while (dpif_port_dump_next(&port_dump, &port)) {
1160 node = shash_find(&init_ofp_ports, port.name);
1161 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1162 garbage = xmalloc(sizeof *garbage);
1163 garbage->odp_port = port.port_no;
1164 list_push_front(&garbage_list, &garbage->list_node);
1167 dpif_port_dump_done(&port_dump);
1169 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1170 dpif_port_del(backer->dpif, garbage->odp_port);
1171 list_remove(&garbage->list_node);
1175 shash_add(&all_dpif_backers, type, backer);
1177 error = dpif_recv_set(backer->dpif, true);
1179 VLOG_ERR("failed to listen on datapath of type %s: %s",
1180 type, strerror(error));
1181 close_dpif_backer(backer);
1189 construct(struct ofproto *ofproto_)
1191 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1192 struct shash_node *node, *next;
1197 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1202 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1203 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1205 ofproto->n_matches = 0;
1207 ofproto->netflow = NULL;
1208 ofproto->sflow = NULL;
1209 ofproto->stp = NULL;
1210 hmap_init(&ofproto->bundles);
1211 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1212 for (i = 0; i < MAX_MIRRORS; i++) {
1213 ofproto->mirrors[i] = NULL;
1215 ofproto->has_bonded_bundles = false;
1217 hmap_init(&ofproto->facets);
1218 hmap_init(&ofproto->subfacets);
1219 ofproto->governor = NULL;
1221 for (i = 0; i < N_TABLES; i++) {
1222 struct table_dpif *table = &ofproto->tables[i];
1224 table->catchall_table = NULL;
1225 table->other_table = NULL;
1226 table->basis = random_uint32();
1229 list_init(&ofproto->completions);
1231 ofproto_dpif_unixctl_init();
1233 ofproto->has_mirrors = false;
1234 ofproto->has_bundle_action = false;
1236 hmap_init(&ofproto->vlandev_map);
1237 hmap_init(&ofproto->realdev_vid_map);
1239 sset_init(&ofproto->ports);
1240 sset_init(&ofproto->ghost_ports);
1241 sset_init(&ofproto->port_poll_set);
1242 ofproto->port_poll_errno = 0;
1244 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1245 struct iface_hint *iface_hint = node->data;
1247 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1248 /* Check if the datapath already has this port. */
1249 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1250 sset_add(&ofproto->ports, node->name);
1253 free(iface_hint->br_name);
1254 free(iface_hint->br_type);
1256 shash_delete(&init_ofp_ports, node);
1260 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1261 hash_string(ofproto->up.name, 0));
1262 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1264 ofproto_init_tables(ofproto_, N_TABLES);
1265 error = add_internal_flows(ofproto);
1266 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1272 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1273 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1275 struct ofputil_flow_mod fm;
1278 match_init_catchall(&fm.match);
1280 match_set_reg(&fm.match, 0, id);
1281 fm.new_cookie = htonll(0);
1282 fm.cookie = htonll(0);
1283 fm.cookie_mask = htonll(0);
1284 fm.table_id = TBL_INTERNAL;
1285 fm.command = OFPFC_ADD;
1286 fm.idle_timeout = 0;
1287 fm.hard_timeout = 0;
1291 fm.ofpacts = ofpacts->data;
1292 fm.ofpacts_len = ofpacts->size;
1294 error = ofproto_flow_mod(&ofproto->up, &fm);
1296 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1297 id, ofperr_to_string(error));
1301 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1302 ovs_assert(*rulep != NULL);
1308 add_internal_flows(struct ofproto_dpif *ofproto)
1310 struct ofpact_controller *controller;
1311 uint64_t ofpacts_stub[128 / 8];
1312 struct ofpbuf ofpacts;
1316 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1319 controller = ofpact_put_CONTROLLER(&ofpacts);
1320 controller->max_len = UINT16_MAX;
1321 controller->controller_id = 0;
1322 controller->reason = OFPR_NO_MATCH;
1323 ofpact_pad(&ofpacts);
1325 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1330 ofpbuf_clear(&ofpacts);
1331 error = add_internal_flow(ofproto, id++, &ofpacts,
1332 &ofproto->no_packet_in_rule);
1337 complete_operations(struct ofproto_dpif *ofproto)
1339 struct dpif_completion *c, *next;
1341 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1342 ofoperation_complete(c->op, 0);
1343 list_remove(&c->list_node);
1349 destruct(struct ofproto *ofproto_)
1351 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1352 struct rule_dpif *rule, *next_rule;
1353 struct oftable *table;
1356 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1357 complete_operations(ofproto);
1359 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1360 struct cls_cursor cursor;
1362 cls_cursor_init(&cursor, &table->cls, NULL);
1363 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1364 ofproto_rule_destroy(&rule->up);
1368 for (i = 0; i < MAX_MIRRORS; i++) {
1369 mirror_destroy(ofproto->mirrors[i]);
1372 netflow_destroy(ofproto->netflow);
1373 dpif_sflow_destroy(ofproto->sflow);
1374 hmap_destroy(&ofproto->bundles);
1375 mac_learning_destroy(ofproto->ml);
1377 hmap_destroy(&ofproto->facets);
1378 hmap_destroy(&ofproto->subfacets);
1379 governor_destroy(ofproto->governor);
1381 hmap_destroy(&ofproto->vlandev_map);
1382 hmap_destroy(&ofproto->realdev_vid_map);
1384 sset_destroy(&ofproto->ports);
1385 sset_destroy(&ofproto->ghost_ports);
1386 sset_destroy(&ofproto->port_poll_set);
1388 close_dpif_backer(ofproto->backer);
1392 run_fast(struct ofproto *ofproto_)
1394 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1395 struct ofport_dpif *ofport;
1397 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1398 port_run_fast(ofport);
1405 run(struct ofproto *ofproto_)
1407 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1408 struct ofport_dpif *ofport;
1409 struct ofbundle *bundle;
1413 complete_operations(ofproto);
1416 error = run_fast(ofproto_);
1421 if (ofproto->netflow) {
1422 if (netflow_run(ofproto->netflow)) {
1423 send_netflow_active_timeouts(ofproto);
1426 if (ofproto->sflow) {
1427 dpif_sflow_run(ofproto->sflow);
1430 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1433 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1438 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1440 /* Check the consistency of a random facet, to aid debugging. */
1441 if (!hmap_is_empty(&ofproto->facets)
1442 && !ofproto->backer->need_revalidate) {
1443 struct facet *facet;
1445 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1446 struct facet, hmap_node);
1447 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1449 if (!facet_check_consistency(facet)) {
1450 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1455 if (ofproto->governor) {
1458 governor_run(ofproto->governor);
1460 /* If the governor has shrunk to its minimum size and the number of
1461 * subfacets has dwindled, then drop the governor entirely.
1463 * For hysteresis, the number of subfacets to drop the governor is
1464 * smaller than the number needed to trigger its creation. */
1465 n_subfacets = hmap_count(&ofproto->subfacets);
1466 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1467 && governor_is_idle(ofproto->governor)) {
1468 governor_destroy(ofproto->governor);
1469 ofproto->governor = NULL;
1477 wait(struct ofproto *ofproto_)
1479 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1480 struct ofport_dpif *ofport;
1481 struct ofbundle *bundle;
1483 if (!clogged && !list_is_empty(&ofproto->completions)) {
1484 poll_immediate_wake();
1487 dpif_wait(ofproto->backer->dpif);
1488 dpif_recv_wait(ofproto->backer->dpif);
1489 if (ofproto->sflow) {
1490 dpif_sflow_wait(ofproto->sflow);
1492 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1493 poll_immediate_wake();
1495 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1498 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1499 bundle_wait(bundle);
1501 if (ofproto->netflow) {
1502 netflow_wait(ofproto->netflow);
1504 mac_learning_wait(ofproto->ml);
1506 if (ofproto->backer->need_revalidate) {
1507 /* Shouldn't happen, but if it does just go around again. */
1508 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1509 poll_immediate_wake();
1511 if (ofproto->governor) {
1512 governor_wait(ofproto->governor);
1517 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1519 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1521 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1522 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1526 flush(struct ofproto *ofproto_)
1528 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1529 struct subfacet *subfacet, *next_subfacet;
1530 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1534 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1535 &ofproto->subfacets) {
1536 if (subfacet->path != SF_NOT_INSTALLED) {
1537 batch[n_batch++] = subfacet;
1538 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1539 subfacet_destroy_batch(ofproto, batch, n_batch);
1543 subfacet_destroy(subfacet);
1548 subfacet_destroy_batch(ofproto, batch, n_batch);
1553 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1554 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1556 *arp_match_ip = true;
1557 *actions = (OFPUTIL_A_OUTPUT |
1558 OFPUTIL_A_SET_VLAN_VID |
1559 OFPUTIL_A_SET_VLAN_PCP |
1560 OFPUTIL_A_STRIP_VLAN |
1561 OFPUTIL_A_SET_DL_SRC |
1562 OFPUTIL_A_SET_DL_DST |
1563 OFPUTIL_A_SET_NW_SRC |
1564 OFPUTIL_A_SET_NW_DST |
1565 OFPUTIL_A_SET_NW_TOS |
1566 OFPUTIL_A_SET_TP_SRC |
1567 OFPUTIL_A_SET_TP_DST |
1572 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1574 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1575 struct dpif_dp_stats s;
1577 strcpy(ots->name, "classifier");
1579 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1581 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1582 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1585 static struct ofport *
1588 struct ofport_dpif *port = xmalloc(sizeof *port);
1593 port_dealloc(struct ofport *port_)
1595 struct ofport_dpif *port = ofport_dpif_cast(port_);
1600 port_construct(struct ofport *port_)
1602 struct ofport_dpif *port = ofport_dpif_cast(port_);
1603 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1604 const struct netdev *netdev = port->up.netdev;
1605 struct dpif_port dpif_port;
1608 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1609 port->bundle = NULL;
1611 port->tag = tag_create_random();
1612 port->may_enable = true;
1613 port->stp_port = NULL;
1614 port->stp_state = STP_DISABLED;
1615 port->tnl_port = NULL;
1616 hmap_init(&port->priorities);
1617 port->realdev_ofp_port = 0;
1618 port->vlandev_vid = 0;
1619 port->carrier_seq = netdev_get_carrier_resets(netdev);
1621 if (netdev_vport_is_patch(netdev)) {
1622 /* XXX By bailing out here, we don't do required sFlow work. */
1623 port->odp_port = OVSP_NONE;
1627 error = dpif_port_query_by_name(ofproto->backer->dpif,
1628 netdev_vport_get_dpif_port(netdev),
1634 port->odp_port = dpif_port.port_no;
1636 if (netdev_get_tunnel_config(netdev)) {
1637 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1639 /* Sanity-check that a mapping doesn't already exist. This
1640 * shouldn't happen for non-tunnel ports. */
1641 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1642 VLOG_ERR("port %s already has an OpenFlow port number",
1644 dpif_port_destroy(&dpif_port);
1648 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1649 hash_int(port->odp_port, 0));
1651 dpif_port_destroy(&dpif_port);
1653 if (ofproto->sflow) {
1654 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1661 port_destruct(struct ofport *port_)
1663 struct ofport_dpif *port = ofport_dpif_cast(port_);
1664 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1665 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1666 const char *devname = netdev_get_name(port->up.netdev);
1668 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1669 /* The underlying device is still there, so delete it. This
1670 * happens when the ofproto is being destroyed, since the caller
1671 * assumes that removal of attached ports will happen as part of
1673 if (!port->tnl_port) {
1674 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1676 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1679 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1680 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1683 tnl_port_del(port->tnl_port);
1684 sset_find_and_delete(&ofproto->ports, devname);
1685 sset_find_and_delete(&ofproto->ghost_ports, devname);
1686 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1687 bundle_remove(port_);
1688 set_cfm(port_, NULL);
1689 if (ofproto->sflow) {
1690 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1693 ofport_clear_priorities(port);
1694 hmap_destroy(&port->priorities);
1698 port_modified(struct ofport *port_)
1700 struct ofport_dpif *port = ofport_dpif_cast(port_);
1702 if (port->bundle && port->bundle->bond) {
1703 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1708 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1710 struct ofport_dpif *port = ofport_dpif_cast(port_);
1711 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1712 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1714 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1715 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1716 OFPUTIL_PC_NO_PACKET_IN)) {
1717 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1719 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1720 bundle_update(port->bundle);
1726 set_sflow(struct ofproto *ofproto_,
1727 const struct ofproto_sflow_options *sflow_options)
1729 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1730 struct dpif_sflow *ds = ofproto->sflow;
1732 if (sflow_options) {
1734 struct ofport_dpif *ofport;
1736 ds = ofproto->sflow = dpif_sflow_create();
1737 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1738 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1740 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1742 dpif_sflow_set_options(ds, sflow_options);
1745 dpif_sflow_destroy(ds);
1746 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1747 ofproto->sflow = NULL;
1754 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1756 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1763 struct ofproto_dpif *ofproto;
1765 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1766 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1767 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1770 if (cfm_configure(ofport->cfm, s)) {
1776 cfm_destroy(ofport->cfm);
1782 get_cfm_fault(const struct ofport *ofport_)
1784 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1786 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1790 get_cfm_opup(const struct ofport *ofport_)
1792 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1794 return ofport->cfm ? cfm_get_opup(ofport->cfm) : -1;
1798 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1801 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1804 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1812 get_cfm_health(const struct ofport *ofport_)
1814 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1816 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1819 /* Spanning Tree. */
1822 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1824 struct ofproto_dpif *ofproto = ofproto_;
1825 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1826 struct ofport_dpif *ofport;
1828 ofport = stp_port_get_aux(sp);
1830 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1831 ofproto->up.name, port_num);
1833 struct eth_header *eth = pkt->l2;
1835 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1836 if (eth_addr_is_zero(eth->eth_src)) {
1837 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1838 "with unknown MAC", ofproto->up.name, port_num);
1840 send_packet(ofport, pkt);
1846 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1848 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1850 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1852 /* Only revalidate flows if the configuration changed. */
1853 if (!s != !ofproto->stp) {
1854 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1858 if (!ofproto->stp) {
1859 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1860 send_bpdu_cb, ofproto);
1861 ofproto->stp_last_tick = time_msec();
1864 stp_set_bridge_id(ofproto->stp, s->system_id);
1865 stp_set_bridge_priority(ofproto->stp, s->priority);
1866 stp_set_hello_time(ofproto->stp, s->hello_time);
1867 stp_set_max_age(ofproto->stp, s->max_age);
1868 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1870 struct ofport *ofport;
1872 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1873 set_stp_port(ofport, NULL);
1876 stp_destroy(ofproto->stp);
1877 ofproto->stp = NULL;
1884 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1886 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1890 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1891 s->designated_root = stp_get_designated_root(ofproto->stp);
1892 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1901 update_stp_port_state(struct ofport_dpif *ofport)
1903 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1904 enum stp_state state;
1906 /* Figure out new state. */
1907 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1911 if (ofport->stp_state != state) {
1912 enum ofputil_port_state of_state;
1915 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1916 netdev_get_name(ofport->up.netdev),
1917 stp_state_name(ofport->stp_state),
1918 stp_state_name(state));
1919 if (stp_learn_in_state(ofport->stp_state)
1920 != stp_learn_in_state(state)) {
1921 /* xxx Learning action flows should also be flushed. */
1922 mac_learning_flush(ofproto->ml,
1923 &ofproto->backer->revalidate_set);
1925 fwd_change = stp_forward_in_state(ofport->stp_state)
1926 != stp_forward_in_state(state);
1928 ofproto->backer->need_revalidate = REV_STP;
1929 ofport->stp_state = state;
1930 ofport->stp_state_entered = time_msec();
1932 if (fwd_change && ofport->bundle) {
1933 bundle_update(ofport->bundle);
1936 /* Update the STP state bits in the OpenFlow port description. */
1937 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1938 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1939 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1940 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1941 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1943 ofproto_port_set_state(&ofport->up, of_state);
1947 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1948 * caller is responsible for assigning STP port numbers and ensuring
1949 * there are no duplicates. */
1951 set_stp_port(struct ofport *ofport_,
1952 const struct ofproto_port_stp_settings *s)
1954 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1955 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1956 struct stp_port *sp = ofport->stp_port;
1958 if (!s || !s->enable) {
1960 ofport->stp_port = NULL;
1961 stp_port_disable(sp);
1962 update_stp_port_state(ofport);
1965 } else if (sp && stp_port_no(sp) != s->port_num
1966 && ofport == stp_port_get_aux(sp)) {
1967 /* The port-id changed, so disable the old one if it's not
1968 * already in use by another port. */
1969 stp_port_disable(sp);
1972 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1973 stp_port_enable(sp);
1975 stp_port_set_aux(sp, ofport);
1976 stp_port_set_priority(sp, s->priority);
1977 stp_port_set_path_cost(sp, s->path_cost);
1979 update_stp_port_state(ofport);
1985 get_stp_port_status(struct ofport *ofport_,
1986 struct ofproto_port_stp_status *s)
1988 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1989 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1990 struct stp_port *sp = ofport->stp_port;
1992 if (!ofproto->stp || !sp) {
1998 s->port_id = stp_port_get_id(sp);
1999 s->state = stp_port_get_state(sp);
2000 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2001 s->role = stp_port_get_role(sp);
2002 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2008 stp_run(struct ofproto_dpif *ofproto)
2011 long long int now = time_msec();
2012 long long int elapsed = now - ofproto->stp_last_tick;
2013 struct stp_port *sp;
2016 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2017 ofproto->stp_last_tick = now;
2019 while (stp_get_changed_port(ofproto->stp, &sp)) {
2020 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2023 update_stp_port_state(ofport);
2027 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2028 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2034 stp_wait(struct ofproto_dpif *ofproto)
2037 poll_timer_wait(1000);
2041 /* Returns true if STP should process 'flow'. */
2043 stp_should_process_flow(const struct flow *flow)
2045 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2049 stp_process_packet(const struct ofport_dpif *ofport,
2050 const struct ofpbuf *packet)
2052 struct ofpbuf payload = *packet;
2053 struct eth_header *eth = payload.data;
2054 struct stp_port *sp = ofport->stp_port;
2056 /* Sink packets on ports that have STP disabled when the bridge has
2058 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2062 /* Trim off padding on payload. */
2063 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2064 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2067 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2068 stp_received_bpdu(sp, payload.data, payload.size);
2072 static struct priority_to_dscp *
2073 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2075 struct priority_to_dscp *pdscp;
2078 hash = hash_int(priority, 0);
2079 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2080 if (pdscp->priority == priority) {
2088 ofport_clear_priorities(struct ofport_dpif *ofport)
2090 struct priority_to_dscp *pdscp, *next;
2092 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2093 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2099 set_queues(struct ofport *ofport_,
2100 const struct ofproto_port_queue *qdscp_list,
2103 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2104 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2105 struct hmap new = HMAP_INITIALIZER(&new);
2108 for (i = 0; i < n_qdscp; i++) {
2109 struct priority_to_dscp *pdscp;
2113 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2114 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2119 pdscp = get_priority(ofport, priority);
2121 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2123 pdscp = xmalloc(sizeof *pdscp);
2124 pdscp->priority = priority;
2126 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2129 if (pdscp->dscp != dscp) {
2131 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2134 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2137 if (!hmap_is_empty(&ofport->priorities)) {
2138 ofport_clear_priorities(ofport);
2139 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2142 hmap_swap(&new, &ofport->priorities);
2150 /* Expires all MAC learning entries associated with 'bundle' and forces its
2151 * ofproto to revalidate every flow.
