2 * Copyright (c) 2009, 2010, 2011, 2012 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"
26 #include "byte-order.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
36 #include "mac-learning.h"
37 #include "meta-flow.h"
38 #include "multipath.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"
53 #include "unaligned.h"
55 #include "vlan-bitmap.h"
58 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
60 COVERAGE_DEFINE(ofproto_dpif_expired);
61 COVERAGE_DEFINE(ofproto_dpif_xlate);
62 COVERAGE_DEFINE(facet_changed_rule);
63 COVERAGE_DEFINE(facet_revalidate);
64 COVERAGE_DEFINE(facet_unexpected);
65 COVERAGE_DEFINE(facet_suppress);
67 /* Maximum depth of flow table recursion (due to resubmit actions) in a
68 * flow translation. */
69 #define MAX_RESUBMIT_RECURSION 32
71 /* Number of implemented OpenFlow tables. */
72 enum { N_TABLES = 255 };
73 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
74 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
84 * - Do include packets and bytes from facets that have been deleted or
85 * whose own statistics have been folded into the rule.
87 * - Do include packets and bytes sent "by hand" that were accounted to
88 * the rule without any facet being involved (this is a rare corner
89 * case in rule_execute()).
91 * - Do not include packet or bytes that can be obtained from any facet's
92 * packet_count or byte_count member or that can be obtained from the
93 * datapath by, e.g., dpif_flow_get() for any subfacet.
95 uint64_t packet_count; /* Number of packets received. */
96 uint64_t byte_count; /* Number of bytes received. */
98 tag_type tag; /* Caches rule_calculate_tag() result. */
100 struct list facets; /* List of "struct facet"s. */
103 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
105 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
108 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
109 const struct flow *);
110 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
114 static void rule_credit_stats(struct rule_dpif *,
115 const struct dpif_flow_stats *);
116 static void flow_push_stats(struct rule_dpif *, const struct flow *,
117 const struct dpif_flow_stats *);
118 static tag_type rule_calculate_tag(const struct flow *,
119 const struct flow_wildcards *,
121 static void rule_invalidate(const struct rule_dpif *);
123 #define MAX_MIRRORS 32
124 typedef uint32_t mirror_mask_t;
125 #define MIRROR_MASK_C(X) UINT32_C(X)
126 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
128 struct ofproto_dpif *ofproto; /* Owning ofproto. */
129 size_t idx; /* In ofproto's "mirrors" array. */
130 void *aux; /* Key supplied by ofproto's client. */
131 char *name; /* Identifier for log messages. */
133 /* Selection criteria. */
134 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
135 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
136 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
138 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
139 struct ofbundle *out; /* Output port or NULL. */
140 int out_vlan; /* Output VLAN or -1. */
141 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
144 int64_t packet_count; /* Number of packets sent. */
145 int64_t byte_count; /* Number of bytes sent. */
148 static void mirror_destroy(struct ofmirror *);
149 static void update_mirror_stats(struct ofproto_dpif *ofproto,
150 mirror_mask_t mirrors,
151 uint64_t packets, uint64_t bytes);
154 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
155 struct ofproto_dpif *ofproto; /* Owning ofproto. */
156 void *aux; /* Key supplied by ofproto's client. */
157 char *name; /* Identifier for log messages. */
160 struct list ports; /* Contains "struct ofport"s. */
161 enum port_vlan_mode vlan_mode; /* VLAN mode */
162 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
163 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
164 * NULL if all VLANs are trunked. */
165 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
166 struct bond *bond; /* Nonnull iff more than one port. */
167 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
170 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
172 /* Port mirroring info. */
173 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
174 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
175 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
178 static void bundle_remove(struct ofport *);
179 static void bundle_update(struct ofbundle *);
180 static void bundle_destroy(struct ofbundle *);
181 static void bundle_del_port(struct ofport_dpif *);
182 static void bundle_run(struct ofbundle *);
183 static void bundle_wait(struct ofbundle *);
184 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
185 uint16_t in_port, bool warn,
186 struct ofport_dpif **in_ofportp);
188 /* A controller may use OFPP_NONE as the ingress port to indicate that
189 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
190 * when an input bundle is needed for validation (e.g., mirroring or
191 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
192 * any 'port' structs, so care must be taken when dealing with it. */
193 static struct ofbundle ofpp_none_bundle = {
195 .vlan_mode = PORT_VLAN_TRUNK
198 static void stp_run(struct ofproto_dpif *ofproto);
199 static void stp_wait(struct ofproto_dpif *ofproto);
200 static int set_stp_port(struct ofport *,
201 const struct ofproto_port_stp_settings *);
203 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
205 struct action_xlate_ctx {
206 /* action_xlate_ctx_init() initializes these members. */
209 struct ofproto_dpif *ofproto;
211 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
212 * this flow when actions change header fields. */
215 /* The packet corresponding to 'flow', or a null pointer if we are
216 * revalidating without a packet to refer to. */
217 const struct ofpbuf *packet;
219 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
220 * actions update the flow table?
222 * We want to update these tables if we are actually processing a packet,
223 * or if we are accounting for packets that the datapath has processed, but
224 * not if we are just revalidating. */
227 /* The rule that we are currently translating, or NULL. */
228 struct rule_dpif *rule;
230 /* Union of the set of TCP flags seen so far in this flow. (Used only by
231 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
235 /* If nonnull, flow translation calls this function just before executing a
236 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
237 * when the recursion depth is exceeded.
239 * 'rule' is the rule being submitted into. It will be null if the
240 * resubmit or OFPP_TABLE action didn't find a matching rule.
242 * This is normally null so the client has to set it manually after
243 * calling action_xlate_ctx_init(). */
244 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule);
246 /* If nonnull, flow translation credits the specified statistics to each
247 * rule reached through a resubmit or OFPP_TABLE action.
249 * This is normally null so the client has to set it manually after
250 * calling action_xlate_ctx_init(). */
251 const struct dpif_flow_stats *resubmit_stats;
253 /* xlate_actions() initializes and uses these members. The client might want
254 * to look at them after it returns. */
256 struct ofpbuf *odp_actions; /* Datapath actions. */
257 tag_type tags; /* Tags associated with actions. */
258 enum slow_path_reason slow; /* 0 if fast path may be used. */
259 bool has_learn; /* Actions include NXAST_LEARN? */
260 bool has_normal; /* Actions output to OFPP_NORMAL? */
261 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
262 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
263 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
265 /* xlate_actions() initializes and uses these members, but the client has no
266 * reason to look at them. */
268 int recurse; /* Recursion level, via xlate_table_action. */
269 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
270 struct flow base_flow; /* Flow at the last commit. */
271 uint32_t orig_skb_priority; /* Priority when packet arrived. */
272 uint8_t table_id; /* OpenFlow table ID where flow was found. */
273 uint32_t sflow_n_outputs; /* Number of output ports. */
274 uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
275 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
276 bool exit; /* No further actions should be processed. */
277 struct flow orig_flow; /* Copy of original flow. */
280 static void action_xlate_ctx_init(struct action_xlate_ctx *,
281 struct ofproto_dpif *, const struct flow *,
282 ovs_be16 initial_tci, struct rule_dpif *,
283 uint8_t tcp_flags, const struct ofpbuf *);
284 static void xlate_actions(struct action_xlate_ctx *,
285 const struct ofpact *ofpacts, size_t ofpacts_len,
286 struct ofpbuf *odp_actions);
287 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
288 const struct ofpact *ofpacts,
291 static size_t put_userspace_action(const struct ofproto_dpif *,
292 struct ofpbuf *odp_actions,
294 const union user_action_cookie *);
296 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
297 enum slow_path_reason,
298 uint64_t *stub, size_t stub_size,
299 const struct nlattr **actionsp,
300 size_t *actions_lenp);
302 /* A subfacet (see "struct subfacet" below) has three possible installation
305 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
306 * case just after the subfacet is created, just before the subfacet is
307 * destroyed, or if the datapath returns an error when we try to install a
310 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
312 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
313 * ofproto_dpif is installed in the datapath.
