2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-provider.h"
25 #include "byte-order.h"
30 #include "dynamic-string.h"
31 #include "fail-open.h"
35 #include "mac-learning.h"
36 #include "meta-flow.h"
37 #include "multipath.h"
38 #include "netdev-vport.h"
45 #include "ofp-actions.h"
46 #include "ofp-parse.h"
47 #include "ofp-print.h"
48 #include "ofproto-dpif-governor.h"
49 #include "ofproto-dpif-sflow.h"
50 #include "poll-loop.h"
55 #include "unaligned.h"
57 #include "vlan-bitmap.h"
60 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
62 COVERAGE_DEFINE(ofproto_dpif_expired);
63 COVERAGE_DEFINE(ofproto_dpif_xlate);
64 COVERAGE_DEFINE(facet_changed_rule);
65 COVERAGE_DEFINE(facet_revalidate);
66 COVERAGE_DEFINE(facet_unexpected);
67 COVERAGE_DEFINE(facet_suppress);
69 /* Maximum depth of flow table recursion (due to resubmit actions) in a
70 * flow translation. */
71 #define MAX_RESUBMIT_RECURSION 64
73 /* Number of implemented OpenFlow tables. */
74 enum { N_TABLES = 255 };
75 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
76 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
87 * - Do include packets and bytes from facets that have been deleted or
88 * whose own statistics have been folded into the rule.
90 * - Do include packets and bytes sent "by hand" that were accounted to
91 * the rule without any facet being involved (this is a rare corner
92 * case in rule_execute()).
94 * - Do not include packet or bytes that can be obtained from any facet's
95 * packet_count or byte_count member or that can be obtained from the
96 * datapath by, e.g., dpif_flow_get() for any subfacet.
98 uint64_t packet_count; /* Number of packets received. */
99 uint64_t byte_count; /* Number of bytes received. */
101 tag_type tag; /* Caches rule_calculate_tag() result. */
103 struct list facets; /* List of "struct facet"s. */
106 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
108 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
111 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
112 const struct flow *);
113 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
116 static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto,
117 const struct flow *flow);
119 static void rule_credit_stats(struct rule_dpif *,
120 const struct dpif_flow_stats *);
121 static void flow_push_stats(struct rule_dpif *, const struct flow *,
122 const struct dpif_flow_stats *);
123 static tag_type rule_calculate_tag(const struct flow *,
124 const struct minimask *, uint32_t basis);
125 static void rule_invalidate(const struct rule_dpif *);
127 #define MAX_MIRRORS 32
128 typedef uint32_t mirror_mask_t;
129 #define MIRROR_MASK_C(X) UINT32_C(X)
130 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
132 struct ofproto_dpif *ofproto; /* Owning ofproto. */
133 size_t idx; /* In ofproto's "mirrors" array. */
134 void *aux; /* Key supplied by ofproto's client. */
135 char *name; /* Identifier for log messages. */
137 /* Selection criteria. */
138 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
139 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
140 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
142 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
143 struct ofbundle *out; /* Output port or NULL. */
144 int out_vlan; /* Output VLAN or -1. */
145 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
148 int64_t packet_count; /* Number of packets sent. */
149 int64_t byte_count; /* Number of bytes sent. */
152 static void mirror_destroy(struct ofmirror *);
153 static void update_mirror_stats(struct ofproto_dpif *ofproto,
154 mirror_mask_t mirrors,
155 uint64_t packets, uint64_t bytes);
158 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
159 struct ofproto_dpif *ofproto; /* Owning ofproto. */
160 void *aux; /* Key supplied by ofproto's client. */
161 char *name; /* Identifier for log messages. */
164 struct list ports; /* Contains "struct ofport"s. */
165 enum port_vlan_mode vlan_mode; /* VLAN mode */
166 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
167 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
168 * NULL if all VLANs are trunked. */
169 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
170 struct bond *bond; /* Nonnull iff more than one port. */
171 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
174 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
176 /* Port mirroring info. */
177 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
178 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
179 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
182 static void bundle_remove(struct ofport *);
183 static void bundle_update(struct ofbundle *);
184 static void bundle_destroy(struct ofbundle *);
185 static void bundle_del_port(struct ofport_dpif *);
186 static void bundle_run(struct ofbundle *);
187 static void bundle_wait(struct ofbundle *);
188 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
189 uint16_t in_port, bool warn,
190 struct ofport_dpif **in_ofportp);
192 /* A controller may use OFPP_NONE as the ingress port to indicate that
193 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
194 * when an input bundle is needed for validation (e.g., mirroring or
195 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
196 * any 'port' structs, so care must be taken when dealing with it. */
197 static struct ofbundle ofpp_none_bundle = {
199 .vlan_mode = PORT_VLAN_TRUNK
202 static void stp_run(struct ofproto_dpif *ofproto);
203 static void stp_wait(struct ofproto_dpif *ofproto);
204 static int set_stp_port(struct ofport *,
205 const struct ofproto_port_stp_settings *);
207 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
209 struct action_xlate_ctx {
210 /* action_xlate_ctx_init() initializes these members. */
213 struct ofproto_dpif *ofproto;
215 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
216 * this flow when actions change header fields. */
219 /* The packet corresponding to 'flow', or a null pointer if we are
220 * revalidating without a packet to refer to. */
221 const struct ofpbuf *packet;
223 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
224 * actions update the flow table?
226 * We want to update these tables if we are actually processing a packet,
227 * or if we are accounting for packets that the datapath has processed, but
228 * not if we are just revalidating. */
231 /* The rule that we are currently translating, or NULL. */
232 struct rule_dpif *rule;
234 /* Union of the set of TCP flags seen so far in this flow. (Used only by
235 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
239 /* If nonnull, flow translation calls this function just before executing a
240 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
241 * when the recursion depth is exceeded.
243 * 'rule' is the rule being submitted into. It will be null if the
244 * resubmit or OFPP_TABLE action didn't find a matching rule.
246 * This is normally null so the client has to set it manually after
247 * calling action_xlate_ctx_init(). */
248 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule);
250 /* If nonnull, flow translation calls this function to report some
251 * significant decision, e.g. to explain why OFPP_NORMAL translation
252 * dropped a packet. */
253 void (*report_hook)(struct action_xlate_ctx *, const char *s);
255 /* If nonnull, flow translation credits the specified statistics to each
256 * rule reached through a resubmit or OFPP_TABLE action.
258 * This is normally null so the client has to set it manually after
259 * calling action_xlate_ctx_init(). */
260 const struct dpif_flow_stats *resubmit_stats;
262 /* xlate_actions() initializes and uses these members. The client might want
263 * to look at them after it returns. */
265 struct ofpbuf *odp_actions; /* Datapath actions. */
266 tag_type tags; /* Tags associated with actions. */
267 enum slow_path_reason slow; /* 0 if fast path may be used. */
268 bool has_learn; /* Actions include NXAST_LEARN? */
269 bool has_normal; /* Actions output to OFPP_NORMAL? */
270 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
271 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
272 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
274 /* xlate_actions() initializes and uses these members, but the client has no
275 * reason to look at them. */
277 int recurse; /* Recursion level, via xlate_table_action. */
278 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
279 struct flow base_flow; /* Flow at the last commit. */
280 uint32_t orig_skb_priority; /* Priority when packet arrived. */
281 uint8_t table_id; /* OpenFlow table ID where flow was found. */
282 uint32_t sflow_n_outputs; /* Number of output ports. */
283 uint32_t sflow_odp_port; /* Output port for composing sFlow action. */
284 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
285 bool exit; /* No further actions should be processed. */
288 static void action_xlate_ctx_init(struct action_xlate_ctx *,
289 struct ofproto_dpif *, const struct flow *,
290 ovs_be16 initial_tci, struct rule_dpif *,
291 uint8_t tcp_flags, const struct ofpbuf *);
292 static void xlate_actions(struct action_xlate_ctx *,
293 const struct ofpact *ofpacts, size_t ofpacts_len,
294 struct ofpbuf *odp_actions);
295 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
296 const struct ofpact *ofpacts,
298 static void xlate_table_action(struct action_xlate_ctx *, uint16_t in_port,
299 uint8_t table_id, bool may_packet_in);
301 static size_t put_userspace_action(const struct ofproto_dpif *,
302 struct ofpbuf *odp_actions,
304 const union user_action_cookie *);
306 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
307 enum slow_path_reason,
308 uint64_t *stub, size_t stub_size,
309 const struct nlattr **actionsp,
310 size_t *actions_lenp);
312 static void xlate_report(struct action_xlate_ctx *ctx, const char *s);
314 /* A subfacet (see "struct subfacet" below) has three possible installation
317 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
318 * case just after the subfacet is created, just before the subfacet is
319 * destroyed, or if the datapath returns an error when we try to install a
322 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
324 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
325 * ofproto_dpif is installed in the datapath.
328 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
329 SF_FAST_PATH, /* Full actions are installed. */
330 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
333 static const char *subfacet_path_to_string(enum subfacet_path);
335 /* A dpif flow and actions associated with a facet.
337 * See also the large comment on struct facet. */
340 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
341 struct list list_node; /* In struct facet's 'facets' list. */
342 struct facet *facet; /* Owning facet. */
346 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
347 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
348 * regenerate the ODP flow key from ->facet->flow. */
349 enum odp_key_fitness key_fitness;
353 long long int used; /* Time last used; time created if not used. */
355 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
356 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
360 * These should be essentially identical for every subfacet in a facet, but
361 * may differ in trivial ways due to VLAN splinters. */
362 size_t actions_len; /* Number of bytes in actions[]. */
363 struct nlattr *actions; /* Datapath actions. */
365 enum slow_path_reason slow; /* 0 if fast path may be used. */
366 enum subfacet_path path; /* Installed in datapath? */
368 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
369 * splinters can cause it to differ. This value should be removed when
370 * the VLAN splinters feature is no longer needed. */
371 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
373 /* Datapath port the packet arrived on. This is needed to remove
374 * flows for ports that are no longer part of the bridge. Since the
375 * flow definition only has the OpenFlow port number and the port is
376 * no longer part of the bridge, we can't determine the datapath port
377 * number needed to delete the flow from the datapath. */
378 uint32_t odp_in_port;
381 #define SUBFACET_DESTROY_MAX_BATCH 50
383 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
385 static struct subfacet *subfacet_find(struct ofproto_dpif *,
386 const struct nlattr *key, size_t key_len,
388 const struct flow *flow);
389 static void subfacet_destroy(struct subfacet *);
390 static void subfacet_destroy__(struct subfacet *);
391 static void subfacet_destroy_batch(struct ofproto_dpif *,
392 struct subfacet **, int n);
393 static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *,
395 static void subfacet_reset_dp_stats(struct subfacet *,
396 struct dpif_flow_stats *);
397 static void subfacet_update_time(struct subfacet *, long long int used);
398 static void subfacet_update_stats(struct subfacet *,
399 const struct dpif_flow_stats *);
400 static void subfacet_make_actions(struct subfacet *,
401 const struct ofpbuf *packet,
402 struct ofpbuf *odp_actions);
403 static int subfacet_install(struct subfacet *,
404 const struct nlattr *actions, size_t actions_len,
405 struct dpif_flow_stats *, enum slow_path_reason);
406 static void subfacet_uninstall(struct subfacet *);
408 static enum subfacet_path subfacet_want_path(enum slow_path_reason);
410 /* An exact-match instantiation of an OpenFlow flow.
412 * A facet associates a "struct flow", which represents the Open vSwitch
413 * userspace idea of an exact-match flow, with one or more subfacets. Each
414 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
415 * the facet. When the kernel module (or other dpif implementation) and Open
416 * vSwitch userspace agree on the definition of a flow key, there is exactly
417 * one subfacet per facet. If the dpif implementation supports more-specific
418 * flow matching than userspace, however, a facet can have more than one
419 * subfacet, each of which corresponds to some distinction in flow that
420 * userspace simply doesn't understand.
422 * Flow expiration works in terms of subfacets, so a facet must have at least
423 * one subfacet or it will never expire, leaking memory. */
426 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
427 struct list list_node; /* In owning rule's 'facets' list. */
428 struct rule_dpif *rule; /* Owning rule. */
431 struct list subfacets;
432 long long int used; /* Time last used; time created if not used. */
439 * - Do include packets and bytes sent "by hand", e.g. with
442 * - Do include packets and bytes that were obtained from the datapath
443 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
444 * DPIF_FP_ZERO_STATS).
446 * - Do not include packets or bytes that can be obtained from the
447 * datapath for any existing subfacet.
449 uint64_t packet_count; /* Number of packets received. */
450 uint64_t byte_count; /* Number of bytes received. */
452 /* Resubmit statistics. */
453 uint64_t prev_packet_count; /* Number of packets from last stats push. */
454 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
455 long long int prev_used; /* Used time from last stats push. */
458 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
459 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
460 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
462 /* Properties of datapath actions.
464 * Every subfacet has its own actions because actions can differ slightly
465 * between splintered and non-splintered subfacets due to the VLAN tag
466 * being initially different (present vs. absent). All of them have these
467 * properties in common so we just store one copy of them here. */
468 bool has_learn; /* Actions include NXAST_LEARN? */
469 bool has_normal; /* Actions output to OFPP_NORMAL? */
470 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
471 tag_type tags; /* Tags that would require revalidation. */
472 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
474 /* Storage for a single subfacet, to reduce malloc() time and space
475 * overhead. (A facet always has at least one subfacet and in the common
476 * case has exactly one subfacet.) */
477 struct subfacet one_subfacet;
480 static struct facet *facet_create(struct rule_dpif *,
481 const struct flow *, uint32_t hash);
482 static void facet_remove(struct facet *);
483 static void facet_free(struct facet *);
485 static struct facet *facet_find(struct ofproto_dpif *,
486 const struct flow *, uint32_t hash);
487 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
488 const struct flow *, uint32_t hash);
489 static void facet_revalidate(struct facet *);
490 static bool facet_check_consistency(struct facet *);
492 static void facet_flush_stats(struct facet *);
494 static void facet_update_time(struct facet *, long long int used);
495 static void facet_reset_counters(struct facet *);
496 static void facet_push_stats(struct facet *);
497 static void facet_learn(struct facet *);
498 static void facet_account(struct facet *);
500 static bool facet_is_controller_flow(struct facet *);
503 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
507 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
508 struct list bundle_node; /* In struct ofbundle's "ports" list. */
509 struct cfm *cfm; /* Connectivity Fault Management, if any. */
510 tag_type tag; /* Tag associated with this port. */
511 bool may_enable; /* May be enabled in bonds. */
512 long long int carrier_seq; /* Carrier status changes. */
513 struct tnl_port *tnl_port; /* Tunnel handle, or null. */
516 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
517 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
518 long long int stp_state_entered;
520 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
522 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
524 * This is deprecated. It is only for compatibility with broken device
525 * drivers in old versions of Linux that do not properly support VLANs when
526 * VLAN devices are not used. When broken device drivers are no longer in
527 * widespread use, we will delete these interfaces. */
528 uint16_t realdev_ofp_port;
532 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
533 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
534 * traffic egressing the 'ofport' with that priority should be marked with. */
535 struct priority_to_dscp {
536 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
537 uint32_t priority; /* Priority of this queue (see struct flow). */
539 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
542 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
544 * This is deprecated. It is only for compatibility with broken device drivers
545 * in old versions of Linux that do not properly support VLANs when VLAN
546 * devices are not used. When broken device drivers are no longer in
547 * widespread use, we will delete these interfaces. */
548 struct vlan_splinter {
549 struct hmap_node realdev_vid_node;
550 struct hmap_node vlandev_node;
551 uint16_t realdev_ofp_port;
552 uint16_t vlandev_ofp_port;
556 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
557 uint32_t realdev, ovs_be16 vlan_tci);
558 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
559 static void vsp_remove(struct ofport_dpif *);
560 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
562 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
564 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
567 static struct ofport_dpif *
568 ofport_dpif_cast(const struct ofport *ofport)
570 ovs_assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
571 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
574 static void port_run(struct ofport_dpif *);
575 static void port_run_fast(struct ofport_dpif *);
576 static void port_wait(struct ofport_dpif *);
577 static int set_cfm(struct ofport *, const struct cfm_settings *);
578 static void ofport_clear_priorities(struct ofport_dpif *);
580 struct dpif_completion {
581 struct list list_node;
582 struct ofoperation *op;
585 /* Extra information about a classifier table.
586 * Currently used just for optimized flow revalidation. */
588 /* If either of these is nonnull, then this table has a form that allows
589 * flows to be tagged to avoid revalidating most flows for the most common
590 * kinds of flow table changes. */
591 struct cls_table *catchall_table; /* Table that wildcards all fields. */
592 struct cls_table *other_table; /* Table with any other wildcard set. */
593 uint32_t basis; /* Keeps each table's tags separate. */
596 /* Reasons that we might need to revalidate every facet, and corresponding
599 * A value of 0 means that there is no need to revalidate.
601 * It would be nice to have some cleaner way to integrate with coverage
602 * counters, but with only a few reasons I guess this is good enough for
604 enum revalidate_reason {
605 REV_RECONFIGURE = 1, /* Switch configuration changed. */
606 REV_STP, /* Spanning tree protocol port status change. */
607 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
608 REV_FLOW_TABLE, /* Flow table changed. */
609 REV_INCONSISTENCY /* Facet self-check failed. */
611 COVERAGE_DEFINE(rev_reconfigure);
612 COVERAGE_DEFINE(rev_stp);
613 COVERAGE_DEFINE(rev_port_toggled);
614 COVERAGE_DEFINE(rev_flow_table);
615 COVERAGE_DEFINE(rev_inconsistency);
617 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
618 * These are datapath flows which have no associated ofproto, if they did we
619 * would use facets. */
621 struct hmap_node hmap_node;
626 /* All datapaths of a given type share a single dpif backer instance. */
631 struct timer next_expiration;
632 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
634 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
636 /* Facet revalidation flags applying to facets which use this backer. */
637 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
638 struct tag_set revalidate_set; /* Revalidate only matching facets. */
640 struct hmap drop_keys; /* Set of dropped odp keys. */
643 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
644 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
646 static void drop_key_clear(struct dpif_backer *);
647 static struct ofport_dpif *
648 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
650 struct ofproto_dpif {
651 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
653 struct dpif_backer *backer;
655 /* Special OpenFlow rules. */
656 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
657 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
663 struct netflow *netflow;
664 struct dpif_sflow *sflow;
665 struct hmap bundles; /* Contains "struct ofbundle"s. */
666 struct mac_learning *ml;
667 struct ofmirror *mirrors[MAX_MIRRORS];
669 bool has_bonded_bundles;
673 struct hmap subfacets;
674 struct governor *governor;
677 struct table_dpif tables[N_TABLES];
679 /* Support for debugging async flow mods. */
680 struct list completions;
682 bool has_bundle_action; /* True when the first bundle action appears. */
683 struct netdev_stats stats; /* To account packets generated and consumed in
688 long long int stp_last_tick;
690 /* VLAN splinters. */
691 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
692 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
695 struct sset ports; /* Set of standard port names. */
696 struct sset ghost_ports; /* Ports with no datapath port. */
697 struct sset port_poll_set; /* Queued names for port_poll() reply. */
698 int port_poll_errno; /* Last errno for port_poll() reply. */
701 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
702 * for debugging the asynchronous flow_mod implementation.) */
705 /* All existing ofproto_dpif instances, indexed by ->up.name. */
706 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
708 static void ofproto_dpif_unixctl_init(void);
710 static struct ofproto_dpif *
711 ofproto_dpif_cast(const struct ofproto *ofproto)
713 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
714 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
717 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
719 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
721 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
722 const struct ofpbuf *, ovs_be16 initial_tci,
725 /* Packet processing. */
726 static void update_learning_table(struct ofproto_dpif *,
727 const struct flow *, int vlan,
730 #define FLOW_MISS_MAX_BATCH 50
731 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
733 /* Flow expiration. */
734 static int expire(struct dpif_backer *);
737 static void send_netflow_active_timeouts(struct ofproto_dpif *);
740 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
741 static size_t compose_sflow_action(const struct ofproto_dpif *,
742 struct ofpbuf *odp_actions,
743 const struct flow *, uint32_t odp_port);
744 static void add_mirror_actions(struct action_xlate_ctx *ctx,
745 const struct flow *flow);
746 /* Global variables. */
747 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
749 /* Initial mappings of port to bridge mappings. */
750 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
752 /* Factory functions. */
755 init(const struct shash *iface_hints)
757 struct shash_node *node;
759 /* Make a local copy, since we don't own 'iface_hints' elements. */
760 SHASH_FOR_EACH(node, iface_hints) {
761 const struct iface_hint *orig_hint = node->data;
762 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
764 new_hint->br_name = xstrdup(orig_hint->br_name);
765 new_hint->br_type = xstrdup(orig_hint->br_type);
766 new_hint->ofp_port = orig_hint->ofp_port;
768 shash_add(&init_ofp_ports, node->name, new_hint);
773 enumerate_types(struct sset *types)
775 dp_enumerate_types(types);
779 enumerate_names(const char *type, struct sset *names)
781 struct ofproto_dpif *ofproto;
784 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
785 if (strcmp(type, ofproto->up.type)) {
788 sset_add(names, ofproto->up.name);
795 del(const char *type, const char *name)
800 error = dpif_open(name, type, &dpif);
802 error = dpif_delete(dpif);
809 port_open_type(const char *datapath_type, const char *port_type)
811 return dpif_port_open_type(datapath_type, port_type);
814 /* Type functions. */
816 static struct ofproto_dpif *
817 lookup_ofproto_dpif_by_port_name(const char *name)
819 struct ofproto_dpif *ofproto;
821 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
822 if (sset_contains(&ofproto->ports, name)) {
831 type_run(const char *type)
833 struct dpif_backer *backer;
837 backer = shash_find_data(&all_dpif_backers, type);
839 /* This is not necessarily a problem, since backers are only
840 * created on demand. */
844 dpif_run(backer->dpif);
846 if (backer->need_revalidate
847 || !tag_set_is_empty(&backer->revalidate_set)) {
848 struct tag_set revalidate_set = backer->revalidate_set;
849 bool need_revalidate = backer->need_revalidate;
850 struct ofproto_dpif *ofproto;
851 struct simap_node *node;
852 struct simap tmp_backers;
854 /* Handle tunnel garbage collection. */
855 simap_init(&tmp_backers);
856 simap_swap(&backer->tnl_backers, &tmp_backers);
858 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
859 struct ofport_dpif *iter;
861 if (backer != ofproto->backer) {
865 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
868 if (!iter->tnl_port) {
872 dp_port = netdev_vport_get_dpif_port(iter->up.netdev);
873 node = simap_find(&tmp_backers, dp_port);
875 simap_put(&backer->tnl_backers, dp_port, node->data);
876 simap_delete(&tmp_backers, node);
877 node = simap_find(&backer->tnl_backers, dp_port);
879 node = simap_find(&backer->tnl_backers, dp_port);
881 uint32_t odp_port = UINT32_MAX;
883 if (!dpif_port_add(backer->dpif, iter->up.netdev,
885 simap_put(&backer->tnl_backers, dp_port, odp_port);
886 node = simap_find(&backer->tnl_backers, dp_port);
891 iter->odp_port = node ? node->data : OVSP_NONE;
892 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
894 backer->need_revalidate = REV_RECONFIGURE;
899 SIMAP_FOR_EACH (node, &tmp_backers) {
900 dpif_port_del(backer->dpif, node->data);
902 simap_destroy(&tmp_backers);
904 switch (backer->need_revalidate) {
905 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
906 case REV_STP: COVERAGE_INC(rev_stp); break;
907 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
908 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
909 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
912 if (backer->need_revalidate) {
913 /* Clear the drop_keys in case we should now be accepting some
914 * formerly dropped flows. */
915 drop_key_clear(backer);
918 /* Clear the revalidation flags. */
919 tag_set_init(&backer->revalidate_set);
920 backer->need_revalidate = 0;
922 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
925 if (ofproto->backer != backer) {
929 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
931 || tag_set_intersects(&revalidate_set, facet->tags)) {
932 facet_revalidate(facet);
938 if (timer_expired(&backer->next_expiration)) {
939 int delay = expire(backer);
940 timer_set_duration(&backer->next_expiration, delay);
943 /* Check for port changes in the dpif. */
944 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
945 struct ofproto_dpif *ofproto;
946 struct dpif_port port;
948 /* Don't report on the datapath's device. */
949 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
953 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
954 &all_ofproto_dpifs) {
955 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
960 ofproto = lookup_ofproto_dpif_by_port_name(devname);
961 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
962 /* The port was removed. If we know the datapath,
963 * report it through poll_set(). If we don't, it may be
964 * notifying us of a removal we initiated, so ignore it.