2153 * Normally MAC learning entries are removed only from the ofproto associated
2154 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2155 * are removed from every ofproto. When patch ports and SLB bonds are in use
2156 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2157 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2158 * with the host from which it migrated. */
2160 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2162 struct ofproto_dpif *ofproto = bundle->ofproto;
2163 struct mac_learning *ml = ofproto->ml;
2164 struct mac_entry *mac, *next_mac;
2166 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2167 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2168 if (mac->port.p == bundle) {
2170 struct ofproto_dpif *o;
2172 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2174 struct mac_entry *e;
2176 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2179 mac_learning_expire(o->ml, e);
2185 mac_learning_expire(ml, mac);
2190 static struct ofbundle *
2191 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2193 struct ofbundle *bundle;
2195 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2196 &ofproto->bundles) {
2197 if (bundle->aux == aux) {
2204 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2205 * ones that are found to 'bundles'. */
2207 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2208 void **auxes, size_t n_auxes,
2209 struct hmapx *bundles)
2213 hmapx_init(bundles);
2214 for (i = 0; i < n_auxes; i++) {
2215 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2217 hmapx_add(bundles, bundle);
2223 bundle_update(struct ofbundle *bundle)
2225 struct ofport_dpif *port;
2227 bundle->floodable = true;
2228 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2229 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2230 || !stp_forward_in_state(port->stp_state)) {
2231 bundle->floodable = false;
2238 bundle_del_port(struct ofport_dpif *port)
2240 struct ofbundle *bundle = port->bundle;
2242 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2244 list_remove(&port->bundle_node);
2245 port->bundle = NULL;
2248 lacp_slave_unregister(bundle->lacp, port);
2251 bond_slave_unregister(bundle->bond, port);
2254 bundle_update(bundle);
2258 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2259 struct lacp_slave_settings *lacp)
2261 struct ofport_dpif *port;
2263 port = get_ofp_port(bundle->ofproto, ofp_port);
2268 if (port->bundle != bundle) {
2269 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2271 bundle_del_port(port);
2274 port->bundle = bundle;
2275 list_push_back(&bundle->ports, &port->bundle_node);
2276 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2277 || !stp_forward_in_state(port->stp_state)) {
2278 bundle->floodable = false;
2282 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2283 lacp_slave_register(bundle->lacp, port, lacp);
2290 bundle_destroy(struct ofbundle *bundle)
2292 struct ofproto_dpif *ofproto;
2293 struct ofport_dpif *port, *next_port;
2300 ofproto = bundle->ofproto;
2301 for (i = 0; i < MAX_MIRRORS; i++) {
2302 struct ofmirror *m = ofproto->mirrors[i];
2304 if (m->out == bundle) {
2306 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2307 || hmapx_find_and_delete(&m->dsts, bundle)) {
2308 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2313 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2314 bundle_del_port(port);
2317 bundle_flush_macs(bundle, true);
2318 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2320 free(bundle->trunks);
2321 lacp_destroy(bundle->lacp);
2322 bond_destroy(bundle->bond);
2327 bundle_set(struct ofproto *ofproto_, void *aux,
2328 const struct ofproto_bundle_settings *s)
2330 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2331 bool need_flush = false;
2332 struct ofport_dpif *port;
2333 struct ofbundle *bundle;
2334 unsigned long *trunks;
2340 bundle_destroy(bundle_lookup(ofproto, aux));
2344 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2345 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2347 bundle = bundle_lookup(ofproto, aux);
2349 bundle = xmalloc(sizeof *bundle);
2351 bundle->ofproto = ofproto;
2352 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2353 hash_pointer(aux, 0));
2355 bundle->name = NULL;
2357 list_init(&bundle->ports);
2358 bundle->vlan_mode = PORT_VLAN_TRUNK;
2360 bundle->trunks = NULL;
2361 bundle->use_priority_tags = s->use_priority_tags;
2362 bundle->lacp = NULL;
2363 bundle->bond = NULL;
2365 bundle->floodable = true;
2367 bundle->src_mirrors = 0;
2368 bundle->dst_mirrors = 0;
2369 bundle->mirror_out = 0;
2372 if (!bundle->name || strcmp(s->name, bundle->name)) {
2374 bundle->name = xstrdup(s->name);
2379 if (!bundle->lacp) {
2380 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2381 bundle->lacp = lacp_create();
2383 lacp_configure(bundle->lacp, s->lacp);
2385 lacp_destroy(bundle->lacp);
2386 bundle->lacp = NULL;
2389 /* Update set of ports. */
2391 for (i = 0; i < s->n_slaves; i++) {
2392 if (!bundle_add_port(bundle, s->slaves[i],
2393 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2397 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2398 struct ofport_dpif *next_port;
2400 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2401 for (i = 0; i < s->n_slaves; i++) {
2402 if (s->slaves[i] == port->up.ofp_port) {
2407 bundle_del_port(port);
2411 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2413 if (list_is_empty(&bundle->ports)) {
2414 bundle_destroy(bundle);
2418 /* Set VLAN tagging mode */
2419 if (s->vlan_mode != bundle->vlan_mode
2420 || s->use_priority_tags != bundle->use_priority_tags) {
2421 bundle->vlan_mode = s->vlan_mode;
2422 bundle->use_priority_tags = s->use_priority_tags;
2427 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2428 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2430 if (vlan != bundle->vlan) {
2431 bundle->vlan = vlan;
2435 /* Get trunked VLANs. */
2436 switch (s->vlan_mode) {
2437 case PORT_VLAN_ACCESS:
2441 case PORT_VLAN_TRUNK:
2442 trunks = CONST_CAST(unsigned long *, s->trunks);
2445 case PORT_VLAN_NATIVE_UNTAGGED:
2446 case PORT_VLAN_NATIVE_TAGGED:
2447 if (vlan != 0 && (!s->trunks
2448 || !bitmap_is_set(s->trunks, vlan)
2449 || bitmap_is_set(s->trunks, 0))) {
2450 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2452 trunks = bitmap_clone(s->trunks, 4096);
2454 trunks = bitmap_allocate1(4096);
2456 bitmap_set1(trunks, vlan);
2457 bitmap_set0(trunks, 0);
2459 trunks = CONST_CAST(unsigned long *, s->trunks);
2466 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2467 free(bundle->trunks);
2468 if (trunks == s->trunks) {
2469 bundle->trunks = vlan_bitmap_clone(trunks);
2471 bundle->trunks = trunks;
2476 if (trunks != s->trunks) {
2481 if (!list_is_short(&bundle->ports)) {
2482 bundle->ofproto->has_bonded_bundles = true;
2484 if (bond_reconfigure(bundle->bond, s->bond)) {
2485 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2488 bundle->bond = bond_create(s->bond);
2489 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2492 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2493 bond_slave_register(bundle->bond, port, port->up.netdev);
2496 bond_destroy(bundle->bond);
2497 bundle->bond = NULL;
2500 /* If we changed something that would affect MAC learning, un-learn
2501 * everything on this port and force flow revalidation. */
2503 bundle_flush_macs(bundle, false);
2510 bundle_remove(struct ofport *port_)
2512 struct ofport_dpif *port = ofport_dpif_cast(port_);
2513 struct ofbundle *bundle = port->bundle;
2516 bundle_del_port(port);
2517 if (list_is_empty(&bundle->ports)) {
2518 bundle_destroy(bundle);
2519 } else if (list_is_short(&bundle->ports)) {
2520 bond_destroy(bundle->bond);
2521 bundle->bond = NULL;
2527 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2529 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2530 struct ofport_dpif *port = port_;
2531 uint8_t ea[ETH_ADDR_LEN];
2534 error = netdev_get_etheraddr(port->up.netdev, ea);
2536 struct ofpbuf packet;
2539 ofpbuf_init(&packet, 0);
2540 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2542 memcpy(packet_pdu, pdu, pdu_size);
2544 send_packet(port, &packet);
2545 ofpbuf_uninit(&packet);
2547 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2548 "%s (%s)", port->bundle->name,
2549 netdev_get_name(port->up.netdev), strerror(error));
2554 bundle_send_learning_packets(struct ofbundle *bundle)
2556 struct ofproto_dpif *ofproto = bundle->ofproto;
2557 int error, n_packets, n_errors;
2558 struct mac_entry *e;
2560 error = n_packets = n_errors = 0;
2561 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2562 if (e->port.p != bundle) {
2563 struct ofpbuf *learning_packet;
2564 struct ofport_dpif *port;
2568 /* The assignment to "port" is unnecessary but makes "grep"ing for
2569 * struct ofport_dpif more effective. */
2570 learning_packet = bond_compose_learning_packet(bundle->bond,
2574 ret = send_packet(port, learning_packet);
2575 ofpbuf_delete(learning_packet);
2585 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2586 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2587 "packets, last error was: %s",
2588 bundle->name, n_errors, n_packets, strerror(error));
2590 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2591 bundle->name, n_packets);
2596 bundle_run(struct ofbundle *bundle)
2599 lacp_run(bundle->lacp, send_pdu_cb);
2602 struct ofport_dpif *port;
2604 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2605 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2608 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2609 lacp_status(bundle->lacp));
2610 if (bond_should_send_learning_packets(bundle->bond)) {
2611 bundle_send_learning_packets(bundle);
2617 bundle_wait(struct ofbundle *bundle)
2620 lacp_wait(bundle->lacp);
2623 bond_wait(bundle->bond);
2630 mirror_scan(struct ofproto_dpif *ofproto)
2634 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2635 if (!ofproto->mirrors[idx]) {
2642 static struct ofmirror *
2643 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2647 for (i = 0; i < MAX_MIRRORS; i++) {
2648 struct ofmirror *mirror = ofproto->mirrors[i];
2649 if (mirror && mirror->aux == aux) {
2657 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2659 mirror_update_dups(struct ofproto_dpif *ofproto)
2663 for (i = 0; i < MAX_MIRRORS; i++) {
2664 struct ofmirror *m = ofproto->mirrors[i];
2667 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2671 for (i = 0; i < MAX_MIRRORS; i++) {
2672 struct ofmirror *m1 = ofproto->mirrors[i];
2679 for (j = i + 1; j < MAX_MIRRORS; j++) {
2680 struct ofmirror *m2 = ofproto->mirrors[j];
2682 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2683 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2684 m2->dup_mirrors |= m1->dup_mirrors;
2691 mirror_set(struct ofproto *ofproto_, void *aux,
2692 const struct ofproto_mirror_settings *s)
2694 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2695 mirror_mask_t mirror_bit;
2696 struct ofbundle *bundle;
2697 struct ofmirror *mirror;
2698 struct ofbundle *out;
2699 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2700 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2703 mirror = mirror_lookup(ofproto, aux);
2705 mirror_destroy(mirror);
2711 idx = mirror_scan(ofproto);
2713 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2715 ofproto->up.name, MAX_MIRRORS, s->name);
2719 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2720 mirror->ofproto = ofproto;
2723 mirror->out_vlan = -1;
2724 mirror->name = NULL;
2727 if (!mirror->name || strcmp(s->name, mirror->name)) {
2729 mirror->name = xstrdup(s->name);
2732 /* Get the new configuration. */
2733 if (s->out_bundle) {
2734 out = bundle_lookup(ofproto, s->out_bundle);
2736 mirror_destroy(mirror);
2742 out_vlan = s->out_vlan;
2744 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2745 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2747 /* If the configuration has not changed, do nothing. */
2748 if (hmapx_equals(&srcs, &mirror->srcs)
2749 && hmapx_equals(&dsts, &mirror->dsts)
2750 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2751 && mirror->out == out
2752 && mirror->out_vlan == out_vlan)
2754 hmapx_destroy(&srcs);
2755 hmapx_destroy(&dsts);
2759 hmapx_swap(&srcs, &mirror->srcs);
2760 hmapx_destroy(&srcs);
2762 hmapx_swap(&dsts, &mirror->dsts);
2763 hmapx_destroy(&dsts);
2765 free(mirror->vlans);
2766 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2769 mirror->out_vlan = out_vlan;
2771 /* Update bundles. */
2772 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2773 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2774 if (hmapx_contains(&mirror->srcs, bundle)) {
2775 bundle->src_mirrors |= mirror_bit;
2777 bundle->src_mirrors &= ~mirror_bit;
2780 if (hmapx_contains(&mirror->dsts, bundle)) {
2781 bundle->dst_mirrors |= mirror_bit;
2783 bundle->dst_mirrors &= ~mirror_bit;
2786 if (mirror->out == bundle) {
2787 bundle->mirror_out |= mirror_bit;
2789 bundle->mirror_out &= ~mirror_bit;
2793 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2794 ofproto->has_mirrors = true;
2795 mac_learning_flush(ofproto->ml,
2796 &ofproto->backer->revalidate_set);
2797 mirror_update_dups(ofproto);
2803 mirror_destroy(struct ofmirror *mirror)
2805 struct ofproto_dpif *ofproto;
2806 mirror_mask_t mirror_bit;
2807 struct ofbundle *bundle;
2814 ofproto = mirror->ofproto;
2815 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2816 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2818 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2819 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2820 bundle->src_mirrors &= ~mirror_bit;
2821 bundle->dst_mirrors &= ~mirror_bit;
2822 bundle->mirror_out &= ~mirror_bit;
2825 hmapx_destroy(&mirror->srcs);
2826 hmapx_destroy(&mirror->dsts);
2827 free(mirror->vlans);
2829 ofproto->mirrors[mirror->idx] = NULL;
2833 mirror_update_dups(ofproto);
2835 ofproto->has_mirrors = false;
2836 for (i = 0; i < MAX_MIRRORS; i++) {
2837 if (ofproto->mirrors[i]) {
2838 ofproto->has_mirrors = true;
2845 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2846 uint64_t *packets, uint64_t *bytes)
2848 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2849 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2852 *packets = *bytes = UINT64_MAX;
2856 *packets = mirror->packet_count;
2857 *bytes = mirror->byte_count;
2863 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2865 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2866 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2867 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2873 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2875 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2876 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2877 return bundle && bundle->mirror_out != 0;
2881 forward_bpdu_changed(struct ofproto *ofproto_)
2883 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2884 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2888 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2891 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2892 mac_learning_set_idle_time(ofproto->ml, idle_time);
2893 mac_learning_set_max_entries(ofproto->ml, max_entries);
2898 static struct ofport_dpif *
2899 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2901 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2902 return ofport ? ofport_dpif_cast(ofport) : NULL;
2905 static struct ofport_dpif *
2906 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2908 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2909 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2913 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2914 struct ofproto_port *ofproto_port,
2915 struct dpif_port *dpif_port)
2917 ofproto_port->name = dpif_port->name;
2918 ofproto_port->type = dpif_port->type;
2919 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2922 static struct ofport_dpif *
2923 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
2925 const struct ofproto_dpif *ofproto;
2928 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
2933 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2934 struct ofport *ofport;
2936 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
2937 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
2938 return ofport_dpif_cast(ofport);
2945 port_run_fast(struct ofport_dpif *ofport)
2947 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2948 struct ofpbuf packet;
2950 ofpbuf_init(&packet, 0);
2951 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2952 send_packet(ofport, &packet);
2953 ofpbuf_uninit(&packet);
2958 port_run(struct ofport_dpif *ofport)
2960 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2961 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2962 bool enable = netdev_get_carrier(ofport->up.netdev);
2964 ofport->carrier_seq = carrier_seq;
2966 port_run_fast(ofport);
2968 if (ofport->tnl_port
2969 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
2970 &ofport->tnl_port)) {
2971 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
2975 int cfm_opup = cfm_get_opup(ofport->cfm);
2977 cfm_run(ofport->cfm);
2978 enable = enable && !cfm_get_fault(ofport->cfm);
2980 if (cfm_opup >= 0) {
2981 enable = enable && cfm_opup;
2985 if (ofport->bundle) {
2986 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2987 if (carrier_changed) {
2988 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2992 if (ofport->may_enable != enable) {
2993 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2995 if (ofproto->has_bundle_action) {
2996 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3000 ofport->may_enable = enable;
3004 port_wait(struct ofport_dpif *ofport)
3007 cfm_wait(ofport->cfm);
3012 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3013 struct ofproto_port *ofproto_port)
3015 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3016 struct dpif_port dpif_port;
3019 if (sset_contains(&ofproto->ghost_ports, devname)) {
3020 const char *type = netdev_get_type_from_name(devname);
3022 /* We may be called before ofproto->up.port_by_name is populated with
3023 * the appropriate ofport. For this reason, we must get the name and
3024 * type from the netdev layer directly. */
3026 const struct ofport *ofport;
3028 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3029 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3030 ofproto_port->name = xstrdup(devname);
3031 ofproto_port->type = xstrdup(type);
3037 if (!sset_contains(&ofproto->ports, devname)) {
3040 error = dpif_port_query_by_name(ofproto->backer->dpif,
3041 devname, &dpif_port);
3043 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3049 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3051 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3052 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3053 const char *devname = netdev_get_name(netdev);
3055 if (netdev_vport_is_patch(netdev)) {
3056 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3060 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3061 uint32_t port_no = UINT32_MAX;
3064 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3068 if (netdev_get_tunnel_config(netdev)) {
3069 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3073 if (netdev_get_tunnel_config(netdev)) {
3074 sset_add(&ofproto->ghost_ports, devname);
3076 sset_add(&ofproto->ports, devname);
3082 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3084 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3085 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3092 sset_find_and_delete(&ofproto->ghost_ports,
3093 netdev_get_name(ofport->up.netdev));
3094 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3095 if (!ofport->tnl_port) {
3096 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3098 /* The caller is going to close ofport->up.netdev. If this is a
3099 * bonded port, then the bond is using that netdev, so remove it
3100 * from the bond. The client will need to reconfigure everything
3101 * after deleting ports, so then the slave will get re-added. */
3102 bundle_remove(&ofport->up);
3109 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3111 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3114 error = netdev_get_stats(ofport->up.netdev, stats);
3116 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3117 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3119 /* ofproto->stats.tx_packets represents packets that we created
3120 * internally and sent to some port (e.g. packets sent with
3121 * send_packet()). Account for them as if they had come from
3122 * OFPP_LOCAL and got forwarded. */
3124 if (stats->rx_packets != UINT64_MAX) {
3125 stats->rx_packets += ofproto->stats.tx_packets;
3128 if (stats->rx_bytes != UINT64_MAX) {
3129 stats->rx_bytes += ofproto->stats.tx_bytes;
3132 /* ofproto->stats.rx_packets represents packets that were received on
3133 * some port and we processed internally and dropped (e.g. STP).
3134 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3136 if (stats->tx_packets != UINT64_MAX) {
3137 stats->tx_packets += ofproto->stats.rx_packets;
3140 if (stats->tx_bytes != UINT64_MAX) {
3141 stats->tx_bytes += ofproto->stats.rx_bytes;
3148 /* Account packets for LOCAL port. */
3150 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
3151 size_t tx_size, size_t rx_size)
3153 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3156 ofproto->stats.rx_packets++;
3157 ofproto->stats.rx_bytes += rx_size;
3160 ofproto->stats.tx_packets++;
3161 ofproto->stats.tx_bytes += tx_size;
3165 struct port_dump_state {
3170 struct ofproto_port port;
3175 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3177 *statep = xzalloc(sizeof(struct port_dump_state));
3182 port_dump_next(const struct ofproto *ofproto_, void *state_,
3183 struct ofproto_port *port)
3185 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3186 struct port_dump_state *state = state_;
3187 const struct sset *sset;
3188 struct sset_node *node;
3190 if (state->has_port) {
3191 ofproto_port_destroy(&state->port);
3192 state->has_port = false;
3194 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3195 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3198 error = port_query_by_name(ofproto_, node->name, &state->port);
3200 *port = state->port;
3201 state->has_port = true;
3203 } else if (error != ENODEV) {
3208 if (!state->ghost) {
3209 state->ghost = true;
3212 return port_dump_next(ofproto_, state_, port);
3219 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3221 struct port_dump_state *state = state_;
3223 if (state->has_port) {
3224 ofproto_port_destroy(&state->port);
3231 port_poll(const struct ofproto *ofproto_, char **devnamep)
3233 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3235 if (ofproto->port_poll_errno) {
3236 int error = ofproto->port_poll_errno;
3237 ofproto->port_poll_errno = 0;
3241 if (sset_is_empty(&ofproto->port_poll_set)) {
3245 *devnamep = sset_pop(&ofproto->port_poll_set);
3250 port_poll_wait(const struct ofproto *ofproto_)
3252 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3253 dpif_port_poll_wait(ofproto->backer->dpif);
3257 port_is_lacp_current(const struct ofport *ofport_)
3259 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3260 return (ofport->bundle && ofport->bundle->lacp
3261 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3265 /* Upcall handling. */
3267 /* Flow miss batching.