316 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
317 SF_FAST_PATH, /* Full actions are installed. */
318 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
321 static const char *subfacet_path_to_string(enum subfacet_path);
323 /* A dpif flow and actions associated with a facet.
325 * See also the large comment on struct facet. */
328 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
329 struct list list_node; /* In struct facet's 'facets' list. */
330 struct facet *facet; /* Owning facet. */
334 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
335 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
336 * regenerate the ODP flow key from ->facet->flow. */
337 enum odp_key_fitness key_fitness;
341 long long int used; /* Time last used; time created if not used. */
343 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
344 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
348 * These should be essentially identical for every subfacet in a facet, but
349 * may differ in trivial ways due to VLAN splinters. */
350 size_t actions_len; /* Number of bytes in actions[]. */
351 struct nlattr *actions; /* Datapath actions. */
353 enum slow_path_reason slow; /* 0 if fast path may be used. */
354 enum subfacet_path path; /* Installed in datapath? */
356 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
357 * splinters can cause it to differ. This value should be removed when
358 * the VLAN splinters feature is no longer needed. */
359 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
362 static struct subfacet *subfacet_create(struct facet *, enum odp_key_fitness,
363 const struct nlattr *key,
364 size_t key_len, ovs_be16 initial_tci);
365 static struct subfacet *subfacet_find(struct ofproto_dpif *,
366 const struct nlattr *key, size_t key_len);
367 static void subfacet_destroy(struct subfacet *);
368 static void subfacet_destroy__(struct subfacet *);
369 static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *,
371 static void subfacet_reset_dp_stats(struct subfacet *,
372 struct dpif_flow_stats *);
373 static void subfacet_update_time(struct subfacet *, long long int used);
374 static void subfacet_update_stats(struct subfacet *,
375 const struct dpif_flow_stats *);
376 static void subfacet_make_actions(struct subfacet *,
377 const struct ofpbuf *packet,
378 struct ofpbuf *odp_actions);
379 static int subfacet_install(struct subfacet *,
380 const struct nlattr *actions, size_t actions_len,
381 struct dpif_flow_stats *, enum slow_path_reason);
382 static void subfacet_uninstall(struct subfacet *);
384 static enum subfacet_path subfacet_want_path(enum slow_path_reason);
386 /* An exact-match instantiation of an OpenFlow flow.
388 * A facet associates a "struct flow", which represents the Open vSwitch
389 * userspace idea of an exact-match flow, with one or more subfacets. Each
390 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
391 * the facet. When the kernel module (or other dpif implementation) and Open
392 * vSwitch userspace agree on the definition of a flow key, there is exactly
393 * one subfacet per facet. If the dpif implementation supports more-specific
394 * flow matching than userspace, however, a facet can have more than one
395 * subfacet, each of which corresponds to some distinction in flow that
396 * userspace simply doesn't understand.
398 * Flow expiration works in terms of subfacets, so a facet must have at least
399 * one subfacet or it will never expire, leaking memory. */
402 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
403 struct list list_node; /* In owning rule's 'facets' list. */
404 struct rule_dpif *rule; /* Owning rule. */
407 struct list subfacets;
408 long long int used; /* Time last used; time created if not used. */
415 * - Do include packets and bytes sent "by hand", e.g. with
418 * - Do include packets and bytes that were obtained from the datapath
419 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
420 * DPIF_FP_ZERO_STATS).
422 * - Do not include packets or bytes that can be obtained from the
423 * datapath for any existing subfacet.
425 uint64_t packet_count; /* Number of packets received. */
426 uint64_t byte_count; /* Number of bytes received. */
428 /* Resubmit statistics. */
429 uint64_t prev_packet_count; /* Number of packets from last stats push. */
430 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
431 long long int prev_used; /* Used time from last stats push. */
434 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
435 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
436 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
438 /* Properties of datapath actions.
440 * Every subfacet has its own actions because actions can differ slightly
441 * between splintered and non-splintered subfacets due to the VLAN tag
442 * being initially different (present vs. absent). All of them have these
443 * properties in common so we just store one copy of them here. */
444 bool has_learn; /* Actions include NXAST_LEARN? */
445 bool has_normal; /* Actions output to OFPP_NORMAL? */
446 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
447 tag_type tags; /* Tags that would require revalidation. */
448 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
450 /* Storage for a single subfacet, to reduce malloc() time and space
451 * overhead. (A facet always has at least one subfacet and in the common
452 * case has exactly one subfacet.) */
453 struct subfacet one_subfacet;
456 static struct facet *facet_create(struct rule_dpif *,
457 const struct flow *, uint32_t hash);
458 static void facet_remove(struct facet *);
459 static void facet_free(struct facet *);
461 static struct facet *facet_find(struct ofproto_dpif *,
462 const struct flow *, uint32_t hash);
463 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
464 const struct flow *, uint32_t hash);
465 static void facet_revalidate(struct facet *);
466 static bool facet_check_consistency(struct facet *);
468 static void facet_flush_stats(struct facet *);
470 static void facet_update_time(struct facet *, long long int used);
471 static void facet_reset_counters(struct facet *);
472 static void facet_push_stats(struct facet *);
473 static void facet_learn(struct facet *);
474 static void facet_account(struct facet *);
476 static bool facet_is_controller_flow(struct facet *);
482 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
483 struct list bundle_node; /* In struct ofbundle's "ports" list. */
484 struct cfm *cfm; /* Connectivity Fault Management, if any. */
485 tag_type tag; /* Tag associated with this port. */
486 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
487 bool may_enable; /* May be enabled in bonds. */
488 long long int carrier_seq; /* Carrier status changes. */
491 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
492 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
493 long long int stp_state_entered;
495 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
497 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
499 * This is deprecated. It is only for compatibility with broken device
500 * drivers in old versions of Linux that do not properly support VLANs when
501 * VLAN devices are not used. When broken device drivers are no longer in
502 * widespread use, we will delete these interfaces. */
503 uint16_t realdev_ofp_port;
507 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
508 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
509 * traffic egressing the 'ofport' with that priority should be marked with. */
510 struct priority_to_dscp {
511 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
512 uint32_t priority; /* Priority of this queue (see struct flow). */
514 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
517 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
519 * This is deprecated. It is only for compatibility with broken device drivers
520 * in old versions of Linux that do not properly support VLANs when VLAN
521 * devices are not used. When broken device drivers are no longer in
522 * widespread use, we will delete these interfaces. */
523 struct vlan_splinter {
524 struct hmap_node realdev_vid_node;
525 struct hmap_node vlandev_node;
526 uint16_t realdev_ofp_port;
527 uint16_t vlandev_ofp_port;
531 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
532 uint32_t realdev, ovs_be16 vlan_tci);
533 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
534 static void vsp_remove(struct ofport_dpif *);
535 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
537 static struct ofport_dpif *
538 ofport_dpif_cast(const struct ofport *ofport)
540 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
541 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
544 static void port_run(struct ofport_dpif *);
545 static void port_run_fast(struct ofport_dpif *);
546 static void port_wait(struct ofport_dpif *);
547 static int set_cfm(struct ofport *, const struct cfm_settings *);
548 static void ofport_clear_priorities(struct ofport_dpif *);
550 struct dpif_completion {
551 struct list list_node;
552 struct ofoperation *op;
555 /* Extra information about a classifier table.