965 * If there's a pending ENOBUFS, let it stand, since
966 * everything will be reevaluated. */
967 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
968 sset_add(&ofproto->port_poll_set, devname);
969 ofproto->port_poll_errno = 0;
971 } else if (!ofproto) {
972 /* The port was added, but we don't know with which
973 * ofproto we should associate it. Delete it. */
974 dpif_port_del(backer->dpif, port.port_no);
976 dpif_port_destroy(&port);
982 if (error != EAGAIN) {
983 struct ofproto_dpif *ofproto;
985 /* There was some sort of error, so propagate it to all
986 * ofprotos that use this backer. */
987 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
988 &all_ofproto_dpifs) {
989 if (ofproto->backer == backer) {
990 sset_clear(&ofproto->port_poll_set);
991 ofproto->port_poll_errno = error;
1000 type_run_fast(const char *type)
1002 struct dpif_backer *backer;
1005 backer = shash_find_data(&all_dpif_backers, type);
1007 /* This is not necessarily a problem, since backers are only
1008 * created on demand. */
1012 /* Handle one or more batches of upcalls, until there's nothing left to do
1013 * or until we do a fixed total amount of work.
1015 * We do work in batches because it can be much cheaper to set up a number
1016 * of flows and fire off their patches all at once. We do multiple batches
1017 * because in some cases handling a packet can cause another packet to be
1018 * queued almost immediately as part of the return flow. Both
1019 * optimizations can make major improvements on some benchmarks and
1020 * presumably for real traffic as well. */
1022 while (work < FLOW_MISS_MAX_BATCH) {
1023 int retval = handle_upcalls(backer, FLOW_MISS_MAX_BATCH - work);
1034 type_wait(const char *type)
1036 struct dpif_backer *backer;
1038 backer = shash_find_data(&all_dpif_backers, type);
1040 /* This is not necessarily a problem, since backers are only
1041 * created on demand. */
1045 timer_wait(&backer->next_expiration);
1048 /* Basic life-cycle. */
1050 static int add_internal_flows(struct ofproto_dpif *);
1052 static struct ofproto *
1055 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1056 return &ofproto->up;
1060 dealloc(struct ofproto *ofproto_)
1062 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1067 close_dpif_backer(struct dpif_backer *backer)
1069 struct shash_node *node;
1071 ovs_assert(backer->refcount > 0);
1073 if (--backer->refcount) {
1077 drop_key_clear(backer);
1078 hmap_destroy(&backer->drop_keys);
1080 simap_destroy(&backer->tnl_backers);
1081 hmap_destroy(&backer->odp_to_ofport_map);
1082 node = shash_find(&all_dpif_backers, backer->type);
1084 shash_delete(&all_dpif_backers, node);
1085 dpif_close(backer->dpif);
1090 /* Datapath port slated for removal from datapath. */
1091 struct odp_garbage {
1092 struct list list_node;
1097 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1099 struct dpif_backer *backer;
1100 struct dpif_port_dump port_dump;
1101 struct dpif_port port;
1102 struct shash_node *node;
1103 struct list garbage_list;
1104 struct odp_garbage *garbage, *next;
1110 backer = shash_find_data(&all_dpif_backers, type);
1117 backer_name = xasprintf("ovs-%s", type);
1119 /* Remove any existing datapaths, since we assume we're the only
1120 * userspace controlling the datapath. */
1122 dp_enumerate_names(type, &names);
1123 SSET_FOR_EACH(name, &names) {
1124 struct dpif *old_dpif;
1126 /* Don't remove our backer if it exists. */
1127 if (!strcmp(name, backer_name)) {
1131 if (dpif_open(name, type, &old_dpif)) {
1132 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1134 dpif_delete(old_dpif);
1135 dpif_close(old_dpif);
1138 sset_destroy(&names);
1140 backer = xmalloc(sizeof *backer);
1142 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1145 VLOG_ERR("failed to open datapath of type %s: %s", type,
1151 backer->type = xstrdup(type);
1152 backer->refcount = 1;
1153 hmap_init(&backer->odp_to_ofport_map);
1154 hmap_init(&backer->drop_keys);
1155 timer_set_duration(&backer->next_expiration, 1000);
1156 backer->need_revalidate = 0;
1157 simap_init(&backer->tnl_backers);
1158 tag_set_init(&backer->revalidate_set);
1161 dpif_flow_flush(backer->dpif);
1163 /* Loop through the ports already on the datapath and remove any
1164 * that we don't need anymore. */
1165 list_init(&garbage_list);
1166 dpif_port_dump_start(&port_dump, backer->dpif);
1167 while (dpif_port_dump_next(&port_dump, &port)) {
1168 node = shash_find(&init_ofp_ports, port.name);
1169 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1170 garbage = xmalloc(sizeof *garbage);
1171 garbage->odp_port = port.port_no;
1172 list_push_front(&garbage_list, &garbage->list_node);
1175 dpif_port_dump_done(&port_dump);
1177 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1178 dpif_port_del(backer->dpif, garbage->odp_port);
1179 list_remove(&garbage->list_node);
1183 shash_add(&all_dpif_backers, type, backer);
1185 error = dpif_recv_set(backer->dpif, true);
1187 VLOG_ERR("failed to listen on datapath of type %s: %s",
1188 type, strerror(error));
1189 close_dpif_backer(backer);
1197 construct(struct ofproto *ofproto_)
1199 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1200 struct shash_node *node, *next;
1205 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1210 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1211 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1213 ofproto->n_matches = 0;
1215 ofproto->netflow = NULL;
1216 ofproto->sflow = NULL;
1217 ofproto->stp = NULL;
1218 hmap_init(&ofproto->bundles);
1219 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1220 for (i = 0; i < MAX_MIRRORS; i++) {
1221 ofproto->mirrors[i] = NULL;
1223 ofproto->has_bonded_bundles = false;
1225 hmap_init(&ofproto->facets);
1226 hmap_init(&ofproto->subfacets);
1227 ofproto->governor = NULL;
1229 for (i = 0; i < N_TABLES; i++) {
1230 struct table_dpif *table = &ofproto->tables[i];
1232 table->catchall_table = NULL;
1233 table->other_table = NULL;
1234 table->basis = random_uint32();
1237 list_init(&ofproto->completions);
1239 ofproto_dpif_unixctl_init();
1241 ofproto->has_mirrors = false;
1242 ofproto->has_bundle_action = false;
1244 hmap_init(&ofproto->vlandev_map);
1245 hmap_init(&ofproto->realdev_vid_map);
1247 sset_init(&ofproto->ports);
1248 sset_init(&ofproto->ghost_ports);
1249 sset_init(&ofproto->port_poll_set);
1250 ofproto->port_poll_errno = 0;
1252 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1253 struct iface_hint *iface_hint = node->data;
1255 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1256 /* Check if the datapath already has this port. */
1257 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1258 sset_add(&ofproto->ports, node->name);
1261 free(iface_hint->br_name);
1262 free(iface_hint->br_type);
1264 shash_delete(&init_ofp_ports, node);
1268 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1269 hash_string(ofproto->up.name, 0));
1270 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1272 ofproto_init_tables(ofproto_, N_TABLES);
1273 error = add_internal_flows(ofproto);
1274 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1280 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1281 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1283 struct ofputil_flow_mod fm;
1286 match_init_catchall(&fm.match);
1288 match_set_reg(&fm.match, 0, id);
1289 fm.new_cookie = htonll(0);
1290 fm.cookie = htonll(0);
1291 fm.cookie_mask = htonll(0);
1292 fm.table_id = TBL_INTERNAL;
1293 fm.command = OFPFC_ADD;
1294 fm.idle_timeout = 0;
1295 fm.hard_timeout = 0;
1299 fm.ofpacts = ofpacts->data;
1300 fm.ofpacts_len = ofpacts->size;
1302 error = ofproto_flow_mod(&ofproto->up, &fm);
1304 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1305 id, ofperr_to_string(error));
1309 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1310 ovs_assert(*rulep != NULL);
1316 add_internal_flows(struct ofproto_dpif *ofproto)
1318 struct ofpact_controller *controller;
1319 uint64_t ofpacts_stub[128 / 8];
1320 struct ofpbuf ofpacts;
1324 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1327 controller = ofpact_put_CONTROLLER(&ofpacts);
1328 controller->max_len = UINT16_MAX;
1329 controller->controller_id = 0;
1330 controller->reason = OFPR_NO_MATCH;
1331 ofpact_pad(&ofpacts);
1333 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1338 ofpbuf_clear(&ofpacts);
1339 error = add_internal_flow(ofproto, id++, &ofpacts,
1340 &ofproto->no_packet_in_rule);
1345 complete_operations(struct ofproto_dpif *ofproto)
1347 struct dpif_completion *c, *next;
1349 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1350 ofoperation_complete(c->op, 0);
1351 list_remove(&c->list_node);
1357 destruct(struct ofproto *ofproto_)
1359 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1360 struct rule_dpif *rule, *next_rule;
1361 struct oftable *table;
1364 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1365 complete_operations(ofproto);
1367 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1368 struct cls_cursor cursor;
1370 cls_cursor_init(&cursor, &table->cls, NULL);
1371 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1372 ofproto_rule_destroy(&rule->up);
1376 for (i = 0; i < MAX_MIRRORS; i++) {
1377 mirror_destroy(ofproto->mirrors[i]);
1380 netflow_destroy(ofproto->netflow);
1381 dpif_sflow_destroy(ofproto->sflow);
1382 hmap_destroy(&ofproto->bundles);
1383 mac_learning_destroy(ofproto->ml);
1385 hmap_destroy(&ofproto->facets);
1386 hmap_destroy(&ofproto->subfacets);
1387 governor_destroy(ofproto->governor);
1389 hmap_destroy(&ofproto->vlandev_map);
1390 hmap_destroy(&ofproto->realdev_vid_map);
1392 sset_destroy(&ofproto->ports);
1393 sset_destroy(&ofproto->ghost_ports);
1394 sset_destroy(&ofproto->port_poll_set);
1396 close_dpif_backer(ofproto->backer);
1400 run_fast(struct ofproto *ofproto_)
1402 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1403 struct ofport_dpif *ofport;
1405 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1406 port_run_fast(ofport);
1413 run(struct ofproto *ofproto_)
1415 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1416 struct ofport_dpif *ofport;
1417 struct ofbundle *bundle;
1421 complete_operations(ofproto);
1424 error = run_fast(ofproto_);
1429 if (ofproto->netflow) {
1430 if (netflow_run(ofproto->netflow)) {
1431 send_netflow_active_timeouts(ofproto);
1434 if (ofproto->sflow) {
1435 dpif_sflow_run(ofproto->sflow);
1438 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1441 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1446 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1448 /* Check the consistency of a random facet, to aid debugging. */
1449 if (!hmap_is_empty(&ofproto->facets)
1450 && !ofproto->backer->need_revalidate) {
1451 struct facet *facet;
1453 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1454 struct facet, hmap_node);
1455 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1457 if (!facet_check_consistency(facet)) {
1458 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1463 if (ofproto->governor) {
1466 governor_run(ofproto->governor);
1468 /* If the governor has shrunk to its minimum size and the number of
1469 * subfacets has dwindled, then drop the governor entirely.
1471 * For hysteresis, the number of subfacets to drop the governor is
1472 * smaller than the number needed to trigger its creation. */
1473 n_subfacets = hmap_count(&ofproto->subfacets);
1474 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1475 && governor_is_idle(ofproto->governor)) {
1476 governor_destroy(ofproto->governor);
1477 ofproto->governor = NULL;
1485 wait(struct ofproto *ofproto_)
1487 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1488 struct ofport_dpif *ofport;
1489 struct ofbundle *bundle;
1491 if (!clogged && !list_is_empty(&ofproto->completions)) {
1492 poll_immediate_wake();
1495 dpif_wait(ofproto->backer->dpif);
1496 dpif_recv_wait(ofproto->backer->dpif);
1497 if (ofproto->sflow) {
1498 dpif_sflow_wait(ofproto->sflow);
1500 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1501 poll_immediate_wake();
1503 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1506 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1507 bundle_wait(bundle);
1509 if (ofproto->netflow) {
1510 netflow_wait(ofproto->netflow);
1512 mac_learning_wait(ofproto->ml);
1514 if (ofproto->backer->need_revalidate) {
1515 /* Shouldn't happen, but if it does just go around again. */
1516 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1517 poll_immediate_wake();
1519 if (ofproto->governor) {
1520 governor_wait(ofproto->governor);
1525 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1527 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1529 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1530 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1534 flush(struct ofproto *ofproto_)
1536 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1537 struct subfacet *subfacet, *next_subfacet;
1538 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1542 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1543 &ofproto->subfacets) {
1544 if (subfacet->path != SF_NOT_INSTALLED) {
1545 batch[n_batch++] = subfacet;
1546 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1547 subfacet_destroy_batch(ofproto, batch, n_batch);
1551 subfacet_destroy(subfacet);
1556 subfacet_destroy_batch(ofproto, batch, n_batch);
1561 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1562 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1564 *arp_match_ip = true;
1565 *actions = (OFPUTIL_A_OUTPUT |
1566 OFPUTIL_A_SET_VLAN_VID |
1567 OFPUTIL_A_SET_VLAN_PCP |
1568 OFPUTIL_A_STRIP_VLAN |
1569 OFPUTIL_A_SET_DL_SRC |
1570 OFPUTIL_A_SET_DL_DST |
1571 OFPUTIL_A_SET_NW_SRC |
1572 OFPUTIL_A_SET_NW_DST |
1573 OFPUTIL_A_SET_NW_TOS |
1574 OFPUTIL_A_SET_TP_SRC |
1575 OFPUTIL_A_SET_TP_DST |
1580 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1582 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1583 struct dpif_dp_stats s;
1585 strcpy(ots->name, "classifier");
1587 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1589 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1590 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1593 static struct ofport *
1596 struct ofport_dpif *port = xmalloc(sizeof *port);
1601 port_dealloc(struct ofport *port_)
1603 struct ofport_dpif *port = ofport_dpif_cast(port_);
1608 port_construct(struct ofport *port_)
1610 struct ofport_dpif *port = ofport_dpif_cast(port_);
1611 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1612 const struct netdev *netdev = port->up.netdev;
1613 struct dpif_port dpif_port;
1616 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1617 port->bundle = NULL;
1619 port->tag = tag_create_random();
1620 port->may_enable = true;
1621 port->stp_port = NULL;
1622 port->stp_state = STP_DISABLED;
1623 port->tnl_port = NULL;
1624 hmap_init(&port->priorities);
1625 port->realdev_ofp_port = 0;
1626 port->vlandev_vid = 0;
1627 port->carrier_seq = netdev_get_carrier_resets(netdev);
1629 if (netdev_vport_is_patch(netdev)) {
1630 /* XXX By bailing out here, we don't do required sFlow work. */
1631 port->odp_port = OVSP_NONE;
1635 error = dpif_port_query_by_name(ofproto->backer->dpif,
1636 netdev_vport_get_dpif_port(netdev),
1642 port->odp_port = dpif_port.port_no;
1644 if (netdev_get_tunnel_config(netdev)) {
1645 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1647 /* Sanity-check that a mapping doesn't already exist. This
1648 * shouldn't happen for non-tunnel ports. */
1649 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1650 VLOG_ERR("port %s already has an OpenFlow port number",
1652 dpif_port_destroy(&dpif_port);
1656 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1657 hash_int(port->odp_port, 0));
1659 dpif_port_destroy(&dpif_port);
1661 if (ofproto->sflow) {
1662 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1669 port_destruct(struct ofport *port_)
1671 struct ofport_dpif *port = ofport_dpif_cast(port_);
1672 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1673 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1674 const char *devname = netdev_get_name(port->up.netdev);
1676 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1677 /* The underlying device is still there, so delete it. This
1678 * happens when the ofproto is being destroyed, since the caller
1679 * assumes that removal of attached ports will happen as part of
1681 if (!port->tnl_port) {
1682 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1684 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1687 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1688 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1691 tnl_port_del(port->tnl_port);
1692 sset_find_and_delete(&ofproto->ports, devname);
1693 sset_find_and_delete(&ofproto->ghost_ports, devname);
1694 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1695 bundle_remove(port_);
1696 set_cfm(port_, NULL);
1697 if (ofproto->sflow) {
1698 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1701 ofport_clear_priorities(port);
1702 hmap_destroy(&port->priorities);
1706 port_modified(struct ofport *port_)
1708 struct ofport_dpif *port = ofport_dpif_cast(port_);
1710 if (port->bundle && port->bundle->bond) {
1711 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1716 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1718 struct ofport_dpif *port = ofport_dpif_cast(port_);
1719 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1720 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1722 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1723 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1724 OFPUTIL_PC_NO_PACKET_IN)) {
1725 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1727 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1728 bundle_update(port->bundle);
1734 set_sflow(struct ofproto *ofproto_,
1735 const struct ofproto_sflow_options *sflow_options)
1737 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1738 struct dpif_sflow *ds = ofproto->sflow;
1740 if (sflow_options) {
1742 struct ofport_dpif *ofport;
1744 ds = ofproto->sflow = dpif_sflow_create();
1745 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1746 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1748 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1750 dpif_sflow_set_options(ds, sflow_options);
1753 dpif_sflow_destroy(ds);
1754 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1755 ofproto->sflow = NULL;
1762 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1764 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1771 struct ofproto_dpif *ofproto;
1773 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1774 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1775 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1778 if (cfm_configure(ofport->cfm, s)) {
1784 cfm_destroy(ofport->cfm);
1790 get_cfm_fault(const struct ofport *ofport_)
1792 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1794 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1798 get_cfm_opup(const struct ofport *ofport_)
1800 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1802 return ofport->cfm ? cfm_get_opup(ofport->cfm) : -1;
1806 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1809 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1812 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1820 get_cfm_health(const struct ofport *ofport_)
1822 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1824 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1827 /* Spanning Tree. */
1830 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1832 struct ofproto_dpif *ofproto = ofproto_;
1833 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1834 struct ofport_dpif *ofport;
1836 ofport = stp_port_get_aux(sp);
1838 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1839 ofproto->up.name, port_num);
1841 struct eth_header *eth = pkt->l2;
1843 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1844 if (eth_addr_is_zero(eth->eth_src)) {
1845 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1846 "with unknown MAC", ofproto->up.name, port_num);
1848 send_packet(ofport, pkt);
1854 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1856 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1858 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1860 /* Only revalidate flows if the configuration changed. */
1861 if (!s != !ofproto->stp) {
1862 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1866 if (!ofproto->stp) {
1867 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1868 send_bpdu_cb, ofproto);
1869 ofproto->stp_last_tick = time_msec();
1872 stp_set_bridge_id(ofproto->stp, s->system_id);
1873 stp_set_bridge_priority(ofproto->stp, s->priority);
1874 stp_set_hello_time(ofproto->stp, s->hello_time);
1875 stp_set_max_age(ofproto->stp, s->max_age);
1876 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1878 struct ofport *ofport;
1880 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1881 set_stp_port(ofport, NULL);
1884 stp_destroy(ofproto->stp);
1885 ofproto->stp = NULL;
1892 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1894 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1898 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1899 s->designated_root = stp_get_designated_root(ofproto->stp);
1900 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1909 update_stp_port_state(struct ofport_dpif *ofport)
1911 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1912 enum stp_state state;
1914 /* Figure out new state. */
1915 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1919 if (ofport->stp_state != state) {
1920 enum ofputil_port_state of_state;
1923 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1924 netdev_get_name(ofport->up.netdev),
1925 stp_state_name(ofport->stp_state),
1926 stp_state_name(state));
1927 if (stp_learn_in_state(ofport->stp_state)
1928 != stp_learn_in_state(state)) {
1929 /* xxx Learning action flows should also be flushed. */
1930 mac_learning_flush(ofproto->ml,
1931 &ofproto->backer->revalidate_set);
1933 fwd_change = stp_forward_in_state(ofport->stp_state)
1934 != stp_forward_in_state(state);
1936 ofproto->backer->need_revalidate = REV_STP;
1937 ofport->stp_state = state;
1938 ofport->stp_state_entered = time_msec();
1940 if (fwd_change && ofport->bundle) {
1941 bundle_update(ofport->bundle);
1944 /* Update the STP state bits in the OpenFlow port description. */
1945 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1946 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1947 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1948 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1949 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1951 ofproto_port_set_state(&ofport->up, of_state);
1955 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1956 * caller is responsible for assigning STP port numbers and ensuring
1957 * there are no duplicates. */
1959 set_stp_port(struct ofport *ofport_,
1960 const struct ofproto_port_stp_settings *s)
1962 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1963 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1964 struct stp_port *sp = ofport->stp_port;
1966 if (!s || !s->enable) {
1968 ofport->stp_port = NULL;
1969 stp_port_disable(sp);
1970 update_stp_port_state(ofport);
1973 } else if (sp && stp_port_no(sp) != s->port_num
1974 && ofport == stp_port_get_aux(sp)) {
1975 /* The port-id changed, so disable the old one if it's not
1976 * already in use by another port. */
1977 stp_port_disable(sp);
1980 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1981 stp_port_enable(sp);
1983 stp_port_set_aux(sp, ofport);
1984 stp_port_set_priority(sp, s->priority);
1985 stp_port_set_path_cost(sp, s->path_cost);
1987 update_stp_port_state(ofport);
1993 get_stp_port_status(struct ofport *ofport_,
1994 struct ofproto_port_stp_status *s)
1996 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1997 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1998 struct stp_port *sp = ofport->stp_port;
2000 if (!ofproto->stp || !sp) {
2006 s->port_id = stp_port_get_id(sp);
2007 s->state = stp_port_get_state(sp);
2008 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2009 s->role = stp_port_get_role(sp);
2010 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2016 stp_run(struct ofproto_dpif *ofproto)
2019 long long int now = time_msec();
2020 long long int elapsed = now - ofproto->stp_last_tick;
2021 struct stp_port *sp;
2024 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2025 ofproto->stp_last_tick = now;
2027 while (stp_get_changed_port(ofproto->stp, &sp)) {
2028 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2031 update_stp_port_state(ofport);
2035 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2036 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2042 stp_wait(struct ofproto_dpif *ofproto)
2045 poll_timer_wait(1000);
2049 /* Returns true if STP should process 'flow'. */
2051 stp_should_process_flow(const struct flow *flow)
2053 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2057 stp_process_packet(const struct ofport_dpif *ofport,
2058 const struct ofpbuf *packet)
2060 struct ofpbuf payload = *packet;
2061 struct eth_header *eth = payload.data;
2062 struct stp_port *sp = ofport->stp_port;
2064 /* Sink packets on ports that have STP disabled when the bridge has
2066 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2070 /* Trim off padding on payload. */
2071 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2072 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2075 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2076 stp_received_bpdu(sp, payload.data, payload.size);
2080 static struct priority_to_dscp *
2081 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2083 struct priority_to_dscp *pdscp;
2086 hash = hash_int(priority, 0);
2087 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2088 if (pdscp->priority == priority) {
2096 ofport_clear_priorities(struct ofport_dpif *ofport)
2098 struct priority_to_dscp *pdscp, *next;
2100 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2101 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2107 set_queues(struct ofport *ofport_,
2108 const struct ofproto_port_queue *qdscp_list,
2111 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2112 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2113 struct hmap new = HMAP_INITIALIZER(&new);
2116 for (i = 0; i < n_qdscp; i++) {
2117 struct priority_to_dscp *pdscp;
2121 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2122 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2127 pdscp = get_priority(ofport, priority);
2129 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2131 pdscp = xmalloc(sizeof *pdscp);
2132 pdscp->priority = priority;
2134 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2137 if (pdscp->dscp != dscp) {
2139 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2142 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2145 if (!hmap_is_empty(&ofport->priorities)) {
2146 ofport_clear_priorities(ofport);
2147 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2150 hmap_swap(&new, &ofport->priorities);
2158 /* Expires all MAC learning entries associated with 'bundle' and forces its
2159 * ofproto to revalidate every flow.