3269 * Some dpifs implement operations faster when you hand them off in a batch.
3270 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3271 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3272 * more packets, plus possibly installing the flow in the dpif.
3274 * So far we only batch the operations that affect flow setup time the most.
3275 * It's possible to batch more than that, but the benefit might be minimal. */
3277 struct hmap_node hmap_node;
3278 struct ofproto_dpif *ofproto;
3280 enum odp_key_fitness key_fitness;
3281 const struct nlattr *key;
3283 ovs_be16 initial_tci;
3284 struct list packets;
3285 enum dpif_upcall_type upcall_type;
3286 uint32_t odp_in_port;
3289 struct flow_miss_op {
3290 struct dpif_op dpif_op;
3291 void *garbage; /* Pointer to pass to free(), NULL if none. */
3292 uint64_t stub[1024 / 8]; /* Temporary buffer. */
3295 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3296 * OpenFlow controller as necessary according to their individual
3297 * configurations. */
3299 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3300 const struct flow *flow)
3302 struct ofputil_packet_in pin;
3304 pin.packet = packet->data;
3305 pin.packet_len = packet->size;
3306 pin.reason = OFPR_NO_MATCH;
3307 pin.controller_id = 0;
3312 pin.send_len = 0; /* not used for flow table misses */
3314 flow_get_metadata(flow, &pin.fmd);
3316 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3319 static enum slow_path_reason
3320 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3321 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3325 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3327 cfm_process_heartbeat(ofport->cfm, packet);
3330 } else if (ofport->bundle && ofport->bundle->lacp
3331 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3333 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3336 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3338 stp_process_packet(ofport, packet);
3346 static struct flow_miss *
3347 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3348 const struct flow *flow, uint32_t hash)
3350 struct flow_miss *miss;
3352 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3353 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3361 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3362 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3363 * 'miss' is associated with a subfacet the caller must also initialize the
3364 * returned op->subfacet, and if anything needs to be freed after processing
3365 * the op, the caller must initialize op->garbage also. */
3367 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3368 struct flow_miss_op *op)
3370 if (miss->flow.vlan_tci != miss->initial_tci) {
3371 /* This packet was received on a VLAN splinter port. We
3372 * added a VLAN to the packet to make the packet resemble
3373 * the flow, but the actions were composed assuming that
3374 * the packet contained no VLAN. So, we must remove the
3375 * VLAN header from the packet before trying to execute the
3377 eth_pop_vlan(packet);
3381 op->dpif_op.type = DPIF_OP_EXECUTE;
3382 op->dpif_op.u.execute.key = miss->key;
3383 op->dpif_op.u.execute.key_len = miss->key_len;
3384 op->dpif_op.u.execute.packet = packet;
3387 /* Helper for handle_flow_miss_without_facet() and
3388 * handle_flow_miss_with_facet(). */
3390 handle_flow_miss_common(struct rule_dpif *rule,
3391 struct ofpbuf *packet, const struct flow *flow)
3393 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3395 ofproto->n_matches++;
3397 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3399 * Extra-special case for fail-open mode.
3401 * We are in fail-open mode and the packet matched the fail-open
3402 * rule, but we are connected to a controller too. We should send
3403 * the packet up to the controller in the hope that it will try to
3404 * set up a flow and thereby allow us to exit fail-open.
3406 * See the top-level comment in fail-open.c for more information.
3408 send_packet_in_miss(ofproto, packet, flow);
3412 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3413 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3414 * installing a datapath flow. The answer is usually "yes" (a return value of
3415 * true). However, for short flows the cost of bookkeeping is much higher than
3416 * the benefits, so when the datapath holds a large number of flows we impose
3417 * some heuristics to decide which flows are likely to be worth tracking. */
3419 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3420 struct flow_miss *miss, uint32_t hash)
3422 if (!ofproto->governor) {
3425 n_subfacets = hmap_count(&ofproto->subfacets);
3426 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3430 ofproto->governor = governor_create(ofproto->up.name);
3433 return governor_should_install_flow(ofproto->governor, hash,
3434 list_size(&miss->packets));
3437 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3438 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3439 * increment '*n_ops'. */
3441 handle_flow_miss_without_facet(struct flow_miss *miss,
3442 struct rule_dpif *rule,
3443 struct flow_miss_op *ops, size_t *n_ops)
3445 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3446 long long int now = time_msec();
3447 struct action_xlate_ctx ctx;
3448 struct ofpbuf *packet;
3450 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3451 struct flow_miss_op *op = &ops[*n_ops];
3452 struct dpif_flow_stats stats;
3453 struct ofpbuf odp_actions;
3455 COVERAGE_INC(facet_suppress);
3457 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3459 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3460 rule_credit_stats(rule, &stats);
3462 action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci,
3464 ctx.resubmit_stats = &stats;
3465 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3468 if (odp_actions.size) {
3469 struct dpif_execute *execute = &op->dpif_op.u.execute;
3471 init_flow_miss_execute_op(miss, packet, op);
3472 execute->actions = odp_actions.data;
3473 execute->actions_len = odp_actions.size;
3474 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3478 ofpbuf_uninit(&odp_actions);
3483 /* Handles 'miss', which matches 'facet'. May add any required datapath
3484 * operations to 'ops', incrementing '*n_ops' for each new op.
3486 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3487 * This is really important only for new facets: if we just called time_msec()
3488 * here, then the new subfacet or its packets could look (occasionally) as
3489 * though it was used some time after the facet was used. That can make a
3490 * one-packet flow look like it has a nonzero duration, which looks odd in
3491 * e.g. NetFlow statistics. */
3493 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3495 struct flow_miss_op *ops, size_t *n_ops)
3497 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3498 enum subfacet_path want_path;
3499 struct subfacet *subfacet;
3500 struct ofpbuf *packet;
3502 subfacet = subfacet_create(facet, miss, now);
3504 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3505 struct flow_miss_op *op = &ops[*n_ops];
3506 struct dpif_flow_stats stats;
3507 struct ofpbuf odp_actions;
3509 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3511 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3512 if (!subfacet->actions || subfacet->slow) {
3513 subfacet_make_actions(subfacet, packet, &odp_actions);
3516 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3517 subfacet_update_stats(subfacet, &stats);
3519 if (subfacet->actions_len) {
3520 struct dpif_execute *execute = &op->dpif_op.u.execute;
3522 init_flow_miss_execute_op(miss, packet, op);
3523 if (!subfacet->slow) {
3524 execute->actions = subfacet->actions;
3525 execute->actions_len = subfacet->actions_len;
3526 ofpbuf_uninit(&odp_actions);
3528 execute->actions = odp_actions.data;
3529 execute->actions_len = odp_actions.size;
3530 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3535 ofpbuf_uninit(&odp_actions);
3539 want_path = subfacet_want_path(subfacet->slow);
3540 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3541 struct flow_miss_op *op = &ops[(*n_ops)++];
3542 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3544 subfacet->path = want_path;
3547 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3548 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3549 put->key = miss->key;
3550 put->key_len = miss->key_len;
3551 if (want_path == SF_FAST_PATH) {
3552 put->actions = subfacet->actions;
3553 put->actions_len = subfacet->actions_len;
3555 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
3556 op->stub, sizeof op->stub,
3557 &put->actions, &put->actions_len);
3563 /* Handles flow miss 'miss'. May add any required datapath operations
3564 * to 'ops', incrementing '*n_ops' for each new op. */
3566 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3569 struct ofproto_dpif *ofproto = miss->ofproto;
3570 struct facet *facet;
3574 /* The caller must ensure that miss->hmap_node.hash contains
3575 * flow_hash(miss->flow, 0). */
3576 hash = miss->hmap_node.hash;
3578 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3580 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
3582 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
3583 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
3587 facet = facet_create(rule, &miss->flow, hash);
3592 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3595 static struct drop_key *
3596 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3599 struct drop_key *drop_key;
3601 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3602 &backer->drop_keys) {
3603 if (drop_key->key_len == key_len
3604 && !memcmp(drop_key->key, key, key_len)) {
3612 drop_key_clear(struct dpif_backer *backer)
3614 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3615 struct drop_key *drop_key, *next;
3617 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3620 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3622 if (error && !VLOG_DROP_WARN(&rl)) {
3623 struct ds ds = DS_EMPTY_INITIALIZER;
3624 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3625 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3630 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3631 free(drop_key->key);
3636 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3637 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3638 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3639 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3640 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3641 * 'packet' ingressed.
3643 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3644 * 'flow''s in_port to OFPP_NONE.
3646 * This function does post-processing on data returned from
3647 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3648 * of the upcall processing logic. In particular, if the extracted in_port is
3649 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3650 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3651 * a VLAN header onto 'packet' (if it is nonnull).
3653 * Optionally, if nonnull, sets '*initial_tci' to the VLAN TCI with which the
3654 * packet was really received, that is, the actual VLAN TCI extracted by
3655 * odp_flow_key_to_flow(). (This differs from the value returned in
3656 * flow->vlan_tci only for packets received on VLAN splinters.)
3658 * Similarly, this function also includes some logic to help with tunnels. It
3659 * may modify 'flow' as necessary to make the tunneling implementation
3660 * transparent to the upcall processing logic.
3662 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3663 * or some other positive errno if there are other problems. */
3665 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3666 const struct nlattr *key, size_t key_len,
3667 struct flow *flow, enum odp_key_fitness *fitnessp,
3668 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3669 ovs_be16 *initial_tci)
3671 const struct ofport_dpif *port;
3672 enum odp_key_fitness fitness;
3675 fitness = odp_flow_key_to_flow(key, key_len, flow);
3676 if (fitness == ODP_FIT_ERROR) {
3682 *initial_tci = flow->vlan_tci;
3686 *odp_in_port = flow->in_port;
3689 if (tnl_port_should_receive(flow)) {
3690 const struct ofport *ofport = tnl_port_receive(flow);
3692 flow->in_port = OFPP_NONE;
3695 port = ofport_dpif_cast(ofport);
3697 /* We can't reproduce 'key' from 'flow'. */
3698 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3700 /* XXX: Since the tunnel module is not scoped per backer, it's
3701 * theoretically possible that we'll receive an ofport belonging to an
3702 * entirely different datapath. In practice, this can't happen because
3703 * no platforms has two separate datapaths which each support
3705 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3707 port = odp_port_to_ofport(backer, flow->in_port);
3709 flow->in_port = OFPP_NONE;
3713 flow->in_port = port->up.ofp_port;
3714 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3716 /* Make the packet resemble the flow, so that it gets sent to
3717 * an OpenFlow controller properly, so that it looks correct
3718 * for sFlow, and so that flow_extract() will get the correct
3719 * vlan_tci if it is called on 'packet'.
3721 * The allocated space inside 'packet' probably also contains
3722 * 'key', that is, both 'packet' and 'key' are probably part of
3723 * a struct dpif_upcall (see the large comment on that
3724 * structure definition), so pushing data on 'packet' is in
3725 * general not a good idea since it could overwrite 'key' or
3726 * free it as a side effect. However, it's OK in this special
3727 * case because we know that 'packet' is inside a Netlink
3728 * attribute: pushing 4 bytes will just overwrite the 4-byte
3729 * "struct nlattr", which is fine since we don't need that
3730 * header anymore. */
3731 eth_push_vlan(packet, flow->vlan_tci);
3733 /* We can't reproduce 'key' from 'flow'. */
3734 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3740 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3745 *fitnessp = fitness;
3751 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3754 struct dpif_upcall *upcall;
3755 struct flow_miss *miss;
3756 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3757 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3758 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3768 /* Construct the to-do list.
3770 * This just amounts to extracting the flow from each packet and sticking
3771 * the packets that have the same flow in the same "flow_miss" structure so
3772 * that we can process them together. */
3775 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3776 struct flow_miss *miss = &misses[n_misses];
3777 struct flow_miss *existing_miss;
3778 struct ofproto_dpif *ofproto;
3779 uint32_t odp_in_port;
3784 error = ofproto_receive(backer, upcall->packet, upcall->key,
3785 upcall->key_len, &flow, &miss->key_fitness,
3786 &ofproto, &odp_in_port, &miss->initial_tci);
3787 if (error == ENODEV) {
3788 struct drop_key *drop_key;
3790 /* Received packet on port for which we couldn't associate
3791 * an ofproto. This can happen if a port is removed while
3792 * traffic is being received. Print a rate-limited message
3793 * in case it happens frequently. Install a drop flow so
3794 * that future packets of the flow are inexpensively dropped
3796 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
3799 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3801 drop_key = xmalloc(sizeof *drop_key);
3802 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3803 drop_key->key_len = upcall->key_len;
3805 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3806 hash_bytes(drop_key->key, drop_key->key_len, 0));
3807 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3808 drop_key->key, drop_key->key_len, NULL, 0, NULL);
3815 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3816 &flow.tunnel, flow.in_port, &miss->flow);
3818 /* Add other packets to a to-do list. */
3819 hash = flow_hash(&miss->flow, 0);
3820 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
3821 if (!existing_miss) {
3822 hmap_insert(&todo, &miss->hmap_node, hash);
3823 miss->ofproto = ofproto;
3824 miss->key = upcall->key;
3825 miss->key_len = upcall->key_len;
3826 miss->upcall_type = upcall->type;
3827 miss->odp_in_port = odp_in_port;
3828 list_init(&miss->packets);
3832 miss = existing_miss;
3834 list_push_back(&miss->packets, &upcall->packet->list_node);
3837 /* Process each element in the to-do list, constructing the set of
3838 * operations to batch. */
3840 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3841 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3843 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3845 /* Execute batch. */
3846 for (i = 0; i < n_ops; i++) {
3847 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3849 dpif_operate(backer->dpif, dpif_ops, n_ops);
3852 for (i = 0; i < n_ops; i++) {
3853 free(flow_miss_ops[i].garbage);
3855 hmap_destroy(&todo);
3858 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
3859 classify_upcall(const struct dpif_upcall *upcall)
3861 union user_action_cookie cookie;
3863 /* First look at the upcall type. */
3864 switch (upcall->type) {
3865 case DPIF_UC_ACTION:
3871 case DPIF_N_UC_TYPES:
3873 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3877 /* "action" upcalls need a closer look. */
3878 if (!upcall->userdata) {
3879 VLOG_WARN_RL(&rl, "action upcall missing cookie");
3882 if (nl_attr_get_size(upcall->userdata) != sizeof(cookie)) {
3883 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
3884 nl_attr_get_size(upcall->userdata));
3887 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof(cookie));
3888 switch (cookie.type) {
3889 case USER_ACTION_COOKIE_SFLOW:
3890 return SFLOW_UPCALL;
3892 case USER_ACTION_COOKIE_SLOW_PATH:
3895 case USER_ACTION_COOKIE_UNSPEC:
3897 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64,
3898 nl_attr_get_u64(upcall->userdata));
3904 handle_sflow_upcall(struct dpif_backer *backer,
3905 const struct dpif_upcall *upcall)
3907 struct ofproto_dpif *ofproto;
3908 union user_action_cookie cookie;
3910 uint32_t odp_in_port;
3912 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3913 &flow, NULL, &ofproto, &odp_in_port, NULL)
3914 || !ofproto->sflow) {
3918 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof(cookie));
3919 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3920 odp_in_port, &cookie);
3924 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3926 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3927 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3928 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3933 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
3936 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3937 struct dpif_upcall *upcall = &misses[n_misses];
3938 struct ofpbuf *buf = &miss_bufs[n_misses];
3941 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3942 sizeof miss_buf_stubs[n_misses]);
3943 error = dpif_recv(backer->dpif, upcall, buf);
3949 switch (classify_upcall(upcall)) {
3951 /* Handle it later. */
3956 handle_sflow_upcall(backer, upcall);
3966 /* Handle deferred MISS_UPCALL processing. */
3967 handle_miss_upcalls(backer, misses, n_misses);
3968 for (i = 0; i < n_misses; i++) {
3969 ofpbuf_uninit(&miss_bufs[i]);
3975 /* Flow expiration. */
3977 static int subfacet_max_idle(const struct ofproto_dpif *);
3978 static void update_stats(struct dpif_backer *);
3979 static void rule_expire(struct rule_dpif *);
3980 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
3982 /* This function is called periodically by run(). Its job is to collect
3983 * updates for the flows that have been installed into the datapath, most
3984 * importantly when they last were used, and then use that information to
3985 * expire flows that have not been used recently.
3987 * Returns the number of milliseconds after which it should be called again. */
3989 expire(struct dpif_backer *backer)
3991 struct ofproto_dpif *ofproto;
3992 int max_idle = INT32_MAX;
3994 /* Periodically clear out the drop keys in an effort to keep them
3995 * relatively few. */
3996 drop_key_clear(backer);
3998 /* Update stats for each flow in the backer. */
3999 update_stats(backer);
4001 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4002 struct rule *rule, *next_rule;
4005 if (ofproto->backer != backer) {
4009 /* Expire subfacets that have been idle too long. */
4010 dp_max_idle = subfacet_max_idle(ofproto);
4011 expire_subfacets(ofproto, dp_max_idle);
4013 max_idle = MIN(max_idle, dp_max_idle);
4015 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4017 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4018 &ofproto->up.expirable) {
4019 rule_expire(rule_dpif_cast(rule));
4022 /* All outstanding data in existing flows has been accounted, so it's a
4023 * good time to do bond rebalancing. */
4024 if (ofproto->has_bonded_bundles) {
4025 struct ofbundle *bundle;
4027 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4029 bond_rebalance(bundle->bond, &backer->revalidate_set);
4035 return MIN(max_idle, 1000);
4038 /* Updates flow table statistics given that the datapath just reported 'stats'
4039 * as 'subfacet''s statistics. */
4041 update_subfacet_stats(struct subfacet *subfacet,
4042 const struct dpif_flow_stats *stats)
4044 struct facet *facet = subfacet->facet;
4046 if (stats->n_packets >= subfacet->dp_packet_count) {
4047 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
4048 facet->packet_count += extra;
4050 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4053 if (stats->n_bytes >= subfacet->dp_byte_count) {
4054 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
4056 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4059 subfacet->dp_packet_count = stats->n_packets;
4060 subfacet->dp_byte_count = stats->n_bytes;
4062 facet->tcp_flags |= stats->tcp_flags;
4064 subfacet_update_time(subfacet, stats->used);
4065 if (facet->accounted_bytes < facet->byte_count) {
4067 facet_account(facet);
4068 facet->accounted_bytes = facet->byte_count;
4070 facet_push_stats(facet);
4073 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4074 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4076 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4077 const struct nlattr *key, size_t key_len)
4079 if (!VLOG_DROP_WARN(&rl)) {
4083 odp_flow_key_format(key, key_len, &s);
4084 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4088 COVERAGE_INC(facet_unexpected);
4089 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4092 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4094 * This function also pushes statistics updates to rules which each facet
4095 * resubmits into. Generally these statistics will be accurate. However, if a
4096 * facet changes the rule it resubmits into at some time in between
4097 * update_stats() runs, it is possible that statistics accrued to the
4098 * old rule will be incorrectly attributed to the new rule. This could be
4099 * avoided by calling update_stats() whenever rules are created or
4100 * deleted. However, the performance impact of making so many calls to the
4101 * datapath do not justify the benefit of having perfectly accurate statistics.