556 * Currently used just for optimized flow revalidation. */
558 /* If either of these is nonnull, then this table has a form that allows
559 * flows to be tagged to avoid revalidating most flows for the most common
560 * kinds of flow table changes. */
561 struct cls_table *catchall_table; /* Table that wildcards all fields. */
562 struct cls_table *other_table; /* Table with any other wildcard set. */
563 uint32_t basis; /* Keeps each table's tags separate. */
566 /* Reasons that we might need to revalidate every facet, and corresponding
569 * A value of 0 means that there is no need to revalidate.
571 * It would be nice to have some cleaner way to integrate with coverage
572 * counters, but with only a few reasons I guess this is good enough for
574 enum revalidate_reason {
575 REV_RECONFIGURE = 1, /* Switch configuration changed. */
576 REV_STP, /* Spanning tree protocol port status change. */
577 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
578 REV_FLOW_TABLE, /* Flow table changed. */
579 REV_INCONSISTENCY /* Facet self-check failed. */
581 COVERAGE_DEFINE(rev_reconfigure);
582 COVERAGE_DEFINE(rev_stp);
583 COVERAGE_DEFINE(rev_port_toggled);
584 COVERAGE_DEFINE(rev_flow_table);
585 COVERAGE_DEFINE(rev_inconsistency);
587 struct ofproto_dpif {
588 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
593 /* Special OpenFlow rules. */
594 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
595 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
601 struct netflow *netflow;
602 struct dpif_sflow *sflow;
603 struct hmap bundles; /* Contains "struct ofbundle"s. */
604 struct mac_learning *ml;
605 struct ofmirror *mirrors[MAX_MIRRORS];
607 bool has_bonded_bundles;
610 struct timer next_expiration;
614 struct hmap subfacets;
615 struct governor *governor;
618 struct table_dpif tables[N_TABLES];
619 enum revalidate_reason need_revalidate;
620 struct tag_set revalidate_set;
622 /* Support for debugging async flow mods. */
623 struct list completions;
625 bool has_bundle_action; /* True when the first bundle action appears. */
626 struct netdev_stats stats; /* To account packets generated and consumed in
631 long long int stp_last_tick;
633 /* VLAN splinters. */
634 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
635 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
638 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
639 * for debugging the asynchronous flow_mod implementation.) */
642 /* All existing ofproto_dpif instances, indexed by ->up.name. */
643 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
645 static void ofproto_dpif_unixctl_init(void);
647 static struct ofproto_dpif *
648 ofproto_dpif_cast(const struct ofproto *ofproto)
650 assert(ofproto->ofproto_class == &ofproto_dpif_class);
651 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
654 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
656 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
658 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
659 const struct ofpbuf *, ovs_be16 initial_tci,
662 /* Packet processing. */
663 static void update_learning_table(struct ofproto_dpif *,
664 const struct flow *, int vlan,
667 #define FLOW_MISS_MAX_BATCH 50
668 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
670 /* Flow expiration. */
671 static int expire(struct ofproto_dpif *);
674 static void send_netflow_active_timeouts(struct ofproto_dpif *);
677 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
678 static size_t compose_sflow_action(const struct ofproto_dpif *,
679 struct ofpbuf *odp_actions,
680 const struct flow *, uint32_t odp_port);
681 static void add_mirror_actions(struct action_xlate_ctx *ctx,
682 const struct flow *flow);
683 /* Global variables. */
684 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
686 /* Factory functions. */
689 enumerate_types(struct sset *types)
691 dp_enumerate_types(types);
695 enumerate_names(const char *type, struct sset *names)
697 return dp_enumerate_names(type, names);
701 del(const char *type, const char *name)
706 error = dpif_open(name, type, &dpif);
708 error = dpif_delete(dpif);
714 /* Basic life-cycle. */
716 static int add_internal_flows(struct ofproto_dpif *);
718 static struct ofproto *
721 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
726 dealloc(struct ofproto *ofproto_)
728 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
733 construct(struct ofproto *ofproto_)
735 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
736 const char *name = ofproto->up.name;
740 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
742 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
746 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
747 ofproto->n_matches = 0;
749 dpif_flow_flush(ofproto->dpif);
750 dpif_recv_purge(ofproto->dpif);
752 error = dpif_recv_set(ofproto->dpif, true);
754 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
755 dpif_close(ofproto->dpif);
759 ofproto->netflow = NULL;
760 ofproto->sflow = NULL;
762 hmap_init(&ofproto->bundles);
763 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
764 for (i = 0; i < MAX_MIRRORS; i++) {
765 ofproto->mirrors[i] = NULL;
767 ofproto->has_bonded_bundles = false;
769 timer_set_duration(&ofproto->next_expiration, 1000);
771 hmap_init(&ofproto->facets);
772 hmap_init(&ofproto->subfacets);
773 ofproto->governor = NULL;
775 for (i = 0; i < N_TABLES; i++) {
776 struct table_dpif *table = &ofproto->tables[i];
778 table->catchall_table = NULL;
779 table->other_table = NULL;
780 table->basis = random_uint32();
782 ofproto->need_revalidate = 0;
783 tag_set_init(&ofproto->revalidate_set);
785 list_init(&ofproto->completions);
787 ofproto_dpif_unixctl_init();
789 ofproto->has_mirrors = false;
790 ofproto->has_bundle_action = false;
792 hmap_init(&ofproto->vlandev_map);
793 hmap_init(&ofproto->realdev_vid_map);
795 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
796 hash_string(ofproto->up.name, 0));
797 memset(&ofproto->stats, 0, sizeof ofproto->stats);
799 ofproto_init_tables(ofproto_, N_TABLES);
800 error = add_internal_flows(ofproto);
801 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
807 add_internal_flow(struct ofproto_dpif *ofproto, int id,
808 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
810 struct ofputil_flow_mod fm;
813 cls_rule_init_catchall(&fm.cr, 0);
814 cls_rule_set_reg(&fm.cr, 0, id);
815 fm.new_cookie = htonll(0);
816 fm.cookie = htonll(0);
817 fm.cookie_mask = htonll(0);
818 fm.table_id = TBL_INTERNAL;
819 fm.command = OFPFC_ADD;
825 fm.ofpacts = ofpacts->data;
826 fm.ofpacts_len = ofpacts->size;
828 error = ofproto_flow_mod(&ofproto->up, &fm);
830 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
831 id, ofperr_to_string(error));
835 *rulep = rule_dpif_lookup__(ofproto, &fm.cr.flow, TBL_INTERNAL);
836 assert(*rulep != NULL);
842 add_internal_flows(struct ofproto_dpif *ofproto)
844 struct ofpact_controller *controller;
845 uint64_t ofpacts_stub[128 / 8];
846 struct ofpbuf ofpacts;
850 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
853 controller = ofpact_put_CONTROLLER(&ofpacts);
854 controller->max_len = UINT16_MAX;
855 controller->controller_id = 0;
856 controller->reason = OFPR_NO_MATCH;
857 ofpact_pad(&ofpacts);
859 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
864 ofpbuf_clear(&ofpacts);
865 error = add_internal_flow(ofproto, id++, &ofpacts,
866 &ofproto->no_packet_in_rule);
871 complete_operations(struct ofproto_dpif *ofproto)
873 struct dpif_completion *c, *next;
875 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
876 ofoperation_complete(c->op, 0);
877 list_remove(&c->list_node);
883 destruct(struct ofproto *ofproto_)
885 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
886 struct rule_dpif *rule, *next_rule;
887 struct oftable *table;
890 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
891 complete_operations(ofproto);
893 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
894 struct cls_cursor cursor;
896 cls_cursor_init(&cursor, &table->cls, NULL);
897 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
898 ofproto_rule_destroy(&rule->up);
902 for (i = 0; i < MAX_MIRRORS; i++) {
903 mirror_destroy(ofproto->mirrors[i]);
906 netflow_destroy(ofproto->netflow);
907 dpif_sflow_destroy(ofproto->sflow);
908 hmap_destroy(&ofproto->bundles);
909 mac_learning_destroy(ofproto->ml);
911 hmap_destroy(&ofproto->facets);
912 hmap_destroy(&ofproto->subfacets);
913 governor_destroy(ofproto->governor);
915 hmap_destroy(&ofproto->vlandev_map);
916 hmap_destroy(&ofproto->realdev_vid_map);
918 dpif_close(ofproto->dpif);
922 run_fast(struct ofproto *ofproto_)
924 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
925 struct ofport_dpif *ofport;
928 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
929 port_run_fast(ofport);
932 /* Handle one or more batches of upcalls, until there's nothing left to do
933 * or until we do a fixed total amount of work.