2161 * Normally MAC learning entries are removed only from the ofproto associated
2162 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2163 * are removed from every ofproto. When patch ports and SLB bonds are in use
2164 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2165 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2166 * with the host from which it migrated. */
2168 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2170 struct ofproto_dpif *ofproto = bundle->ofproto;
2171 struct mac_learning *ml = ofproto->ml;
2172 struct mac_entry *mac, *next_mac;
2174 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2175 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2176 if (mac->port.p == bundle) {
2178 struct ofproto_dpif *o;
2180 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2182 struct mac_entry *e;
2184 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2187 mac_learning_expire(o->ml, e);
2193 mac_learning_expire(ml, mac);
2198 static struct ofbundle *
2199 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2201 struct ofbundle *bundle;
2203 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2204 &ofproto->bundles) {
2205 if (bundle->aux == aux) {
2212 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2213 * ones that are found to 'bundles'. */
2215 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2216 void **auxes, size_t n_auxes,
2217 struct hmapx *bundles)
2221 hmapx_init(bundles);
2222 for (i = 0; i < n_auxes; i++) {
2223 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2225 hmapx_add(bundles, bundle);
2231 bundle_update(struct ofbundle *bundle)
2233 struct ofport_dpif *port;
2235 bundle->floodable = true;
2236 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2237 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2238 || !stp_forward_in_state(port->stp_state)) {
2239 bundle->floodable = false;
2246 bundle_del_port(struct ofport_dpif *port)
2248 struct ofbundle *bundle = port->bundle;
2250 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2252 list_remove(&port->bundle_node);
2253 port->bundle = NULL;
2256 lacp_slave_unregister(bundle->lacp, port);
2259 bond_slave_unregister(bundle->bond, port);
2262 bundle_update(bundle);
2266 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2267 struct lacp_slave_settings *lacp)
2269 struct ofport_dpif *port;
2271 port = get_ofp_port(bundle->ofproto, ofp_port);
2276 if (port->bundle != bundle) {
2277 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2279 bundle_del_port(port);
2282 port->bundle = bundle;
2283 list_push_back(&bundle->ports, &port->bundle_node);
2284 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2285 || !stp_forward_in_state(port->stp_state)) {
2286 bundle->floodable = false;
2290 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2291 lacp_slave_register(bundle->lacp, port, lacp);
2298 bundle_destroy(struct ofbundle *bundle)
2300 struct ofproto_dpif *ofproto;
2301 struct ofport_dpif *port, *next_port;
2308 ofproto = bundle->ofproto;
2309 for (i = 0; i < MAX_MIRRORS; i++) {
2310 struct ofmirror *m = ofproto->mirrors[i];
2312 if (m->out == bundle) {
2314 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2315 || hmapx_find_and_delete(&m->dsts, bundle)) {
2316 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2321 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2322 bundle_del_port(port);
2325 bundle_flush_macs(bundle, true);
2326 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2328 free(bundle->trunks);
2329 lacp_destroy(bundle->lacp);
2330 bond_destroy(bundle->bond);
2335 bundle_set(struct ofproto *ofproto_, void *aux,
2336 const struct ofproto_bundle_settings *s)
2338 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2339 bool need_flush = false;
2340 struct ofport_dpif *port;
2341 struct ofbundle *bundle;
2342 unsigned long *trunks;
2348 bundle_destroy(bundle_lookup(ofproto, aux));
2352 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2353 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2355 bundle = bundle_lookup(ofproto, aux);
2357 bundle = xmalloc(sizeof *bundle);
2359 bundle->ofproto = ofproto;
2360 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2361 hash_pointer(aux, 0));
2363 bundle->name = NULL;
2365 list_init(&bundle->ports);
2366 bundle->vlan_mode = PORT_VLAN_TRUNK;
2368 bundle->trunks = NULL;
2369 bundle->use_priority_tags = s->use_priority_tags;
2370 bundle->lacp = NULL;
2371 bundle->bond = NULL;
2373 bundle->floodable = true;
2375 bundle->src_mirrors = 0;
2376 bundle->dst_mirrors = 0;
2377 bundle->mirror_out = 0;
2380 if (!bundle->name || strcmp(s->name, bundle->name)) {
2382 bundle->name = xstrdup(s->name);
2387 if (!bundle->lacp) {
2388 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2389 bundle->lacp = lacp_create();
2391 lacp_configure(bundle->lacp, s->lacp);
2393 lacp_destroy(bundle->lacp);
2394 bundle->lacp = NULL;
2397 /* Update set of ports. */
2399 for (i = 0; i < s->n_slaves; i++) {
2400 if (!bundle_add_port(bundle, s->slaves[i],
2401 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2405 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2406 struct ofport_dpif *next_port;
2408 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2409 for (i = 0; i < s->n_slaves; i++) {
2410 if (s->slaves[i] == port->up.ofp_port) {
2415 bundle_del_port(port);
2419 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2421 if (list_is_empty(&bundle->ports)) {
2422 bundle_destroy(bundle);
2426 /* Set VLAN tagging mode */
2427 if (s->vlan_mode != bundle->vlan_mode
2428 || s->use_priority_tags != bundle->use_priority_tags) {
2429 bundle->vlan_mode = s->vlan_mode;
2430 bundle->use_priority_tags = s->use_priority_tags;
2435 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2436 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2438 if (vlan != bundle->vlan) {
2439 bundle->vlan = vlan;
2443 /* Get trunked VLANs. */
2444 switch (s->vlan_mode) {
2445 case PORT_VLAN_ACCESS:
2449 case PORT_VLAN_TRUNK:
2450 trunks = CONST_CAST(unsigned long *, s->trunks);
2453 case PORT_VLAN_NATIVE_UNTAGGED:
2454 case PORT_VLAN_NATIVE_TAGGED:
2455 if (vlan != 0 && (!s->trunks
2456 || !bitmap_is_set(s->trunks, vlan)
2457 || bitmap_is_set(s->trunks, 0))) {
2458 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2460 trunks = bitmap_clone(s->trunks, 4096);
2462 trunks = bitmap_allocate1(4096);
2464 bitmap_set1(trunks, vlan);
2465 bitmap_set0(trunks, 0);
2467 trunks = CONST_CAST(unsigned long *, s->trunks);
2474 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2475 free(bundle->trunks);
2476 if (trunks == s->trunks) {
2477 bundle->trunks = vlan_bitmap_clone(trunks);
2479 bundle->trunks = trunks;
2484 if (trunks != s->trunks) {
2489 if (!list_is_short(&bundle->ports)) {
2490 bundle->ofproto->has_bonded_bundles = true;
2492 if (bond_reconfigure(bundle->bond, s->bond)) {
2493 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2496 bundle->bond = bond_create(s->bond);
2497 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2500 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2501 bond_slave_register(bundle->bond, port, port->up.netdev);
2504 bond_destroy(bundle->bond);
2505 bundle->bond = NULL;
2508 /* If we changed something that would affect MAC learning, un-learn
2509 * everything on this port and force flow revalidation. */
2511 bundle_flush_macs(bundle, false);
2518 bundle_remove(struct ofport *port_)
2520 struct ofport_dpif *port = ofport_dpif_cast(port_);
2521 struct ofbundle *bundle = port->bundle;
2524 bundle_del_port(port);
2525 if (list_is_empty(&bundle->ports)) {
2526 bundle_destroy(bundle);
2527 } else if (list_is_short(&bundle->ports)) {
2528 bond_destroy(bundle->bond);
2529 bundle->bond = NULL;
2535 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2537 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2538 struct ofport_dpif *port = port_;
2539 uint8_t ea[ETH_ADDR_LEN];
2542 error = netdev_get_etheraddr(port->up.netdev, ea);
2544 struct ofpbuf packet;
2547 ofpbuf_init(&packet, 0);
2548 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2550 memcpy(packet_pdu, pdu, pdu_size);
2552 send_packet(port, &packet);
2553 ofpbuf_uninit(&packet);
2555 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2556 "%s (%s)", port->bundle->name,
2557 netdev_get_name(port->up.netdev), strerror(error));
2562 bundle_send_learning_packets(struct ofbundle *bundle)
2564 struct ofproto_dpif *ofproto = bundle->ofproto;
2565 int error, n_packets, n_errors;
2566 struct mac_entry *e;
2568 error = n_packets = n_errors = 0;
2569 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2570 if (e->port.p != bundle) {
2571 struct ofpbuf *learning_packet;
2572 struct ofport_dpif *port;
2576 /* The assignment to "port" is unnecessary but makes "grep"ing for
2577 * struct ofport_dpif more effective. */
2578 learning_packet = bond_compose_learning_packet(bundle->bond,
2582 ret = send_packet(port, learning_packet);
2583 ofpbuf_delete(learning_packet);
2593 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2594 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2595 "packets, last error was: %s",
2596 bundle->name, n_errors, n_packets, strerror(error));
2598 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2599 bundle->name, n_packets);
2604 bundle_run(struct ofbundle *bundle)
2607 lacp_run(bundle->lacp, send_pdu_cb);
2610 struct ofport_dpif *port;
2612 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2613 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2616 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2617 lacp_status(bundle->lacp));
2618 if (bond_should_send_learning_packets(bundle->bond)) {
2619 bundle_send_learning_packets(bundle);
2625 bundle_wait(struct ofbundle *bundle)
2628 lacp_wait(bundle->lacp);
2631 bond_wait(bundle->bond);
2638 mirror_scan(struct ofproto_dpif *ofproto)
2642 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2643 if (!ofproto->mirrors[idx]) {
2650 static struct ofmirror *
2651 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2655 for (i = 0; i < MAX_MIRRORS; i++) {
2656 struct ofmirror *mirror = ofproto->mirrors[i];
2657 if (mirror && mirror->aux == aux) {
2665 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2667 mirror_update_dups(struct ofproto_dpif *ofproto)
2671 for (i = 0; i < MAX_MIRRORS; i++) {
2672 struct ofmirror *m = ofproto->mirrors[i];
2675 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2679 for (i = 0; i < MAX_MIRRORS; i++) {
2680 struct ofmirror *m1 = ofproto->mirrors[i];
2687 for (j = i + 1; j < MAX_MIRRORS; j++) {
2688 struct ofmirror *m2 = ofproto->mirrors[j];
2690 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2691 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2692 m2->dup_mirrors |= m1->dup_mirrors;
2699 mirror_set(struct ofproto *ofproto_, void *aux,
2700 const struct ofproto_mirror_settings *s)
2702 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2703 mirror_mask_t mirror_bit;
2704 struct ofbundle *bundle;
2705 struct ofmirror *mirror;
2706 struct ofbundle *out;
2707 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2708 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2711 mirror = mirror_lookup(ofproto, aux);
2713 mirror_destroy(mirror);
2719 idx = mirror_scan(ofproto);
2721 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2723 ofproto->up.name, MAX_MIRRORS, s->name);
2727 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2728 mirror->ofproto = ofproto;
2731 mirror->out_vlan = -1;
2732 mirror->name = NULL;
2735 if (!mirror->name || strcmp(s->name, mirror->name)) {
2737 mirror->name = xstrdup(s->name);
2740 /* Get the new configuration. */
2741 if (s->out_bundle) {
2742 out = bundle_lookup(ofproto, s->out_bundle);
2744 mirror_destroy(mirror);
2750 out_vlan = s->out_vlan;
2752 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2753 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2755 /* If the configuration has not changed, do nothing. */
2756 if (hmapx_equals(&srcs, &mirror->srcs)
2757 && hmapx_equals(&dsts, &mirror->dsts)
2758 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2759 && mirror->out == out
2760 && mirror->out_vlan == out_vlan)
2762 hmapx_destroy(&srcs);
2763 hmapx_destroy(&dsts);
2767 hmapx_swap(&srcs, &mirror->srcs);
2768 hmapx_destroy(&srcs);
2770 hmapx_swap(&dsts, &mirror->dsts);
2771 hmapx_destroy(&dsts);
2773 free(mirror->vlans);
2774 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2777 mirror->out_vlan = out_vlan;
2779 /* Update bundles. */
2780 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2781 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2782 if (hmapx_contains(&mirror->srcs, bundle)) {
2783 bundle->src_mirrors |= mirror_bit;
2785 bundle->src_mirrors &= ~mirror_bit;
2788 if (hmapx_contains(&mirror->dsts, bundle)) {
2789 bundle->dst_mirrors |= mirror_bit;
2791 bundle->dst_mirrors &= ~mirror_bit;
2794 if (mirror->out == bundle) {
2795 bundle->mirror_out |= mirror_bit;
2797 bundle->mirror_out &= ~mirror_bit;
2801 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2802 ofproto->has_mirrors = true;
2803 mac_learning_flush(ofproto->ml,
2804 &ofproto->backer->revalidate_set);
2805 mirror_update_dups(ofproto);
2811 mirror_destroy(struct ofmirror *mirror)
2813 struct ofproto_dpif *ofproto;
2814 mirror_mask_t mirror_bit;
2815 struct ofbundle *bundle;
2822 ofproto = mirror->ofproto;
2823 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2824 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2826 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2827 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2828 bundle->src_mirrors &= ~mirror_bit;
2829 bundle->dst_mirrors &= ~mirror_bit;
2830 bundle->mirror_out &= ~mirror_bit;
2833 hmapx_destroy(&mirror->srcs);
2834 hmapx_destroy(&mirror->dsts);
2835 free(mirror->vlans);
2837 ofproto->mirrors[mirror->idx] = NULL;
2841 mirror_update_dups(ofproto);
2843 ofproto->has_mirrors = false;
2844 for (i = 0; i < MAX_MIRRORS; i++) {
2845 if (ofproto->mirrors[i]) {
2846 ofproto->has_mirrors = true;
2853 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2854 uint64_t *packets, uint64_t *bytes)
2856 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2857 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2860 *packets = *bytes = UINT64_MAX;
2864 *packets = mirror->packet_count;
2865 *bytes = mirror->byte_count;
2871 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2873 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2874 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2875 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2881 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2883 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2884 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2885 return bundle && bundle->mirror_out != 0;
2889 forward_bpdu_changed(struct ofproto *ofproto_)
2891 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2892 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2896 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2899 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2900 mac_learning_set_idle_time(ofproto->ml, idle_time);
2901 mac_learning_set_max_entries(ofproto->ml, max_entries);
2906 static struct ofport_dpif *
2907 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2909 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2910 return ofport ? ofport_dpif_cast(ofport) : NULL;
2913 static struct ofport_dpif *
2914 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2916 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2917 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2921 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2922 struct ofproto_port *ofproto_port,
2923 struct dpif_port *dpif_port)
2925 ofproto_port->name = dpif_port->name;
2926 ofproto_port->type = dpif_port->type;
2927 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2930 static struct ofport_dpif *
2931 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
2933 const struct ofproto_dpif *ofproto;
2936 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
2941 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2942 struct ofport *ofport;
2944 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
2945 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
2946 return ofport_dpif_cast(ofport);
2953 port_run_fast(struct ofport_dpif *ofport)
2955 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2956 struct ofpbuf packet;
2958 ofpbuf_init(&packet, 0);
2959 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2960 send_packet(ofport, &packet);
2961 ofpbuf_uninit(&packet);
2966 port_run(struct ofport_dpif *ofport)
2968 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2969 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2970 bool enable = netdev_get_carrier(ofport->up.netdev);
2972 ofport->carrier_seq = carrier_seq;
2974 port_run_fast(ofport);
2976 if (ofport->tnl_port
2977 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
2978 &ofport->tnl_port)) {
2979 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
2983 int cfm_opup = cfm_get_opup(ofport->cfm);
2985 cfm_run(ofport->cfm);
2986 enable = enable && !cfm_get_fault(ofport->cfm);
2988 if (cfm_opup >= 0) {
2989 enable = enable && cfm_opup;
2993 if (ofport->bundle) {
2994 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2995 if (carrier_changed) {
2996 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3000 if (ofport->may_enable != enable) {
3001 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3003 if (ofproto->has_bundle_action) {
3004 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3008 ofport->may_enable = enable;
3012 port_wait(struct ofport_dpif *ofport)
3015 cfm_wait(ofport->cfm);
3020 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3021 struct ofproto_port *ofproto_port)
3023 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3024 struct dpif_port dpif_port;
3027 if (sset_contains(&ofproto->ghost_ports, devname)) {
3028 const char *type = netdev_get_type_from_name(devname);
3030 /* We may be called before ofproto->up.port_by_name is populated with
3031 * the appropriate ofport. For this reason, we must get the name and
3032 * type from the netdev layer directly. */
3034 const struct ofport *ofport;
3036 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3037 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3038 ofproto_port->name = xstrdup(devname);
3039 ofproto_port->type = xstrdup(type);
3045 if (!sset_contains(&ofproto->ports, devname)) {
3048 error = dpif_port_query_by_name(ofproto->backer->dpif,
3049 devname, &dpif_port);
3051 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3057 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3059 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3060 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3061 const char *devname = netdev_get_name(netdev);
3063 if (netdev_vport_is_patch(netdev)) {
3064 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3068 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3069 uint32_t port_no = UINT32_MAX;
3072 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3076 if (netdev_get_tunnel_config(netdev)) {
3077 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3081 if (netdev_get_tunnel_config(netdev)) {
3082 sset_add(&ofproto->ghost_ports, devname);
3084 sset_add(&ofproto->ports, devname);
3090 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3092 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3093 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3100 sset_find_and_delete(&ofproto->ghost_ports,
3101 netdev_get_name(ofport->up.netdev));
3102 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3103 if (!ofport->tnl_port) {
3104 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3106 /* The caller is going to close ofport->up.netdev. If this is a
3107 * bonded port, then the bond is using that netdev, so remove it
3108 * from the bond. The client will need to reconfigure everything
3109 * after deleting ports, so then the slave will get re-added. */
3110 bundle_remove(&ofport->up);
3117 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3119 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3122 error = netdev_get_stats(ofport->up.netdev, stats);
3124 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3125 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3127 /* ofproto->stats.tx_packets represents packets that we created
3128 * internally and sent to some port (e.g. packets sent with
3129 * send_packet()). Account for them as if they had come from
3130 * OFPP_LOCAL and got forwarded. */
3132 if (stats->rx_packets != UINT64_MAX) {
3133 stats->rx_packets += ofproto->stats.tx_packets;
3136 if (stats->rx_bytes != UINT64_MAX) {
3137 stats->rx_bytes += ofproto->stats.tx_bytes;
3140 /* ofproto->stats.rx_packets represents packets that were received on
3141 * some port and we processed internally and dropped (e.g. STP).
3142 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3144 if (stats->tx_packets != UINT64_MAX) {
3145 stats->tx_packets += ofproto->stats.rx_packets;
3148 if (stats->tx_bytes != UINT64_MAX) {
3149 stats->tx_bytes += ofproto->stats.rx_bytes;
3156 /* Account packets for LOCAL port. */
3158 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
3159 size_t tx_size, size_t rx_size)
3161 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3164 ofproto->stats.rx_packets++;
3165 ofproto->stats.rx_bytes += rx_size;
3168 ofproto->stats.tx_packets++;
3169 ofproto->stats.tx_bytes += tx_size;
3173 struct port_dump_state {
3178 struct ofproto_port port;
3183 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3185 *statep = xzalloc(sizeof(struct port_dump_state));
3190 port_dump_next(const struct ofproto *ofproto_, void *state_,
3191 struct ofproto_port *port)
3193 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3194 struct port_dump_state *state = state_;
3195 const struct sset *sset;
3196 struct sset_node *node;
3198 if (state->has_port) {
3199 ofproto_port_destroy(&state->port);
3200 state->has_port = false;
3202 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3203 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3206 error = port_query_by_name(ofproto_, node->name, &state->port);
3208 *port = state->port;
3209 state->has_port = true;
3211 } else if (error != ENODEV) {
3216 if (!state->ghost) {
3217 state->ghost = true;
3220 return port_dump_next(ofproto_, state_, port);
3227 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3229 struct port_dump_state *state = state_;
3231 if (state->has_port) {
3232 ofproto_port_destroy(&state->port);
3239 port_poll(const struct ofproto *ofproto_, char **devnamep)
3241 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3243 if (ofproto->port_poll_errno) {
3244 int error = ofproto->port_poll_errno;
3245 ofproto->port_poll_errno = 0;
3249 if (sset_is_empty(&ofproto->port_poll_set)) {
3253 *devnamep = sset_pop(&ofproto->port_poll_set);
3258 port_poll_wait(const struct ofproto *ofproto_)
3260 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3261 dpif_port_poll_wait(ofproto->backer->dpif);
3265 port_is_lacp_current(const struct ofport *ofport_)
3267 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3268 return (ofport->bundle && ofport->bundle->lacp
3269 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3273 /* Upcall handling. */
3275 /* Flow miss batching.
3277 * Some dpifs implement operations faster when you hand them off in a batch.
3278 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3279 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3280 * more packets, plus possibly installing the flow in the dpif.
3282 * So far we only batch the operations that affect flow setup time the most.