4104 update_stats(struct dpif_backer *backer)
4106 const struct dpif_flow_stats *stats;
4107 struct dpif_flow_dump dump;
4108 const struct nlattr *key;
4111 dpif_flow_dump_start(&dump, backer->dpif);
4112 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4114 struct subfacet *subfacet;
4115 struct ofproto_dpif *ofproto;
4116 struct ofport_dpif *ofport;
4119 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4124 ofport = get_ofp_port(ofproto, flow.in_port);
4125 if (ofport && ofport->tnl_port) {
4126 netdev_vport_inc_rx(ofport->up.netdev, stats);
4129 key_hash = odp_flow_key_hash(key, key_len);
4130 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4131 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4133 update_subfacet_stats(subfacet, stats);
4137 /* Stats are updated per-packet. */
4140 case SF_NOT_INSTALLED:
4142 delete_unexpected_flow(ofproto, key, key_len);
4146 dpif_flow_dump_done(&dump);
4149 /* Calculates and returns the number of milliseconds of idle time after which
4150 * subfacets should expire from the datapath. When a subfacet expires, we fold
4151 * its statistics into its facet, and when a facet's last subfacet expires, we
4152 * fold its statistic into its rule. */
4154 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4157 * Idle time histogram.
4159 * Most of the time a switch has a relatively small number of subfacets.
4160 * When this is the case we might as well keep statistics for all of them
4161 * in userspace and to cache them in the kernel datapath for performance as
4164 * As the number of subfacets increases, the memory required to maintain
4165 * statistics about them in userspace and in the kernel becomes
4166 * significant. However, with a large number of subfacets it is likely
4167 * that only a few of them are "heavy hitters" that consume a large amount
4168 * of bandwidth. At this point, only heavy hitters are worth caching in
4169 * the kernel and maintaining in userspaces; other subfacets we can
4172 * The technique used to compute the idle time is to build a histogram with
4173 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4174 * that is installed in the kernel gets dropped in the appropriate bucket.
4175 * After the histogram has been built, we compute the cutoff so that only
4176 * the most-recently-used 1% of subfacets (but at least
4177 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4178 * the most-recently-used bucket of subfacets is kept, so actually an
4179 * arbitrary number of subfacets can be kept in any given expiration run
4180 * (though the next run will delete most of those unless they receive
4183 * This requires a second pass through the subfacets, in addition to the
4184 * pass made by update_stats(), because the former function never looks at
4185 * uninstallable subfacets.
4187 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4188 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4189 int buckets[N_BUCKETS] = { 0 };
4190 int total, subtotal, bucket;
4191 struct subfacet *subfacet;
4195 total = hmap_count(&ofproto->subfacets);
4196 if (total <= ofproto->up.flow_eviction_threshold) {
4197 return N_BUCKETS * BUCKET_WIDTH;
4200 /* Build histogram. */
4202 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4203 long long int idle = now - subfacet->used;
4204 int bucket = (idle <= 0 ? 0
4205 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4206 : (unsigned int) idle / BUCKET_WIDTH);
4210 /* Find the first bucket whose flows should be expired. */
4211 subtotal = bucket = 0;
4213 subtotal += buckets[bucket++];
4214 } while (bucket < N_BUCKETS &&
4215 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4217 if (VLOG_IS_DBG_ENABLED()) {
4221 ds_put_cstr(&s, "keep");
4222 for (i = 0; i < N_BUCKETS; i++) {
4224 ds_put_cstr(&s, ", drop");
4227 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4230 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4234 return bucket * BUCKET_WIDTH;
4238 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4240 /* Cutoff time for most flows. */
4241 long long int normal_cutoff = time_msec() - dp_max_idle;
4243 /* We really want to keep flows for special protocols around, so use a more
4244 * conservative cutoff. */
4245 long long int special_cutoff = time_msec() - 10000;
4247 struct subfacet *subfacet, *next_subfacet;
4248 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4252 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4253 &ofproto->subfacets) {
4254 long long int cutoff;
4256 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
4259 if (subfacet->used < cutoff) {
4260 if (subfacet->path != SF_NOT_INSTALLED) {
4261 batch[n_batch++] = subfacet;
4262 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4263 subfacet_destroy_batch(ofproto, batch, n_batch);
4267 subfacet_destroy(subfacet);
4273 subfacet_destroy_batch(ofproto, batch, n_batch);
4277 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4278 * then delete it entirely. */
4280 rule_expire(struct rule_dpif *rule)
4282 struct facet *facet, *next_facet;
4286 if (rule->up.pending) {
4287 /* We'll have to expire it later. */
4291 /* Has 'rule' expired? */
4293 if (rule->up.hard_timeout
4294 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4295 reason = OFPRR_HARD_TIMEOUT;
4296 } else if (rule->up.idle_timeout
4297 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4298 reason = OFPRR_IDLE_TIMEOUT;
4303 COVERAGE_INC(ofproto_dpif_expired);
4305 /* Update stats. (This is a no-op if the rule expired due to an idle
4306 * timeout, because that only happens when the rule has no facets left.) */
4307 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4308 facet_remove(facet);
4311 /* Get rid of the rule. */
4312 ofproto_rule_expire(&rule->up, reason);
4317 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
4319 * The caller must already have determined that no facet with an identical
4320 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
4321 * the ofproto's classifier table.
4323 * 'hash' must be the return value of flow_hash(flow, 0).
4325 * The facet will initially have no subfacets. The caller should create (at
4326 * least) one subfacet with subfacet_create(). */
4327 static struct facet *
4328 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
4330 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4331 struct facet *facet;
4333 facet = xzalloc(sizeof *facet);
4334 facet->used = time_msec();
4335 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4336 list_push_back(&rule->facets, &facet->list_node);
4338 facet->flow = *flow;
4339 list_init(&facet->subfacets);
4340 netflow_flow_init(&facet->nf_flow);
4341 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4347 facet_free(struct facet *facet)
4352 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4353 * 'packet', which arrived on 'in_port'. */
4355 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4356 const struct nlattr *odp_actions, size_t actions_len,
4357 struct ofpbuf *packet)
4359 struct odputil_keybuf keybuf;
4363 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4364 odp_flow_key_from_flow(&key, flow,
4365 ofp_port_to_odp_port(ofproto, flow->in_port));
4367 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4368 odp_actions, actions_len, packet);
4372 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4374 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4375 * rule's statistics, via subfacet_uninstall().
4377 * - Removes 'facet' from its rule and from ofproto->facets.
4380 facet_remove(struct facet *facet)
4382 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4383 struct subfacet *subfacet, *next_subfacet;
4385 ovs_assert(!list_is_empty(&facet->subfacets));
4387 /* First uninstall all of the subfacets to get final statistics. */
4388 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4389 subfacet_uninstall(subfacet);
4392 /* Flush the final stats to the rule.
4394 * This might require us to have at least one subfacet around so that we
4395 * can use its actions for accounting in facet_account(), which is why we
4396 * have uninstalled but not yet destroyed the subfacets. */
4397 facet_flush_stats(facet);
4399 /* Now we're really all done so destroy everything. */
4400 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4401 &facet->subfacets) {
4402 subfacet_destroy__(subfacet);
4404 hmap_remove(&ofproto->facets, &facet->hmap_node);
4405 list_remove(&facet->list_node);
4409 /* Feed information from 'facet' back into the learning table to keep it in
4410 * sync with what is actually flowing through the datapath. */
4412 facet_learn(struct facet *facet)
4414 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4415 struct action_xlate_ctx ctx;
4417 if (!facet->has_learn
4418 && !facet->has_normal
4419 && (!facet->has_fin_timeout
4420 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4424 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4425 facet->flow.vlan_tci,
4426 facet->rule, facet->tcp_flags, NULL);
4427 ctx.may_learn = true;
4428 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
4429 facet->rule->up.ofpacts_len);
4433 facet_account(struct facet *facet)
4435 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4436 struct subfacet *subfacet;
4437 const struct nlattr *a;
4442 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
4445 n_bytes = facet->byte_count - facet->accounted_bytes;
4447 /* This loop feeds byte counters to bond_account() for rebalancing to use
4448 * as a basis. We also need to track the actual VLAN on which the packet
4449 * is going to be sent to ensure that it matches the one passed to
4450 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4453 * We use the actions from an arbitrary subfacet because they should all
4454 * be equally valid for our purpose. */
4455 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
4456 struct subfacet, list_node);
4457 vlan_tci = facet->flow.vlan_tci;
4458 NL_ATTR_FOR_EACH_UNSAFE (a, left,
4459 subfacet->actions, subfacet->actions_len) {
4460 const struct ovs_action_push_vlan *vlan;
4461 struct ofport_dpif *port;
4463 switch (nl_attr_type(a)) {
4464 case OVS_ACTION_ATTR_OUTPUT:
4465 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4466 if (port && port->bundle && port->bundle->bond) {
4467 bond_account(port->bundle->bond, &facet->flow,
4468 vlan_tci_to_vid(vlan_tci), n_bytes);
4472 case OVS_ACTION_ATTR_POP_VLAN:
4473 vlan_tci = htons(0);
4476 case OVS_ACTION_ATTR_PUSH_VLAN:
4477 vlan = nl_attr_get(a);
4478 vlan_tci = vlan->vlan_tci;
4484 /* Returns true if the only action for 'facet' is to send to the controller.
4485 * (We don't report NetFlow expiration messages for such facets because they
4486 * are just part of the control logic for the network, not real traffic). */
4488 facet_is_controller_flow(struct facet *facet)
4491 const struct rule *rule = &facet->rule->up;
4492 const struct ofpact *ofpacts = rule->ofpacts;
4493 size_t ofpacts_len = rule->ofpacts_len;
4495 if (ofpacts_len > 0 &&
4496 ofpacts->type == OFPACT_CONTROLLER &&
4497 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4504 /* Folds all of 'facet''s statistics into its rule. Also updates the
4505 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4506 * 'facet''s statistics in the datapath should have been zeroed and folded into
4507 * its packet and byte counts before this function is called. */
4509 facet_flush_stats(struct facet *facet)
4511 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4512 struct subfacet *subfacet;
4514 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4515 ovs_assert(!subfacet->dp_byte_count);
4516 ovs_assert(!subfacet->dp_packet_count);
4519 facet_push_stats(facet);
4520 if (facet->accounted_bytes < facet->byte_count) {
4521 facet_account(facet);
4522 facet->accounted_bytes = facet->byte_count;
4525 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4526 struct ofexpired expired;
4527 expired.flow = facet->flow;
4528 expired.packet_count = facet->packet_count;
4529 expired.byte_count = facet->byte_count;
4530 expired.used = facet->used;
4531 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4534 facet->rule->packet_count += facet->packet_count;
4535 facet->rule->byte_count += facet->byte_count;
4537 /* Reset counters to prevent double counting if 'facet' ever gets
4539 facet_reset_counters(facet);
4541 netflow_flow_clear(&facet->nf_flow);
4542 facet->tcp_flags = 0;
4545 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4546 * Returns it if found, otherwise a null pointer.
4548 * 'hash' must be the return value of flow_hash(flow, 0).
4550 * The returned facet might need revalidation; use facet_lookup_valid()
4551 * instead if that is important. */
4552 static struct facet *
4553 facet_find(struct ofproto_dpif *ofproto,
4554 const struct flow *flow, uint32_t hash)
4556 struct facet *facet;
4558 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4559 if (flow_equal(flow, &facet->flow)) {
4567 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4568 * Returns it if found, otherwise a null pointer.
4570 * 'hash' must be the return value of flow_hash(flow, 0).
4572 * The returned facet is guaranteed to be valid. */
4573 static struct facet *
4574 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4577 struct facet *facet;
4579 facet = facet_find(ofproto, flow, hash);
4581 && (ofproto->backer->need_revalidate
4582 || tag_set_intersects(&ofproto->backer->revalidate_set,
4584 facet_revalidate(facet);
4586 /* facet_revalidate() may have destroyed 'facet'. */
4587 facet = facet_find(ofproto, flow, hash);
4594 subfacet_path_to_string(enum subfacet_path path)
4597 case SF_NOT_INSTALLED:
4598 return "not installed";
4600 return "in fast path";
4602 return "in slow path";
4608 /* Returns the path in which a subfacet should be installed if its 'slow'
4609 * member has the specified value. */
4610 static enum subfacet_path
4611 subfacet_want_path(enum slow_path_reason slow)
4613 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
4616 /* Returns true if 'subfacet' needs to have its datapath flow updated,
4617 * supposing that its actions have been recalculated as 'want_actions' and that
4618 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
4620 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
4621 const struct ofpbuf *want_actions)
4623 enum subfacet_path want_path = subfacet_want_path(slow);
4624 return (want_path != subfacet->path
4625 || (want_path == SF_FAST_PATH
4626 && (subfacet->actions_len != want_actions->size
4627 || memcmp(subfacet->actions, want_actions->data,
4628 subfacet->actions_len))));
4632 facet_check_consistency(struct facet *facet)
4634 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4636 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4638 uint64_t odp_actions_stub[1024 / 8];
4639 struct ofpbuf odp_actions;
4641 struct rule_dpif *rule;
4642 struct subfacet *subfacet;
4643 bool may_log = false;
4646 /* Check the rule for consistency. */
4647 rule = rule_dpif_lookup(ofproto, &facet->flow);
4648 ok = rule == facet->rule;
4650 may_log = !VLOG_DROP_WARN(&rl);
4655 flow_format(&s, &facet->flow);
4656 ds_put_format(&s, ": facet associated with wrong rule (was "
4657 "table=%"PRIu8",", facet->rule->up.table_id);
4658 cls_rule_format(&facet->rule->up.cr, &s);
4659 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4661 cls_rule_format(&rule->up.cr, &s);
4662 ds_put_char(&s, ')');
4664 VLOG_WARN("%s", ds_cstr(&s));
4669 /* Check the datapath actions for consistency. */
4670 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4671 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4672 enum subfacet_path want_path;
4673 struct action_xlate_ctx ctx;
4676 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4677 subfacet->initial_tci, rule, 0, NULL);
4678 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
4681 if (subfacet->path == SF_NOT_INSTALLED) {
4682 /* This only happens if the datapath reported an error when we
4683 * tried to install the flow. Don't flag another error here. */
4687 want_path = subfacet_want_path(subfacet->slow);
4688 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
4689 /* The actions for slow-path flows may legitimately vary from one
4690 * packet to the next. We're done. */
4694 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
4698 /* Inconsistency! */
4700 may_log = !VLOG_DROP_WARN(&rl);
4704 /* Rate-limited, skip reporting. */
4709 odp_flow_key_format(subfacet->key, subfacet->key_len, &s);
4711 ds_put_cstr(&s, ": inconsistency in subfacet");
4712 if (want_path != subfacet->path) {
4713 enum odp_key_fitness fitness = subfacet->key_fitness;
4715 ds_put_format(&s, " (%s, fitness=%s)",
4716 subfacet_path_to_string(subfacet->path),
4717 odp_key_fitness_to_string(fitness));
4718 ds_put_format(&s, " (should have been %s)",
4719 subfacet_path_to_string(want_path));
4720 } else if (want_path == SF_FAST_PATH) {
4721 ds_put_cstr(&s, " (actions were: ");
4722 format_odp_actions(&s, subfacet->actions,
4723 subfacet->actions_len);
4724 ds_put_cstr(&s, ") (correct actions: ");
4725 format_odp_actions(&s, odp_actions.data, odp_actions.size);
4726 ds_put_char(&s, ')');
4728 ds_put_cstr(&s, " (actions: ");
4729 format_odp_actions(&s, subfacet->actions,
4730 subfacet->actions_len);
4731 ds_put_char(&s, ')');
4733 VLOG_WARN("%s", ds_cstr(&s));
4736 ofpbuf_uninit(&odp_actions);
4741 /* Re-searches the classifier for 'facet':
4743 * - If the rule found is different from 'facet''s current rule, moves
4744 * 'facet' to the new rule and recompiles its actions.
4746 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4747 * where it is and recompiles its actions anyway.
4749 * - If any of 'facet''s subfacets correspond to a new flow according to
4750 * ofproto_receive(), 'facet' is removed. */
4752 facet_revalidate(struct facet *facet)
4754 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4756 struct nlattr *odp_actions;
4759 struct actions *new_actions;
4761 struct action_xlate_ctx ctx;
4762 uint64_t odp_actions_stub[1024 / 8];
4763 struct ofpbuf odp_actions;
4765 struct rule_dpif *new_rule;
4766 struct subfacet *subfacet;
4769 COVERAGE_INC(facet_revalidate);
4771 /* Check that child subfacets still correspond to this facet. Tunnel
4772 * configuration changes could cause a subfacet's OpenFlow in_port to
4774 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4775 struct ofproto_dpif *recv_ofproto;
4776 struct flow recv_flow;
4779 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
4780 subfacet->key_len, &recv_flow, NULL,
4781 &recv_ofproto, NULL, NULL);
4783 || recv_ofproto != ofproto
4784 || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) {
4785 facet_remove(facet);
4790 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
4792 /* Calculate new datapath actions.