935 * We do work in batches because it can be much cheaper to set up a number
936 * of flows and fire off their patches all at once. We do multiple batches
937 * because in some cases handling a packet can cause another packet to be
938 * queued almost immediately as part of the return flow. Both
939 * optimizations can make major improvements on some benchmarks and
940 * presumably for real traffic as well. */
942 while (work < FLOW_MISS_MAX_BATCH) {
943 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
953 run(struct ofproto *ofproto_)
955 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
956 struct ofport_dpif *ofport;
957 struct ofbundle *bundle;
961 complete_operations(ofproto);
963 dpif_run(ofproto->dpif);
965 error = run_fast(ofproto_);
970 if (timer_expired(&ofproto->next_expiration)) {
971 int delay = expire(ofproto);
972 timer_set_duration(&ofproto->next_expiration, delay);
975 if (ofproto->netflow) {
976 if (netflow_run(ofproto->netflow)) {
977 send_netflow_active_timeouts(ofproto);
980 if (ofproto->sflow) {
981 dpif_sflow_run(ofproto->sflow);
984 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
987 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
992 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
994 /* Now revalidate if there's anything to do. */
995 if (ofproto->need_revalidate
996 || !tag_set_is_empty(&ofproto->revalidate_set)) {
997 struct tag_set revalidate_set = ofproto->revalidate_set;
998 bool revalidate_all = ofproto->need_revalidate;
1001 switch (ofproto->need_revalidate) {
1002 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
1003 case REV_STP: COVERAGE_INC(rev_stp); break;
1004 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
1005 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
1006 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
1009 /* Clear the revalidation flags. */
1010 tag_set_init(&ofproto->revalidate_set);
1011 ofproto->need_revalidate = 0;
1013 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
1015 || tag_set_intersects(&revalidate_set, facet->tags)) {
1016 facet_revalidate(facet);
1021 /* Check the consistency of a random facet, to aid debugging. */
1022 if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) {
1023 struct facet *facet;
1025 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1026 struct facet, hmap_node);
1027 if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) {
1028 if (!facet_check_consistency(facet)) {
1029 ofproto->need_revalidate = REV_INCONSISTENCY;
1034 if (ofproto->governor) {
1037 governor_run(ofproto->governor);
1039 /* If the governor has shrunk to its minimum size and the number of
1040 * subfacets has dwindled, then drop the governor entirely.
1042 * For hysteresis, the number of subfacets to drop the governor is
1043 * smaller than the number needed to trigger its creation. */
1044 n_subfacets = hmap_count(&ofproto->subfacets);
1045 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1046 && governor_is_idle(ofproto->governor)) {
1047 governor_destroy(ofproto->governor);
1048 ofproto->governor = NULL;
1056 wait(struct ofproto *ofproto_)
1058 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1059 struct ofport_dpif *ofport;
1060 struct ofbundle *bundle;
1062 if (!clogged && !list_is_empty(&ofproto->completions)) {
1063 poll_immediate_wake();
1066 dpif_wait(ofproto->dpif);
1067 dpif_recv_wait(ofproto->dpif);
1068 if (ofproto->sflow) {
1069 dpif_sflow_wait(ofproto->sflow);
1071 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
1072 poll_immediate_wake();
1074 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1077 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1078 bundle_wait(bundle);
1080 if (ofproto->netflow) {
1081 netflow_wait(ofproto->netflow);
1083 mac_learning_wait(ofproto->ml);
1085 if (ofproto->need_revalidate) {
1086 /* Shouldn't happen, but if it does just go around again. */
1087 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1088 poll_immediate_wake();
1090 timer_wait(&ofproto->next_expiration);
1092 if (ofproto->governor) {
1093 governor_wait(ofproto->governor);
1098 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1100 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1102 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1103 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1107 flush(struct ofproto *ofproto_)
1109 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1110 struct facet *facet, *next_facet;
1112 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1113 /* Mark the facet as not installed so that facet_remove() doesn't
1114 * bother trying to uninstall it. There is no point in uninstalling it
1115 * individually since we are about to blow away all the facets with
1116 * dpif_flow_flush(). */
1117 struct subfacet *subfacet;
1119 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
1120 subfacet->path = SF_NOT_INSTALLED;
1121 subfacet->dp_packet_count = 0;
1122 subfacet->dp_byte_count = 0;
1124 facet_remove(facet);
1126 dpif_flow_flush(ofproto->dpif);
1130 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1131 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1133 *arp_match_ip = true;
1134 *actions = (OFPUTIL_A_OUTPUT |
1135 OFPUTIL_A_SET_VLAN_VID |
1136 OFPUTIL_A_SET_VLAN_PCP |
1137 OFPUTIL_A_STRIP_VLAN |
1138 OFPUTIL_A_SET_DL_SRC |
1139 OFPUTIL_A_SET_DL_DST |
1140 OFPUTIL_A_SET_NW_SRC |
1141 OFPUTIL_A_SET_NW_DST |
1142 OFPUTIL_A_SET_NW_TOS |
1143 OFPUTIL_A_SET_TP_SRC |
1144 OFPUTIL_A_SET_TP_DST |
1149 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
1151 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1152 struct dpif_dp_stats s;
1154 strcpy(ots->name, "classifier");
1156 dpif_get_dp_stats(ofproto->dpif, &s);
1157 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
1158 put_32aligned_be64(&ots->matched_count,
1159 htonll(s.n_hit + ofproto->n_matches));
1162 static struct ofport *
1165 struct ofport_dpif *port = xmalloc(sizeof *port);
1170 port_dealloc(struct ofport *port_)
1172 struct ofport_dpif *port = ofport_dpif_cast(port_);
1177 port_construct(struct ofport *port_)
1179 struct ofport_dpif *port = ofport_dpif_cast(port_);
1180 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1182 ofproto->need_revalidate = REV_RECONFIGURE;
1183 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
1184 port->bundle = NULL;
1186 port->tag = tag_create_random();
1187 port->may_enable = true;
1188 port->stp_port = NULL;
1189 port->stp_state = STP_DISABLED;
1190 hmap_init(&port->priorities);
1191 port->realdev_ofp_port = 0;
1192 port->vlandev_vid = 0;
1193 port->carrier_seq = netdev_get_carrier_resets(port->up.netdev);
1195 if (ofproto->sflow) {
1196 dpif_sflow_add_port(ofproto->sflow, port_);
1203 port_destruct(struct ofport *port_)
1205 struct ofport_dpif *port = ofport_dpif_cast(port_);
1206 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1208 ofproto->need_revalidate = REV_RECONFIGURE;
1209 bundle_remove(port_);
1210 set_cfm(port_, NULL);
1211 if (ofproto->sflow) {
1212 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1215 ofport_clear_priorities(port);
1216 hmap_destroy(&port->priorities);
1220 port_modified(struct ofport *port_)
1222 struct ofport_dpif *port = ofport_dpif_cast(port_);
1224 if (port->bundle && port->bundle->bond) {
1225 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1230 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1232 struct ofport_dpif *port = ofport_dpif_cast(port_);
1233 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1234 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1236 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1237 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1238 OFPUTIL_PC_NO_PACKET_IN)) {
1239 ofproto->need_revalidate = REV_RECONFIGURE;
1241 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1242 bundle_update(port->bundle);
1248 set_sflow(struct ofproto *ofproto_,
1249 const struct ofproto_sflow_options *sflow_options)
1251 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1252 struct dpif_sflow *ds = ofproto->sflow;
1254 if (sflow_options) {
1256 struct ofport_dpif *ofport;
1258 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1259 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1260 dpif_sflow_add_port(ds, &ofport->up);
1262 ofproto->need_revalidate = REV_RECONFIGURE;
1264 dpif_sflow_set_options(ds, sflow_options);
1267 dpif_sflow_destroy(ds);
1268 ofproto->need_revalidate = REV_RECONFIGURE;
1269 ofproto->sflow = NULL;