3283 * It's possible to batch more than that, but the benefit might be minimal. */
3285 struct hmap_node hmap_node;
3286 struct ofproto_dpif *ofproto;
3288 enum odp_key_fitness key_fitness;
3289 const struct nlattr *key;
3291 ovs_be16 initial_tci;
3292 struct list packets;
3293 enum dpif_upcall_type upcall_type;
3294 uint32_t odp_in_port;
3297 struct flow_miss_op {
3298 struct dpif_op dpif_op;
3299 struct subfacet *subfacet; /* Subfacet */
3300 void *garbage; /* Pointer to pass to free(), NULL if none. */
3301 uint64_t stub[1024 / 8]; /* Temporary buffer. */
3304 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3305 * OpenFlow controller as necessary according to their individual
3306 * configurations. */
3308 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3309 const struct flow *flow)
3311 struct ofputil_packet_in pin;
3313 pin.packet = packet->data;
3314 pin.packet_len = packet->size;
3315 pin.reason = OFPR_NO_MATCH;
3316 pin.controller_id = 0;
3321 pin.send_len = 0; /* not used for flow table misses */
3323 flow_get_metadata(flow, &pin.fmd);
3325 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3328 static enum slow_path_reason
3329 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3330 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3334 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3336 cfm_process_heartbeat(ofport->cfm, packet);
3339 } else if (ofport->bundle && ofport->bundle->lacp
3340 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3342 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3345 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3347 stp_process_packet(ofport, packet);
3355 static struct flow_miss *
3356 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3357 const struct flow *flow, uint32_t hash)
3359 struct flow_miss *miss;
3361 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3362 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3370 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3371 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3372 * 'miss' is associated with a subfacet the caller must also initialize the
3373 * returned op->subfacet, and if anything needs to be freed after processing
3374 * the op, the caller must initialize op->garbage also. */
3376 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3377 struct flow_miss_op *op)
3379 if (miss->flow.vlan_tci != miss->initial_tci) {
3380 /* This packet was received on a VLAN splinter port. We
3381 * added a VLAN to the packet to make the packet resemble
3382 * the flow, but the actions were composed assuming that
3383 * the packet contained no VLAN. So, we must remove the
3384 * VLAN header from the packet before trying to execute the
3386 eth_pop_vlan(packet);
3389 op->subfacet = NULL;
3391 op->dpif_op.type = DPIF_OP_EXECUTE;
3392 op->dpif_op.u.execute.key = miss->key;
3393 op->dpif_op.u.execute.key_len = miss->key_len;
3394 op->dpif_op.u.execute.packet = packet;
3397 /* Helper for handle_flow_miss_without_facet() and
3398 * handle_flow_miss_with_facet(). */
3400 handle_flow_miss_common(struct rule_dpif *rule,
3401 struct ofpbuf *packet, const struct flow *flow)
3403 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3405 ofproto->n_matches++;
3407 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3409 * Extra-special case for fail-open mode.
3411 * We are in fail-open mode and the packet matched the fail-open
3412 * rule, but we are connected to a controller too. We should send
3413 * the packet up to the controller in the hope that it will try to
3414 * set up a flow and thereby allow us to exit fail-open.
3416 * See the top-level comment in fail-open.c for more information.
3418 send_packet_in_miss(ofproto, packet, flow);
3422 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3423 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3424 * installing a datapath flow. The answer is usually "yes" (a return value of
3425 * true). However, for short flows the cost of bookkeeping is much higher than
3426 * the benefits, so when the datapath holds a large number of flows we impose
3427 * some heuristics to decide which flows are likely to be worth tracking. */
3429 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3430 struct flow_miss *miss, uint32_t hash)
3432 if (!ofproto->governor) {
3435 n_subfacets = hmap_count(&ofproto->subfacets);
3436 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3440 ofproto->governor = governor_create(ofproto->up.name);
3443 return governor_should_install_flow(ofproto->governor, hash,
3444 list_size(&miss->packets));
3447 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3448 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3449 * increment '*n_ops'. */
3451 handle_flow_miss_without_facet(struct flow_miss *miss,
3452 struct rule_dpif *rule,
3453 struct flow_miss_op *ops, size_t *n_ops)
3455 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3456 long long int now = time_msec();
3457 struct action_xlate_ctx ctx;
3458 struct ofpbuf *packet;
3460 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3461 struct flow_miss_op *op = &ops[*n_ops];
3462 struct dpif_flow_stats stats;
3463 struct ofpbuf odp_actions;
3465 COVERAGE_INC(facet_suppress);
3467 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3469 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3470 rule_credit_stats(rule, &stats);
3472 action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci,
3474 ctx.resubmit_stats = &stats;
3475 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3478 if (odp_actions.size) {
3479 struct dpif_execute *execute = &op->dpif_op.u.execute;
3481 init_flow_miss_execute_op(miss, packet, op);
3482 execute->actions = odp_actions.data;
3483 execute->actions_len = odp_actions.size;
3484 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3488 ofpbuf_uninit(&odp_actions);
3493 /* Handles 'miss', which matches 'facet'. May add any required datapath
3494 * operations to 'ops', incrementing '*n_ops' for each new op.
3496 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3497 * This is really important only for new facets: if we just called time_msec()
3498 * here, then the new subfacet or its packets could look (occasionally) as
3499 * though it was used some time after the facet was used. That can make a
3500 * one-packet flow look like it has a nonzero duration, which looks odd in
3501 * e.g. NetFlow statistics. */
3503 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3505 struct flow_miss_op *ops, size_t *n_ops)
3507 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3508 enum subfacet_path want_path;
3509 struct subfacet *subfacet;
3510 struct ofpbuf *packet;
3512 subfacet = subfacet_create(facet, miss, now);
3514 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3515 struct flow_miss_op *op = &ops[*n_ops];
3516 struct dpif_flow_stats stats;
3517 struct ofpbuf odp_actions;
3519 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3521 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3522 if (!subfacet->actions || subfacet->slow) {
3523 subfacet_make_actions(subfacet, packet, &odp_actions);
3526 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3527 subfacet_update_stats(subfacet, &stats);
3529 if (subfacet->actions_len) {
3530 struct dpif_execute *execute = &op->dpif_op.u.execute;
3532 init_flow_miss_execute_op(miss, packet, op);
3533 op->subfacet = subfacet;
3534 if (!subfacet->slow) {
3535 execute->actions = subfacet->actions;
3536 execute->actions_len = subfacet->actions_len;
3537 ofpbuf_uninit(&odp_actions);
3539 execute->actions = odp_actions.data;
3540 execute->actions_len = odp_actions.size;
3541 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3546 ofpbuf_uninit(&odp_actions);
3550 want_path = subfacet_want_path(subfacet->slow);
3551 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3552 struct flow_miss_op *op = &ops[(*n_ops)++];
3553 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3555 op->subfacet = subfacet;
3557 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3558 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3559 put->key = miss->key;
3560 put->key_len = miss->key_len;
3561 if (want_path == SF_FAST_PATH) {
3562 put->actions = subfacet->actions;
3563 put->actions_len = subfacet->actions_len;
3565 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
3566 op->stub, sizeof op->stub,
3567 &put->actions, &put->actions_len);
3573 /* Handles flow miss 'miss'. May add any required datapath operations
3574 * to 'ops', incrementing '*n_ops' for each new op. */
3576 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3579 struct ofproto_dpif *ofproto = miss->ofproto;
3580 struct facet *facet;
3584 /* The caller must ensure that miss->hmap_node.hash contains
3585 * flow_hash(miss->flow, 0). */
3586 hash = miss->hmap_node.hash;
3588 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3590 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
3592 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
3593 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
3597 facet = facet_create(rule, &miss->flow, hash);
3602 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3605 static struct drop_key *
3606 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3609 struct drop_key *drop_key;
3611 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3612 &backer->drop_keys) {
3613 if (drop_key->key_len == key_len
3614 && !memcmp(drop_key->key, key, key_len)) {
3622 drop_key_clear(struct dpif_backer *backer)
3624 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3625 struct drop_key *drop_key, *next;
3627 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3630 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3632 if (error && !VLOG_DROP_WARN(&rl)) {
3633 struct ds ds = DS_EMPTY_INITIALIZER;
3634 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3635 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3640 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3641 free(drop_key->key);
3646 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3647 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3648 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3649 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3650 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3651 * 'packet' ingressed.
3653 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3654 * 'flow''s in_port to OFPP_NONE.
3656 * This function does post-processing on data returned from
3657 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3658 * of the upcall processing logic. In particular, if the extracted in_port is
3659 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3660 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3661 * a VLAN header onto 'packet' (if it is nonnull).
3663 * Optionally, if nonnull, sets '*initial_tci' to the VLAN TCI with which the
3664 * packet was really received, that is, the actual VLAN TCI extracted by
3665 * odp_flow_key_to_flow(). (This differs from the value returned in
3666 * flow->vlan_tci only for packets received on VLAN splinters.)
3668 * Similarly, this function also includes some logic to help with tunnels. It
3669 * may modify 'flow' as necessary to make the tunneling implementation
3670 * transparent to the upcall processing logic.
3672 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3673 * or some other positive errno if there are other problems. */
3675 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3676 const struct nlattr *key, size_t key_len,
3677 struct flow *flow, enum odp_key_fitness *fitnessp,
3678 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3679 ovs_be16 *initial_tci)
3681 const struct ofport_dpif *port;
3682 enum odp_key_fitness fitness;
3685 fitness = odp_flow_key_to_flow(key, key_len, flow);
3686 if (fitness == ODP_FIT_ERROR) {
3692 *initial_tci = flow->vlan_tci;
3696 *odp_in_port = flow->in_port;
3699 if (tnl_port_should_receive(flow)) {
3700 const struct ofport *ofport = tnl_port_receive(flow);
3702 flow->in_port = OFPP_NONE;
3705 port = ofport_dpif_cast(ofport);
3707 /* We can't reproduce 'key' from 'flow'. */
3708 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3710 /* XXX: Since the tunnel module is not scoped per backer, it's
3711 * theoretically possible that we'll receive an ofport belonging to an
3712 * entirely different datapath. In practice, this can't happen because
3713 * no platforms has two separate datapaths which each support
3715 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3717 port = odp_port_to_ofport(backer, flow->in_port);
3719 flow->in_port = OFPP_NONE;
3723 flow->in_port = port->up.ofp_port;
3724 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3726 /* Make the packet resemble the flow, so that it gets sent to
3727 * an OpenFlow controller properly, so that it looks correct
3728 * for sFlow, and so that flow_extract() will get the correct
3729 * vlan_tci if it is called on 'packet'.
3731 * The allocated space inside 'packet' probably also contains
3732 * 'key', that is, both 'packet' and 'key' are probably part of
3733 * a struct dpif_upcall (see the large comment on that
3734 * structure definition), so pushing data on 'packet' is in
3735 * general not a good idea since it could overwrite 'key' or
3736 * free it as a side effect. However, it's OK in this special
3737 * case because we know that 'packet' is inside a Netlink
3738 * attribute: pushing 4 bytes will just overwrite the 4-byte
3739 * "struct nlattr", which is fine since we don't need that
3740 * header anymore. */
3741 eth_push_vlan(packet, flow->vlan_tci);
3743 /* We can't reproduce 'key' from 'flow'. */
3744 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3750 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3755 *fitnessp = fitness;
3761 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3764 struct dpif_upcall *upcall;
3765 struct flow_miss *miss;
3766 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3767 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3768 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3778 /* Construct the to-do list.
3780 * This just amounts to extracting the flow from each packet and sticking
3781 * the packets that have the same flow in the same "flow_miss" structure so
3782 * that we can process them together. */
3785 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3786 struct flow_miss *miss = &misses[n_misses];
3787 struct flow_miss *existing_miss;
3788 struct ofproto_dpif *ofproto;
3789 uint32_t odp_in_port;
3794 error = ofproto_receive(backer, upcall->packet, upcall->key,
3795 upcall->key_len, &flow, &miss->key_fitness,
3796 &ofproto, &odp_in_port, &miss->initial_tci);
3797 if (error == ENODEV) {
3798 struct drop_key *drop_key;
3800 /* Received packet on port for which we couldn't associate
3801 * an ofproto. This can happen if a port is removed while
3802 * traffic is being received. Print a rate-limited message
3803 * in case it happens frequently. Install a drop flow so
3804 * that future packets of the flow are inexpensively dropped
3806 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
3809 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3811 drop_key = xmalloc(sizeof *drop_key);
3812 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3813 drop_key->key_len = upcall->key_len;
3815 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3816 hash_bytes(drop_key->key, drop_key->key_len, 0));
3817 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3818 drop_key->key, drop_key->key_len, NULL, 0, NULL);
3825 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3826 &flow.tunnel, flow.in_port, &miss->flow);
3828 /* Add other packets to a to-do list. */
3829 hash = flow_hash(&miss->flow, 0);
3830 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
3831 if (!existing_miss) {
3832 hmap_insert(&todo, &miss->hmap_node, hash);
3833 miss->ofproto = ofproto;
3834 miss->key = upcall->key;
3835 miss->key_len = upcall->key_len;
3836 miss->upcall_type = upcall->type;
3837 miss->odp_in_port = odp_in_port;
3838 list_init(&miss->packets);
3842 miss = existing_miss;
3844 list_push_back(&miss->packets, &upcall->packet->list_node);
3847 /* Process each element in the to-do list, constructing the set of
3848 * operations to batch. */
3850 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3851 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3853 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3855 /* Execute batch. */
3856 for (i = 0; i < n_ops; i++) {
3857 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3859 dpif_operate(backer->dpif, dpif_ops, n_ops);
3861 /* Free memory and update facets. */
3862 for (i = 0; i < n_ops; i++) {
3863 struct flow_miss_op *op = &flow_miss_ops[i];
3865 switch (op->dpif_op.type) {
3866 case DPIF_OP_EXECUTE:
3869 case DPIF_OP_FLOW_PUT:
3870 if (!op->dpif_op.error) {
3871 op->subfacet->path = subfacet_want_path(op->subfacet->slow);
3875 case DPIF_OP_FLOW_DEL:
3881 hmap_destroy(&todo);
3884 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
3885 classify_upcall(const struct dpif_upcall *upcall)
3887 union user_action_cookie cookie;
3889 /* First look at the upcall type. */
3890 switch (upcall->type) {
3891 case DPIF_UC_ACTION:
3897 case DPIF_N_UC_TYPES:
3899 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3903 /* "action" upcalls need a closer look. */
3904 if (!upcall->userdata) {
3905 VLOG_WARN_RL(&rl, "action upcall missing cookie");
3908 if (nl_attr_get_size(upcall->userdata) != sizeof(cookie)) {
3909 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
3910 nl_attr_get_size(upcall->userdata));
3913 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof(cookie));
3914 switch (cookie.type) {
3915 case USER_ACTION_COOKIE_SFLOW:
3916 return SFLOW_UPCALL;
3918 case USER_ACTION_COOKIE_SLOW_PATH:
3921 case USER_ACTION_COOKIE_UNSPEC:
3923 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64,
3924 nl_attr_get_u64(upcall->userdata));
3930 handle_sflow_upcall(struct dpif_backer *backer,
3931 const struct dpif_upcall *upcall)
3933 struct ofproto_dpif *ofproto;
3934 union user_action_cookie cookie;
3936 uint32_t odp_in_port;
3938 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3939 &flow, NULL, &ofproto, &odp_in_port, NULL)
3940 || !ofproto->sflow) {
3944 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof(cookie));
3945 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3946 odp_in_port, &cookie);
3950 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3952 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3953 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3954 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3959 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
3962 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3963 struct dpif_upcall *upcall = &misses[n_misses];
3964 struct ofpbuf *buf = &miss_bufs[n_misses];
3967 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3968 sizeof miss_buf_stubs[n_misses]);
3969 error = dpif_recv(backer->dpif, upcall, buf);
3975 switch (classify_upcall(upcall)) {
3977 /* Handle it later. */
3982 handle_sflow_upcall(backer, upcall);
3992 /* Handle deferred MISS_UPCALL processing. */
3993 handle_miss_upcalls(backer, misses, n_misses);
3994 for (i = 0; i < n_misses; i++) {
3995 ofpbuf_uninit(&miss_bufs[i]);
4001 /* Flow expiration. */
4003 static int subfacet_max_idle(const struct ofproto_dpif *);
4004 static void update_stats(struct dpif_backer *);
4005 static void rule_expire(struct rule_dpif *);
4006 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
4008 /* This function is called periodically by run(). Its job is to collect
4009 * updates for the flows that have been installed into the datapath, most
4010 * importantly when they last were used, and then use that information to
4011 * expire flows that have not been used recently.
4013 * Returns the number of milliseconds after which it should be called again. */
4015 expire(struct dpif_backer *backer)
4017 struct ofproto_dpif *ofproto;
4018 int max_idle = INT32_MAX;
4020 /* Periodically clear out the drop keys in an effort to keep them
4021 * relatively few. */
4022 drop_key_clear(backer);
4024 /* Update stats for each flow in the backer. */
4025 update_stats(backer);
4027 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4028 struct rule *rule, *next_rule;
4031 if (ofproto->backer != backer) {
4035 /* Expire subfacets that have been idle too long. */
4036 dp_max_idle = subfacet_max_idle(ofproto);
4037 expire_subfacets(ofproto, dp_max_idle);
4039 max_idle = MIN(max_idle, dp_max_idle);
4041 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4043 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4044 &ofproto->up.expirable) {
4045 rule_expire(rule_dpif_cast(rule));
4048 /* All outstanding data in existing flows has been accounted, so it's a
4049 * good time to do bond rebalancing. */
4050 if (ofproto->has_bonded_bundles) {
4051 struct ofbundle *bundle;
4053 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4055 bond_rebalance(bundle->bond, &backer->revalidate_set);
4061 return MIN(max_idle, 1000);
4064 /* Updates flow table statistics given that the datapath just reported 'stats'
4065 * as 'subfacet''s statistics. */
4067 update_subfacet_stats(struct subfacet *subfacet,
4068 const struct dpif_flow_stats *stats)
4070 struct facet *facet = subfacet->facet;
4072 if (stats->n_packets >= subfacet->dp_packet_count) {
4073 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
4074 facet->packet_count += extra;
4076 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4079 if (stats->n_bytes >= subfacet->dp_byte_count) {
4080 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
4082 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4085 subfacet->dp_packet_count = stats->n_packets;
4086 subfacet->dp_byte_count = stats->n_bytes;
4088 facet->tcp_flags |= stats->tcp_flags;
4090 subfacet_update_time(subfacet, stats->used);
4091 if (facet->accounted_bytes < facet->byte_count) {
4093 facet_account(facet);
4094 facet->accounted_bytes = facet->byte_count;
4096 facet_push_stats(facet);
4099 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4100 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4102 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4103 const struct nlattr *key, size_t key_len)
4105 if (!VLOG_DROP_WARN(&rl)) {
4109 odp_flow_key_format(key, key_len, &s);
4110 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4114 COVERAGE_INC(facet_unexpected);
4115 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4118 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4120 * This function also pushes statistics updates to rules which each facet
4121 * resubmits into. Generally these statistics will be accurate. However, if a
4122 * facet changes the rule it resubmits into at some time in between
4123 * update_stats() runs, it is possible that statistics accrued to the
4124 * old rule will be incorrectly attributed to the new rule. This could be
4125 * avoided by calling update_stats() whenever rules are created or
4126 * deleted. However, the performance impact of making so many calls to the
4127 * datapath do not justify the benefit of having perfectly accurate statistics.
4130 update_stats(struct dpif_backer *backer)
4132 const struct dpif_flow_stats *stats;
4133 struct dpif_flow_dump dump;
4134 const struct nlattr *key;
4137 dpif_flow_dump_start(&dump, backer->dpif);
4138 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4140 struct subfacet *subfacet;
4141 struct ofproto_dpif *ofproto;
4142 struct ofport_dpif *ofport;
4145 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4150 ofport = get_ofp_port(ofproto, flow.in_port);
4151 if (ofport && ofport->tnl_port) {
4152 netdev_vport_inc_rx(ofport->up.netdev, stats);
4155 key_hash = odp_flow_key_hash(key, key_len);
4156 subfacet = subfacet_find(ofproto, key, key_len, key_hash, &flow);
4157 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4159 update_subfacet_stats(subfacet, stats);
4163 /* Stats are updated per-packet. */
4166 case SF_NOT_INSTALLED:
4168 delete_unexpected_flow(ofproto, key, key_len);
4172 dpif_flow_dump_done(&dump);
4175 /* Calculates and returns the number of milliseconds of idle time after which
4176 * subfacets should expire from the datapath. When a subfacet expires, we fold
4177 * its statistics into its facet, and when a facet's last subfacet expires, we
4178 * fold its statistic into its rule. */
4180 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4183 * Idle time histogram.
4185 * Most of the time a switch has a relatively small number of subfacets.
4186 * When this is the case we might as well keep statistics for all of them
4187 * in userspace and to cache them in the kernel datapath for performance as
4190 * As the number of subfacets increases, the memory required to maintain
4191 * statistics about them in userspace and in the kernel becomes
4192 * significant. However, with a large number of subfacets it is likely
4193 * that only a few of them are "heavy hitters" that consume a large amount
4194 * of bandwidth. At this point, only heavy hitters are worth caching in
4195 * the kernel and maintaining in userspaces; other subfacets we can
4198 * The technique used to compute the idle time is to build a histogram with
4199 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4200 * that is installed in the kernel gets dropped in the appropriate bucket.
4201 * After the histogram has been built, we compute the cutoff so that only
4202 * the most-recently-used 1% of subfacets (but at least
4203 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4204 * the most-recently-used bucket of subfacets is kept, so actually an
4205 * arbitrary number of subfacets can be kept in any given expiration run
4206 * (though the next run will delete most of those unless they receive
4209 * This requires a second pass through the subfacets, in addition to the
4210 * pass made by update_stats(), because the former function never looks at
4211 * uninstallable subfacets.
4213 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4214 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4215 int buckets[N_BUCKETS] = { 0 };
4216 int total, subtotal, bucket;
4217 struct subfacet *subfacet;
4221 total = hmap_count(&ofproto->subfacets);
4222 if (total <= ofproto->up.flow_eviction_threshold) {
4223 return N_BUCKETS * BUCKET_WIDTH;
4226 /* Build histogram. */
4228 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4229 long long int idle = now - subfacet->used;
4230 int bucket = (idle <= 0 ? 0
4231 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4232 : (unsigned int) idle / BUCKET_WIDTH);
4236 /* Find the first bucket whose flows should be expired. */
4237 subtotal = bucket = 0;
4239 subtotal += buckets[bucket++];
4240 } while (bucket < N_BUCKETS &&
4241 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4243 if (VLOG_IS_DBG_ENABLED()) {
4247 ds_put_cstr(&s, "keep");
4248 for (i = 0; i < N_BUCKETS; i++) {
4250 ds_put_cstr(&s, ", drop");
4253 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4256 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4260 return bucket * BUCKET_WIDTH;
4264 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4266 /* Cutoff time for most flows. */
4267 long long int normal_cutoff = time_msec() - dp_max_idle;
4269 /* We really want to keep flows for special protocols around, so use a more
4270 * conservative cutoff. */
4271 long long int special_cutoff = time_msec() - 10000;
4273 struct subfacet *subfacet, *next_subfacet;
4274 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4278 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4279 &ofproto->subfacets) {
4280 long long int cutoff;
4282 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
4285 if (subfacet->used < cutoff) {
4286 if (subfacet->path != SF_NOT_INSTALLED) {
4287 batch[n_batch++] = subfacet;
4288 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4289 subfacet_destroy_batch(ofproto, batch, n_batch);
4293 subfacet_destroy(subfacet);
4299 subfacet_destroy_batch(ofproto, batch, n_batch);
4303 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4304 * then delete it entirely. */
4306 rule_expire(struct rule_dpif *rule)
4308 struct facet *facet, *next_facet;
4312 if (rule->up.pending) {
4313 /* We'll have to expire it later. */
4317 /* Has 'rule' expired? */
4319 if (rule->up.hard_timeout
4320 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4321 reason = OFPRR_HARD_TIMEOUT;
4322 } else if (rule->up.idle_timeout
4323 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4324 reason = OFPRR_IDLE_TIMEOUT;
4329 COVERAGE_INC(ofproto_dpif_expired);
4331 /* Update stats. (This is a no-op if the rule expired due to an idle
4332 * timeout, because that only happens when the rule has no facets left.) */
4333 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4334 facet_remove(facet);
4337 /* Get rid of the rule. */
4338 ofproto_rule_expire(&rule->up, reason);
4343 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
4345 * The caller must already have determined that no facet with an identical
4346 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
4347 * the ofproto's classifier table.