4794 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4795 * emit a NetFlow expiration and, if so, we need to have the old state
4796 * around to properly compose it. */
4798 /* If the datapath actions changed or the installability changed,
4799 * then we need to talk to the datapath. */
4802 memset(&ctx, 0, sizeof ctx);
4803 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4804 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4805 enum slow_path_reason slow;
4807 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4808 subfacet->initial_tci, new_rule, 0, NULL);
4809 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
4812 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4813 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
4814 struct dpif_flow_stats stats;
4816 subfacet_install(subfacet,
4817 odp_actions.data, odp_actions.size, &stats, slow);
4818 subfacet_update_stats(subfacet, &stats);
4821 new_actions = xcalloc(list_size(&facet->subfacets),
4822 sizeof *new_actions);
4824 new_actions[i].odp_actions = xmemdup(odp_actions.data,
4826 new_actions[i].actions_len = odp_actions.size;
4831 ofpbuf_uninit(&odp_actions);
4834 facet_flush_stats(facet);
4837 /* Update 'facet' now that we've taken care of all the old state. */
4838 facet->tags = ctx.tags;
4839 facet->nf_flow.output_iface = ctx.nf_output_iface;
4840 facet->has_learn = ctx.has_learn;
4841 facet->has_normal = ctx.has_normal;
4842 facet->has_fin_timeout = ctx.has_fin_timeout;
4843 facet->mirrors = ctx.mirrors;
4846 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4847 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4849 if (new_actions && new_actions[i].odp_actions) {
4850 free(subfacet->actions);
4851 subfacet->actions = new_actions[i].odp_actions;
4852 subfacet->actions_len = new_actions[i].actions_len;
4858 if (facet->rule != new_rule) {
4859 COVERAGE_INC(facet_changed_rule);
4860 list_remove(&facet->list_node);
4861 list_push_back(&new_rule->facets, &facet->list_node);
4862 facet->rule = new_rule;
4863 facet->used = new_rule->up.created;
4864 facet->prev_used = facet->used;
4868 /* Updates 'facet''s used time. Caller is responsible for calling
4869 * facet_push_stats() to update the flows which 'facet' resubmits into. */
4871 facet_update_time(struct facet *facet, long long int used)
4873 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4874 if (used > facet->used) {
4876 ofproto_rule_update_used(&facet->rule->up, used);
4877 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
4882 facet_reset_counters(struct facet *facet)
4884 facet->packet_count = 0;
4885 facet->byte_count = 0;
4886 facet->prev_packet_count = 0;
4887 facet->prev_byte_count = 0;
4888 facet->accounted_bytes = 0;
4892 facet_push_stats(struct facet *facet)
4894 struct dpif_flow_stats stats;
4896 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4897 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4898 ovs_assert(facet->used >= facet->prev_used);
4900 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4901 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4902 stats.used = facet->used;
4903 stats.tcp_flags = 0;
4905 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
4906 facet->prev_packet_count = facet->packet_count;
4907 facet->prev_byte_count = facet->byte_count;
4908 facet->prev_used = facet->used;
4910 flow_push_stats(facet->rule, &facet->flow, &stats);
4912 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
4913 facet->mirrors, stats.n_packets, stats.n_bytes);
4918 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4920 rule->packet_count += stats->n_packets;
4921 rule->byte_count += stats->n_bytes;
4922 ofproto_rule_update_used(&rule->up, stats->used);
4925 /* Pushes flow statistics to the rules which 'flow' resubmits into given
4926 * 'rule''s actions and mirrors. */
4928 flow_push_stats(struct rule_dpif *rule,
4929 const struct flow *flow, const struct dpif_flow_stats *stats)
4931 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4932 struct action_xlate_ctx ctx;
4934 ofproto_rule_update_used(&rule->up, stats->used);
4936 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule,
4938 ctx.resubmit_stats = stats;
4939 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
4940 rule->up.ofpacts_len);
4945 static struct subfacet *
4946 subfacet_find(struct ofproto_dpif *ofproto,
4947 const struct nlattr *key, size_t key_len, uint32_t key_hash)
4949 struct subfacet *subfacet;
4951 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4952 &ofproto->subfacets) {
4953 if (subfacet->key_len == key_len
4954 && !memcmp(key, subfacet->key, key_len)) {
4962 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4963 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4964 * existing subfacet if there is one, otherwise creates and returns a
4967 * If the returned subfacet is new, then subfacet->actions will be NULL, in
4968 * which case the caller must populate the actions with
4969 * subfacet_make_actions(). */
4970 static struct subfacet *
4971 subfacet_create(struct facet *facet, struct flow_miss *miss,
4974 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4975 enum odp_key_fitness key_fitness = miss->key_fitness;
4976 const struct nlattr *key = miss->key;
4977 size_t key_len = miss->key_len;
4979 struct subfacet *subfacet;
4981 key_hash = odp_flow_key_hash(key, key_len);
4983 if (list_is_empty(&facet->subfacets)) {
4984 subfacet = &facet->one_subfacet;
4986 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4988 if (subfacet->facet == facet) {
4992 /* This shouldn't happen. */
4993 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4994 subfacet_destroy(subfacet);
4997 subfacet = xmalloc(sizeof *subfacet);
5000 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
5001 list_push_back(&facet->subfacets, &subfacet->list_node);
5002 subfacet->facet = facet;
5003 subfacet->key_fitness = key_fitness;
5004 subfacet->key = xmemdup(key, key_len);
5005 subfacet->key_len = key_len;
5006 subfacet->used = now;
5007 subfacet->dp_packet_count = 0;
5008 subfacet->dp_byte_count = 0;
5009 subfacet->actions_len = 0;
5010 subfacet->actions = NULL;
5011 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
5014 subfacet->path = SF_NOT_INSTALLED;
5015 subfacet->initial_tci = miss->initial_tci;
5016 subfacet->odp_in_port = miss->odp_in_port;
5021 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5022 * its facet within 'ofproto', and frees it. */
5024 subfacet_destroy__(struct subfacet *subfacet)
5026 struct facet *facet = subfacet->facet;
5027 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5029 subfacet_uninstall(subfacet);
5030 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5031 list_remove(&subfacet->list_node);
5032 free(subfacet->key);
5033 free(subfacet->actions);
5034 if (subfacet != &facet->one_subfacet) {
5039 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5040 * last remaining subfacet in its facet destroys the facet too. */
5042 subfacet_destroy(struct subfacet *subfacet)
5044 struct facet *facet = subfacet->facet;
5046 if (list_is_singleton(&facet->subfacets)) {
5047 /* facet_remove() needs at least one subfacet (it will remove it). */
5048 facet_remove(facet);
5050 subfacet_destroy__(subfacet);
5055 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5056 struct subfacet **subfacets, int n)
5058 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5059 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5060 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5063 for (i = 0; i < n; i++) {
5064 ops[i].type = DPIF_OP_FLOW_DEL;
5065 ops[i].u.flow_del.key = subfacets[i]->key;
5066 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5067 ops[i].u.flow_del.stats = &stats[i];
5071 dpif_operate(ofproto->backer->dpif, opsp, n);
5072 for (i = 0; i < n; i++) {
5073 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5074 subfacets[i]->path = SF_NOT_INSTALLED;
5075 subfacet_destroy(subfacets[i]);
5079 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
5080 * Translates the actions into 'odp_actions', which the caller must have
5081 * initialized and is responsible for uninitializing. */
5083 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
5084 struct ofpbuf *odp_actions)
5086 struct facet *facet = subfacet->facet;
5087 struct rule_dpif *rule = facet->rule;
5088 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5090 struct action_xlate_ctx ctx;
5092 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
5094 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
5095 facet->tags = ctx.tags;
5096 facet->has_learn = ctx.has_learn;
5097 facet->has_normal = ctx.has_normal;
5098 facet->has_fin_timeout = ctx.has_fin_timeout;
5099 facet->nf_flow.output_iface = ctx.nf_output_iface;
5100 facet->mirrors = ctx.mirrors;
5102 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
5103 if (subfacet->actions_len != odp_actions->size
5104 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
5105 free(subfacet->actions);
5106 subfacet->actions_len = odp_actions->size;
5107 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
5111 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5112 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5113 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5114 * since 'subfacet' was last updated.
5116 * Returns 0 if successful, otherwise a positive errno value. */
5118 subfacet_install(struct subfacet *subfacet,
5119 const struct nlattr *actions, size_t actions_len,
5120 struct dpif_flow_stats *stats,
5121 enum slow_path_reason slow)
5123 struct facet *facet = subfacet->facet;
5124 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5125 enum subfacet_path path = subfacet_want_path(slow);
5126 uint64_t slow_path_stub[128 / 8];
5127 enum dpif_flow_put_flags flags;
5130 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5132 flags |= DPIF_FP_ZERO_STATS;
5135 if (path == SF_SLOW_PATH) {
5136 compose_slow_path(ofproto, &facet->flow, slow,
5137 slow_path_stub, sizeof slow_path_stub,
5138 &actions, &actions_len);
5141 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5142 subfacet->key_len, actions, actions_len, stats);
5145 subfacet_reset_dp_stats(subfacet, stats);
5149 subfacet->path = path;
5155 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
5157 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
5158 stats, subfacet->slow);
5161 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5163 subfacet_uninstall(struct subfacet *subfacet)
5165 if (subfacet->path != SF_NOT_INSTALLED) {
5166 struct rule_dpif *rule = subfacet->facet->rule;
5167 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5168 struct dpif_flow_stats stats;
5171 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5172 subfacet->key_len, &stats);
5173 subfacet_reset_dp_stats(subfacet, &stats);
5175 subfacet_update_stats(subfacet, &stats);
5177 subfacet->path = SF_NOT_INSTALLED;
5179 ovs_assert(subfacet->dp_packet_count == 0);
5180 ovs_assert(subfacet->dp_byte_count == 0);
5184 /* Resets 'subfacet''s datapath statistics counters. This should be called
5185 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5186 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5187 * was reset in the datapath. 'stats' will be modified to include only
5188 * statistics new since 'subfacet' was last updated. */
5190 subfacet_reset_dp_stats(struct subfacet *subfacet,
5191 struct dpif_flow_stats *stats)
5194 && subfacet->dp_packet_count <= stats->n_packets
5195 && subfacet->dp_byte_count <= stats->n_bytes) {
5196 stats->n_packets -= subfacet->dp_packet_count;
5197 stats->n_bytes -= subfacet->dp_byte_count;
5200 subfacet->dp_packet_count = 0;
5201 subfacet->dp_byte_count = 0;
5204 /* Updates 'subfacet''s used time. The caller is responsible for calling
5205 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
5207 subfacet_update_time(struct subfacet *subfacet, long long int used)
5209 if (used > subfacet->used) {
5210 subfacet->used = used;
5211 facet_update_time(subfacet->facet, used);
5215 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5217 * Because of the meaning of a subfacet's counters, it only makes sense to do
5218 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5219 * represents a packet that was sent by hand or if it represents statistics
5220 * that have been cleared out of the datapath. */
5222 subfacet_update_stats(struct subfacet *subfacet,
5223 const struct dpif_flow_stats *stats)
5225 if (stats->n_packets || stats->used > subfacet->used) {
5226 struct facet *facet = subfacet->facet;
5228 subfacet_update_time(subfacet, stats->used);
5229 facet->packet_count += stats->n_packets;
5230 facet->byte_count += stats->n_bytes;
5231 facet->tcp_flags |= stats->tcp_flags;
5232 facet_push_stats(facet);
5233 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
5239 static struct rule_dpif *
5240 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5242 struct rule_dpif *rule;
5244 rule = rule_dpif_lookup__(ofproto, flow, 0);
5249 return rule_dpif_miss_rule(ofproto, flow);
5252 static struct rule_dpif *
5253 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5256 struct cls_rule *cls_rule;
5257 struct classifier *cls;
5259 if (table_id >= N_TABLES) {
5263 cls = &ofproto->up.tables[table_id].cls;
5264 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5265 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5266 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5267 * are unavailable. */
5268 struct flow ofpc_normal_flow = *flow;
5269 ofpc_normal_flow.tp_src = htons(0);
5270 ofpc_normal_flow.tp_dst = htons(0);
5271 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5273 cls_rule = classifier_lookup(cls, flow);
5275 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5278 static struct rule_dpif *
5279 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5281 struct ofport_dpif *port;
5283 port = get_ofp_port(ofproto, flow->in_port);
5285 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5286 return ofproto->miss_rule;
5289 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5290 return ofproto->no_packet_in_rule;
5292 return ofproto->miss_rule;
5296 complete_operation(struct rule_dpif *rule)
5298 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5300 rule_invalidate(rule);
5302 struct dpif_completion *c = xmalloc(sizeof *c);
5303 c->op = rule->up.pending;
5304 list_push_back(&ofproto->completions, &c->list_node);
5306 ofoperation_complete(rule->up.pending, 0);
5310 static struct rule *
5313 struct rule_dpif *rule = xmalloc(sizeof *rule);
5318 rule_dealloc(struct rule *rule_)
5320 struct rule_dpif *rule = rule_dpif_cast(rule_);
5325 rule_construct(struct rule *rule_)
5327 struct rule_dpif *rule = rule_dpif_cast(rule_);
5328 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5329 struct rule_dpif *victim;
5332 rule->packet_count = 0;
5333 rule->byte_count = 0;
5335 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5336 if (victim && !list_is_empty(&victim->facets)) {
5337 struct facet *facet;
5339 rule->facets = victim->facets;
5340 list_moved(&rule->facets);
5341 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5342 /* XXX: We're only clearing our local counters here. It's possible
5343 * that quite a few packets are unaccounted for in the datapath
5344 * statistics. These will be accounted to the new rule instead of
5345 * cleared as required. This could be fixed by clearing out the
5346 * datapath statistics for this facet, but currently it doesn't
5348 facet_reset_counters(facet);
5352 /* Must avoid list_moved() in this case. */
5353 list_init(&rule->facets);
5356 table_id = rule->up.table_id;
5358 rule->tag = victim->tag;
5359 } else if (table_id == 0) {
5364 miniflow_expand(&rule->up.cr.match.flow, &flow);
5365 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5366 ofproto->tables[table_id].basis);
5369 complete_operation(rule);
5374 rule_destruct(struct rule *rule_)
5376 struct rule_dpif *rule = rule_dpif_cast(rule_);
5377 struct facet *facet, *next_facet;
5379 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5380 facet_revalidate(facet);
5383 complete_operation(rule);
5387 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5389 struct rule_dpif *rule = rule_dpif_cast(rule_);
5390 struct facet *facet;
5392 /* Start from historical data for 'rule' itself that are no longer tracked
5393 * in facets. This counts, for example, facets that have expired. */
5394 *packets = rule->packet_count;
5395 *bytes = rule->byte_count;
5397 /* Add any statistics that are tracked by facets. This includes
5398 * statistical data recently updated by ofproto_update_stats() as well as
5399 * stats for packets that were executed "by hand" via dpif_execute(). */
5400 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5401 *packets += facet->packet_count;
5402 *bytes += facet->byte_count;
5407 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5408 struct ofpbuf *packet)
5410 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5412 struct dpif_flow_stats stats;
5414 struct action_xlate_ctx ctx;
5415 uint64_t odp_actions_stub[1024 / 8];
5416 struct ofpbuf odp_actions;
5418 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5419 rule_credit_stats(rule, &stats);
5421 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5422 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
5423 rule, stats.tcp_flags, packet);
5424 ctx.resubmit_stats = &stats;
5425 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
5427 execute_odp_actions(ofproto, flow, odp_actions.data,
5428 odp_actions.size, packet);
5430 ofpbuf_uninit(&odp_actions);
5434 rule_execute(struct rule *rule, const struct flow *flow,
5435 struct ofpbuf *packet)
5437 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5438 ofpbuf_delete(packet);
5443 rule_modify_actions(struct rule *rule_)
5445 struct rule_dpif *rule = rule_dpif_cast(rule_);
5447 complete_operation(rule);
5450 /* Sends 'packet' out 'ofport'.
5451 * May modify 'packet'.
5452 * Returns 0 if successful, otherwise a positive errno value. */
5454 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5456 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5457 uint64_t odp_actions_stub[1024 / 8];
5458 struct ofpbuf key, odp_actions;
5459 struct odputil_keybuf keybuf;
5464 flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow);
5465 if (netdev_vport_is_patch(ofport->up.netdev)) {
5466 struct ofproto_dpif *peer_ofproto;
5467 struct dpif_flow_stats stats;
5468 struct ofport_dpif *peer;
5469 struct rule_dpif *rule;
5471 peer = ofport_get_peer(ofport);
5476 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5477 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5478 netdev_vport_inc_rx(peer->up.netdev, &stats);
5480 flow.in_port = peer->up.ofp_port;
5481 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5482 rule = rule_dpif_lookup(peer_ofproto, &flow);
5483 rule_dpif_execute(rule, &flow, packet);
5488 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5490 if (ofport->tnl_port) {
5491 struct dpif_flow_stats stats;
5493 odp_port = tnl_port_send(ofport->tnl_port, &flow);
5494 if (odp_port == OVSP_NONE) {
5498 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5499 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5500 odp_put_tunnel_action(&flow.tunnel, &odp_actions);
5501 odp_put_skb_mark_action(flow.skb_mark, &odp_actions);
5503 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5505 if (odp_port != ofport->odp_port) {
5506 eth_pop_vlan(packet);
5507 flow.vlan_tci = htons(0);
5511 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5512 odp_flow_key_from_flow(&key, &flow,
5513 ofp_port_to_odp_port(ofproto, flow.in_port));
5515 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5517 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5518 error = dpif_execute(ofproto->backer->dpif,
5520 odp_actions.data, odp_actions.size,
5522 ofpbuf_uninit(&odp_actions);
5525 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5526 ofproto->up.name, odp_port, strerror(error));
5528 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5532 /* OpenFlow to datapath action translation. */
5534 static bool may_receive(const struct ofport_dpif *, struct action_xlate_ctx *);
5535 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5536 struct action_xlate_ctx *);
5537 static void xlate_normal(struct action_xlate_ctx *);
5539 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5540 * The action will state 'slow' as the reason that the action is in the slow
5541 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5542 * dump-flows" output to see why a flow is in the slow path.)
5544 * The 'stub_size' bytes in 'stub' will be used to store the action.
5545 * 'stub_size' must be large enough for the action.
5547 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5550 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5551 enum slow_path_reason slow,
5552 uint64_t *stub, size_t stub_size,
5553 const struct nlattr **actionsp, size_t *actions_lenp)
5555 union user_action_cookie cookie;
5558 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5559 cookie.slow_path.unused = 0;
5560 cookie.slow_path.reason = slow;
5562 ofpbuf_use_stack(&buf, stub, stub_size);
5563 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
5564 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5565 odp_put_userspace_action(pid, &cookie, sizeof cookie, &buf);
5567 put_userspace_action(ofproto, &buf, flow, &cookie);
5569 *actionsp = buf.data;
5570 *actions_lenp = buf.size;
5574 put_userspace_action(const struct ofproto_dpif *ofproto,
5575 struct ofpbuf *odp_actions,
5576 const struct flow *flow,
5577 const union user_action_cookie *cookie)
5581 pid = dpif_port_get_pid(ofproto->backer->dpif,
5582 ofp_port_to_odp_port(ofproto, flow->in_port));
5584 return odp_put_userspace_action(pid, cookie, sizeof *cookie, odp_actions);
5588 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5589 ovs_be16 vlan_tci, uint32_t odp_port,
5590 unsigned int n_outputs, union user_action_cookie *cookie)
5594 cookie->type = USER_ACTION_COOKIE_SFLOW;
5595 cookie->sflow.vlan_tci = vlan_tci;
5597 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5598 * port information") for the interpretation of cookie->output. */
5599 switch (n_outputs) {
5601 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5602 cookie->sflow.output = 0x40000000 | 256;
5606 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5608 cookie->sflow.output = ifindex;
5613 /* 0x80000000 means "multiple output ports. */
5614 cookie->sflow.output = 0x80000000 | n_outputs;
5619 /* Compose SAMPLE action for sFlow. */
5621 compose_sflow_action(const struct ofproto_dpif *ofproto,
5622 struct ofpbuf *odp_actions,
5623 const struct flow *flow,
5626 uint32_t probability;
5627 union user_action_cookie cookie;
5628 size_t sample_offset, actions_offset;
5631 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5635 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5637 /* Number of packets out of UINT_MAX to sample. */
5638 probability = dpif_sflow_get_probability(ofproto->sflow);
5639 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5641 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5642 compose_sflow_cookie(ofproto, htons(0), odp_port,
5643 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5644 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
5646 nl_msg_end_nested(odp_actions, actions_offset);
5647 nl_msg_end_nested(odp_actions, sample_offset);
5648 return cookie_offset;
5651 /* SAMPLE action must be first action in any given list of actions.
5652 * At this point we do not have all information required to build it. So try to
5653 * build sample action as complete as possible. */
5655 add_sflow_action(struct action_xlate_ctx *ctx)
5657 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5659 &ctx->flow, OVSP_NONE);
5660 ctx->sflow_odp_port = 0;
5661 ctx->sflow_n_outputs = 0;
5664 /* Fix SAMPLE action according to data collected while composing ODP actions.