1276 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1278 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1285 struct ofproto_dpif *ofproto;
1287 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1288 ofproto->need_revalidate = REV_RECONFIGURE;
1289 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1292 if (cfm_configure(ofport->cfm, s)) {
1298 cfm_destroy(ofport->cfm);
1304 get_cfm_fault(const struct ofport *ofport_)
1306 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1308 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1312 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1315 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1318 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1326 get_cfm_health(const struct ofport *ofport_)
1328 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1330 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1333 /* Spanning Tree. */
1336 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1338 struct ofproto_dpif *ofproto = ofproto_;
1339 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1340 struct ofport_dpif *ofport;
1342 ofport = stp_port_get_aux(sp);
1344 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1345 ofproto->up.name, port_num);
1347 struct eth_header *eth = pkt->l2;
1349 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1350 if (eth_addr_is_zero(eth->eth_src)) {
1351 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1352 "with unknown MAC", ofproto->up.name, port_num);
1354 send_packet(ofport, pkt);
1360 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1362 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1364 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1366 /* Only revalidate flows if the configuration changed. */
1367 if (!s != !ofproto->stp) {
1368 ofproto->need_revalidate = REV_RECONFIGURE;
1372 if (!ofproto->stp) {
1373 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1374 send_bpdu_cb, ofproto);
1375 ofproto->stp_last_tick = time_msec();
1378 stp_set_bridge_id(ofproto->stp, s->system_id);
1379 stp_set_bridge_priority(ofproto->stp, s->priority);
1380 stp_set_hello_time(ofproto->stp, s->hello_time);
1381 stp_set_max_age(ofproto->stp, s->max_age);
1382 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1384 struct ofport *ofport;
1386 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1387 set_stp_port(ofport, NULL);
1390 stp_destroy(ofproto->stp);
1391 ofproto->stp = NULL;
1398 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1400 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1404 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1405 s->designated_root = stp_get_designated_root(ofproto->stp);
1406 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1415 update_stp_port_state(struct ofport_dpif *ofport)
1417 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1418 enum stp_state state;
1420 /* Figure out new state. */
1421 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1425 if (ofport->stp_state != state) {
1426 enum ofputil_port_state of_state;
1429 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1430 netdev_get_name(ofport->up.netdev),
1431 stp_state_name(ofport->stp_state),
1432 stp_state_name(state));
1433 if (stp_learn_in_state(ofport->stp_state)
1434 != stp_learn_in_state(state)) {
1435 /* xxx Learning action flows should also be flushed. */
1436 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1438 fwd_change = stp_forward_in_state(ofport->stp_state)
1439 != stp_forward_in_state(state);
1441 ofproto->need_revalidate = REV_STP;
1442 ofport->stp_state = state;
1443 ofport->stp_state_entered = time_msec();
1445 if (fwd_change && ofport->bundle) {
1446 bundle_update(ofport->bundle);
1449 /* Update the STP state bits in the OpenFlow port description. */
1450 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1451 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1452 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1453 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1454 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1456 ofproto_port_set_state(&ofport->up, of_state);
1460 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1461 * caller is responsible for assigning STP port numbers and ensuring
1462 * there are no duplicates. */
1464 set_stp_port(struct ofport *ofport_,
1465 const struct ofproto_port_stp_settings *s)
1467 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1468 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1469 struct stp_port *sp = ofport->stp_port;
1471 if (!s || !s->enable) {
1473 ofport->stp_port = NULL;
1474 stp_port_disable(sp);
1475 update_stp_port_state(ofport);
1478 } else if (sp && stp_port_no(sp) != s->port_num
1479 && ofport == stp_port_get_aux(sp)) {
1480 /* The port-id changed, so disable the old one if it's not
1481 * already in use by another port. */
1482 stp_port_disable(sp);
1485 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1486 stp_port_enable(sp);
1488 stp_port_set_aux(sp, ofport);
1489 stp_port_set_priority(sp, s->priority);
1490 stp_port_set_path_cost(sp, s->path_cost);
1492 update_stp_port_state(ofport);
1498 get_stp_port_status(struct ofport *ofport_,
1499 struct ofproto_port_stp_status *s)
1501 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1502 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1503 struct stp_port *sp = ofport->stp_port;
1505 if (!ofproto->stp || !sp) {
1511 s->port_id = stp_port_get_id(sp);
1512 s->state = stp_port_get_state(sp);
1513 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1514 s->role = stp_port_get_role(sp);
1515 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1521 stp_run(struct ofproto_dpif *ofproto)
1524 long long int now = time_msec();
1525 long long int elapsed = now - ofproto->stp_last_tick;
1526 struct stp_port *sp;
1529 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1530 ofproto->stp_last_tick = now;
1532 while (stp_get_changed_port(ofproto->stp, &sp)) {
1533 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1536 update_stp_port_state(ofport);
1540 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1541 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1547 stp_wait(struct ofproto_dpif *ofproto)
1550 poll_timer_wait(1000);
1554 /* Returns true if STP should process 'flow'. */
1556 stp_should_process_flow(const struct flow *flow)
1558 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1562 stp_process_packet(const struct ofport_dpif *ofport,
1563 const struct ofpbuf *packet)
1565 struct ofpbuf payload = *packet;
1566 struct eth_header *eth = payload.data;
1567 struct stp_port *sp = ofport->stp_port;
1569 /* Sink packets on ports that have STP disabled when the bridge has
1571 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1575 /* Trim off padding on payload. */
1576 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1577 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1580 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1581 stp_received_bpdu(sp, payload.data, payload.size);
1585 static struct priority_to_dscp *
1586 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1588 struct priority_to_dscp *pdscp;
1591 hash = hash_int(priority, 0);
1592 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1593 if (pdscp->priority == priority) {
1601 ofport_clear_priorities(struct ofport_dpif *ofport)
1603 struct priority_to_dscp *pdscp, *next;
1605 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1606 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1612 set_queues(struct ofport *ofport_,
1613 const struct ofproto_port_queue *qdscp_list,
1616 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1617 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1618 struct hmap new = HMAP_INITIALIZER(&new);
1621 for (i = 0; i < n_qdscp; i++) {
1622 struct priority_to_dscp *pdscp;
1626 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1627 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1632 pdscp = get_priority(ofport, priority);
1634 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1636 pdscp = xmalloc(sizeof *pdscp);
1637 pdscp->priority = priority;
1639 ofproto->need_revalidate = REV_RECONFIGURE;
1642 if (pdscp->dscp != dscp) {
1644 ofproto->need_revalidate = REV_RECONFIGURE;
1647 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1650 if (!hmap_is_empty(&ofport->priorities)) {
1651 ofport_clear_priorities(ofport);
1652 ofproto->need_revalidate = REV_RECONFIGURE;
1655 hmap_swap(&new, &ofport->priorities);
1663 /* Expires all MAC learning entries associated with 'bundle' and forces its
1664 * ofproto to revalidate every flow.