4349 * 'hash' must be the return value of flow_hash(flow, 0).
4351 * The facet will initially have no subfacets. The caller should create (at
4352 * least) one subfacet with subfacet_create(). */
4353 static struct facet *
4354 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
4356 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4357 struct facet *facet;
4359 facet = xzalloc(sizeof *facet);
4360 facet->used = time_msec();
4361 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4362 list_push_back(&rule->facets, &facet->list_node);
4364 facet->flow = *flow;
4365 list_init(&facet->subfacets);
4366 netflow_flow_init(&facet->nf_flow);
4367 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4373 facet_free(struct facet *facet)
4378 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4379 * 'packet', which arrived on 'in_port'. */
4381 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4382 const struct nlattr *odp_actions, size_t actions_len,
4383 struct ofpbuf *packet)
4385 struct odputil_keybuf keybuf;
4389 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4390 odp_flow_key_from_flow(&key, flow,
4391 ofp_port_to_odp_port(ofproto, flow->in_port));
4393 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4394 odp_actions, actions_len, packet);
4398 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4400 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4401 * rule's statistics, via subfacet_uninstall().
4403 * - Removes 'facet' from its rule and from ofproto->facets.
4406 facet_remove(struct facet *facet)
4408 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4409 struct subfacet *subfacet, *next_subfacet;
4411 ovs_assert(!list_is_empty(&facet->subfacets));
4413 /* First uninstall all of the subfacets to get final statistics. */
4414 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4415 subfacet_uninstall(subfacet);
4418 /* Flush the final stats to the rule.
4420 * This might require us to have at least one subfacet around so that we
4421 * can use its actions for accounting in facet_account(), which is why we
4422 * have uninstalled but not yet destroyed the subfacets. */
4423 facet_flush_stats(facet);
4425 /* Now we're really all done so destroy everything. */
4426 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4427 &facet->subfacets) {
4428 subfacet_destroy__(subfacet);
4430 hmap_remove(&ofproto->facets, &facet->hmap_node);
4431 list_remove(&facet->list_node);
4435 /* Feed information from 'facet' back into the learning table to keep it in
4436 * sync with what is actually flowing through the datapath. */
4438 facet_learn(struct facet *facet)
4440 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4441 struct action_xlate_ctx ctx;
4443 if (!facet->has_learn
4444 && !facet->has_normal
4445 && (!facet->has_fin_timeout
4446 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4450 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4451 facet->flow.vlan_tci,
4452 facet->rule, facet->tcp_flags, NULL);
4453 ctx.may_learn = true;
4454 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
4455 facet->rule->up.ofpacts_len);
4459 facet_account(struct facet *facet)
4461 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4462 struct subfacet *subfacet;
4463 const struct nlattr *a;
4468 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
4471 n_bytes = facet->byte_count - facet->accounted_bytes;
4473 /* This loop feeds byte counters to bond_account() for rebalancing to use
4474 * as a basis. We also need to track the actual VLAN on which the packet
4475 * is going to be sent to ensure that it matches the one passed to
4476 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4479 * We use the actions from an arbitrary subfacet because they should all
4480 * be equally valid for our purpose. */
4481 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
4482 struct subfacet, list_node);
4483 vlan_tci = facet->flow.vlan_tci;
4484 NL_ATTR_FOR_EACH_UNSAFE (a, left,
4485 subfacet->actions, subfacet->actions_len) {
4486 const struct ovs_action_push_vlan *vlan;
4487 struct ofport_dpif *port;
4489 switch (nl_attr_type(a)) {
4490 case OVS_ACTION_ATTR_OUTPUT:
4491 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4492 if (port && port->bundle && port->bundle->bond) {
4493 bond_account(port->bundle->bond, &facet->flow,
4494 vlan_tci_to_vid(vlan_tci), n_bytes);
4498 case OVS_ACTION_ATTR_POP_VLAN:
4499 vlan_tci = htons(0);
4502 case OVS_ACTION_ATTR_PUSH_VLAN:
4503 vlan = nl_attr_get(a);
4504 vlan_tci = vlan->vlan_tci;
4510 /* Returns true if the only action for 'facet' is to send to the controller.
4511 * (We don't report NetFlow expiration messages for such facets because they
4512 * are just part of the control logic for the network, not real traffic). */
4514 facet_is_controller_flow(struct facet *facet)
4517 const struct rule *rule = &facet->rule->up;
4518 const struct ofpact *ofpacts = rule->ofpacts;
4519 size_t ofpacts_len = rule->ofpacts_len;
4521 if (ofpacts_len > 0 &&
4522 ofpacts->type == OFPACT_CONTROLLER &&
4523 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4530 /* Folds all of 'facet''s statistics into its rule. Also updates the
4531 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4532 * 'facet''s statistics in the datapath should have been zeroed and folded into
4533 * its packet and byte counts before this function is called. */
4535 facet_flush_stats(struct facet *facet)
4537 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4538 struct subfacet *subfacet;
4540 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4541 ovs_assert(!subfacet->dp_byte_count);
4542 ovs_assert(!subfacet->dp_packet_count);
4545 facet_push_stats(facet);
4546 if (facet->accounted_bytes < facet->byte_count) {
4547 facet_account(facet);
4548 facet->accounted_bytes = facet->byte_count;
4551 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4552 struct ofexpired expired;
4553 expired.flow = facet->flow;
4554 expired.packet_count = facet->packet_count;
4555 expired.byte_count = facet->byte_count;
4556 expired.used = facet->used;
4557 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4560 facet->rule->packet_count += facet->packet_count;
4561 facet->rule->byte_count += facet->byte_count;
4563 /* Reset counters to prevent double counting if 'facet' ever gets
4565 facet_reset_counters(facet);
4567 netflow_flow_clear(&facet->nf_flow);
4568 facet->tcp_flags = 0;
4571 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4572 * Returns it if found, otherwise a null pointer.
4574 * 'hash' must be the return value of flow_hash(flow, 0).
4576 * The returned facet might need revalidation; use facet_lookup_valid()
4577 * instead if that is important. */
4578 static struct facet *
4579 facet_find(struct ofproto_dpif *ofproto,
4580 const struct flow *flow, uint32_t hash)
4582 struct facet *facet;
4584 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4585 if (flow_equal(flow, &facet->flow)) {
4593 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4594 * Returns it if found, otherwise a null pointer.
4596 * 'hash' must be the return value of flow_hash(flow, 0).
4598 * The returned facet is guaranteed to be valid. */
4599 static struct facet *
4600 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4603 struct facet *facet;
4605 facet = facet_find(ofproto, flow, hash);
4607 && (ofproto->backer->need_revalidate
4608 || tag_set_intersects(&ofproto->backer->revalidate_set,
4610 facet_revalidate(facet);
4617 subfacet_path_to_string(enum subfacet_path path)
4620 case SF_NOT_INSTALLED:
4621 return "not installed";
4623 return "in fast path";
4625 return "in slow path";
4631 /* Returns the path in which a subfacet should be installed if its 'slow'
4632 * member has the specified value. */
4633 static enum subfacet_path
4634 subfacet_want_path(enum slow_path_reason slow)
4636 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
4639 /* Returns true if 'subfacet' needs to have its datapath flow updated,
4640 * supposing that its actions have been recalculated as 'want_actions' and that
4641 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
4643 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
4644 const struct ofpbuf *want_actions)
4646 enum subfacet_path want_path = subfacet_want_path(slow);
4647 return (want_path != subfacet->path
4648 || (want_path == SF_FAST_PATH
4649 && (subfacet->actions_len != want_actions->size
4650 || memcmp(subfacet->actions, want_actions->data,
4651 subfacet->actions_len))));
4655 facet_check_consistency(struct facet *facet)
4657 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4659 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4661 uint64_t odp_actions_stub[1024 / 8];
4662 struct ofpbuf odp_actions;
4664 struct rule_dpif *rule;
4665 struct subfacet *subfacet;
4666 bool may_log = false;
4669 /* Check the rule for consistency. */
4670 rule = rule_dpif_lookup(ofproto, &facet->flow);
4671 ok = rule == facet->rule;
4673 may_log = !VLOG_DROP_WARN(&rl);
4678 flow_format(&s, &facet->flow);
4679 ds_put_format(&s, ": facet associated with wrong rule (was "
4680 "table=%"PRIu8",", facet->rule->up.table_id);
4681 cls_rule_format(&facet->rule->up.cr, &s);
4682 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4684 cls_rule_format(&rule->up.cr, &s);
4685 ds_put_char(&s, ')');
4687 VLOG_WARN("%s", ds_cstr(&s));
4692 /* Check the datapath actions for consistency. */
4693 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4694 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4695 enum subfacet_path want_path;
4696 struct odputil_keybuf keybuf;
4697 struct action_xlate_ctx ctx;
4701 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4702 subfacet->initial_tci, rule, 0, NULL);
4703 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
4706 if (subfacet->path == SF_NOT_INSTALLED) {
4707 /* This only happens if the datapath reported an error when we
4708 * tried to install the flow. Don't flag another error here. */
4712 want_path = subfacet_want_path(subfacet->slow);
4713 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
4714 /* The actions for slow-path flows may legitimately vary from one
4715 * packet to the next. We're done. */
4719 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
4723 /* Inconsistency! */
4725 may_log = !VLOG_DROP_WARN(&rl);
4729 /* Rate-limited, skip reporting. */
4734 subfacet_get_key(subfacet, &keybuf, &key);
4735 odp_flow_key_format(key.data, key.size, &s);
4737 ds_put_cstr(&s, ": inconsistency in subfacet");
4738 if (want_path != subfacet->path) {
4739 enum odp_key_fitness fitness = subfacet->key_fitness;
4741 ds_put_format(&s, " (%s, fitness=%s)",
4742 subfacet_path_to_string(subfacet->path),
4743 odp_key_fitness_to_string(fitness));
4744 ds_put_format(&s, " (should have been %s)",
4745 subfacet_path_to_string(want_path));
4746 } else if (want_path == SF_FAST_PATH) {
4747 ds_put_cstr(&s, " (actions were: ");
4748 format_odp_actions(&s, subfacet->actions,
4749 subfacet->actions_len);
4750 ds_put_cstr(&s, ") (correct actions: ");
4751 format_odp_actions(&s, odp_actions.data, odp_actions.size);
4752 ds_put_char(&s, ')');
4754 ds_put_cstr(&s, " (actions: ");
4755 format_odp_actions(&s, subfacet->actions,
4756 subfacet->actions_len);
4757 ds_put_char(&s, ')');
4759 VLOG_WARN("%s", ds_cstr(&s));
4762 ofpbuf_uninit(&odp_actions);
4767 /* Re-searches the classifier for 'facet':
4769 * - If the rule found is different from 'facet''s current rule, moves
4770 * 'facet' to the new rule and recompiles its actions.
4772 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4773 * where it is and recompiles its actions anyway. */
4775 facet_revalidate(struct facet *facet)
4777 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4779 struct nlattr *odp_actions;
4782 struct actions *new_actions;
4784 struct action_xlate_ctx ctx;
4785 uint64_t odp_actions_stub[1024 / 8];
4786 struct ofpbuf odp_actions;
4788 struct rule_dpif *new_rule;
4789 struct subfacet *subfacet;
4792 COVERAGE_INC(facet_revalidate);
4794 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
4796 /* Calculate new datapath actions.
4798 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4799 * emit a NetFlow expiration and, if so, we need to have the old state
4800 * around to properly compose it. */
4802 /* If the datapath actions changed or the installability changed,
4803 * then we need to talk to the datapath. */
4806 memset(&ctx, 0, sizeof ctx);
4807 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4808 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4809 enum slow_path_reason slow;
4811 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4812 subfacet->initial_tci, new_rule, 0, NULL);
4813 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
4816 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4817 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
4818 struct dpif_flow_stats stats;
4820 subfacet_install(subfacet,
4821 odp_actions.data, odp_actions.size, &stats, slow);
4822 subfacet_update_stats(subfacet, &stats);
4825 new_actions = xcalloc(list_size(&facet->subfacets),
4826 sizeof *new_actions);
4828 new_actions[i].odp_actions = xmemdup(odp_actions.data,
4830 new_actions[i].actions_len = odp_actions.size;
4835 ofpbuf_uninit(&odp_actions);
4838 facet_flush_stats(facet);
4841 /* Update 'facet' now that we've taken care of all the old state. */
4842 facet->tags = ctx.tags;
4843 facet->nf_flow.output_iface = ctx.nf_output_iface;
4844 facet->has_learn = ctx.has_learn;
4845 facet->has_normal = ctx.has_normal;
4846 facet->has_fin_timeout = ctx.has_fin_timeout;
4847 facet->mirrors = ctx.mirrors;
4850 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4851 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4853 if (new_actions && new_actions[i].odp_actions) {
4854 free(subfacet->actions);
4855 subfacet->actions = new_actions[i].odp_actions;
4856 subfacet->actions_len = new_actions[i].actions_len;
4862 if (facet->rule != new_rule) {
4863 COVERAGE_INC(facet_changed_rule);
4864 list_remove(&facet->list_node);
4865 list_push_back(&new_rule->facets, &facet->list_node);
4866 facet->rule = new_rule;
4867 facet->used = new_rule->up.created;
4868 facet->prev_used = facet->used;
4872 /* Updates 'facet''s used time. Caller is responsible for calling
4873 * facet_push_stats() to update the flows which 'facet' resubmits into. */
4875 facet_update_time(struct facet *facet, long long int used)
4877 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4878 if (used > facet->used) {
4880 ofproto_rule_update_used(&facet->rule->up, used);
4881 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
4886 facet_reset_counters(struct facet *facet)
4888 facet->packet_count = 0;
4889 facet->byte_count = 0;
4890 facet->prev_packet_count = 0;
4891 facet->prev_byte_count = 0;
4892 facet->accounted_bytes = 0;
4896 facet_push_stats(struct facet *facet)
4898 struct dpif_flow_stats stats;
4900 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4901 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4902 ovs_assert(facet->used >= facet->prev_used);
4904 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4905 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4906 stats.used = facet->used;
4907 stats.tcp_flags = 0;
4909 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
4910 facet->prev_packet_count = facet->packet_count;
4911 facet->prev_byte_count = facet->byte_count;
4912 facet->prev_used = facet->used;
4914 flow_push_stats(facet->rule, &facet->flow, &stats);
4916 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
4917 facet->mirrors, stats.n_packets, stats.n_bytes);
4922 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4924 rule->packet_count += stats->n_packets;
4925 rule->byte_count += stats->n_bytes;
4926 ofproto_rule_update_used(&rule->up, stats->used);
4929 /* Pushes flow statistics to the rules which 'flow' resubmits into given
4930 * 'rule''s actions and mirrors. */
4932 flow_push_stats(struct rule_dpif *rule,
4933 const struct flow *flow, const struct dpif_flow_stats *stats)
4935 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4936 struct action_xlate_ctx ctx;
4938 ofproto_rule_update_used(&rule->up, stats->used);
4940 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule,
4942 ctx.resubmit_stats = stats;
4943 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
4944 rule->up.ofpacts_len);
4949 static struct subfacet *
4950 subfacet_find(struct ofproto_dpif *ofproto,
4951 const struct nlattr *key, size_t key_len, uint32_t key_hash,
4952 const struct flow *flow)
4954 struct subfacet *subfacet;
4956 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4957 &ofproto->subfacets) {
4959 ? (subfacet->key_len == key_len
4960 && !memcmp(key, subfacet->key, key_len))
4961 : flow_equal(flow, &subfacet->facet->flow)) {
4969 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4970 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4971 * existing subfacet if there is one, otherwise creates and returns a
4974 * If the returned subfacet is new, then subfacet->actions will be NULL, in
4975 * which case the caller must populate the actions with
4976 * subfacet_make_actions(). */
4977 static struct subfacet *
4978 subfacet_create(struct facet *facet, struct flow_miss *miss,
4981 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4982 enum odp_key_fitness key_fitness = miss->key_fitness;
4983 const struct nlattr *key = miss->key;
4984 size_t key_len = miss->key_len;
4986 struct subfacet *subfacet;
4988 key_hash = odp_flow_key_hash(key, key_len);
4990 if (list_is_empty(&facet->subfacets)) {
4991 subfacet = &facet->one_subfacet;
4993 subfacet = subfacet_find(ofproto, key, key_len, key_hash,
4996 if (subfacet->facet == facet) {
5000 /* This shouldn't happen. */
5001 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
5002 subfacet_destroy(subfacet);
5005 subfacet = xmalloc(sizeof *subfacet);
5008 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
5009 list_push_back(&facet->subfacets, &subfacet->list_node);
5010 subfacet->facet = facet;
5011 subfacet->key_fitness = key_fitness;
5012 if (key_fitness != ODP_FIT_PERFECT) {
5013 subfacet->key = xmemdup(key, key_len);
5014 subfacet->key_len = key_len;
5016 subfacet->key = NULL;
5017 subfacet->key_len = 0;
5019 subfacet->used = now;
5020 subfacet->dp_packet_count = 0;
5021 subfacet->dp_byte_count = 0;
5022 subfacet->actions_len = 0;
5023 subfacet->actions = NULL;
5024 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
5027 subfacet->path = SF_NOT_INSTALLED;
5028 subfacet->initial_tci = miss->initial_tci;
5029 subfacet->odp_in_port = miss->odp_in_port;
5034 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5035 * its facet within 'ofproto', and frees it. */
5037 subfacet_destroy__(struct subfacet *subfacet)
5039 struct facet *facet = subfacet->facet;
5040 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5042 subfacet_uninstall(subfacet);
5043 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5044 list_remove(&subfacet->list_node);
5045 free(subfacet->key);
5046 free(subfacet->actions);
5047 if (subfacet != &facet->one_subfacet) {
5052 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5053 * last remaining subfacet in its facet destroys the facet too. */
5055 subfacet_destroy(struct subfacet *subfacet)
5057 struct facet *facet = subfacet->facet;
5059 if (list_is_singleton(&facet->subfacets)) {
5060 /* facet_remove() needs at least one subfacet (it will remove it). */
5061 facet_remove(facet);
5063 subfacet_destroy__(subfacet);
5068 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5069 struct subfacet **subfacets, int n)
5071 struct odputil_keybuf keybufs[SUBFACET_DESTROY_MAX_BATCH];
5072 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5073 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5074 struct ofpbuf keys[SUBFACET_DESTROY_MAX_BATCH];
5075 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5078 for (i = 0; i < n; i++) {
5079 ops[i].type = DPIF_OP_FLOW_DEL;
5080 subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]);
5081 ops[i].u.flow_del.key = keys[i].data;
5082 ops[i].u.flow_del.key_len = keys[i].size;
5083 ops[i].u.flow_del.stats = &stats[i];
5087 dpif_operate(ofproto->backer->dpif, opsp, n);
5088 for (i = 0; i < n; i++) {
5089 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5090 subfacets[i]->path = SF_NOT_INSTALLED;
5091 subfacet_destroy(subfacets[i]);
5095 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
5096 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
5097 * for use as temporary storage. */
5099 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
5103 if (!subfacet->key) {
5104 struct flow *flow = &subfacet->facet->flow;
5106 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
5107 odp_flow_key_from_flow(key, flow, subfacet->odp_in_port);
5109 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
5113 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
5114 * Translates the actions into 'odp_actions', which the caller must have
5115 * initialized and is responsible for uninitializing. */
5117 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
5118 struct ofpbuf *odp_actions)
5120 struct facet *facet = subfacet->facet;
5121 struct rule_dpif *rule = facet->rule;
5122 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5124 struct action_xlate_ctx ctx;
5126 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
5128 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
5129 facet->tags = ctx.tags;
5130 facet->has_learn = ctx.has_learn;
5131 facet->has_normal = ctx.has_normal;
5132 facet->has_fin_timeout = ctx.has_fin_timeout;
5133 facet->nf_flow.output_iface = ctx.nf_output_iface;
5134 facet->mirrors = ctx.mirrors;
5136 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
5137 if (subfacet->actions_len != odp_actions->size
5138 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
5139 free(subfacet->actions);
5140 subfacet->actions_len = odp_actions->size;
5141 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
5145 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5146 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5147 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5148 * since 'subfacet' was last updated.