5665 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5666 * USERSPACE action's user-cookie which is required for sflow. */
5668 fix_sflow_action(struct action_xlate_ctx *ctx)
5670 const struct flow *base = &ctx->base_flow;
5671 union user_action_cookie *cookie;
5673 if (!ctx->user_cookie_offset) {
5677 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
5679 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5681 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5682 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5686 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
5689 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5690 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
5691 ovs_be64 flow_tun_id = ctx->flow.tunnel.tun_id;
5692 uint8_t flow_nw_tos = ctx->flow.nw_tos;
5693 struct priority_to_dscp *pdscp;
5694 uint32_t out_port, odp_port;
5696 /* If 'struct flow' gets additional metadata, we'll need to zero it out
5697 * before traversing a patch port. */
5698 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 19);
5701 xlate_report(ctx, "Nonexistent output port");
5703 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5704 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5706 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5707 xlate_report(ctx, "STP not in forwarding state, skipping output");
5711 if (netdev_vport_is_patch(ofport->up.netdev)) {
5712 struct ofport_dpif *peer = ofport_get_peer(ofport);
5713 struct flow old_flow = ctx->flow;
5714 const struct ofproto_dpif *peer_ofproto;
5715 enum slow_path_reason special;
5716 struct ofport_dpif *in_port;
5719 xlate_report(ctx, "Nonexistent patch port peer");
5723 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5724 if (peer_ofproto->backer != ctx->ofproto->backer) {
5725 xlate_report(ctx, "Patch port peer on a different datapath");
5729 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
5730 ctx->flow.in_port = peer->up.ofp_port;
5731 ctx->flow.metadata = htonll(0);
5732 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
5733 memset(ctx->flow.regs, 0, sizeof ctx->flow.regs);
5735 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5736 special = process_special(ctx->ofproto, &ctx->flow, in_port,
5739 ctx->slow |= special;
5740 } else if (!in_port || may_receive(in_port, ctx)) {
5741 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
5742 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5744 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
5745 * learning action look at the packet, then drop it. */
5746 struct flow old_base_flow = ctx->base_flow;
5747 size_t old_size = ctx->odp_actions->size;
5748 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5749 ctx->base_flow = old_base_flow;
5750 ctx->odp_actions->size = old_size;
5754 ctx->flow = old_flow;
5755 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5757 if (ctx->resubmit_stats) {
5758 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
5759 netdev_vport_inc_rx(peer->up.netdev, ctx->resubmit_stats);
5765 pdscp = get_priority(ofport, ctx->flow.skb_priority);
5767 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5768 ctx->flow.nw_tos |= pdscp->dscp;
5771 odp_port = ofp_port_to_odp_port(ctx->ofproto, ofp_port);
5772 if (ofport->tnl_port) {
5773 odp_port = tnl_port_send(ofport->tnl_port, &ctx->flow);
5774 if (odp_port == OVSP_NONE) {
5775 xlate_report(ctx, "Tunneling decided against output");
5779 if (ctx->resubmit_stats) {
5780 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
5782 out_port = odp_port;
5783 commit_odp_tunnel_action(&ctx->flow, &ctx->base_flow,
5786 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
5787 ctx->flow.vlan_tci);
5788 if (out_port != odp_port) {
5789 ctx->flow.vlan_tci = htons(0);
5792 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
5793 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
5795 ctx->sflow_odp_port = odp_port;
5796 ctx->sflow_n_outputs++;
5797 ctx->nf_output_iface = ofp_port;
5798 ctx->flow.tunnel.tun_id = flow_tun_id;
5799 ctx->flow.vlan_tci = flow_vlan_tci;
5800 ctx->flow.nw_tos = flow_nw_tos;
5804 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
5806 compose_output_action__(ctx, ofp_port, true);
5810 xlate_table_action(struct action_xlate_ctx *ctx,
5811 uint16_t in_port, uint8_t table_id, bool may_packet_in)
5813 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
5814 struct ofproto_dpif *ofproto = ctx->ofproto;
5815 struct rule_dpif *rule;
5816 uint16_t old_in_port;
5817 uint8_t old_table_id;
5819 old_table_id = ctx->table_id;
5820 ctx->table_id = table_id;
5822 /* Look up a flow with 'in_port' as the input port. */
5823 old_in_port = ctx->flow.in_port;
5824 ctx->flow.in_port = in_port;
5825 rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id);
5828 if (table_id > 0 && table_id < N_TABLES) {
5829 struct table_dpif *table = &ofproto->tables[table_id];
5830 if (table->other_table) {
5831 ctx->tags |= (rule && rule->tag
5833 : rule_calculate_tag(&ctx->flow,
5834 &table->other_table->mask,
5839 /* Restore the original input port. Otherwise OFPP_NORMAL and
5840 * OFPP_IN_PORT will have surprising behavior. */
5841 ctx->flow.in_port = old_in_port;
5843 if (ctx->resubmit_hook) {
5844 ctx->resubmit_hook(ctx, rule);
5847 if (rule == NULL && may_packet_in) {
5849 * check if table configuration flags
5850 * OFPTC_TABLE_MISS_CONTROLLER, default.
5851 * OFPTC_TABLE_MISS_CONTINUE,
5852 * OFPTC_TABLE_MISS_DROP
5853 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
5855 rule = rule_dpif_miss_rule(ofproto, &ctx->flow);
5859 struct rule_dpif *old_rule = ctx->rule;
5861 if (ctx->resubmit_stats) {
5862 rule_credit_stats(rule, ctx->resubmit_stats);
5867 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
5868 ctx->rule = old_rule;
5872 ctx->table_id = old_table_id;
5874 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5876 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
5877 MAX_RESUBMIT_RECURSION);
5878 ctx->max_resubmit_trigger = true;
5883 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
5884 const struct ofpact_resubmit *resubmit)
5889 in_port = resubmit->in_port;
5890 if (in_port == OFPP_IN_PORT) {
5891 in_port = ctx->flow.in_port;
5894 table_id = resubmit->table_id;
5895 if (table_id == 255) {
5896 table_id = ctx->table_id;
5899 xlate_table_action(ctx, in_port, table_id, false);
5903 flood_packets(struct action_xlate_ctx *ctx, bool all)
5905 struct ofport_dpif *ofport;
5907 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
5908 uint16_t ofp_port = ofport->up.ofp_port;
5910 if (ofp_port == ctx->flow.in_port) {
5915 compose_output_action__(ctx, ofp_port, false);
5916 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
5917 compose_output_action(ctx, ofp_port);
5921 ctx->nf_output_iface = NF_OUT_FLOOD;
5925 execute_controller_action(struct action_xlate_ctx *ctx, int len,
5926 enum ofp_packet_in_reason reason,
5927 uint16_t controller_id)
5929 struct ofputil_packet_in pin;
5930 struct ofpbuf *packet;
5932 ctx->slow |= SLOW_CONTROLLER;
5937 packet = ofpbuf_clone(ctx->packet);
5939 if (packet->l2 && packet->l3) {
5940 struct eth_header *eh;
5941 uint16_t mpls_depth;
5943 eth_pop_vlan(packet);
5946 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
5947 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
5949 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
5950 eth_push_vlan(packet, ctx->flow.vlan_tci);
5953 mpls_depth = eth_mpls_depth(packet);
5955 if (mpls_depth < ctx->flow.mpls_depth) {
5956 push_mpls(packet, ctx->flow.dl_type, ctx->flow.mpls_lse);
5957 } else if (mpls_depth > ctx->flow.mpls_depth) {
5958 pop_mpls(packet, ctx->flow.dl_type);
5959 } else if (mpls_depth) {
5960 set_mpls_lse(packet, ctx->flow.mpls_lse);
5964 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5965 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
5966 ctx->flow.nw_tos, ctx->flow.nw_ttl);
5970 if (ctx->flow.nw_proto == IPPROTO_TCP) {
5971 packet_set_tcp_port(packet, ctx->flow.tp_src,
5973 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
5974 packet_set_udp_port(packet, ctx->flow.tp_src,
5981 pin.packet = packet->data;
5982 pin.packet_len = packet->size;
5983 pin.reason = reason;
5984 pin.controller_id = controller_id;
5985 pin.table_id = ctx->table_id;
5986 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
5989 flow_get_metadata(&ctx->flow, &pin.fmd);
5991 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
5992 ofpbuf_delete(packet);
5996 execute_mpls_push_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
5998 ovs_assert(eth_type_mpls(eth_type));
6000 if (ctx->base_flow.mpls_depth) {
6001 ctx->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK);
6002 ctx->flow.mpls_depth++;
6007 if (ctx->flow.dl_type == htons(ETH_TYPE_IPV6)) {
6008 label = htonl(0x2); /* IPV6 Explicit Null. */
6010 label = htonl(0x0); /* IPV4 Explicit Null. */
6012 tc = (ctx->flow.nw_tos & IP_DSCP_MASK) >> 2;
6013 ttl = ctx->flow.nw_ttl ? ctx->flow.nw_ttl : 0x40;
6014 ctx->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
6015 ctx->flow.encap_dl_type = ctx->flow.dl_type;
6016 ctx->flow.mpls_depth = 1;
6018 ctx->flow.dl_type = eth_type;
6022 execute_mpls_pop_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
6024 ovs_assert(eth_type_mpls(ctx->flow.dl_type));
6025 ovs_assert(!eth_type_mpls(eth_type));
6027 if (ctx->flow.mpls_depth) {
6028 ctx->flow.mpls_depth--;
6029 ctx->flow.mpls_lse = htonl(0);
6030 if (!ctx->flow.mpls_depth) {
6031 ctx->flow.dl_type = eth_type;
6032 ctx->flow.encap_dl_type = htons(0);
6038 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6040 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
6041 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6045 if (ctx->flow.nw_ttl > 1) {
6051 for (i = 0; i < ids->n_controllers; i++) {
6052 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6056 /* Stop processing for current table. */
6062 execute_dec_mpls_ttl_action(struct action_xlate_ctx *ctx)
6064 uint8_t ttl = mpls_lse_to_ttl(ctx->flow.mpls_lse);
6066 if (!eth_type_mpls(ctx->flow.dl_type)) {
6072 set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl);
6075 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
6077 /* Stop processing for current table. */
6083 xlate_output_action(struct action_xlate_ctx *ctx,
6084 uint16_t port, uint16_t max_len, bool may_packet_in)
6086 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
6088 ctx->nf_output_iface = NF_OUT_DROP;
6092 compose_output_action(ctx, ctx->flow.in_port);
6095 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
6101 flood_packets(ctx, false);
6104 flood_packets(ctx, true);
6106 case OFPP_CONTROLLER:
6107 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6113 if (port != ctx->flow.in_port) {
6114 compose_output_action(ctx, port);
6116 xlate_report(ctx, "skipping output to input port");
6121 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6122 ctx->nf_output_iface = NF_OUT_FLOOD;
6123 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
6124 ctx->nf_output_iface = prev_nf_output_iface;
6125 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6126 ctx->nf_output_iface != NF_OUT_FLOOD) {
6127 ctx->nf_output_iface = NF_OUT_MULTI;
6132 xlate_output_reg_action(struct action_xlate_ctx *ctx,
6133 const struct ofpact_output_reg *or)
6135 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
6136 if (port <= UINT16_MAX) {
6137 xlate_output_action(ctx, port, or->max_len, false);
6142 xlate_enqueue_action(struct action_xlate_ctx *ctx,
6143 const struct ofpact_enqueue *enqueue)
6145 uint16_t ofp_port = enqueue->port;
6146 uint32_t queue_id = enqueue->queue;
6147 uint32_t flow_priority, priority;
6150 /* Translate queue to priority. */
6151 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6152 queue_id, &priority);
6154 /* Fall back to ordinary output action. */
6155 xlate_output_action(ctx, enqueue->port, 0, false);
6159 /* Check output port. */
6160 if (ofp_port == OFPP_IN_PORT) {
6161 ofp_port = ctx->flow.in_port;
6162 } else if (ofp_port == ctx->flow.in_port) {
6166 /* Add datapath actions. */
6167 flow_priority = ctx->flow.skb_priority;
6168 ctx->flow.skb_priority = priority;
6169 compose_output_action(ctx, ofp_port);
6170 ctx->flow.skb_priority = flow_priority;
6172 /* Update NetFlow output port. */
6173 if (ctx->nf_output_iface == NF_OUT_DROP) {
6174 ctx->nf_output_iface = ofp_port;
6175 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
6176 ctx->nf_output_iface = NF_OUT_MULTI;
6181 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
6183 uint32_t skb_priority;
6185 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6186 queue_id, &skb_priority)) {
6187 ctx->flow.skb_priority = skb_priority;
6189 /* Couldn't translate queue to a priority. Nothing to do. A warning
6190 * has already been logged. */
6194 struct xlate_reg_state {
6200 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6202 struct ofproto_dpif *ofproto = ofproto_;
6203 struct ofport_dpif *port;
6213 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6216 port = get_ofp_port(ofproto, ofp_port);
6217 return port ? port->may_enable : false;
6222 xlate_bundle_action(struct action_xlate_ctx *ctx,
6223 const struct ofpact_bundle *bundle)
6227 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
6228 if (bundle->dst.field) {
6229 nxm_reg_load(&bundle->dst, port, &ctx->flow);
6231 xlate_output_action(ctx, port, 0, false);
6236 xlate_learn_action(struct action_xlate_ctx *ctx,
6237 const struct ofpact_learn *learn)
6239 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6240 struct ofputil_flow_mod fm;
6241 uint64_t ofpacts_stub[1024 / 8];
6242 struct ofpbuf ofpacts;
6245 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6246 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
6248 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6249 if (error && !VLOG_DROP_WARN(&rl)) {
6250 VLOG_WARN("learning action failed to modify flow table (%s)",
6251 ofperr_get_name(error));
6254 ofpbuf_uninit(&ofpacts);
6257 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6258 * means "infinite". */
6260 reduce_timeout(uint16_t max, uint16_t *timeout)
6262 if (max && (!*timeout || *timeout > max)) {
6268 xlate_fin_timeout(struct action_xlate_ctx *ctx,
6269 const struct ofpact_fin_timeout *oft)
6271 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6272 struct rule_dpif *rule = ctx->rule;
6274 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6275 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6280 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
6282 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
6283 ? OFPUTIL_PC_NO_RECV_STP
6284 : OFPUTIL_PC_NO_RECV)) {
6288 /* Only drop packets here if both forwarding and learning are
6289 * disabled. If just learning is enabled, we need to have
6290 * OFPP_NORMAL and the learning action have a look at the packet
6291 * before we can drop it. */
6292 if (!stp_forward_in_state(port->stp_state)
6293 && !stp_learn_in_state(port->stp_state)) {
6301 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6302 struct action_xlate_ctx *ctx)
6304 bool was_evictable = true;
6305 const struct ofpact *a;
6308 /* Don't let the rule we're working on get evicted underneath us. */
6309 was_evictable = ctx->rule->up.evictable;
6310 ctx->rule->up.evictable = false;
6312 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6313 struct ofpact_controller *controller;
6314 const struct ofpact_metadata *metadata;
6322 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6323 ofpact_get_OUTPUT(a)->max_len, true);
6326 case OFPACT_CONTROLLER:
6327 controller = ofpact_get_CONTROLLER(a);
6328 execute_controller_action(ctx, controller->max_len,
6330 controller->controller_id);
6333 case OFPACT_ENQUEUE:
6334 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6337 case OFPACT_SET_VLAN_VID:
6338 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6339 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6343 case OFPACT_SET_VLAN_PCP:
6344 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6345 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
6350 case OFPACT_STRIP_VLAN:
6351 ctx->flow.vlan_tci = htons(0);
6354 case OFPACT_PUSH_VLAN:
6355 /* XXX 802.1AD(QinQ) */
6356 ctx->flow.vlan_tci = htons(VLAN_CFI);
6359 case OFPACT_SET_ETH_SRC:
6360 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6364 case OFPACT_SET_ETH_DST:
6365 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6369 case OFPACT_SET_IPV4_SRC:
6370 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6373 case OFPACT_SET_IPV4_DST:
6374 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6377 case OFPACT_SET_IPV4_DSCP:
6378 /* OpenFlow 1.0 only supports IPv4. */
6379 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6380 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
6381 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6385 case OFPACT_SET_L4_SRC_PORT:
6386 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6389 case OFPACT_SET_L4_DST_PORT:
6390 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6393 case OFPACT_RESUBMIT:
6394 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6397 case OFPACT_SET_TUNNEL:
6398 ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6401 case OFPACT_SET_QUEUE:
6402 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6405 case OFPACT_POP_QUEUE:
6406 ctx->flow.skb_priority = ctx->orig_skb_priority;
6409 case OFPACT_REG_MOVE:
6410 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
6413 case OFPACT_REG_LOAD:
6414 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
6417 case OFPACT_PUSH_MPLS:
6418 execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
6421 case OFPACT_POP_MPLS:
6422 execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
6425 case OFPACT_DEC_MPLS_TTL:
6426 if (execute_dec_mpls_ttl_action(ctx)) {
6431 case OFPACT_DEC_TTL:
6432 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6438 /* Nothing to do. */
6441 case OFPACT_MULTIPATH:
6442 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
6446 ctx->ofproto->has_bundle_action = true;
6447 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6450 case OFPACT_OUTPUT_REG:
6451 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6455 ctx->has_learn = true;
6456 if (ctx->may_learn) {
6457 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6465 case OFPACT_FIN_TIMEOUT:
6466 ctx->has_fin_timeout = true;
6467 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6470 case OFPACT_CLEAR_ACTIONS:
6472 * Nothing to do because writa-actions is not supported for now.
6473 * When writa-actions is supported, clear-actions also must
6474 * be supported at the same time.
6478 case OFPACT_WRITE_METADATA:
6479 metadata = ofpact_get_WRITE_METADATA(a);
6480 ctx->flow.metadata &= ~metadata->mask;
6481 ctx->flow.metadata |= metadata->metadata & metadata->mask;
6484 case OFPACT_GOTO_TABLE: {
6485 /* XXX remove recursion */
6486 /* It is assumed that goto-table is last action */
6487 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6488 ovs_assert(ctx->table_id < ogt->table_id);
6489 xlate_table_action(ctx, ctx->flow.in_port, ogt->table_id, true);
6497 ctx->rule->up.evictable = was_evictable;
6502 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
6503 struct ofproto_dpif *ofproto, const struct flow *flow,
6504 ovs_be16 initial_tci, struct rule_dpif *rule,
6505 uint8_t tcp_flags, const struct ofpbuf *packet)
6507 ovs_be64 initial_tun_id = flow->tunnel.tun_id;
6509 /* Flow initialization rules:
6510 * - 'base_flow' must match the kernel's view of the packet at the
6511 * time that action processing starts. 'flow' represents any
6512 * transformations we wish to make through actions.
6513 * - By default 'base_flow' and 'flow' are the same since the input
6514 * packet matches the output before any actions are applied.
6515 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
6516 * of the received packet as seen by the kernel. If we later output
6517 * to another device without any modifications this will cause us to
6518 * insert a new tag since the original one was stripped off by the
6520 * - Tunnel 'flow' is largely cleared when transitioning between
6521 * the input and output stages since it does not make sense to output
6522 * a packet with the exact headers that it was received with (i.e.