1666 * Normally MAC learning entries are removed only from the ofproto associated
1667 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1668 * are removed from every ofproto. When patch ports and SLB bonds are in use
1669 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1670 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1671 * with the host from which it migrated. */
1673 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1675 struct ofproto_dpif *ofproto = bundle->ofproto;
1676 struct mac_learning *ml = ofproto->ml;
1677 struct mac_entry *mac, *next_mac;
1679 ofproto->need_revalidate = REV_RECONFIGURE;
1680 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1681 if (mac->port.p == bundle) {
1683 struct ofproto_dpif *o;
1685 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1687 struct mac_entry *e;
1689 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1692 tag_set_add(&o->revalidate_set, e->tag);
1693 mac_learning_expire(o->ml, e);
1699 mac_learning_expire(ml, mac);
1704 static struct ofbundle *
1705 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1707 struct ofbundle *bundle;
1709 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1710 &ofproto->bundles) {
1711 if (bundle->aux == aux) {
1718 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1719 * ones that are found to 'bundles'. */
1721 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1722 void **auxes, size_t n_auxes,
1723 struct hmapx *bundles)
1727 hmapx_init(bundles);
1728 for (i = 0; i < n_auxes; i++) {
1729 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1731 hmapx_add(bundles, bundle);
1737 bundle_update(struct ofbundle *bundle)
1739 struct ofport_dpif *port;
1741 bundle->floodable = true;
1742 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1743 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1744 || !stp_forward_in_state(port->stp_state)) {
1745 bundle->floodable = false;
1752 bundle_del_port(struct ofport_dpif *port)
1754 struct ofbundle *bundle = port->bundle;
1756 bundle->ofproto->need_revalidate = REV_RECONFIGURE;
1758 list_remove(&port->bundle_node);
1759 port->bundle = NULL;
1762 lacp_slave_unregister(bundle->lacp, port);
1765 bond_slave_unregister(bundle->bond, port);
1768 bundle_update(bundle);
1772 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1773 struct lacp_slave_settings *lacp,
1774 uint32_t bond_stable_id)
1776 struct ofport_dpif *port;
1778 port = get_ofp_port(bundle->ofproto, ofp_port);
1783 if (port->bundle != bundle) {
1784 bundle->ofproto->need_revalidate = REV_RECONFIGURE;
1786 bundle_del_port(port);
1789 port->bundle = bundle;
1790 list_push_back(&bundle->ports, &port->bundle_node);
1791 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1792 || !stp_forward_in_state(port->stp_state)) {
1793 bundle->floodable = false;
1797 port->bundle->ofproto->need_revalidate = REV_RECONFIGURE;
1798 lacp_slave_register(bundle->lacp, port, lacp);
1801 port->bond_stable_id = bond_stable_id;
1807 bundle_destroy(struct ofbundle *bundle)
1809 struct ofproto_dpif *ofproto;
1810 struct ofport_dpif *port, *next_port;
1817 ofproto = bundle->ofproto;
1818 for (i = 0; i < MAX_MIRRORS; i++) {
1819 struct ofmirror *m = ofproto->mirrors[i];
1821 if (m->out == bundle) {
1823 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1824 || hmapx_find_and_delete(&m->dsts, bundle)) {
1825 ofproto->need_revalidate = REV_RECONFIGURE;
1830 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1831 bundle_del_port(port);
1834 bundle_flush_macs(bundle, true);
1835 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1837 free(bundle->trunks);
1838 lacp_destroy(bundle->lacp);
1839 bond_destroy(bundle->bond);
1844 bundle_set(struct ofproto *ofproto_, void *aux,
1845 const struct ofproto_bundle_settings *s)
1847 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1848 bool need_flush = false;
1849 struct ofport_dpif *port;
1850 struct ofbundle *bundle;
1851 unsigned long *trunks;
1857 bundle_destroy(bundle_lookup(ofproto, aux));
1861 assert(s->n_slaves == 1 || s->bond != NULL);
1862 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1864 bundle = bundle_lookup(ofproto, aux);
1866 bundle = xmalloc(sizeof *bundle);
1868 bundle->ofproto = ofproto;
1869 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1870 hash_pointer(aux, 0));
1872 bundle->name = NULL;
1874 list_init(&bundle->ports);
1875 bundle->vlan_mode = PORT_VLAN_TRUNK;
1877 bundle->trunks = NULL;
1878 bundle->use_priority_tags = s->use_priority_tags;
1879 bundle->lacp = NULL;
1880 bundle->bond = NULL;
1882 bundle->floodable = true;
1884 bundle->src_mirrors = 0;
1885 bundle->dst_mirrors = 0;
1886 bundle->mirror_out = 0;
1889 if (!bundle->name || strcmp(s->name, bundle->name)) {
1891 bundle->name = xstrdup(s->name);
1896 if (!bundle->lacp) {
1897 ofproto->need_revalidate = REV_RECONFIGURE;
1898 bundle->lacp = lacp_create();
1900 lacp_configure(bundle->lacp, s->lacp);
1902 lacp_destroy(bundle->lacp);
1903 bundle->lacp = NULL;
1906 /* Update set of ports. */
1908 for (i = 0; i < s->n_slaves; i++) {
1909 if (!bundle_add_port(bundle, s->slaves[i],
1910 s->lacp ? &s->lacp_slaves[i] : NULL,
1911 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1915 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1916 struct ofport_dpif *next_port;
1918 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1919 for (i = 0; i < s->n_slaves; i++) {
1920 if (s->slaves[i] == port->up.ofp_port) {
1925 bundle_del_port(port);
1929 assert(list_size(&bundle->ports) <= s->n_slaves);
1931 if (list_is_empty(&bundle->ports)) {
1932 bundle_destroy(bundle);
1936 /* Set VLAN tagging mode */
1937 if (s->vlan_mode != bundle->vlan_mode
1938 || s->use_priority_tags != bundle->use_priority_tags) {
1939 bundle->vlan_mode = s->vlan_mode;
1940 bundle->use_priority_tags = s->use_priority_tags;
1945 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1946 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1948 if (vlan != bundle->vlan) {
1949 bundle->vlan = vlan;
1953 /* Get trunked VLANs. */
1954 switch (s->vlan_mode) {
1955 case PORT_VLAN_ACCESS:
1959 case PORT_VLAN_TRUNK:
1960 trunks = (unsigned long *) s->trunks;
1963 case PORT_VLAN_NATIVE_UNTAGGED:
1964 case PORT_VLAN_NATIVE_TAGGED:
1965 if (vlan != 0 && (!s->trunks
1966 || !bitmap_is_set(s->trunks, vlan)
1967 || bitmap_is_set(s->trunks, 0))) {
1968 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1970 trunks = bitmap_clone(s->trunks, 4096);
1972 trunks = bitmap_allocate1(4096);
1974 bitmap_set1(trunks, vlan);
1975 bitmap_set0(trunks, 0);
1977 trunks = (unsigned long *) s->trunks;
1984 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1985 free(bundle->trunks);
1986 if (trunks == s->trunks) {
1987 bundle->trunks = vlan_bitmap_clone(trunks);
1989 bundle->trunks = trunks;
1994 if (trunks != s->trunks) {
1999 if (!list_is_short(&bundle->ports)) {
2000 bundle->ofproto->has_bonded_bundles = true;
2002 if (bond_reconfigure(bundle->bond, s->bond)) {
2003 ofproto->need_revalidate = REV_RECONFIGURE;
2006 bundle->bond = bond_create(s->bond);
2007 ofproto->need_revalidate = REV_RECONFIGURE;
2010 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2011 bond_slave_register(bundle->bond, port, port->bond_stable_id,
2015 bond_destroy(bundle->bond);
2016 bundle->bond = NULL;
2019 /* If we changed something that would affect MAC learning, un-learn
2020 * everything on this port and force flow revalidation. */
2022 bundle_flush_macs(bundle, false);
2029 bundle_remove(struct ofport *port_)
2031 struct ofport_dpif *port = ofport_dpif_cast(port_);
2032 struct ofbundle *bundle = port->bundle;
2035 bundle_del_port(port);
2036 if (list_is_empty(&bundle->ports)) {
2037 bundle_destroy(bundle);
2038 } else if (list_is_short(&bundle->ports)) {
2039 bond_destroy(bundle->bond);
2040 bundle->bond = NULL;
2046 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2048 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2049 struct ofport_dpif *port = port_;
2050 uint8_t ea[ETH_ADDR_LEN];
2053 error = netdev_get_etheraddr(port->up.netdev, ea);
2055 struct ofpbuf packet;
2058 ofpbuf_init(&packet, 0);