5150 * Returns 0 if successful, otherwise a positive errno value. */
5152 subfacet_install(struct subfacet *subfacet,
5153 const struct nlattr *actions, size_t actions_len,
5154 struct dpif_flow_stats *stats,
5155 enum slow_path_reason slow)
5157 struct facet *facet = subfacet->facet;
5158 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5159 enum subfacet_path path = subfacet_want_path(slow);
5160 uint64_t slow_path_stub[128 / 8];
5161 struct odputil_keybuf keybuf;
5162 enum dpif_flow_put_flags flags;
5166 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5168 flags |= DPIF_FP_ZERO_STATS;
5171 if (path == SF_SLOW_PATH) {
5172 compose_slow_path(ofproto, &facet->flow, slow,
5173 slow_path_stub, sizeof slow_path_stub,
5174 &actions, &actions_len);
5177 subfacet_get_key(subfacet, &keybuf, &key);
5178 ret = dpif_flow_put(ofproto->backer->dpif, flags, key.data, key.size,
5179 actions, actions_len, stats);
5182 subfacet_reset_dp_stats(subfacet, stats);
5186 subfacet->path = path;
5192 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
5194 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
5195 stats, subfacet->slow);
5198 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5200 subfacet_uninstall(struct subfacet *subfacet)
5202 if (subfacet->path != SF_NOT_INSTALLED) {
5203 struct rule_dpif *rule = subfacet->facet->rule;
5204 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5205 struct odputil_keybuf keybuf;
5206 struct dpif_flow_stats stats;
5210 subfacet_get_key(subfacet, &keybuf, &key);
5211 error = dpif_flow_del(ofproto->backer->dpif,
5212 key.data, key.size, &stats);
5213 subfacet_reset_dp_stats(subfacet, &stats);
5215 subfacet_update_stats(subfacet, &stats);
5217 subfacet->path = SF_NOT_INSTALLED;
5219 ovs_assert(subfacet->dp_packet_count == 0);
5220 ovs_assert(subfacet->dp_byte_count == 0);
5224 /* Resets 'subfacet''s datapath statistics counters. This should be called
5225 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5226 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5227 * was reset in the datapath. 'stats' will be modified to include only
5228 * statistics new since 'subfacet' was last updated. */
5230 subfacet_reset_dp_stats(struct subfacet *subfacet,
5231 struct dpif_flow_stats *stats)
5234 && subfacet->dp_packet_count <= stats->n_packets
5235 && subfacet->dp_byte_count <= stats->n_bytes) {
5236 stats->n_packets -= subfacet->dp_packet_count;
5237 stats->n_bytes -= subfacet->dp_byte_count;
5240 subfacet->dp_packet_count = 0;
5241 subfacet->dp_byte_count = 0;
5244 /* Updates 'subfacet''s used time. The caller is responsible for calling
5245 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
5247 subfacet_update_time(struct subfacet *subfacet, long long int used)
5249 if (used > subfacet->used) {
5250 subfacet->used = used;
5251 facet_update_time(subfacet->facet, used);
5255 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5257 * Because of the meaning of a subfacet's counters, it only makes sense to do
5258 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5259 * represents a packet that was sent by hand or if it represents statistics
5260 * that have been cleared out of the datapath. */
5262 subfacet_update_stats(struct subfacet *subfacet,
5263 const struct dpif_flow_stats *stats)
5265 if (stats->n_packets || stats->used > subfacet->used) {
5266 struct facet *facet = subfacet->facet;
5268 subfacet_update_time(subfacet, stats->used);
5269 facet->packet_count += stats->n_packets;
5270 facet->byte_count += stats->n_bytes;
5271 facet->tcp_flags |= stats->tcp_flags;
5272 facet_push_stats(facet);
5273 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
5279 static struct rule_dpif *
5280 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5282 struct rule_dpif *rule;
5284 rule = rule_dpif_lookup__(ofproto, flow, 0);
5289 return rule_dpif_miss_rule(ofproto, flow);
5292 static struct rule_dpif *
5293 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5296 struct cls_rule *cls_rule;
5297 struct classifier *cls;
5299 if (table_id >= N_TABLES) {
5303 cls = &ofproto->up.tables[table_id].cls;
5304 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5305 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5306 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5307 * are unavailable. */
5308 struct flow ofpc_normal_flow = *flow;
5309 ofpc_normal_flow.tp_src = htons(0);
5310 ofpc_normal_flow.tp_dst = htons(0);
5311 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5313 cls_rule = classifier_lookup(cls, flow);
5315 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5318 static struct rule_dpif *
5319 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5321 struct ofport_dpif *port;
5323 port = get_ofp_port(ofproto, flow->in_port);
5325 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5326 return ofproto->miss_rule;
5329 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5330 return ofproto->no_packet_in_rule;
5332 return ofproto->miss_rule;
5336 complete_operation(struct rule_dpif *rule)
5338 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5340 rule_invalidate(rule);
5342 struct dpif_completion *c = xmalloc(sizeof *c);
5343 c->op = rule->up.pending;
5344 list_push_back(&ofproto->completions, &c->list_node);
5346 ofoperation_complete(rule->up.pending, 0);
5350 static struct rule *
5353 struct rule_dpif *rule = xmalloc(sizeof *rule);
5358 rule_dealloc(struct rule *rule_)
5360 struct rule_dpif *rule = rule_dpif_cast(rule_);
5365 rule_construct(struct rule *rule_)
5367 struct rule_dpif *rule = rule_dpif_cast(rule_);
5368 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5369 struct rule_dpif *victim;
5372 rule->packet_count = 0;
5373 rule->byte_count = 0;
5375 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5376 if (victim && !list_is_empty(&victim->facets)) {
5377 struct facet *facet;
5379 rule->facets = victim->facets;
5380 list_moved(&rule->facets);
5381 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5382 /* XXX: We're only clearing our local counters here. It's possible
5383 * that quite a few packets are unaccounted for in the datapath
5384 * statistics. These will be accounted to the new rule instead of
5385 * cleared as required. This could be fixed by clearing out the
5386 * datapath statistics for this facet, but currently it doesn't
5388 facet_reset_counters(facet);
5392 /* Must avoid list_moved() in this case. */
5393 list_init(&rule->facets);
5396 table_id = rule->up.table_id;
5398 rule->tag = victim->tag;
5399 } else if (table_id == 0) {
5404 miniflow_expand(&rule->up.cr.match.flow, &flow);
5405 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5406 ofproto->tables[table_id].basis);
5409 complete_operation(rule);
5414 rule_destruct(struct rule *rule_)
5416 struct rule_dpif *rule = rule_dpif_cast(rule_);
5417 struct facet *facet, *next_facet;
5419 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5420 facet_revalidate(facet);
5423 complete_operation(rule);
5427 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5429 struct rule_dpif *rule = rule_dpif_cast(rule_);
5430 struct facet *facet;
5432 /* Start from historical data for 'rule' itself that are no longer tracked
5433 * in facets. This counts, for example, facets that have expired. */
5434 *packets = rule->packet_count;
5435 *bytes = rule->byte_count;
5437 /* Add any statistics that are tracked by facets. This includes
5438 * statistical data recently updated by ofproto_update_stats() as well as
5439 * stats for packets that were executed "by hand" via dpif_execute(). */
5440 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5441 *packets += facet->packet_count;
5442 *bytes += facet->byte_count;
5447 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5448 struct ofpbuf *packet)
5450 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5452 struct dpif_flow_stats stats;
5454 struct action_xlate_ctx ctx;
5455 uint64_t odp_actions_stub[1024 / 8];
5456 struct ofpbuf odp_actions;
5458 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5459 rule_credit_stats(rule, &stats);
5461 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5462 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
5463 rule, stats.tcp_flags, packet);
5464 ctx.resubmit_stats = &stats;
5465 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
5467 execute_odp_actions(ofproto, flow, odp_actions.data,
5468 odp_actions.size, packet);
5470 ofpbuf_uninit(&odp_actions);
5474 rule_execute(struct rule *rule, const struct flow *flow,
5475 struct ofpbuf *packet)
5477 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5478 ofpbuf_delete(packet);
5483 rule_modify_actions(struct rule *rule_)
5485 struct rule_dpif *rule = rule_dpif_cast(rule_);
5487 complete_operation(rule);
5490 /* Sends 'packet' out 'ofport'.
5491 * May modify 'packet'.
5492 * Returns 0 if successful, otherwise a positive errno value. */
5494 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5496 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5497 uint64_t odp_actions_stub[1024 / 8];
5498 struct ofpbuf key, odp_actions;
5499 struct odputil_keybuf keybuf;
5504 flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow);
5505 if (netdev_vport_is_patch(ofport->up.netdev)) {
5506 struct ofproto_dpif *peer_ofproto;
5507 struct dpif_flow_stats stats;
5508 struct ofport_dpif *peer;
5509 struct rule_dpif *rule;
5511 peer = ofport_get_peer(ofport);
5516 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5517 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5518 netdev_vport_inc_rx(peer->up.netdev, &stats);
5520 flow.in_port = peer->up.ofp_port;
5521 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5522 rule = rule_dpif_lookup(peer_ofproto, &flow);
5523 rule_dpif_execute(rule, &flow, packet);
5528 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5530 if (ofport->tnl_port) {
5531 struct dpif_flow_stats stats;
5533 odp_port = tnl_port_send(ofport->tnl_port, &flow);
5534 if (odp_port == OVSP_NONE) {
5538 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5539 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5540 odp_put_tunnel_action(&flow.tunnel, &odp_actions);
5541 odp_put_skb_mark_action(flow.skb_mark, &odp_actions);
5543 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5545 if (odp_port != ofport->odp_port) {
5546 eth_pop_vlan(packet);
5547 flow.vlan_tci = htons(0);
5551 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5552 odp_flow_key_from_flow(&key, &flow,
5553 ofp_port_to_odp_port(ofproto, flow.in_port));
5555 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5557 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5558 error = dpif_execute(ofproto->backer->dpif,
5560 odp_actions.data, odp_actions.size,
5562 ofpbuf_uninit(&odp_actions);
5565 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5566 ofproto->up.name, odp_port, strerror(error));
5568 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5572 /* OpenFlow to datapath action translation. */
5574 static bool may_receive(const struct ofport_dpif *, struct action_xlate_ctx *);
5575 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5576 struct action_xlate_ctx *);
5577 static void xlate_normal(struct action_xlate_ctx *);
5579 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5580 * The action will state 'slow' as the reason that the action is in the slow
5581 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5582 * dump-flows" output to see why a flow is in the slow path.)
5584 * The 'stub_size' bytes in 'stub' will be used to store the action.
5585 * 'stub_size' must be large enough for the action.
5587 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5590 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5591 enum slow_path_reason slow,
5592 uint64_t *stub, size_t stub_size,
5593 const struct nlattr **actionsp, size_t *actions_lenp)
5595 union user_action_cookie cookie;
5598 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5599 cookie.slow_path.unused = 0;
5600 cookie.slow_path.reason = slow;
5602 ofpbuf_use_stack(&buf, stub, stub_size);
5603 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
5604 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5605 odp_put_userspace_action(pid, &cookie, sizeof cookie, &buf);
5607 put_userspace_action(ofproto, &buf, flow, &cookie);
5609 *actionsp = buf.data;
5610 *actions_lenp = buf.size;
5614 put_userspace_action(const struct ofproto_dpif *ofproto,
5615 struct ofpbuf *odp_actions,
5616 const struct flow *flow,
5617 const union user_action_cookie *cookie)
5621 pid = dpif_port_get_pid(ofproto->backer->dpif,
5622 ofp_port_to_odp_port(ofproto, flow->in_port));
5624 return odp_put_userspace_action(pid, cookie, sizeof *cookie, odp_actions);
5628 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5629 ovs_be16 vlan_tci, uint32_t odp_port,
5630 unsigned int n_outputs, union user_action_cookie *cookie)
5634 cookie->type = USER_ACTION_COOKIE_SFLOW;
5635 cookie->sflow.vlan_tci = vlan_tci;
5637 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5638 * port information") for the interpretation of cookie->output. */
5639 switch (n_outputs) {
5641 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5642 cookie->sflow.output = 0x40000000 | 256;
5646 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5648 cookie->sflow.output = ifindex;
5653 /* 0x80000000 means "multiple output ports. */
5654 cookie->sflow.output = 0x80000000 | n_outputs;
5659 /* Compose SAMPLE action for sFlow. */
5661 compose_sflow_action(const struct ofproto_dpif *ofproto,
5662 struct ofpbuf *odp_actions,
5663 const struct flow *flow,
5666 uint32_t probability;
5667 union user_action_cookie cookie;
5668 size_t sample_offset, actions_offset;
5671 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5675 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5677 /* Number of packets out of UINT_MAX to sample. */
5678 probability = dpif_sflow_get_probability(ofproto->sflow);
5679 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5681 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5682 compose_sflow_cookie(ofproto, htons(0), odp_port,
5683 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5684 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
5686 nl_msg_end_nested(odp_actions, actions_offset);
5687 nl_msg_end_nested(odp_actions, sample_offset);
5688 return cookie_offset;
5691 /* SAMPLE action must be first action in any given list of actions.
5692 * At this point we do not have all information required to build it. So try to
5693 * build sample action as complete as possible. */
5695 add_sflow_action(struct action_xlate_ctx *ctx)
5697 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5699 &ctx->flow, OVSP_NONE);
5700 ctx->sflow_odp_port = 0;
5701 ctx->sflow_n_outputs = 0;
5704 /* Fix SAMPLE action according to data collected while composing ODP actions.
5705 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5706 * USERSPACE action's user-cookie which is required for sflow. */
5708 fix_sflow_action(struct action_xlate_ctx *ctx)
5710 const struct flow *base = &ctx->base_flow;
5711 union user_action_cookie *cookie;
5713 if (!ctx->user_cookie_offset) {
5717 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
5719 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5721 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5722 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5726 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
5729 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5730 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
5731 ovs_be64 flow_tun_id = ctx->flow.tunnel.tun_id;
5732 uint8_t flow_nw_tos = ctx->flow.nw_tos;
5733 struct priority_to_dscp *pdscp;
5734 uint32_t out_port, odp_port;
5736 /* If 'struct flow' gets additional metadata, we'll need to zero it out
5737 * before traversing a patch port. */
5738 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 19);
5741 xlate_report(ctx, "Nonexistent output port");
5743 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5744 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5746 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5747 xlate_report(ctx, "STP not in forwarding state, skipping output");
5751 if (netdev_vport_is_patch(ofport->up.netdev)) {
5752 struct ofport_dpif *peer = ofport_get_peer(ofport);
5753 struct flow old_flow = ctx->flow;
5754 const struct ofproto_dpif *peer_ofproto;
5755 enum slow_path_reason special;
5756 struct ofport_dpif *in_port;
5759 xlate_report(ctx, "Nonexistent patch port peer");
5763 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5764 if (peer_ofproto->backer != ctx->ofproto->backer) {
5765 xlate_report(ctx, "Patch port peer on a different datapath");
5769 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
5770 ctx->flow.in_port = peer->up.ofp_port;
5771 ctx->flow.metadata = htonll(0);
5772 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
5773 memset(ctx->flow.regs, 0, sizeof ctx->flow.regs);
5775 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5776 special = process_special(ctx->ofproto, &ctx->flow, in_port,
5779 ctx->slow |= special;
5780 } else if (!in_port || may_receive(in_port, ctx)) {
5781 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
5782 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5784 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
5785 * learning action look at the packet, then drop it. */
5786 struct flow old_base_flow = ctx->base_flow;
5787 size_t old_size = ctx->odp_actions->size;
5788 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5789 ctx->base_flow = old_base_flow;
5790 ctx->odp_actions->size = old_size;
5794 ctx->flow = old_flow;
5795 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5797 if (ctx->resubmit_stats) {
5798 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
5799 netdev_vport_inc_rx(peer->up.netdev, ctx->resubmit_stats);
5805 pdscp = get_priority(ofport, ctx->flow.skb_priority);
5807 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5808 ctx->flow.nw_tos |= pdscp->dscp;
5811 odp_port = ofp_port_to_odp_port(ctx->ofproto, ofp_port);
5812 if (ofport->tnl_port) {
5813 odp_port = tnl_port_send(ofport->tnl_port, &ctx->flow);
5814 if (odp_port == OVSP_NONE) {
5815 xlate_report(ctx, "Tunneling decided against output");
5819 if (ctx->resubmit_stats) {
5820 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
5822 out_port = odp_port;
5823 commit_odp_tunnel_action(&ctx->flow, &ctx->base_flow,
5826 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
5827 ctx->flow.vlan_tci);
5828 if (out_port != odp_port) {
5829 ctx->flow.vlan_tci = htons(0);
5832 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
5833 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
5835 ctx->sflow_odp_port = odp_port;
5836 ctx->sflow_n_outputs++;
5837 ctx->nf_output_iface = ofp_port;
5838 ctx->flow.tunnel.tun_id = flow_tun_id;
5839 ctx->flow.vlan_tci = flow_vlan_tci;
5840 ctx->flow.nw_tos = flow_nw_tos;
5844 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
5846 compose_output_action__(ctx, ofp_port, true);
5850 xlate_table_action(struct action_xlate_ctx *ctx,
5851 uint16_t in_port, uint8_t table_id, bool may_packet_in)
5853 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
5854 struct ofproto_dpif *ofproto = ctx->ofproto;
5855 struct rule_dpif *rule;
5856 uint16_t old_in_port;
5857 uint8_t old_table_id;
5859 old_table_id = ctx->table_id;
5860 ctx->table_id = table_id;
5862 /* Look up a flow with 'in_port' as the input port. */
5863 old_in_port = ctx->flow.in_port;
5864 ctx->flow.in_port = in_port;
5865 rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id);
5868 if (table_id > 0 && table_id < N_TABLES) {
5869 struct table_dpif *table = &ofproto->tables[table_id];
5870 if (table->other_table) {
5871 ctx->tags |= (rule && rule->tag
5873 : rule_calculate_tag(&ctx->flow,
5874 &table->other_table->mask,
5879 /* Restore the original input port. Otherwise OFPP_NORMAL and
5880 * OFPP_IN_PORT will have surprising behavior. */
5881 ctx->flow.in_port = old_in_port;
5883 if (ctx->resubmit_hook) {
5884 ctx->resubmit_hook(ctx, rule);
5887 if (rule == NULL && may_packet_in) {
5889 * check if table configuration flags
5890 * OFPTC_TABLE_MISS_CONTROLLER, default.
5891 * OFPTC_TABLE_MISS_CONTINUE,
5892 * OFPTC_TABLE_MISS_DROP
5893 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
5895 rule = rule_dpif_miss_rule(ofproto, &ctx->flow);
5899 struct rule_dpif *old_rule = ctx->rule;
5901 if (ctx->resubmit_stats) {
5902 rule_credit_stats(rule, ctx->resubmit_stats);
5907 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
5908 ctx->rule = old_rule;
5912 ctx->table_id = old_table_id;
5914 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5916 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
5917 MAX_RESUBMIT_RECURSION);
5918 ctx->max_resubmit_trigger = true;
5923 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
5924 const struct ofpact_resubmit *resubmit)
5929 in_port = resubmit->in_port;
5930 if (in_port == OFPP_IN_PORT) {
5931 in_port = ctx->flow.in_port;
5934 table_id = resubmit->table_id;
5935 if (table_id == 255) {
5936 table_id = ctx->table_id;
5939 xlate_table_action(ctx, in_port, table_id, false);
5943 flood_packets(struct action_xlate_ctx *ctx, bool all)
5945 struct ofport_dpif *ofport;
5947 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
5948 uint16_t ofp_port = ofport->up.ofp_port;
5950 if (ofp_port == ctx->flow.in_port) {
5955 compose_output_action__(ctx, ofp_port, false);
5956 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
5957 compose_output_action(ctx, ofp_port);
5961 ctx->nf_output_iface = NF_OUT_FLOOD;
5965 execute_controller_action(struct action_xlate_ctx *ctx, int len,
5966 enum ofp_packet_in_reason reason,
5967 uint16_t controller_id)
5969 struct ofputil_packet_in pin;
5970 struct ofpbuf *packet;
5972 ctx->slow |= SLOW_CONTROLLER;
5977 packet = ofpbuf_clone(ctx->packet);
5979 if (packet->l2 && packet->l3) {
5980 struct eth_header *eh;
5981 uint16_t mpls_depth;
5983 eth_pop_vlan(packet);
5986 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
5987 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
5989 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
5990 eth_push_vlan(packet, ctx->flow.vlan_tci);
5993 mpls_depth = eth_mpls_depth(packet);
5995 if (mpls_depth < ctx->flow.mpls_depth) {
5996 push_mpls(packet, ctx->flow.dl_type, ctx->flow.mpls_lse);
5997 } else if (mpls_depth > ctx->flow.mpls_depth) {
5998 pop_mpls(packet, ctx->flow.dl_type);
5999 } else if (mpls_depth) {
6000 set_mpls_lse(packet, ctx->flow.mpls_lse);
6004 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6005 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
6006 ctx->flow.nw_tos, ctx->flow.nw_ttl);
6010 if (ctx->flow.nw_proto == IPPROTO_TCP) {
6011 packet_set_tcp_port(packet, ctx->flow.tp_src,
6013 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
6014 packet_set_udp_port(packet, ctx->flow.tp_src,
6021 pin.packet = packet->data;
6022 pin.packet_len = packet->size;
6023 pin.reason = reason;
6024 pin.controller_id = controller_id;
6025 pin.table_id = ctx->table_id;
6026 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
6029 flow_get_metadata(&ctx->flow, &pin.fmd);
6031 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6032 ofpbuf_delete(packet);
6036 execute_mpls_push_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
6038 ovs_assert(eth_type_mpls(eth_type));
6040 if (ctx->base_flow.mpls_depth) {
6041 ctx->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK);
6042 ctx->flow.mpls_depth++;
6047 if (ctx->flow.dl_type == htons(ETH_TYPE_IPV6)) {
6048 label = htonl(0x2); /* IPV6 Explicit Null. */
6050 label = htonl(0x0); /* IPV4 Explicit Null. */
6052 tc = (ctx->flow.nw_tos & IP_DSCP_MASK) >> 2;
6053 ttl = ctx->flow.nw_ttl ? ctx->flow.nw_ttl : 0x40;
6054 ctx->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
6055 ctx->flow.encap_dl_type = ctx->flow.dl_type;
6056 ctx->flow.mpls_depth = 1;
6058 ctx->flow.dl_type = eth_type;
6062 execute_mpls_pop_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
6064 ovs_assert(eth_type_mpls(ctx->flow.dl_type));
6065 ovs_assert(!eth_type_mpls(eth_type));
6067 if (ctx->flow.mpls_depth) {
6068 ctx->flow.mpls_depth--;
6069 ctx->flow.mpls_lse = htonl(0);
6070 if (!ctx->flow.mpls_depth) {
6071 ctx->flow.dl_type = eth_type;
6072 ctx->flow.encap_dl_type = htons(0);
6078 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6080 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
6081 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6085 if (ctx->flow.nw_ttl > 1) {
6091 for (i = 0; i < ids->n_controllers; i++) {
6092 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6096 /* Stop processing for current table. */
6102 xlate_output_action(struct action_xlate_ctx *ctx,
6103 uint16_t port, uint16_t max_len, bool may_packet_in)
6105 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
6107 ctx->nf_output_iface = NF_OUT_DROP;
6111 compose_output_action(ctx, ctx->flow.in_port);
6114 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
6120 flood_packets(ctx, false);
6123 flood_packets(ctx, true);
6125 case OFPP_CONTROLLER:
6126 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6132 if (port != ctx->flow.in_port) {
6133 compose_output_action(ctx, port);
6135 xlate_report(ctx, "skipping output to input port");
6140 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6141 ctx->nf_output_iface = NF_OUT_FLOOD;
6142 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
6143 ctx->nf_output_iface = prev_nf_output_iface;
6144 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6145 ctx->nf_output_iface != NF_OUT_FLOOD) {
6146 ctx->nf_output_iface = NF_OUT_MULTI;
6151 xlate_output_reg_action(struct action_xlate_ctx *ctx,
6152 const struct ofpact_output_reg *or)
6154 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
6155 if (port <= UINT16_MAX) {
6156 xlate_output_action(ctx, port, or->max_len, false);
6161 xlate_enqueue_action(struct action_xlate_ctx *ctx,
6162 const struct ofpact_enqueue *enqueue)
6164 uint16_t ofp_port = enqueue->port;
6165 uint32_t queue_id = enqueue->queue;
6166 uint32_t flow_priority, priority;
6169 /* Translate queue to priority. */
6170 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6171 queue_id, &priority);
6173 /* Fall back to ordinary output action. */
6174 xlate_output_action(ctx, enqueue->port, 0, false);
6178 /* Check output port. */
6179 if (ofp_port == OFPP_IN_PORT) {
6180 ofp_port = ctx->flow.in_port;
6181 } else if (ofp_port == ctx->flow.in_port) {
6185 /* Add datapath actions. */
6186 flow_priority = ctx->flow.skb_priority;
6187 ctx->flow.skb_priority = priority;
6188 compose_output_action(ctx, ofp_port);
6189 ctx->flow.skb_priority = flow_priority;
6191 /* Update NetFlow output port. */
6192 if (ctx->nf_output_iface == NF_OUT_DROP) {
6193 ctx->nf_output_iface = ofp_port;
6194 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
6195 ctx->nf_output_iface = NF_OUT_MULTI;
6200 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
6202 uint32_t skb_priority;
6204 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6205 queue_id, &skb_priority)) {
6206 ctx->flow.skb_priority = skb_priority;
6208 /* Couldn't translate queue to a priority. Nothing to do. A warning
6209 * has already been logged. */
6213 struct xlate_reg_state {
6219 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6221 struct ofproto_dpif *ofproto = ofproto_;
6222 struct ofport_dpif *port;
6232 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6235 port = get_ofp_port(ofproto, ofp_port);
6236 return port ? port->may_enable : false;
6241 xlate_bundle_action(struct action_xlate_ctx *ctx,
6242 const struct ofpact_bundle *bundle)
6246 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
6247 if (bundle->dst.field) {
6248 nxm_reg_load(&bundle->dst, port, &ctx->flow);
6250 xlate_output_action(ctx, port, 0, false);
6255 xlate_learn_action(struct action_xlate_ctx *ctx,
6256 const struct ofpact_learn *learn)
6258 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6259 struct ofputil_flow_mod fm;
6260 uint64_t ofpacts_stub[1024 / 8];
6261 struct ofpbuf ofpacts;
6264 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6265 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
6267 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6268 if (error && !VLOG_DROP_WARN(&rl)) {
6269 VLOG_WARN("learning action failed to modify flow table (%s)",
6270 ofperr_get_name(error));
6273 ofpbuf_uninit(&ofpacts);
6276 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6277 * means "infinite". */
6279 reduce_timeout(uint16_t max, uint16_t *timeout)
6281 if (max && (!*timeout || *timeout > max)) {
6287 xlate_fin_timeout(struct action_xlate_ctx *ctx,
6288 const struct ofpact_fin_timeout *oft)
6290 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6291 struct rule_dpif *rule = ctx->rule;
6293 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6294 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6299 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
6301 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
6302 ? OFPUTIL_PC_NO_RECV_STP
6303 : OFPUTIL_PC_NO_RECV)) {
6307 /* Only drop packets here if both forwarding and learning are
6308 * disabled. If just learning is enabled, we need to have
6309 * OFPP_NORMAL and the learning action have a look at the packet
6310 * before we can drop it. */
6311 if (!stp_forward_in_state(port->stp_state)
6312 && !stp_learn_in_state(port->stp_state)) {
6320 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6321 struct action_xlate_ctx *ctx)
6323 bool was_evictable = true;
6324 const struct ofpact *a;
6327 /* Don't let the rule we're working on get evicted underneath us. */
6328 was_evictable = ctx->rule->up.evictable;
6329 ctx->rule->up.evictable = false;
6331 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6332 struct ofpact_controller *controller;
6333 const struct ofpact_metadata *metadata;
6341 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6342 ofpact_get_OUTPUT(a)->max_len, true);
6345 case OFPACT_CONTROLLER:
6346 controller = ofpact_get_CONTROLLER(a);
6347 execute_controller_action(ctx, controller->max_len,
6349 controller->controller_id);
6352 case OFPACT_ENQUEUE:
6353 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6356 case OFPACT_SET_VLAN_VID:
6357 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6358 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6362 case OFPACT_SET_VLAN_PCP:
6363 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6364 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
6369 case OFPACT_STRIP_VLAN:
6370 ctx->flow.vlan_tci = htons(0);
6373 case OFPACT_PUSH_VLAN:
6374 /* XXX 802.1AD(QinQ) */
6375 ctx->flow.vlan_tci = htons(VLAN_CFI);
6378 case OFPACT_SET_ETH_SRC:
6379 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6383 case OFPACT_SET_ETH_DST:
6384 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6388 case OFPACT_SET_IPV4_SRC:
6389 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6392 case OFPACT_SET_IPV4_DST:
6393 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6396 case OFPACT_SET_IPV4_DSCP:
6397 /* OpenFlow 1.0 only supports IPv4. */
6398 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6399 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
6400 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6404 case OFPACT_SET_L4_SRC_PORT:
6405 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6408 case OFPACT_SET_L4_DST_PORT:
6409 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6412 case OFPACT_RESUBMIT:
6413 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6416 case OFPACT_SET_TUNNEL:
6417 ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6420 case OFPACT_SET_QUEUE:
6421 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6424 case OFPACT_POP_QUEUE:
6425 ctx->flow.skb_priority = ctx->orig_skb_priority;
6428 case OFPACT_REG_MOVE:
6429 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
6432 case OFPACT_REG_LOAD:
6433 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
6436 case OFPACT_PUSH_MPLS:
6437 execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
6440 case OFPACT_POP_MPLS:
6441 execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
6444 case OFPACT_DEC_TTL:
6445 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6451 /* Nothing to do. */
6454 case OFPACT_MULTIPATH:
6455 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
6459 ctx->ofproto->has_bundle_action = true;
6460 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6463 case OFPACT_OUTPUT_REG:
6464 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6468 ctx->has_learn = true;
6469 if (ctx->may_learn) {
6470 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6478 case OFPACT_FIN_TIMEOUT:
6479 ctx->has_fin_timeout = true;
6480 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6483 case OFPACT_CLEAR_ACTIONS:
6485 * Nothing to do because writa-actions is not supported for now.