6523 * the destination IP is us). The one exception is the tun_id, which
6524 * is preserved to allow use in later resubmit lookups and loads into
6526 * - Tunnel 'base_flow' is completely cleared since that is what the
6527 * kernel does. If we wish to maintain the original values an action
6528 * needs to be generated. */
6530 ctx->ofproto = ofproto;
6532 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
6533 ctx->base_flow = ctx->flow;
6534 ctx->base_flow.vlan_tci = initial_tci;
6535 ctx->flow.tunnel.tun_id = initial_tun_id;
6537 ctx->packet = packet;
6538 ctx->may_learn = packet != NULL;
6539 ctx->tcp_flags = tcp_flags;
6540 ctx->resubmit_hook = NULL;
6541 ctx->report_hook = NULL;
6542 ctx->resubmit_stats = NULL;
6545 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6546 * into datapath actions in 'odp_actions', using 'ctx'. */
6548 xlate_actions(struct action_xlate_ctx *ctx,
6549 const struct ofpact *ofpacts, size_t ofpacts_len,
6550 struct ofpbuf *odp_actions)
6552 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6553 * that in the future we always keep a copy of the original flow for
6554 * tracing purposes. */
6555 static bool hit_resubmit_limit;
6557 enum slow_path_reason special;
6558 struct ofport_dpif *in_port;
6559 struct flow orig_flow;
6561 COVERAGE_INC(ofproto_dpif_xlate);
6563 ofpbuf_clear(odp_actions);
6564 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
6566 ctx->odp_actions = odp_actions;
6569 ctx->has_learn = false;
6570 ctx->has_normal = false;
6571 ctx->has_fin_timeout = false;
6572 ctx->nf_output_iface = NF_OUT_DROP;
6575 ctx->max_resubmit_trigger = false;
6576 ctx->orig_skb_priority = ctx->flow.skb_priority;
6580 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
6581 /* Do this conditionally because the copy is expensive enough that it
6582 * shows up in profiles. */
6583 orig_flow = ctx->flow;
6586 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6587 switch (ctx->ofproto->up.frag_handling) {
6588 case OFPC_FRAG_NORMAL:
6589 /* We must pretend that transport ports are unavailable. */
6590 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
6591 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
6594 case OFPC_FRAG_DROP:
6597 case OFPC_FRAG_REASM:
6600 case OFPC_FRAG_NX_MATCH:
6601 /* Nothing to do. */
6604 case OFPC_INVALID_TTL_TO_CONTROLLER:
6609 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
6610 special = process_special(ctx->ofproto, &ctx->flow, in_port, ctx->packet);
6612 ctx->slow |= special;
6614 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6615 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
6616 uint32_t local_odp_port;
6618 add_sflow_action(ctx);
6620 if (!in_port || may_receive(in_port, ctx)) {
6621 do_xlate_actions(ofpacts, ofpacts_len, ctx);
6623 /* We've let OFPP_NORMAL and the learning action look at the
6624 * packet, so drop it now if forwarding is disabled. */
6625 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
6626 ofpbuf_clear(ctx->odp_actions);
6627 add_sflow_action(ctx);
6631 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
6632 if (!hit_resubmit_limit) {
6633 /* We didn't record the original flow. Make sure we do from
6635 hit_resubmit_limit = true;
6636 } else if (!VLOG_DROP_ERR(&trace_rl)) {
6637 struct ds ds = DS_EMPTY_INITIALIZER;
6639 ofproto_trace(ctx->ofproto, &orig_flow, ctx->packet,
6641 VLOG_ERR("Trace triggered by excessive resubmit "
6642 "recursion:\n%s", ds_cstr(&ds));
6647 local_odp_port = ofp_port_to_odp_port(ctx->ofproto, OFPP_LOCAL);
6648 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
6650 ctx->odp_actions->data,
6651 ctx->odp_actions->size)) {
6652 ctx->slow |= SLOW_IN_BAND;
6654 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
6656 compose_output_action(ctx, OFPP_LOCAL);
6659 if (ctx->ofproto->has_mirrors) {
6660 add_mirror_actions(ctx, &orig_flow);
6662 fix_sflow_action(ctx);
6666 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
6667 * into datapath actions, using 'ctx', and discards the datapath actions. */
6669 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
6670 const struct ofpact *ofpacts,
6673 uint64_t odp_actions_stub[1024 / 8];
6674 struct ofpbuf odp_actions;
6676 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
6677 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
6678 ofpbuf_uninit(&odp_actions);
6682 xlate_report(struct action_xlate_ctx *ctx, const char *s)
6684 if (ctx->report_hook) {
6685 ctx->report_hook(ctx, s);
6689 /* OFPP_NORMAL implementation. */
6691 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
6693 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
6694 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
6695 * the bundle on which the packet was received, returns the VLAN to which the
6698 * Both 'vid' and the return value are in the range 0...4095. */
6700 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
6702 switch (in_bundle->vlan_mode) {
6703 case PORT_VLAN_ACCESS:
6704 return in_bundle->vlan;
6707 case PORT_VLAN_TRUNK:
6710 case PORT_VLAN_NATIVE_UNTAGGED:
6711 case PORT_VLAN_NATIVE_TAGGED:
6712 return vid ? vid : in_bundle->vlan;
6719 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
6720 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
6723 * 'vid' should be the VID obtained from the 802.1Q header that was received as
6724 * part of a packet (specify 0 if there was no 802.1Q header), in the range
6727 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
6729 /* Allow any VID on the OFPP_NONE port. */
6730 if (in_bundle == &ofpp_none_bundle) {
6734 switch (in_bundle->vlan_mode) {
6735 case PORT_VLAN_ACCESS:
6738 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6739 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
6740 "packet received on port %s configured as VLAN "
6741 "%"PRIu16" access port",
6742 in_bundle->ofproto->up.name, vid,
6743 in_bundle->name, in_bundle->vlan);
6749 case PORT_VLAN_NATIVE_UNTAGGED:
6750 case PORT_VLAN_NATIVE_TAGGED:
6752 /* Port must always carry its native VLAN. */
6756 case PORT_VLAN_TRUNK:
6757 if (!ofbundle_includes_vlan(in_bundle, vid)) {
6759 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6760 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
6761 "received on port %s not configured for trunking "
6763 in_bundle->ofproto->up.name, vid,
6764 in_bundle->name, vid);
6776 /* Given 'vlan', the VLAN that a packet belongs to, and
6777 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
6778 * that should be included in the 802.1Q header. (If the return value is 0,
6779 * then the 802.1Q header should only be included in the packet if there is a
6782 * Both 'vlan' and the return value are in the range 0...4095. */
6784 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
6786 switch (out_bundle->vlan_mode) {
6787 case PORT_VLAN_ACCESS:
6790 case PORT_VLAN_TRUNK:
6791 case PORT_VLAN_NATIVE_TAGGED:
6794 case PORT_VLAN_NATIVE_UNTAGGED:
6795 return vlan == out_bundle->vlan ? 0 : vlan;
6803 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
6806 struct ofport_dpif *port;
6808 ovs_be16 tci, old_tci;
6810 vid = output_vlan_to_vid(out_bundle, vlan);
6811 if (!out_bundle->bond) {
6812 port = ofbundle_get_a_port(out_bundle);
6814 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
6817 /* No slaves enabled, so drop packet. */
6822 old_tci = ctx->flow.vlan_tci;
6824 if (tci || out_bundle->use_priority_tags) {
6825 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
6827 tci |= htons(VLAN_CFI);
6830 ctx->flow.vlan_tci = tci;
6832 compose_output_action(ctx, port->up.ofp_port);
6833 ctx->flow.vlan_tci = old_tci;
6837 mirror_mask_ffs(mirror_mask_t mask)
6839 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
6844 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
6846 return (bundle->vlan_mode != PORT_VLAN_ACCESS
6847 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
6851 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
6853 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
6856 /* Returns an arbitrary interface within 'bundle'. */
6857 static struct ofport_dpif *
6858 ofbundle_get_a_port(const struct ofbundle *bundle)
6860 return CONTAINER_OF(list_front(&bundle->ports),
6861 struct ofport_dpif, bundle_node);
6865 vlan_is_mirrored(const struct ofmirror *m, int vlan)
6867 return !m->vlans || bitmap_is_set(m->vlans, vlan);
6871 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
6873 struct ofproto_dpif *ofproto = ctx->ofproto;
6874 mirror_mask_t mirrors;
6875 struct ofbundle *in_bundle;
6878 const struct nlattr *a;
6881 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
6882 ctx->packet != NULL, NULL);
6886 mirrors = in_bundle->src_mirrors;
6888 /* Drop frames on bundles reserved for mirroring. */
6889 if (in_bundle->mirror_out) {
6890 if (ctx->packet != NULL) {
6891 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6892 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6893 "%s, which is reserved exclusively for mirroring",
6894 ctx->ofproto->up.name, in_bundle->name);
6900 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
6901 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6904 vlan = input_vid_to_vlan(in_bundle, vid);
6906 /* Look at the output ports to check for destination selections. */
6908 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
6909 ctx->odp_actions->size) {
6910 enum ovs_action_attr type = nl_attr_type(a);
6911 struct ofport_dpif *ofport;
6913 if (type != OVS_ACTION_ATTR_OUTPUT) {
6917 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
6918 if (ofport && ofport->bundle) {
6919 mirrors |= ofport->bundle->dst_mirrors;
6927 /* Restore the original packet before adding the mirror actions. */
6928 ctx->flow = *orig_flow;
6933 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6935 if (!vlan_is_mirrored(m, vlan)) {
6936 mirrors = zero_rightmost_1bit(mirrors);
6940 mirrors &= ~m->dup_mirrors;
6941 ctx->mirrors |= m->dup_mirrors;
6943 output_normal(ctx, m->out, vlan);
6944 } else if (vlan != m->out_vlan
6945 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
6946 struct ofbundle *bundle;
6948 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
6949 if (ofbundle_includes_vlan(bundle, m->out_vlan)
6950 && !bundle->mirror_out) {
6951 output_normal(ctx, bundle, m->out_vlan);
6959 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
6960 uint64_t packets, uint64_t bytes)
6966 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
6969 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6972 /* In normal circumstances 'm' will not be NULL. However,
6973 * if mirrors are reconfigured, we can temporarily get out
6974 * of sync in facet_revalidate(). We could "correct" the
6975 * mirror list before reaching here, but doing that would
6976 * not properly account the traffic stats we've currently
6977 * accumulated for previous mirror configuration. */
6981 m->packet_count += packets;
6982 m->byte_count += bytes;
6986 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
6987 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
6988 * indicate this; newer upstream kernels use gratuitous ARP requests. */
6990 is_gratuitous_arp(const struct flow *flow)
6992 return (flow->dl_type == htons(ETH_TYPE_ARP)
6993 && eth_addr_is_broadcast(flow->dl_dst)
6994 && (flow->nw_proto == ARP_OP_REPLY
6995 || (flow->nw_proto == ARP_OP_REQUEST
6996 && flow->nw_src == flow->nw_dst)));
7000 update_learning_table(struct ofproto_dpif *ofproto,
7001 const struct flow *flow, int vlan,
7002 struct ofbundle *in_bundle)
7004 struct mac_entry *mac;
7006 /* Don't learn the OFPP_NONE port. */
7007 if (in_bundle == &ofpp_none_bundle) {
7011 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7015 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7016 if (is_gratuitous_arp(flow)) {
7017 /* We don't want to learn from gratuitous ARP packets that are
7018 * reflected back over bond slaves so we lock the learning table. */
7019 if (!in_bundle->bond) {
7020 mac_entry_set_grat_arp_lock(mac);
7021 } else if (mac_entry_is_grat_arp_locked(mac)) {
7026 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7027 /* The log messages here could actually be useful in debugging,
7028 * so keep the rate limit relatively high. */
7029 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7030 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7031 "on port %s in VLAN %d",
7032 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7033 in_bundle->name, vlan);
7035 mac->port.p = in_bundle;
7036 tag_set_add(&ofproto->backer->revalidate_set,
7037 mac_learning_changed(ofproto->ml, mac));
7041 static struct ofbundle *
7042 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7043 bool warn, struct ofport_dpif **in_ofportp)
7045 struct ofport_dpif *ofport;
7047 /* Find the port and bundle for the received packet. */
7048 ofport = get_ofp_port(ofproto, in_port);
7050 *in_ofportp = ofport;
7052 if (ofport && ofport->bundle) {
7053 return ofport->bundle;
7056 /* Special-case OFPP_NONE, which a controller may use as the ingress
7057 * port for traffic that it is sourcing. */
7058 if (in_port == OFPP_NONE) {
7059 return &ofpp_none_bundle;
7062 /* Odd. A few possible reasons here:
7064 * - We deleted a port but there are still a few packets queued up
7067 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7068 * we don't know about.
7070 * - The ofproto client didn't configure the port as part of a bundle.
7071 * This is particularly likely to happen if a packet was received on the
7072 * port after it was created, but before the client had a chance to
7073 * configure its bundle.
7076 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7078 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7079 "port %"PRIu16, ofproto->up.name, in_port);
7084 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7085 * dropped. Returns true if they may be forwarded, false if they should be
7088 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7089 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7091 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7092 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7093 * checked by input_vid_is_valid().
7095 * May also add tags to '*tags', although the current implementation only does
7096 * so in one special case.
7099 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
7102 struct ofproto_dpif *ofproto = ctx->ofproto;
7103 struct flow *flow = &ctx->flow;
7104 struct ofbundle *in_bundle = in_port->bundle;
7106 /* Drop frames for reserved multicast addresses
7107 * only if forward_bpdu option is absent. */
7108 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7109 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7113 if (in_bundle->bond) {
7114 struct mac_entry *mac;
7116 switch (bond_check_admissibility(in_bundle->bond, in_port,
7117 flow->dl_dst, &ctx->tags)) {
7122 xlate_report(ctx, "bonding refused admissibility, dropping");
7125 case BV_DROP_IF_MOVED:
7126 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7127 if (mac && mac->port.p != in_bundle &&
7128 (!is_gratuitous_arp(flow)
7129 || mac_entry_is_grat_arp_locked(mac))) {
7130 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7142 xlate_normal(struct action_xlate_ctx *ctx)
7144 struct ofport_dpif *in_port;
7145 struct ofbundle *in_bundle;
7146 struct mac_entry *mac;
7150 ctx->has_normal = true;
7152 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
7153 ctx->packet != NULL, &in_port);
7155 xlate_report(ctx, "no input bundle, dropping");
7159 /* Drop malformed frames. */
7160 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7161 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
7162 if (ctx->packet != NULL) {
7163 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7164 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7165 "VLAN tag received on port %s",
7166 ctx->ofproto->up.name, in_bundle->name);
7168 xlate_report(ctx, "partial VLAN tag, dropping");
7172 /* Drop frames on bundles reserved for mirroring. */
7173 if (in_bundle->mirror_out) {
7174 if (ctx->packet != NULL) {
7175 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7176 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7177 "%s, which is reserved exclusively for mirroring",
7178 ctx->ofproto->up.name, in_bundle->name);
7180 xlate_report(ctx, "input port is mirror output port, dropping");
7185 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
7186 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7187 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7190 vlan = input_vid_to_vlan(in_bundle, vid);
7192 /* Check other admissibility requirements. */
7193 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7197 /* Learn source MAC. */
7198 if (ctx->may_learn) {
7199 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
7202 /* Determine output bundle. */
7203 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
7206 if (mac->port.p != in_bundle) {
7207 xlate_report(ctx, "forwarding to learned port");
7208 output_normal(ctx, mac->port.p, vlan);
7210 xlate_report(ctx, "learned port is input port, dropping");
7213 struct ofbundle *bundle;
7215 xlate_report(ctx, "no learned MAC for destination, flooding");
7216 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7217 if (bundle != in_bundle
7218 && ofbundle_includes_vlan(bundle, vlan)
7219 && bundle->floodable
7220 && !bundle->mirror_out) {
7221 output_normal(ctx, bundle, vlan);
7224 ctx->nf_output_iface = NF_OUT_FLOOD;
7228 /* Optimized flow revalidation.
7230 * It's a difficult problem, in general, to tell which facets need to have
7231 * their actions recalculated whenever the OpenFlow flow table changes. We
7232 * don't try to solve that general problem: for most kinds of OpenFlow flow
7233 * table changes, we recalculate the actions for every facet. This is
7234 * relatively expensive, but it's good enough if the OpenFlow flow table
7235 * doesn't change very often.
7237 * However, we can expect one particular kind of OpenFlow flow table change to
7238 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7239 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7240 * table, we add a special case that applies to flow tables in which every rule
7241 * has the same form (that is, the same wildcards), except that the table is
7242 * also allowed to have a single "catch-all" flow that matches all packets. We
7243 * optimize this case by tagging all of the facets that resubmit into the table
7244 * and invalidating the same tag whenever a flow changes in that table. The
7245 * end result is that we revalidate just the facets that need it (and sometimes
7246 * a few more, but not all of the facets or even all of the facets that
7247 * resubmit to the table modified by MAC learning). */
7249 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7250 * into an OpenFlow table with the given 'basis'. */
7252 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7255 if (minimask_is_catchall(mask)) {
7258 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7259 return tag_create_deterministic(hash);
7263 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7264 * taggability of that table.
7266 * This function must be called after *each* change to a flow table. If you
7267 * skip calling it on some changes then the pointer comparisons at the end can
7268 * be invalid if you get unlucky. For example, if a flow removal causes a
7269 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7270 * different wildcards to be created with the same address, then this function
7271 * will incorrectly skip revalidation. */
7273 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7275 struct table_dpif *table = &ofproto->tables[table_id];
7276 const struct oftable *oftable = &ofproto->up.tables[table_id];
7277 struct cls_table *catchall, *other;
7278 struct cls_table *t;
7280 catchall = other = NULL;
7282 switch (hmap_count(&oftable->cls.tables)) {
7284 /* We could tag this OpenFlow table but it would make the logic a
7285 * little harder and it's a corner case that doesn't seem worth it
7291 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7292 if (cls_table_is_catchall(t)) {
7294 } else if (!other) {
7297 /* Indicate that we can't tag this by setting both tables to
7298 * NULL. (We know that 'catchall' is already NULL.) */
7305 /* Can't tag this table. */
7309 if (table->catchall_table != catchall || table->other_table != other) {
7310 table->catchall_table = catchall;
7311 table->other_table = other;
7312 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7316 /* Given 'rule' that has changed in some way (either it is a rule being
7317 * inserted, a rule being deleted, or a rule whose actions are being
7318 * modified), marks facets for revalidation to ensure that packets will be
7319 * forwarded correctly according to the new state of the flow table.