2059 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2061 memcpy(packet_pdu, pdu, pdu_size);
2063 send_packet(port, &packet);
2064 ofpbuf_uninit(&packet);
2066 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2067 "%s (%s)", port->bundle->name,
2068 netdev_get_name(port->up.netdev), strerror(error));
2073 bundle_send_learning_packets(struct ofbundle *bundle)
2075 struct ofproto_dpif *ofproto = bundle->ofproto;
2076 int error, n_packets, n_errors;
2077 struct mac_entry *e;
2079 error = n_packets = n_errors = 0;
2080 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2081 if (e->port.p != bundle) {
2082 struct ofpbuf *learning_packet;
2083 struct ofport_dpif *port;
2087 /* The assignment to "port" is unnecessary but makes "grep"ing for
2088 * struct ofport_dpif more effective. */
2089 learning_packet = bond_compose_learning_packet(bundle->bond,
2093 ret = send_packet(port, learning_packet);
2094 ofpbuf_delete(learning_packet);
2104 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2105 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2106 "packets, last error was: %s",
2107 bundle->name, n_errors, n_packets, strerror(error));
2109 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2110 bundle->name, n_packets);
2115 bundle_run(struct ofbundle *bundle)
2118 lacp_run(bundle->lacp, send_pdu_cb);
2121 struct ofport_dpif *port;
2123 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2124 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2127 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
2128 lacp_status(bundle->lacp));
2129 if (bond_should_send_learning_packets(bundle->bond)) {
2130 bundle_send_learning_packets(bundle);
2136 bundle_wait(struct ofbundle *bundle)
2139 lacp_wait(bundle->lacp);
2142 bond_wait(bundle->bond);
2149 mirror_scan(struct ofproto_dpif *ofproto)
2153 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2154 if (!ofproto->mirrors[idx]) {
2161 static struct ofmirror *
2162 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2166 for (i = 0; i < MAX_MIRRORS; i++) {
2167 struct ofmirror *mirror = ofproto->mirrors[i];
2168 if (mirror && mirror->aux == aux) {
2176 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2178 mirror_update_dups(struct ofproto_dpif *ofproto)
2182 for (i = 0; i < MAX_MIRRORS; i++) {
2183 struct ofmirror *m = ofproto->mirrors[i];
2186 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2190 for (i = 0; i < MAX_MIRRORS; i++) {
2191 struct ofmirror *m1 = ofproto->mirrors[i];
2198 for (j = i + 1; j < MAX_MIRRORS; j++) {
2199 struct ofmirror *m2 = ofproto->mirrors[j];
2201 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2202 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2203 m2->dup_mirrors |= m1->dup_mirrors;
2210 mirror_set(struct ofproto *ofproto_, void *aux,
2211 const struct ofproto_mirror_settings *s)
2213 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2214 mirror_mask_t mirror_bit;
2215 struct ofbundle *bundle;
2216 struct ofmirror *mirror;
2217 struct ofbundle *out;
2218 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2219 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2222 mirror = mirror_lookup(ofproto, aux);
2224 mirror_destroy(mirror);
2230 idx = mirror_scan(ofproto);
2232 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2234 ofproto->up.name, MAX_MIRRORS, s->name);
2238 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2239 mirror->ofproto = ofproto;
2242 mirror->out_vlan = -1;
2243 mirror->name = NULL;
2246 if (!mirror->name || strcmp(s->name, mirror->name)) {
2248 mirror->name = xstrdup(s->name);
2251 /* Get the new configuration. */
2252 if (s->out_bundle) {
2253 out = bundle_lookup(ofproto, s->out_bundle);
2255 mirror_destroy(mirror);
2261 out_vlan = s->out_vlan;
2263 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2264 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2266 /* If the configuration has not changed, do nothing. */
2267 if (hmapx_equals(&srcs, &mirror->srcs)
2268 && hmapx_equals(&dsts, &mirror->dsts)
2269 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2270 && mirror->out == out
2271 && mirror->out_vlan == out_vlan)
2273 hmapx_destroy(&srcs);
2274 hmapx_destroy(&dsts);
2278 hmapx_swap(&srcs, &mirror->srcs);
2279 hmapx_destroy(&srcs);
2281 hmapx_swap(&dsts, &mirror->dsts);
2282 hmapx_destroy(&dsts);
2284 free(mirror->vlans);
2285 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2288 mirror->out_vlan = out_vlan;
2290 /* Update bundles. */
2291 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2292 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2293 if (hmapx_contains(&mirror->srcs, bundle)) {
2294 bundle->src_mirrors |= mirror_bit;
2296 bundle->src_mirrors &= ~mirror_bit;
2299 if (hmapx_contains(&mirror->dsts, bundle)) {
2300 bundle->dst_mirrors |= mirror_bit;
2302 bundle->dst_mirrors &= ~mirror_bit;
2305 if (mirror->out == bundle) {
2306 bundle->mirror_out |= mirror_bit;
2308 bundle->mirror_out &= ~mirror_bit;
2312 ofproto->need_revalidate = REV_RECONFIGURE;
2313 ofproto->has_mirrors = true;
2314 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2315 mirror_update_dups(ofproto);
2321 mirror_destroy(struct ofmirror *mirror)
2323 struct ofproto_dpif *ofproto;
2324 mirror_mask_t mirror_bit;
2325 struct ofbundle *bundle;
2332 ofproto = mirror->ofproto;
2333 ofproto->need_revalidate = REV_RECONFIGURE;
2334 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2336 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2337 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2338 bundle->src_mirrors &= ~mirror_bit;
2339 bundle->dst_mirrors &= ~mirror_bit;
2340 bundle->mirror_out &= ~mirror_bit;
2343 hmapx_destroy(&mirror->srcs);
2344 hmapx_destroy(&mirror->dsts);
2345 free(mirror->vlans);
2347 ofproto->mirrors[mirror->idx] = NULL;
2351 mirror_update_dups(ofproto);
2353 ofproto->has_mirrors = false;
2354 for (i = 0; i < MAX_MIRRORS; i++) {
2355 if (ofproto->mirrors[i]) {
2356 ofproto->has_mirrors = true;
2363 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2364 uint64_t *packets, uint64_t *bytes)
2366 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2367 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2370 *packets = *bytes = UINT64_MAX;
2374 *packets = mirror->packet_count;
2375 *bytes = mirror->byte_count;
2381 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2383 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2384 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2385 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2391 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2393 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2394 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2395 return bundle && bundle->mirror_out != 0;
2399 forward_bpdu_changed(struct ofproto *ofproto_)
2401 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2402 ofproto->need_revalidate = REV_RECONFIGURE;
2406 set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time)
2408 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2409 mac_learning_set_idle_time(ofproto->ml, idle_time);
2414 static struct ofport_dpif *
2415 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2417 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2418 return ofport ? ofport_dpif_cast(ofport) : NULL;
2421 static struct ofport_dpif *
2422 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2424 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2428 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2429 struct dpif_port *dpif_port)
2431 ofproto_port->name = dpif_port->name;
2432 ofproto_port->type = dpif_port->type;
2433 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2437 port_run_fast(struct ofport_dpif *ofport)
2439 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2440 struct ofpbuf packet;
2442 ofpbuf_init(&packet, 0);
2443 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2444 send_packet(ofport, &packet);
2445 ofpbuf_uninit(&packet);
2450 port_run(struct ofport_dpif *ofport)