6486 * When writa-actions is supported, clear-actions also must
6487 * be supported at the same time.
6491 case OFPACT_WRITE_METADATA:
6492 metadata = ofpact_get_WRITE_METADATA(a);
6493 ctx->flow.metadata &= ~metadata->mask;
6494 ctx->flow.metadata |= metadata->metadata & metadata->mask;
6497 case OFPACT_GOTO_TABLE: {
6498 /* XXX remove recursion */
6499 /* It is assumed that goto-table is last action */
6500 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6501 ovs_assert(ctx->table_id < ogt->table_id);
6502 xlate_table_action(ctx, ctx->flow.in_port, ogt->table_id, true);
6510 ctx->rule->up.evictable = was_evictable;
6515 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
6516 struct ofproto_dpif *ofproto, const struct flow *flow,
6517 ovs_be16 initial_tci, struct rule_dpif *rule,
6518 uint8_t tcp_flags, const struct ofpbuf *packet)
6520 ovs_be64 initial_tun_id = flow->tunnel.tun_id;
6522 /* Flow initialization rules:
6523 * - 'base_flow' must match the kernel's view of the packet at the
6524 * time that action processing starts. 'flow' represents any
6525 * transformations we wish to make through actions.
6526 * - By default 'base_flow' and 'flow' are the same since the input
6527 * packet matches the output before any actions are applied.
6528 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
6529 * of the received packet as seen by the kernel. If we later output
6530 * to another device without any modifications this will cause us to
6531 * insert a new tag since the original one was stripped off by the
6533 * - Tunnel 'flow' is largely cleared when transitioning between
6534 * the input and output stages since it does not make sense to output
6535 * a packet with the exact headers that it was received with (i.e.
6536 * the destination IP is us). The one exception is the tun_id, which
6537 * is preserved to allow use in later resubmit lookups and loads into
6539 * - Tunnel 'base_flow' is completely cleared since that is what the
6540 * kernel does. If we wish to maintain the original values an action
6541 * needs to be generated. */
6543 ctx->ofproto = ofproto;
6545 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
6546 ctx->base_flow = ctx->flow;
6547 ctx->base_flow.vlan_tci = initial_tci;
6548 ctx->flow.tunnel.tun_id = initial_tun_id;
6550 ctx->packet = packet;
6551 ctx->may_learn = packet != NULL;
6552 ctx->tcp_flags = tcp_flags;
6553 ctx->resubmit_hook = NULL;
6554 ctx->report_hook = NULL;
6555 ctx->resubmit_stats = NULL;
6558 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6559 * into datapath actions in 'odp_actions', using 'ctx'. */
6561 xlate_actions(struct action_xlate_ctx *ctx,
6562 const struct ofpact *ofpacts, size_t ofpacts_len,
6563 struct ofpbuf *odp_actions)
6565 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6566 * that in the future we always keep a copy of the original flow for
6567 * tracing purposes. */
6568 static bool hit_resubmit_limit;
6570 enum slow_path_reason special;
6571 struct ofport_dpif *in_port;
6572 struct flow orig_flow;
6574 COVERAGE_INC(ofproto_dpif_xlate);
6576 ofpbuf_clear(odp_actions);
6577 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
6579 ctx->odp_actions = odp_actions;
6582 ctx->has_learn = false;
6583 ctx->has_normal = false;
6584 ctx->has_fin_timeout = false;
6585 ctx->nf_output_iface = NF_OUT_DROP;
6588 ctx->max_resubmit_trigger = false;
6589 ctx->orig_skb_priority = ctx->flow.skb_priority;
6593 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
6594 /* Do this conditionally because the copy is expensive enough that it
6595 * shows up in profiles. */
6596 orig_flow = ctx->flow;
6599 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6600 switch (ctx->ofproto->up.frag_handling) {
6601 case OFPC_FRAG_NORMAL:
6602 /* We must pretend that transport ports are unavailable. */
6603 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
6604 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
6607 case OFPC_FRAG_DROP:
6610 case OFPC_FRAG_REASM:
6613 case OFPC_FRAG_NX_MATCH:
6614 /* Nothing to do. */
6617 case OFPC_INVALID_TTL_TO_CONTROLLER:
6622 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
6623 special = process_special(ctx->ofproto, &ctx->flow, in_port, ctx->packet);
6625 ctx->slow |= special;
6627 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6628 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
6629 uint32_t local_odp_port;
6631 add_sflow_action(ctx);
6633 if (!in_port || may_receive(in_port, ctx)) {
6634 do_xlate_actions(ofpacts, ofpacts_len, ctx);
6636 /* We've let OFPP_NORMAL and the learning action look at the
6637 * packet, so drop it now if forwarding is disabled. */
6638 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
6639 ofpbuf_clear(ctx->odp_actions);
6640 add_sflow_action(ctx);
6644 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
6645 if (!hit_resubmit_limit) {
6646 /* We didn't record the original flow. Make sure we do from
6648 hit_resubmit_limit = true;
6649 } else if (!VLOG_DROP_ERR(&trace_rl)) {
6650 struct ds ds = DS_EMPTY_INITIALIZER;
6652 ofproto_trace(ctx->ofproto, &orig_flow, ctx->packet,
6654 VLOG_ERR("Trace triggered by excessive resubmit "
6655 "recursion:\n%s", ds_cstr(&ds));
6660 local_odp_port = ofp_port_to_odp_port(ctx->ofproto, OFPP_LOCAL);
6661 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
6663 ctx->odp_actions->data,
6664 ctx->odp_actions->size)) {
6665 ctx->slow |= SLOW_IN_BAND;
6667 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
6669 compose_output_action(ctx, OFPP_LOCAL);
6672 if (ctx->ofproto->has_mirrors) {
6673 add_mirror_actions(ctx, &orig_flow);
6675 fix_sflow_action(ctx);
6679 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
6680 * into datapath actions, using 'ctx', and discards the datapath actions. */
6682 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
6683 const struct ofpact *ofpacts,
6686 uint64_t odp_actions_stub[1024 / 8];
6687 struct ofpbuf odp_actions;
6689 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
6690 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
6691 ofpbuf_uninit(&odp_actions);
6695 xlate_report(struct action_xlate_ctx *ctx, const char *s)
6697 if (ctx->report_hook) {
6698 ctx->report_hook(ctx, s);
6702 /* OFPP_NORMAL implementation. */
6704 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
6706 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
6707 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
6708 * the bundle on which the packet was received, returns the VLAN to which the
6711 * Both 'vid' and the return value are in the range 0...4095. */
6713 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
6715 switch (in_bundle->vlan_mode) {
6716 case PORT_VLAN_ACCESS:
6717 return in_bundle->vlan;
6720 case PORT_VLAN_TRUNK:
6723 case PORT_VLAN_NATIVE_UNTAGGED:
6724 case PORT_VLAN_NATIVE_TAGGED:
6725 return vid ? vid : in_bundle->vlan;
6732 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
6733 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
6736 * 'vid' should be the VID obtained from the 802.1Q header that was received as
6737 * part of a packet (specify 0 if there was no 802.1Q header), in the range
6740 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
6742 /* Allow any VID on the OFPP_NONE port. */
6743 if (in_bundle == &ofpp_none_bundle) {
6747 switch (in_bundle->vlan_mode) {
6748 case PORT_VLAN_ACCESS:
6751 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6752 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
6753 "packet received on port %s configured as VLAN "
6754 "%"PRIu16" access port",
6755 in_bundle->ofproto->up.name, vid,
6756 in_bundle->name, in_bundle->vlan);
6762 case PORT_VLAN_NATIVE_UNTAGGED:
6763 case PORT_VLAN_NATIVE_TAGGED:
6765 /* Port must always carry its native VLAN. */
6769 case PORT_VLAN_TRUNK:
6770 if (!ofbundle_includes_vlan(in_bundle, vid)) {
6772 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6773 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
6774 "received on port %s not configured for trunking "
6776 in_bundle->ofproto->up.name, vid,
6777 in_bundle->name, vid);
6789 /* Given 'vlan', the VLAN that a packet belongs to, and
6790 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
6791 * that should be included in the 802.1Q header. (If the return value is 0,
6792 * then the 802.1Q header should only be included in the packet if there is a
6795 * Both 'vlan' and the return value are in the range 0...4095. */
6797 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
6799 switch (out_bundle->vlan_mode) {
6800 case PORT_VLAN_ACCESS:
6803 case PORT_VLAN_TRUNK:
6804 case PORT_VLAN_NATIVE_TAGGED:
6807 case PORT_VLAN_NATIVE_UNTAGGED:
6808 return vlan == out_bundle->vlan ? 0 : vlan;
6816 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
6819 struct ofport_dpif *port;
6821 ovs_be16 tci, old_tci;
6823 vid = output_vlan_to_vid(out_bundle, vlan);
6824 if (!out_bundle->bond) {
6825 port = ofbundle_get_a_port(out_bundle);
6827 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
6830 /* No slaves enabled, so drop packet. */
6835 old_tci = ctx->flow.vlan_tci;
6837 if (tci || out_bundle->use_priority_tags) {
6838 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
6840 tci |= htons(VLAN_CFI);
6843 ctx->flow.vlan_tci = tci;
6845 compose_output_action(ctx, port->up.ofp_port);
6846 ctx->flow.vlan_tci = old_tci;
6850 mirror_mask_ffs(mirror_mask_t mask)
6852 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
6857 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
6859 return (bundle->vlan_mode != PORT_VLAN_ACCESS
6860 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
6864 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
6866 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
6869 /* Returns an arbitrary interface within 'bundle'. */
6870 static struct ofport_dpif *
6871 ofbundle_get_a_port(const struct ofbundle *bundle)
6873 return CONTAINER_OF(list_front(&bundle->ports),
6874 struct ofport_dpif, bundle_node);
6878 vlan_is_mirrored(const struct ofmirror *m, int vlan)
6880 return !m->vlans || bitmap_is_set(m->vlans, vlan);
6884 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
6886 struct ofproto_dpif *ofproto = ctx->ofproto;
6887 mirror_mask_t mirrors;
6888 struct ofbundle *in_bundle;
6891 const struct nlattr *a;
6894 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
6895 ctx->packet != NULL, NULL);
6899 mirrors = in_bundle->src_mirrors;
6901 /* Drop frames on bundles reserved for mirroring. */
6902 if (in_bundle->mirror_out) {
6903 if (ctx->packet != NULL) {
6904 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6905 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6906 "%s, which is reserved exclusively for mirroring",
6907 ctx->ofproto->up.name, in_bundle->name);
6913 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
6914 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6917 vlan = input_vid_to_vlan(in_bundle, vid);
6919 /* Look at the output ports to check for destination selections. */
6921 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
6922 ctx->odp_actions->size) {
6923 enum ovs_action_attr type = nl_attr_type(a);
6924 struct ofport_dpif *ofport;
6926 if (type != OVS_ACTION_ATTR_OUTPUT) {
6930 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
6931 if (ofport && ofport->bundle) {
6932 mirrors |= ofport->bundle->dst_mirrors;
6940 /* Restore the original packet before adding the mirror actions. */
6941 ctx->flow = *orig_flow;
6946 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6948 if (!vlan_is_mirrored(m, vlan)) {
6949 mirrors = zero_rightmost_1bit(mirrors);
6953 mirrors &= ~m->dup_mirrors;
6954 ctx->mirrors |= m->dup_mirrors;
6956 output_normal(ctx, m->out, vlan);
6957 } else if (vlan != m->out_vlan
6958 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
6959 struct ofbundle *bundle;
6961 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
6962 if (ofbundle_includes_vlan(bundle, m->out_vlan)
6963 && !bundle->mirror_out) {
6964 output_normal(ctx, bundle, m->out_vlan);
6972 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
6973 uint64_t packets, uint64_t bytes)
6979 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
6982 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6985 /* In normal circumstances 'm' will not be NULL. However,
6986 * if mirrors are reconfigured, we can temporarily get out
6987 * of sync in facet_revalidate(). We could "correct" the
6988 * mirror list before reaching here, but doing that would
6989 * not properly account the traffic stats we've currently
6990 * accumulated for previous mirror configuration. */
6994 m->packet_count += packets;
6995 m->byte_count += bytes;
6999 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
7000 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
7001 * indicate this; newer upstream kernels use gratuitous ARP requests. */
7003 is_gratuitous_arp(const struct flow *flow)
7005 return (flow->dl_type == htons(ETH_TYPE_ARP)
7006 && eth_addr_is_broadcast(flow->dl_dst)
7007 && (flow->nw_proto == ARP_OP_REPLY
7008 || (flow->nw_proto == ARP_OP_REQUEST
7009 && flow->nw_src == flow->nw_dst)));
7013 update_learning_table(struct ofproto_dpif *ofproto,
7014 const struct flow *flow, int vlan,
7015 struct ofbundle *in_bundle)
7017 struct mac_entry *mac;
7019 /* Don't learn the OFPP_NONE port. */
7020 if (in_bundle == &ofpp_none_bundle) {
7024 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7028 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7029 if (is_gratuitous_arp(flow)) {
7030 /* We don't want to learn from gratuitous ARP packets that are
7031 * reflected back over bond slaves so we lock the learning table. */
7032 if (!in_bundle->bond) {
7033 mac_entry_set_grat_arp_lock(mac);
7034 } else if (mac_entry_is_grat_arp_locked(mac)) {
7039 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7040 /* The log messages here could actually be useful in debugging,
7041 * so keep the rate limit relatively high. */
7042 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7043 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7044 "on port %s in VLAN %d",
7045 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7046 in_bundle->name, vlan);
7048 mac->port.p = in_bundle;
7049 tag_set_add(&ofproto->backer->revalidate_set,
7050 mac_learning_changed(ofproto->ml, mac));
7054 static struct ofbundle *
7055 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7056 bool warn, struct ofport_dpif **in_ofportp)
7058 struct ofport_dpif *ofport;
7060 /* Find the port and bundle for the received packet. */
7061 ofport = get_ofp_port(ofproto, in_port);
7063 *in_ofportp = ofport;
7065 if (ofport && ofport->bundle) {
7066 return ofport->bundle;
7069 /* Special-case OFPP_NONE, which a controller may use as the ingress
7070 * port for traffic that it is sourcing. */
7071 if (in_port == OFPP_NONE) {
7072 return &ofpp_none_bundle;
7075 /* Odd. A few possible reasons here:
7077 * - We deleted a port but there are still a few packets queued up
7080 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7081 * we don't know about.
7083 * - The ofproto client didn't configure the port as part of a bundle.
7084 * This is particularly likely to happen if a packet was received on the
7085 * port after it was created, but before the client had a chance to
7086 * configure its bundle.
7089 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7091 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7092 "port %"PRIu16, ofproto->up.name, in_port);
7097 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7098 * dropped. Returns true if they may be forwarded, false if they should be
7101 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7102 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7104 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7105 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7106 * checked by input_vid_is_valid().
7108 * May also add tags to '*tags', although the current implementation only does
7109 * so in one special case.
7112 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
7115 struct ofproto_dpif *ofproto = ctx->ofproto;
7116 struct flow *flow = &ctx->flow;
7117 struct ofbundle *in_bundle = in_port->bundle;
7119 /* Drop frames for reserved multicast addresses
7120 * only if forward_bpdu option is absent. */
7121 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7122 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7126 if (in_bundle->bond) {
7127 struct mac_entry *mac;
7129 switch (bond_check_admissibility(in_bundle->bond, in_port,
7130 flow->dl_dst, &ctx->tags)) {
7135 xlate_report(ctx, "bonding refused admissibility, dropping");
7138 case BV_DROP_IF_MOVED:
7139 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7140 if (mac && mac->port.p != in_bundle &&
7141 (!is_gratuitous_arp(flow)
7142 || mac_entry_is_grat_arp_locked(mac))) {
7143 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7155 xlate_normal(struct action_xlate_ctx *ctx)
7157 struct ofport_dpif *in_port;
7158 struct ofbundle *in_bundle;
7159 struct mac_entry *mac;
7163 ctx->has_normal = true;
7165 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
7166 ctx->packet != NULL, &in_port);
7168 xlate_report(ctx, "no input bundle, dropping");
7172 /* Drop malformed frames. */
7173 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7174 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
7175 if (ctx->packet != NULL) {
7176 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7177 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7178 "VLAN tag received on port %s",
7179 ctx->ofproto->up.name, in_bundle->name);
7181 xlate_report(ctx, "partial VLAN tag, dropping");
7185 /* Drop frames on bundles reserved for mirroring. */
7186 if (in_bundle->mirror_out) {
7187 if (ctx->packet != NULL) {
7188 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7189 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7190 "%s, which is reserved exclusively for mirroring",
7191 ctx->ofproto->up.name, in_bundle->name);
7193 xlate_report(ctx, "input port is mirror output port, dropping");
7198 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
7199 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7200 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7203 vlan = input_vid_to_vlan(in_bundle, vid);
7205 /* Check other admissibility requirements. */
7206 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7210 /* Learn source MAC. */
7211 if (ctx->may_learn) {
7212 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
7215 /* Determine output bundle. */
7216 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
7219 if (mac->port.p != in_bundle) {
7220 xlate_report(ctx, "forwarding to learned port");
7221 output_normal(ctx, mac->port.p, vlan);
7223 xlate_report(ctx, "learned port is input port, dropping");
7226 struct ofbundle *bundle;
7228 xlate_report(ctx, "no learned MAC for destination, flooding");
7229 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7230 if (bundle != in_bundle
7231 && ofbundle_includes_vlan(bundle, vlan)
7232 && bundle->floodable
7233 && !bundle->mirror_out) {
7234 output_normal(ctx, bundle, vlan);
7237 ctx->nf_output_iface = NF_OUT_FLOOD;
7241 /* Optimized flow revalidation.
7243 * It's a difficult problem, in general, to tell which facets need to have
7244 * their actions recalculated whenever the OpenFlow flow table changes. We
7245 * don't try to solve that general problem: for most kinds of OpenFlow flow
7246 * table changes, we recalculate the actions for every facet. This is
7247 * relatively expensive, but it's good enough if the OpenFlow flow table
7248 * doesn't change very often.
7250 * However, we can expect one particular kind of OpenFlow flow table change to
7251 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7252 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7253 * table, we add a special case that applies to flow tables in which every rule
7254 * has the same form (that is, the same wildcards), except that the table is
7255 * also allowed to have a single "catch-all" flow that matches all packets. We
7256 * optimize this case by tagging all of the facets that resubmit into the table
7257 * and invalidating the same tag whenever a flow changes in that table. The
7258 * end result is that we revalidate just the facets that need it (and sometimes
7259 * a few more, but not all of the facets or even all of the facets that
7260 * resubmit to the table modified by MAC learning). */
7262 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7263 * into an OpenFlow table with the given 'basis'. */
7265 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7268 if (minimask_is_catchall(mask)) {
7271 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7272 return tag_create_deterministic(hash);
7276 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7277 * taggability of that table.
7279 * This function must be called after *each* change to a flow table. If you
7280 * skip calling it on some changes then the pointer comparisons at the end can
7281 * be invalid if you get unlucky. For example, if a flow removal causes a
7282 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7283 * different wildcards to be created with the same address, then this function
7284 * will incorrectly skip revalidation. */
7286 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7288 struct table_dpif *table = &ofproto->tables[table_id];
7289 const struct oftable *oftable = &ofproto->up.tables[table_id];
7290 struct cls_table *catchall, *other;
7291 struct cls_table *t;
7293 catchall = other = NULL;
7295 switch (hmap_count(&oftable->cls.tables)) {
7297 /* We could tag this OpenFlow table but it would make the logic a
7298 * little harder and it's a corner case that doesn't seem worth it
7304 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7305 if (cls_table_is_catchall(t)) {
7307 } else if (!other) {
7310 /* Indicate that we can't tag this by setting both tables to
7311 * NULL. (We know that 'catchall' is already NULL.) */
7318 /* Can't tag this table. */
7322 if (table->catchall_table != catchall || table->other_table != other) {
7323 table->catchall_table = catchall;
7324 table->other_table = other;
7325 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7329 /* Given 'rule' that has changed in some way (either it is a rule being
7330 * inserted, a rule being deleted, or a rule whose actions are being
7331 * modified), marks facets for revalidation to ensure that packets will be
7332 * forwarded correctly according to the new state of the flow table.