7321 * This function must be called after *each* change to a flow table. See
7322 * the comment on table_update_taggable() for more information. */
7324 rule_invalidate(const struct rule_dpif *rule)
7326 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7328 table_update_taggable(ofproto, rule->up.table_id);
7330 if (!ofproto->backer->need_revalidate) {
7331 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7333 if (table->other_table && rule->tag) {
7334 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7336 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7342 set_frag_handling(struct ofproto *ofproto_,
7343 enum ofp_config_flags frag_handling)
7345 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7346 if (frag_handling != OFPC_FRAG_REASM) {
7347 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7355 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7356 const struct flow *flow,
7357 const struct ofpact *ofpacts, size_t ofpacts_len)
7359 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7360 struct odputil_keybuf keybuf;
7361 struct dpif_flow_stats stats;
7365 struct action_xlate_ctx ctx;
7366 uint64_t odp_actions_stub[1024 / 8];
7367 struct ofpbuf odp_actions;
7369 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7370 odp_flow_key_from_flow(&key, flow,
7371 ofp_port_to_odp_port(ofproto, flow->in_port));
7373 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7375 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL,
7376 packet_get_tcp_flags(packet, flow), packet);
7377 ctx.resubmit_stats = &stats;
7379 ofpbuf_use_stub(&odp_actions,
7380 odp_actions_stub, sizeof odp_actions_stub);
7381 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
7382 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7383 odp_actions.data, odp_actions.size, packet);
7384 ofpbuf_uninit(&odp_actions);
7392 set_netflow(struct ofproto *ofproto_,
7393 const struct netflow_options *netflow_options)
7395 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7397 if (netflow_options) {
7398 if (!ofproto->netflow) {
7399 ofproto->netflow = netflow_create();
7401 return netflow_set_options(ofproto->netflow, netflow_options);
7403 netflow_destroy(ofproto->netflow);
7404 ofproto->netflow = NULL;
7410 get_netflow_ids(const struct ofproto *ofproto_,
7411 uint8_t *engine_type, uint8_t *engine_id)
7413 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7415 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7419 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7421 if (!facet_is_controller_flow(facet) &&
7422 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7423 struct subfacet *subfacet;
7424 struct ofexpired expired;
7426 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7427 if (subfacet->path == SF_FAST_PATH) {
7428 struct dpif_flow_stats stats;
7430 subfacet_reinstall(subfacet, &stats);
7431 subfacet_update_stats(subfacet, &stats);
7435 expired.flow = facet->flow;
7436 expired.packet_count = facet->packet_count;
7437 expired.byte_count = facet->byte_count;
7438 expired.used = facet->used;
7439 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
7444 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7446 struct facet *facet;
7448 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7449 send_active_timeout(ofproto, facet);
7453 static struct ofproto_dpif *
7454 ofproto_dpif_lookup(const char *name)
7456 struct ofproto_dpif *ofproto;
7458 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7459 hash_string(name, 0), &all_ofproto_dpifs) {
7460 if (!strcmp(ofproto->up.name, name)) {
7468 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7469 const char *argv[], void *aux OVS_UNUSED)
7471 struct ofproto_dpif *ofproto;
7474 ofproto = ofproto_dpif_lookup(argv[1]);
7476 unixctl_command_reply_error(conn, "no such bridge");
7479 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7481 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7482 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7486 unixctl_command_reply(conn, "table successfully flushed");
7490 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7491 const char *argv[], void *aux OVS_UNUSED)
7493 struct ds ds = DS_EMPTY_INITIALIZER;
7494 const struct ofproto_dpif *ofproto;
7495 const struct mac_entry *e;
7497 ofproto = ofproto_dpif_lookup(argv[1]);
7499 unixctl_command_reply_error(conn, "no such bridge");
7503 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7504 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7505 struct ofbundle *bundle = e->port.p;
7506 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7507 ofbundle_get_a_port(bundle)->odp_port,
7508 e->vlan, ETH_ADDR_ARGS(e->mac),
7509 mac_entry_age(ofproto->ml, e));
7511 unixctl_command_reply(conn, ds_cstr(&ds));
7516 struct action_xlate_ctx ctx;
7522 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7523 const struct rule_dpif *rule)
7525 ds_put_char_multiple(result, '\t', level);
7527 ds_put_cstr(result, "No match\n");
7531 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7532 table_id, ntohll(rule->up.flow_cookie));
7533 cls_rule_format(&rule->up.cr, result);
7534 ds_put_char(result, '\n');
7536 ds_put_char_multiple(result, '\t', level);
7537 ds_put_cstr(result, "OpenFlow ");
7538 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7539 ds_put_char(result, '\n');
7543 trace_format_flow(struct ds *result, int level, const char *title,
7544 struct trace_ctx *trace)
7546 ds_put_char_multiple(result, '\t', level);
7547 ds_put_format(result, "%s: ", title);
7548 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
7549 ds_put_cstr(result, "unchanged");
7551 flow_format(result, &trace->ctx.flow);
7552 trace->flow = trace->ctx.flow;
7554 ds_put_char(result, '\n');
7558 trace_format_regs(struct ds *result, int level, const char *title,
7559 struct trace_ctx *trace)
7563 ds_put_char_multiple(result, '\t', level);
7564 ds_put_format(result, "%s:", title);
7565 for (i = 0; i < FLOW_N_REGS; i++) {
7566 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7568 ds_put_char(result, '\n');
7572 trace_format_odp(struct ds *result, int level, const char *title,
7573 struct trace_ctx *trace)
7575 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
7577 ds_put_char_multiple(result, '\t', level);
7578 ds_put_format(result, "%s: ", title);
7579 format_odp_actions(result, odp_actions->data, odp_actions->size);
7580 ds_put_char(result, '\n');
7584 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
7586 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7587 struct ds *result = trace->result;
7589 ds_put_char(result, '\n');
7590 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7591 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
7592 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
7593 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
7597 trace_report(struct action_xlate_ctx *ctx, const char *s)
7599 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7600 struct ds *result = trace->result;
7602 ds_put_char_multiple(result, '\t', ctx->recurse);
7603 ds_put_cstr(result, s);
7604 ds_put_char(result, '\n');
7608 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
7609 void *aux OVS_UNUSED)
7611 const char *dpname = argv[1];
7612 struct ofproto_dpif *ofproto;
7613 struct ofpbuf odp_key;
7614 struct ofpbuf *packet;
7615 ovs_be16 initial_tci;
7621 ofpbuf_init(&odp_key, 0);
7624 ofproto = ofproto_dpif_lookup(dpname);
7626 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
7630 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
7631 /* ofproto/trace dpname flow [-generate] */
7632 const char *flow_s = argv[2];
7633 const char *generate_s = argv[3];
7635 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
7636 * flow. We guess which type it is based on whether 'flow_s' contains
7637 * an '(', since a datapath flow always contains '(') but an
7638 * OpenFlow-like flow should not (in fact it's allowed but I believe
7639 * that's not documented anywhere).
7641 * An alternative would be to try to parse 'flow_s' both ways, but then
7642 * it would be tricky giving a sensible error message. After all, do
7643 * you just say "syntax error" or do you present both error messages?
7644 * Both choices seem lousy. */
7645 if (strchr(flow_s, '(')) {
7648 /* Convert string to datapath key. */
7649 ofpbuf_init(&odp_key, 0);
7650 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
7652 unixctl_command_reply_error(conn, "Bad flow syntax");
7656 /* XXX: Since we allow the user to specify an ofproto, it's
7657 * possible they will specify a different ofproto than the one the
7658 * port actually belongs too. Ideally we should simply remove the
7659 * ability to specify the ofproto. */
7660 if (ofproto_receive(ofproto->backer, NULL, odp_key.data,
7661 odp_key.size, &flow, NULL, NULL, NULL,
7663 unixctl_command_reply_error(conn, "Invalid flow");
7669 error_s = parse_ofp_exact_flow(&flow, argv[2]);
7671 unixctl_command_reply_error(conn, error_s);
7676 initial_tci = flow.vlan_tci;
7679 /* Generate a packet, if requested. */
7681 packet = ofpbuf_new(0);
7682 flow_compose(packet, &flow);
7684 } else if (argc == 7) {
7685 /* ofproto/trace dpname priority tun_id in_port mark packet */
7686 const char *priority_s = argv[2];
7687 const char *tun_id_s = argv[3];
7688 const char *in_port_s = argv[4];
7689 const char *mark_s = argv[5];
7690 const char *packet_s = argv[6];
7691 uint32_t in_port = atoi(in_port_s);
7692 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
7693 uint32_t priority = atoi(priority_s);
7694 uint32_t mark = atoi(mark_s);
7697 msg = eth_from_hex(packet_s, &packet);
7699 unixctl_command_reply_error(conn, msg);
7703 ds_put_cstr(&result, "Packet: ");
7704 s = ofp_packet_to_string(packet->data, packet->size);
7705 ds_put_cstr(&result, s);
7708 flow_extract(packet, priority, mark, NULL, in_port, &flow);
7709 flow.tunnel.tun_id = tun_id;
7710 initial_tci = flow.vlan_tci;
7712 unixctl_command_reply_error(conn, "Bad command syntax");
7716 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
7717 unixctl_command_reply(conn, ds_cstr(&result));
7720 ds_destroy(&result);
7721 ofpbuf_delete(packet);
7722 ofpbuf_uninit(&odp_key);
7726 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
7727 const struct ofpbuf *packet, ovs_be16 initial_tci,
7730 struct rule_dpif *rule;
7732 ds_put_cstr(ds, "Flow: ");
7733 flow_format(ds, flow);
7734 ds_put_char(ds, '\n');
7736 rule = rule_dpif_lookup(ofproto, flow);
7738 trace_format_rule(ds, 0, 0, rule);
7739 if (rule == ofproto->miss_rule) {
7740 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
7741 } else if (rule == ofproto->no_packet_in_rule) {
7742 ds_put_cstr(ds, "\nNo match, packets dropped because "
7743 "OFPPC_NO_PACKET_IN is set on in_port.\n");
7747 uint64_t odp_actions_stub[1024 / 8];
7748 struct ofpbuf odp_actions;
7750 struct trace_ctx trace;
7753 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
7756 ofpbuf_use_stub(&odp_actions,
7757 odp_actions_stub, sizeof odp_actions_stub);
7758 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
7759 rule, tcp_flags, packet);
7760 trace.ctx.resubmit_hook = trace_resubmit;
7761 trace.ctx.report_hook = trace_report;
7762 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
7765 ds_put_char(ds, '\n');
7766 trace_format_flow(ds, 0, "Final flow", &trace);
7767 ds_put_cstr(ds, "Datapath actions: ");
7768 format_odp_actions(ds, odp_actions.data, odp_actions.size);
7769 ofpbuf_uninit(&odp_actions);
7771 if (trace.ctx.slow) {
7772 enum slow_path_reason slow;
7774 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
7775 "slow path because it:");
7776 for (slow = trace.ctx.slow; slow; ) {
7777 enum slow_path_reason bit = rightmost_1bit(slow);
7781 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
7784 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
7787 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
7790 ds_put_cstr(ds, "\n\t- Needs in-band special case "
7793 ds_put_cstr(ds, "\n\t (The datapath actions are "
7794 "incomplete--for complete actions, "
7795 "please supply a packet.)");
7798 case SLOW_CONTROLLER:
7799 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
7800 "to the OpenFlow controller.");
7803 ds_put_cstr(ds, "\n\t- Needs more specific matching "
7804 "than the datapath supports.");
7811 if (slow & ~SLOW_MATCH) {
7812 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
7813 "the special slow-path processing.");
7820 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7821 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7824 unixctl_command_reply(conn, NULL);
7828 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7829 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7832 unixctl_command_reply(conn, NULL);
7835 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
7836 * 'reply' describing the results. */
7838 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
7840 struct facet *facet;
7844 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7845 if (!facet_check_consistency(facet)) {
7850 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
7854 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
7855 ofproto->up.name, errors);
7857 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
7862 ofproto_dpif_self_check(struct unixctl_conn *conn,
7863 int argc, const char *argv[], void *aux OVS_UNUSED)
7865 struct ds reply = DS_EMPTY_INITIALIZER;
7866 struct ofproto_dpif *ofproto;
7869 ofproto = ofproto_dpif_lookup(argv[1]);
7871 unixctl_command_reply_error(conn, "Unknown ofproto (use "
7872 "ofproto/list for help)");
7875 ofproto_dpif_self_check__(ofproto, &reply);
7877 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7878 ofproto_dpif_self_check__(ofproto, &reply);
7882 unixctl_command_reply(conn, ds_cstr(&reply));
7886 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
7887 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
7888 * to destroy 'ofproto_shash' and free the returned value. */
7889 static const struct shash_node **
7890 get_ofprotos(struct shash *ofproto_shash)
7892 const struct ofproto_dpif *ofproto;
7894 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7895 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
7896 shash_add_nocopy(ofproto_shash, name, ofproto);
7899 return shash_sort(ofproto_shash);
7903 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
7904 const char *argv[] OVS_UNUSED,
7905 void *aux OVS_UNUSED)
7907 struct ds ds = DS_EMPTY_INITIALIZER;
7908 struct shash ofproto_shash;
7909 const struct shash_node **sorted_ofprotos;
7912 shash_init(&ofproto_shash);
7913 sorted_ofprotos = get_ofprotos(&ofproto_shash);
7914 for (i = 0; i < shash_count(&ofproto_shash); i++) {
7915 const struct shash_node *node = sorted_ofprotos[i];
7916 ds_put_format(&ds, "%s\n", node->name);
7919 shash_destroy(&ofproto_shash);
7920 free(sorted_ofprotos);
7922 unixctl_command_reply(conn, ds_cstr(&ds));
7927 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
7929 struct dpif_dp_stats s;
7930 const struct shash_node **ports;
7933 dpif_get_dp_stats(ofproto->backer->dpif, &s);
7935 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
7936 dpif_name(ofproto->backer->dpif));
7937 /* xxx It would be better to show bridge-specific stats instead
7938 * xxx of dp ones. */
7940 "\tlookups: hit:%"PRIu64" missed:%"PRIu64" lost:%"PRIu64"\n",
7941 s.n_hit, s.n_missed, s.n_lost);
7942 ds_put_format(ds, "\tflows: %zu\n",
7943 hmap_count(&ofproto->subfacets));
7945 ports = shash_sort(&ofproto->up.port_by_name);
7946 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
7947 const struct shash_node *node = ports[i];
7948 struct ofport *ofport = node->data;
7949 const char *name = netdev_get_name(ofport->netdev);
7950 const char *type = netdev_get_type(ofport->netdev);
7953 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
7955 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
7956 if (odp_port != OVSP_NONE) {
7957 ds_put_format(ds, "%"PRIu32":", odp_port);
7959 ds_put_cstr(ds, "none:");
7962 if (strcmp(type, "system")) {
7963 struct netdev *netdev;
7966 ds_put_format(ds, " (%s", type);
7968 error = netdev_open(name, type, &netdev);
7973 error = netdev_get_config(netdev, &config);
7975 const struct smap_node **nodes;
7978 nodes = smap_sort(&config);
7979 for (i = 0; i < smap_count(&config); i++) {
7980 const struct smap_node *node = nodes[i];
7981 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
7982 node->key, node->value);
7986 smap_destroy(&config);
7988 netdev_close(netdev);
7990 ds_put_char(ds, ')');
7992 ds_put_char(ds, '\n');
7998 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
7999 const char *argv[], void *aux OVS_UNUSED)
8001 struct ds ds = DS_EMPTY_INITIALIZER;
8002 const struct ofproto_dpif *ofproto;
8006 for (i = 1; i < argc; i++) {
8007 ofproto = ofproto_dpif_lookup(argv[i]);
8009 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
8010 "for help)", argv[i]);
8011 unixctl_command_reply_error(conn, ds_cstr(&ds));
8014 show_dp_format(ofproto, &ds);
8017 struct shash ofproto_shash;
8018 const struct shash_node **sorted_ofprotos;
8021 shash_init(&ofproto_shash);
8022 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8023 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8024 const struct shash_node *node = sorted_ofprotos[i];
8025 show_dp_format(node->data, &ds);
8028 shash_destroy(&ofproto_shash);
8029 free(sorted_ofprotos);
8032 unixctl_command_reply(conn, ds_cstr(&ds));
8037 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8038 int argc OVS_UNUSED, const char *argv[],
8039 void *aux OVS_UNUSED)
8041 struct ds ds = DS_EMPTY_INITIALIZER;
8042 const struct ofproto_dpif *ofproto;
8043 struct subfacet *subfacet;
8045 ofproto = ofproto_dpif_lookup(argv[1]);
8047 unixctl_command_reply_error(conn, "no such bridge");
8051 update_stats(ofproto->backer);
8053 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8054 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
8056 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8057 subfacet->dp_packet_count, subfacet->dp_byte_count);
8058 if (subfacet->used) {
8059 ds_put_format(&ds, "%.3fs",
8060 (time_msec() - subfacet->used) / 1000.0);
8062 ds_put_format(&ds, "never");
8064 if (subfacet->facet->tcp_flags) {
8065 ds_put_cstr(&ds, ", flags:");
8066 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8069 ds_put_cstr(&ds, ", actions:");
8070 format_odp_actions(&ds, subfacet->actions, subfacet->actions_len);
8071 ds_put_char(&ds, '\n');
8074 unixctl_command_reply(conn, ds_cstr(&ds));
8079 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8080 int argc OVS_UNUSED, const char *argv[],
8081 void *aux OVS_UNUSED)
8083 struct ds ds = DS_EMPTY_INITIALIZER;
8084 struct ofproto_dpif *ofproto;
8086 ofproto = ofproto_dpif_lookup(argv[1]);
8088 unixctl_command_reply_error(conn, "no such bridge");
8092 flush(&ofproto->up);
8094 unixctl_command_reply(conn, ds_cstr(&ds));
8099 ofproto_dpif_unixctl_init(void)
8101 static bool registered;
8107 unixctl_command_register(
8109 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
8110 2, 6, ofproto_unixctl_trace, NULL);
8111 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8112 ofproto_unixctl_fdb_flush, NULL);
8113 unixctl_command_register("fdb/show", "bridge", 1, 1,
8114 ofproto_unixctl_fdb_show, NULL);
8115 unixctl_command_register("ofproto/clog", "", 0, 0,
8116 ofproto_dpif_clog, NULL);
8117 unixctl_command_register("ofproto/unclog", "", 0, 0,
8118 ofproto_dpif_unclog, NULL);
8119 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8120 ofproto_dpif_self_check, NULL);
8121 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8122 ofproto_unixctl_dpif_dump_dps, NULL);
8123 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8124 ofproto_unixctl_dpif_show, NULL);
8125 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8126 ofproto_unixctl_dpif_dump_flows, NULL);
8127 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8128 ofproto_unixctl_dpif_del_flows, NULL);
8131 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8133 * This is deprecated. It is only for compatibility with broken device drivers
8134 * in old versions of Linux that do not properly support VLANs when VLAN
8135 * devices are not used. When broken device drivers are no longer in
8136 * widespread use, we will delete these interfaces. */
8139 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8141 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8142 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8144 if (realdev_ofp_port == ofport->realdev_ofp_port
8145 && vid == ofport->vlandev_vid) {
8149 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8151 if (ofport->realdev_ofp_port) {
8154 if (realdev_ofp_port && ofport->bundle) {
8155 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8156 * themselves be part of a bundle. */
8157 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8160 ofport->realdev_ofp_port = realdev_ofp_port;
8161 ofport->vlandev_vid = vid;
8163 if (realdev_ofp_port) {
8164 vsp_add(ofport, realdev_ofp_port, vid);
8171 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8173 return hash_2words(realdev_ofp_port, vid);
8176 /* Returns the ODP port number of the Linux VLAN device that corresponds to
8177 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
8178 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
8179 * it would return the port number of eth0.9.
8181 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
8182 * function just returns its 'realdev_odp_port' argument. */
8184 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8185 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
8187 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8188 uint16_t realdev_ofp_port;
8189 int vid = vlan_tci_to_vid(vlan_tci);
8190 const struct vlan_splinter *vsp;
8192 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
8193 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8194 hash_realdev_vid(realdev_ofp_port, vid),
8195 &ofproto->realdev_vid_map) {
8196 if (vsp->realdev_ofp_port == realdev_ofp_port
8197 && vsp->vid == vid) {
8198 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
8202 return realdev_odp_port;
8205 static struct vlan_splinter *
8206 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8208 struct vlan_splinter *vsp;
8210 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8211 &ofproto->vlandev_map) {
8212 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8220 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8221 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8222 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8223 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8224 * eth0 and store 9 in '*vid'.
8226 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8227 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8230 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8231 uint16_t vlandev_ofp_port, int *vid)
8233 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8234 const struct vlan_splinter *vsp;
8236 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8241 return vsp->realdev_ofp_port;
8247 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8248 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8249 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8250 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8251 * always the case unless VLAN splinters are enabled), returns false without
8252 * making any changes. */
8254 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8259 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8264 /* Cause the flow to be processed as if it came in on the real device with
8265 * the VLAN device's VLAN ID. */
8266 flow->in_port = realdev;
8267 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8272 vsp_remove(struct ofport_dpif *port)
8274 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8275 struct vlan_splinter *vsp;
8277 vsp = vlandev_find(ofproto, port->up.ofp_port);
8279 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8280 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8283 port->realdev_ofp_port = 0;
8285 VLOG_ERR("missing vlan device record");
8290 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8292 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8294 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8295 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8296 == realdev_ofp_port)) {
8297 struct vlan_splinter *vsp;
8299 vsp = xmalloc(sizeof *vsp);
8300 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8301 hash_int(port->up.ofp_port, 0));
8302 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8303 hash_realdev_vid(realdev_ofp_port, vid));
8304 vsp->realdev_ofp_port = realdev_ofp_port;
8305 vsp->vlandev_ofp_port = port->up.ofp_port;
8308 port->realdev_ofp_port = realdev_ofp_port;
8310 VLOG_ERR("duplicate vlan device record");
8315 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8317 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8318 return ofport ? ofport->odp_port : OVSP_NONE;
8321 static struct ofport_dpif *
8322 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8324 struct ofport_dpif *port;
8326 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8327 hash_int(odp_port, 0),
8328 &backer->odp_to_ofport_map) {
8329 if (port->odp_port == odp_port) {
8338 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8340 struct ofport_dpif *port;
8342 port = odp_port_to_ofport(ofproto->backer, odp_port);
8343 if (port && &ofproto->up == port->up.ofproto) {
8344 return port->up.ofp_port;
8350 const struct ofproto_class ofproto_dpif_class = {
8385 port_is_lacp_current,
8386 NULL, /* rule_choose_table */
8393 rule_modify_actions,
8402 get_cfm_remote_mpids,
8407 get_stp_port_status,
8414 is_mirror_output_bundle,
8415 forward_bpdu_changed,
8416 set_mac_table_config,