7334 * This function must be called after *each* change to a flow table. See
7335 * the comment on table_update_taggable() for more information. */
7337 rule_invalidate(const struct rule_dpif *rule)
7339 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7341 table_update_taggable(ofproto, rule->up.table_id);
7343 if (!ofproto->backer->need_revalidate) {
7344 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7346 if (table->other_table && rule->tag) {
7347 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7349 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7355 set_frag_handling(struct ofproto *ofproto_,
7356 enum ofp_config_flags frag_handling)
7358 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7359 if (frag_handling != OFPC_FRAG_REASM) {
7360 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7368 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7369 const struct flow *flow,
7370 const struct ofpact *ofpacts, size_t ofpacts_len)
7372 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7373 struct odputil_keybuf keybuf;
7374 struct dpif_flow_stats stats;
7378 struct action_xlate_ctx ctx;
7379 uint64_t odp_actions_stub[1024 / 8];
7380 struct ofpbuf odp_actions;
7382 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7383 odp_flow_key_from_flow(&key, flow,
7384 ofp_port_to_odp_port(ofproto, flow->in_port));
7386 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7388 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL,
7389 packet_get_tcp_flags(packet, flow), packet);
7390 ctx.resubmit_stats = &stats;
7392 ofpbuf_use_stub(&odp_actions,
7393 odp_actions_stub, sizeof odp_actions_stub);
7394 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
7395 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7396 odp_actions.data, odp_actions.size, packet);
7397 ofpbuf_uninit(&odp_actions);
7405 set_netflow(struct ofproto *ofproto_,
7406 const struct netflow_options *netflow_options)
7408 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7410 if (netflow_options) {
7411 if (!ofproto->netflow) {
7412 ofproto->netflow = netflow_create();
7414 return netflow_set_options(ofproto->netflow, netflow_options);
7416 netflow_destroy(ofproto->netflow);
7417 ofproto->netflow = NULL;
7423 get_netflow_ids(const struct ofproto *ofproto_,
7424 uint8_t *engine_type, uint8_t *engine_id)
7426 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7428 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7432 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7434 if (!facet_is_controller_flow(facet) &&
7435 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7436 struct subfacet *subfacet;
7437 struct ofexpired expired;
7439 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7440 if (subfacet->path == SF_FAST_PATH) {
7441 struct dpif_flow_stats stats;
7443 subfacet_reinstall(subfacet, &stats);
7444 subfacet_update_stats(subfacet, &stats);
7448 expired.flow = facet->flow;
7449 expired.packet_count = facet->packet_count;
7450 expired.byte_count = facet->byte_count;
7451 expired.used = facet->used;
7452 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
7457 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7459 struct facet *facet;
7461 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7462 send_active_timeout(ofproto, facet);
7466 static struct ofproto_dpif *
7467 ofproto_dpif_lookup(const char *name)
7469 struct ofproto_dpif *ofproto;
7471 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7472 hash_string(name, 0), &all_ofproto_dpifs) {
7473 if (!strcmp(ofproto->up.name, name)) {
7481 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7482 const char *argv[], void *aux OVS_UNUSED)
7484 struct ofproto_dpif *ofproto;
7487 ofproto = ofproto_dpif_lookup(argv[1]);
7489 unixctl_command_reply_error(conn, "no such bridge");
7492 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7494 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7495 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7499 unixctl_command_reply(conn, "table successfully flushed");
7503 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7504 const char *argv[], void *aux OVS_UNUSED)
7506 struct ds ds = DS_EMPTY_INITIALIZER;
7507 const struct ofproto_dpif *ofproto;
7508 const struct mac_entry *e;
7510 ofproto = ofproto_dpif_lookup(argv[1]);
7512 unixctl_command_reply_error(conn, "no such bridge");
7516 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7517 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7518 struct ofbundle *bundle = e->port.p;
7519 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7520 ofbundle_get_a_port(bundle)->odp_port,
7521 e->vlan, ETH_ADDR_ARGS(e->mac),
7522 mac_entry_age(ofproto->ml, e));
7524 unixctl_command_reply(conn, ds_cstr(&ds));
7529 struct action_xlate_ctx ctx;
7535 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7536 const struct rule_dpif *rule)
7538 ds_put_char_multiple(result, '\t', level);
7540 ds_put_cstr(result, "No match\n");
7544 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7545 table_id, ntohll(rule->up.flow_cookie));
7546 cls_rule_format(&rule->up.cr, result);
7547 ds_put_char(result, '\n');
7549 ds_put_char_multiple(result, '\t', level);
7550 ds_put_cstr(result, "OpenFlow ");
7551 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7552 ds_put_char(result, '\n');
7556 trace_format_flow(struct ds *result, int level, const char *title,
7557 struct trace_ctx *trace)
7559 ds_put_char_multiple(result, '\t', level);
7560 ds_put_format(result, "%s: ", title);
7561 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
7562 ds_put_cstr(result, "unchanged");
7564 flow_format(result, &trace->ctx.flow);
7565 trace->flow = trace->ctx.flow;
7567 ds_put_char(result, '\n');
7571 trace_format_regs(struct ds *result, int level, const char *title,
7572 struct trace_ctx *trace)
7576 ds_put_char_multiple(result, '\t', level);
7577 ds_put_format(result, "%s:", title);
7578 for (i = 0; i < FLOW_N_REGS; i++) {
7579 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7581 ds_put_char(result, '\n');
7585 trace_format_odp(struct ds *result, int level, const char *title,
7586 struct trace_ctx *trace)
7588 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
7590 ds_put_char_multiple(result, '\t', level);
7591 ds_put_format(result, "%s: ", title);
7592 format_odp_actions(result, odp_actions->data, odp_actions->size);
7593 ds_put_char(result, '\n');
7597 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
7599 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7600 struct ds *result = trace->result;
7602 ds_put_char(result, '\n');
7603 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7604 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
7605 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
7606 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
7610 trace_report(struct action_xlate_ctx *ctx, const char *s)
7612 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7613 struct ds *result = trace->result;
7615 ds_put_char_multiple(result, '\t', ctx->recurse);
7616 ds_put_cstr(result, s);
7617 ds_put_char(result, '\n');
7621 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
7622 void *aux OVS_UNUSED)
7624 const char *dpname = argv[1];
7625 struct ofproto_dpif *ofproto;
7626 struct ofpbuf odp_key;
7627 struct ofpbuf *packet;
7628 ovs_be16 initial_tci;
7634 ofpbuf_init(&odp_key, 0);
7637 ofproto = ofproto_dpif_lookup(dpname);
7639 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
7643 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
7644 /* ofproto/trace dpname flow [-generate] */
7645 const char *flow_s = argv[2];
7646 const char *generate_s = argv[3];
7648 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
7649 * flow. We guess which type it is based on whether 'flow_s' contains
7650 * an '(', since a datapath flow always contains '(') but an
7651 * OpenFlow-like flow should not (in fact it's allowed but I believe
7652 * that's not documented anywhere).
7654 * An alternative would be to try to parse 'flow_s' both ways, but then
7655 * it would be tricky giving a sensible error message. After all, do
7656 * you just say "syntax error" or do you present both error messages?
7657 * Both choices seem lousy. */
7658 if (strchr(flow_s, '(')) {
7661 /* Convert string to datapath key. */
7662 ofpbuf_init(&odp_key, 0);
7663 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
7665 unixctl_command_reply_error(conn, "Bad flow syntax");
7669 /* XXX: Since we allow the user to specify an ofproto, it's
7670 * possible they will specify a different ofproto than the one the
7671 * port actually belongs too. Ideally we should simply remove the
7672 * ability to specify the ofproto. */
7673 if (ofproto_receive(ofproto->backer, NULL, odp_key.data,
7674 odp_key.size, &flow, NULL, NULL, NULL,
7676 unixctl_command_reply_error(conn, "Invalid flow");
7682 error_s = parse_ofp_exact_flow(&flow, argv[2]);
7684 unixctl_command_reply_error(conn, error_s);
7689 initial_tci = flow.vlan_tci;
7692 /* Generate a packet, if requested. */
7694 packet = ofpbuf_new(0);
7695 flow_compose(packet, &flow);
7697 } else if (argc == 7) {
7698 /* ofproto/trace dpname priority tun_id in_port mark packet */
7699 const char *priority_s = argv[2];
7700 const char *tun_id_s = argv[3];
7701 const char *in_port_s = argv[4];
7702 const char *mark_s = argv[5];
7703 const char *packet_s = argv[6];
7704 uint32_t in_port = atoi(in_port_s);
7705 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
7706 uint32_t priority = atoi(priority_s);
7707 uint32_t mark = atoi(mark_s);
7710 msg = eth_from_hex(packet_s, &packet);
7712 unixctl_command_reply_error(conn, msg);
7716 ds_put_cstr(&result, "Packet: ");
7717 s = ofp_packet_to_string(packet->data, packet->size);
7718 ds_put_cstr(&result, s);
7721 flow_extract(packet, priority, mark, NULL, in_port, &flow);
7722 flow.tunnel.tun_id = tun_id;
7723 initial_tci = flow.vlan_tci;
7725 unixctl_command_reply_error(conn, "Bad command syntax");
7729 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
7730 unixctl_command_reply(conn, ds_cstr(&result));
7733 ds_destroy(&result);
7734 ofpbuf_delete(packet);
7735 ofpbuf_uninit(&odp_key);
7739 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
7740 const struct ofpbuf *packet, ovs_be16 initial_tci,
7743 struct rule_dpif *rule;
7745 ds_put_cstr(ds, "Flow: ");
7746 flow_format(ds, flow);
7747 ds_put_char(ds, '\n');
7749 rule = rule_dpif_lookup(ofproto, flow);
7751 trace_format_rule(ds, 0, 0, rule);
7752 if (rule == ofproto->miss_rule) {
7753 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
7754 } else if (rule == ofproto->no_packet_in_rule) {
7755 ds_put_cstr(ds, "\nNo match, packets dropped because "
7756 "OFPPC_NO_PACKET_IN is set on in_port.\n");
7760 uint64_t odp_actions_stub[1024 / 8];
7761 struct ofpbuf odp_actions;
7763 struct trace_ctx trace;
7766 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
7769 ofpbuf_use_stub(&odp_actions,
7770 odp_actions_stub, sizeof odp_actions_stub);
7771 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
7772 rule, tcp_flags, packet);
7773 trace.ctx.resubmit_hook = trace_resubmit;
7774 trace.ctx.report_hook = trace_report;
7775 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
7778 ds_put_char(ds, '\n');
7779 trace_format_flow(ds, 0, "Final flow", &trace);
7780 ds_put_cstr(ds, "Datapath actions: ");
7781 format_odp_actions(ds, odp_actions.data, odp_actions.size);
7782 ofpbuf_uninit(&odp_actions);
7784 if (trace.ctx.slow) {
7785 enum slow_path_reason slow;
7787 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
7788 "slow path because it:");
7789 for (slow = trace.ctx.slow; slow; ) {
7790 enum slow_path_reason bit = rightmost_1bit(slow);
7794 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
7797 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
7800 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
7803 ds_put_cstr(ds, "\n\t- Needs in-band special case "
7806 ds_put_cstr(ds, "\n\t (The datapath actions are "
7807 "incomplete--for complete actions, "
7808 "please supply a packet.)");
7811 case SLOW_CONTROLLER:
7812 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
7813 "to the OpenFlow controller.");
7816 ds_put_cstr(ds, "\n\t- Needs more specific matching "
7817 "than the datapath supports.");
7824 if (slow & ~SLOW_MATCH) {
7825 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
7826 "the special slow-path processing.");
7833 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7834 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7837 unixctl_command_reply(conn, NULL);
7841 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7842 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7845 unixctl_command_reply(conn, NULL);
7848 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
7849 * 'reply' describing the results. */
7851 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
7853 struct facet *facet;
7857 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7858 if (!facet_check_consistency(facet)) {
7863 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
7867 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
7868 ofproto->up.name, errors);
7870 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
7875 ofproto_dpif_self_check(struct unixctl_conn *conn,
7876 int argc, const char *argv[], void *aux OVS_UNUSED)
7878 struct ds reply = DS_EMPTY_INITIALIZER;
7879 struct ofproto_dpif *ofproto;
7882 ofproto = ofproto_dpif_lookup(argv[1]);
7884 unixctl_command_reply_error(conn, "Unknown ofproto (use "
7885 "ofproto/list for help)");
7888 ofproto_dpif_self_check__(ofproto, &reply);
7890 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7891 ofproto_dpif_self_check__(ofproto, &reply);
7895 unixctl_command_reply(conn, ds_cstr(&reply));
7899 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
7900 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
7901 * to destroy 'ofproto_shash' and free the returned value. */
7902 static const struct shash_node **
7903 get_ofprotos(struct shash *ofproto_shash)
7905 const struct ofproto_dpif *ofproto;
7907 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7908 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
7909 shash_add_nocopy(ofproto_shash, name, ofproto);
7912 return shash_sort(ofproto_shash);
7916 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
7917 const char *argv[] OVS_UNUSED,
7918 void *aux OVS_UNUSED)
7920 struct ds ds = DS_EMPTY_INITIALIZER;
7921 struct shash ofproto_shash;
7922 const struct shash_node **sorted_ofprotos;
7925 shash_init(&ofproto_shash);
7926 sorted_ofprotos = get_ofprotos(&ofproto_shash);
7927 for (i = 0; i < shash_count(&ofproto_shash); i++) {
7928 const struct shash_node *node = sorted_ofprotos[i];
7929 ds_put_format(&ds, "%s\n", node->name);
7932 shash_destroy(&ofproto_shash);
7933 free(sorted_ofprotos);
7935 unixctl_command_reply(conn, ds_cstr(&ds));
7940 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
7942 struct dpif_dp_stats s;
7943 const struct shash_node **ports;
7946 dpif_get_dp_stats(ofproto->backer->dpif, &s);
7948 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
7949 dpif_name(ofproto->backer->dpif));
7950 /* xxx It would be better to show bridge-specific stats instead
7951 * xxx of dp ones. */
7953 "\tlookups: hit:%"PRIu64" missed:%"PRIu64" lost:%"PRIu64"\n",
7954 s.n_hit, s.n_missed, s.n_lost);
7955 ds_put_format(ds, "\tflows: %zu\n",
7956 hmap_count(&ofproto->subfacets));
7958 ports = shash_sort(&ofproto->up.port_by_name);
7959 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
7960 const struct shash_node *node = ports[i];
7961 struct ofport *ofport = node->data;
7962 const char *name = netdev_get_name(ofport->netdev);
7963 const char *type = netdev_get_type(ofport->netdev);
7966 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
7968 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
7969 if (odp_port != OVSP_NONE) {
7970 ds_put_format(ds, "%"PRIu32":", odp_port);
7972 ds_put_cstr(ds, "none:");
7975 if (strcmp(type, "system")) {
7976 struct netdev *netdev;
7979 ds_put_format(ds, " (%s", type);
7981 error = netdev_open(name, type, &netdev);
7986 error = netdev_get_config(netdev, &config);
7988 const struct smap_node **nodes;
7991 nodes = smap_sort(&config);
7992 for (i = 0; i < smap_count(&config); i++) {
7993 const struct smap_node *node = nodes[i];
7994 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
7995 node->key, node->value);
7999 smap_destroy(&config);
8001 netdev_close(netdev);
8003 ds_put_char(ds, ')');
8005 ds_put_char(ds, '\n');
8011 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
8012 const char *argv[], void *aux OVS_UNUSED)
8014 struct ds ds = DS_EMPTY_INITIALIZER;
8015 const struct ofproto_dpif *ofproto;
8019 for (i = 1; i < argc; i++) {
8020 ofproto = ofproto_dpif_lookup(argv[i]);
8022 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
8023 "for help)", argv[i]);
8024 unixctl_command_reply_error(conn, ds_cstr(&ds));
8027 show_dp_format(ofproto, &ds);
8030 struct shash ofproto_shash;
8031 const struct shash_node **sorted_ofprotos;
8034 shash_init(&ofproto_shash);
8035 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8036 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8037 const struct shash_node *node = sorted_ofprotos[i];
8038 show_dp_format(node->data, &ds);
8041 shash_destroy(&ofproto_shash);
8042 free(sorted_ofprotos);
8045 unixctl_command_reply(conn, ds_cstr(&ds));
8050 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8051 int argc OVS_UNUSED, const char *argv[],
8052 void *aux OVS_UNUSED)
8054 struct ds ds = DS_EMPTY_INITIALIZER;
8055 const struct ofproto_dpif *ofproto;
8056 struct subfacet *subfacet;
8058 ofproto = ofproto_dpif_lookup(argv[1]);
8060 unixctl_command_reply_error(conn, "no such bridge");
8064 update_stats(ofproto->backer);
8066 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8067 struct odputil_keybuf keybuf;
8070 subfacet_get_key(subfacet, &keybuf, &key);
8071 odp_flow_key_format(key.data, key.size, &ds);
8073 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8074 subfacet->dp_packet_count, subfacet->dp_byte_count);
8075 if (subfacet->used) {
8076 ds_put_format(&ds, "%.3fs",
8077 (time_msec() - subfacet->used) / 1000.0);
8079 ds_put_format(&ds, "never");
8081 if (subfacet->facet->tcp_flags) {
8082 ds_put_cstr(&ds, ", flags:");
8083 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8086 ds_put_cstr(&ds, ", actions:");
8087 format_odp_actions(&ds, subfacet->actions, subfacet->actions_len);
8088 ds_put_char(&ds, '\n');
8091 unixctl_command_reply(conn, ds_cstr(&ds));
8096 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8097 int argc OVS_UNUSED, const char *argv[],
8098 void *aux OVS_UNUSED)
8100 struct ds ds = DS_EMPTY_INITIALIZER;
8101 struct ofproto_dpif *ofproto;
8103 ofproto = ofproto_dpif_lookup(argv[1]);
8105 unixctl_command_reply_error(conn, "no such bridge");
8109 flush(&ofproto->up);
8111 unixctl_command_reply(conn, ds_cstr(&ds));
8116 ofproto_dpif_unixctl_init(void)
8118 static bool registered;
8124 unixctl_command_register(
8126 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
8127 2, 6, ofproto_unixctl_trace, NULL);
8128 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8129 ofproto_unixctl_fdb_flush, NULL);
8130 unixctl_command_register("fdb/show", "bridge", 1, 1,
8131 ofproto_unixctl_fdb_show, NULL);
8132 unixctl_command_register("ofproto/clog", "", 0, 0,
8133 ofproto_dpif_clog, NULL);
8134 unixctl_command_register("ofproto/unclog", "", 0, 0,
8135 ofproto_dpif_unclog, NULL);
8136 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8137 ofproto_dpif_self_check, NULL);
8138 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8139 ofproto_unixctl_dpif_dump_dps, NULL);
8140 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8141 ofproto_unixctl_dpif_show, NULL);
8142 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8143 ofproto_unixctl_dpif_dump_flows, NULL);
8144 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8145 ofproto_unixctl_dpif_del_flows, NULL);
8148 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8150 * This is deprecated. It is only for compatibility with broken device drivers
8151 * in old versions of Linux that do not properly support VLANs when VLAN
8152 * devices are not used. When broken device drivers are no longer in
8153 * widespread use, we will delete these interfaces. */
8156 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8158 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8159 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8161 if (realdev_ofp_port == ofport->realdev_ofp_port
8162 && vid == ofport->vlandev_vid) {
8166 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8168 if (ofport->realdev_ofp_port) {
8171 if (realdev_ofp_port && ofport->bundle) {
8172 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8173 * themselves be part of a bundle. */
8174 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8177 ofport->realdev_ofp_port = realdev_ofp_port;
8178 ofport->vlandev_vid = vid;
8180 if (realdev_ofp_port) {
8181 vsp_add(ofport, realdev_ofp_port, vid);
8188 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8190 return hash_2words(realdev_ofp_port, vid);
8193 /* Returns the ODP port number of the Linux VLAN device that corresponds to
8194 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
8195 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
8196 * it would return the port number of eth0.9.
8198 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
8199 * function just returns its 'realdev_odp_port' argument. */
8201 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8202 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
8204 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8205 uint16_t realdev_ofp_port;
8206 int vid = vlan_tci_to_vid(vlan_tci);
8207 const struct vlan_splinter *vsp;
8209 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
8210 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8211 hash_realdev_vid(realdev_ofp_port, vid),
8212 &ofproto->realdev_vid_map) {
8213 if (vsp->realdev_ofp_port == realdev_ofp_port
8214 && vsp->vid == vid) {
8215 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
8219 return realdev_odp_port;
8222 static struct vlan_splinter *
8223 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8225 struct vlan_splinter *vsp;
8227 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8228 &ofproto->vlandev_map) {
8229 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8237 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8238 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8239 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8240 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8241 * eth0 and store 9 in '*vid'.
8243 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8244 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8247 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8248 uint16_t vlandev_ofp_port, int *vid)
8250 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8251 const struct vlan_splinter *vsp;
8253 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8258 return vsp->realdev_ofp_port;
8264 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8265 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8266 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8267 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8268 * always the case unless VLAN splinters are enabled), returns false without
8269 * making any changes. */
8271 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8276 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8281 /* Cause the flow to be processed as if it came in on the real device with
8282 * the VLAN device's VLAN ID. */
8283 flow->in_port = realdev;
8284 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8289 vsp_remove(struct ofport_dpif *port)
8291 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8292 struct vlan_splinter *vsp;
8294 vsp = vlandev_find(ofproto, port->up.ofp_port);
8296 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8297 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8300 port->realdev_ofp_port = 0;
8302 VLOG_ERR("missing vlan device record");
8307 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8309 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8311 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8312 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8313 == realdev_ofp_port)) {
8314 struct vlan_splinter *vsp;
8316 vsp = xmalloc(sizeof *vsp);
8317 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8318 hash_int(port->up.ofp_port, 0));
8319 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8320 hash_realdev_vid(realdev_ofp_port, vid));
8321 vsp->realdev_ofp_port = realdev_ofp_port;
8322 vsp->vlandev_ofp_port = port->up.ofp_port;
8325 port->realdev_ofp_port = realdev_ofp_port;
8327 VLOG_ERR("duplicate vlan device record");
8332 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8334 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8335 return ofport ? ofport->odp_port : OVSP_NONE;
8338 static struct ofport_dpif *
8339 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8341 struct ofport_dpif *port;
8343 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8344 hash_int(odp_port, 0),
8345 &backer->odp_to_ofport_map) {
8346 if (port->odp_port == odp_port) {
8355 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8357 struct ofport_dpif *port;
8359 port = odp_port_to_ofport(ofproto->backer, odp_port);
8360 if (port && &ofproto->up == port->up.ofproto) {
8361 return port->up.ofp_port;
8367 const struct ofproto_class ofproto_dpif_class = {
8402 port_is_lacp_current,
8403 NULL, /* rule_choose_table */
8410 rule_modify_actions,
8419 get_cfm_remote_mpids,
8424 get_stp_port_status,
8431 is_mirror_output_bundle,
8432 forward_bpdu_changed,
8433 set_mac_table_config,