2452 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2453 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2454 bool enable = netdev_get_carrier(ofport->up.netdev);
2456 ofport->carrier_seq = carrier_seq;
2458 port_run_fast(ofport);
2460 cfm_run(ofport->cfm);
2461 enable = enable && !cfm_get_fault(ofport->cfm)
2462 && cfm_get_opup(ofport->cfm);
2465 if (ofport->bundle) {
2466 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2467 if (carrier_changed) {
2468 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2472 if (ofport->may_enable != enable) {
2473 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2475 if (ofproto->has_bundle_action) {
2476 ofproto->need_revalidate = REV_PORT_TOGGLED;
2480 ofport->may_enable = enable;
2484 port_wait(struct ofport_dpif *ofport)
2487 cfm_wait(ofport->cfm);
2492 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2493 struct ofproto_port *ofproto_port)
2495 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2496 struct dpif_port dpif_port;
2499 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2501 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2507 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2509 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2513 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2515 *ofp_portp = odp_port_to_ofp_port(odp_port);
2521 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2523 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2526 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2528 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2530 /* The caller is going to close ofport->up.netdev. If this is a
2531 * bonded port, then the bond is using that netdev, so remove it
2532 * from the bond. The client will need to reconfigure everything
2533 * after deleting ports, so then the slave will get re-added. */
2534 bundle_remove(&ofport->up);
2541 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2543 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2546 error = netdev_get_stats(ofport->up.netdev, stats);
2548 if (!error && ofport->odp_port == OVSP_LOCAL) {
2549 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2551 /* ofproto->stats.tx_packets represents packets that we created
2552 * internally and sent to some port (e.g. packets sent with
2553 * send_packet()). Account for them as if they had come from
2554 * OFPP_LOCAL and got forwarded. */
2556 if (stats->rx_packets != UINT64_MAX) {
2557 stats->rx_packets += ofproto->stats.tx_packets;
2560 if (stats->rx_bytes != UINT64_MAX) {
2561 stats->rx_bytes += ofproto->stats.tx_bytes;
2564 /* ofproto->stats.rx_packets represents packets that were received on
2565 * some port and we processed internally and dropped (e.g. STP).
2566 * Account fro them as if they had been forwarded to OFPP_LOCAL. */
2568 if (stats->tx_packets != UINT64_MAX) {
2569 stats->tx_packets += ofproto->stats.rx_packets;
2572 if (stats->tx_bytes != UINT64_MAX) {
2573 stats->tx_bytes += ofproto->stats.rx_bytes;
2580 /* Account packets for LOCAL port. */
2582 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2583 size_t tx_size, size_t rx_size)
2585 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2588 ofproto->stats.rx_packets++;
2589 ofproto->stats.rx_bytes += rx_size;
2592 ofproto->stats.tx_packets++;
2593 ofproto->stats.tx_bytes += tx_size;
2597 struct port_dump_state {
2598 struct dpif_port_dump dump;
2603 port_dump_start(const struct ofproto *ofproto_, void **statep)
2605 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2606 struct port_dump_state *state;
2608 *statep = state = xmalloc(sizeof *state);
2609 dpif_port_dump_start(&state->dump, ofproto->dpif);
2610 state->done = false;
2615 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2616 struct ofproto_port *port)
2618 struct port_dump_state *state = state_;
2619 struct dpif_port dpif_port;
2621 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2622 ofproto_port_from_dpif_port(port, &dpif_port);
2625 int error = dpif_port_dump_done(&state->dump);
2627 return error ? error : EOF;
2632 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2634 struct port_dump_state *state = state_;
2637 dpif_port_dump_done(&state->dump);
2644 port_poll(const struct ofproto *ofproto_, char **devnamep)
2646 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2647 return dpif_port_poll(ofproto->dpif, devnamep);
2651 port_poll_wait(const struct ofproto *ofproto_)
2653 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2654 dpif_port_poll_wait(ofproto->dpif);
2658 port_is_lacp_current(const struct ofport *ofport_)
2660 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2661 return (ofport->bundle && ofport->bundle->lacp
2662 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2666 /* Upcall handling. */
2668 /* Flow miss batching.
2670 * Some dpifs implement operations faster when you hand them off in a batch.
2671 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2672 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2673 * more packets, plus possibly installing the flow in the dpif.
2675 * So far we only batch the operations that affect flow setup time the most.
2676 * It's possible to batch more than that, but the benefit might be minimal. */
2678 struct hmap_node hmap_node;
2680 enum odp_key_fitness key_fitness;
2681 const struct nlattr *key;
2683 ovs_be16 initial_tci;
2684 struct list packets;
2685 enum dpif_upcall_type upcall_type;
2688 struct flow_miss_op {
2689 struct dpif_op dpif_op;
2690 struct subfacet *subfacet; /* Subfacet */
2691 void *garbage; /* Pointer to pass to free(), NULL if none. */
2692 uint64_t stub[1024 / 8]; /* Temporary buffer. */
2695 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2696 * OpenFlow controller as necessary according to their individual
2697 * configurations. */
2699 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
2700 const struct flow *flow)
2702 struct ofputil_packet_in pin;
2704 pin.packet = packet->data;
2705 pin.packet_len = packet->size;
2706 pin.reason = OFPR_NO_MATCH;
2707 pin.controller_id = 0;
2712 pin.send_len = 0; /* not used for flow table misses */
2714 flow_get_metadata(flow, &pin.fmd);
2716 /* Registers aren't meaningful on a miss. */
2717 memset(pin.fmd.reg_masks, 0, sizeof pin.fmd.reg_masks);
2719 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
2722 static enum slow_path_reason
2723 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2724 const struct ofpbuf *packet)
2726 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2732 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2734 cfm_process_heartbeat(ofport->cfm, packet);
2737 } else if (ofport->bundle && ofport->bundle->lacp
2738 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2740 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2743 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2745 stp_process_packet(ofport, packet);
2752 static struct flow_miss *
2753 flow_miss_find(struct hmap *todo, const struct flow *flow, uint32_t hash)
2755 struct flow_miss *miss;
2757 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2758 if (flow_equal(&miss->flow, flow)) {
2766 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
2767 * 'packet'. The caller must initialize op->actions and op->actions_len. If
2768 * 'miss' is associated with a subfacet the caller must also initialize the
2769 * returned op->subfacet, and if anything needs to be freed after processing
2770 * the op, the caller must initialize op->garbage also. */
2772 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
2773 struct flow_miss_op *op)
2775 if (miss->flow.vlan_tci != miss->initial_tci) {
2776 /* This packet was received on a VLAN splinter port. We
2777 * added a VLAN to the packet to make the packet resemble
2778 * the flow, but the actions were composed assuming that
2779 * the packet contained no VLAN. So, we must remove the
2780 * VLAN header from the packet before trying to execute the
2782 eth_pop_vlan(packet);
2785 op->subfacet = NULL;
2787 op->dpif_op.type = DPIF_OP_EXECUTE;
2788 op->dpif_op.u.execute.key = miss->key;
2789 op->dpif_op.u.execute.key_len = miss->key_len;
2790 op->dpif_op.u.execute.packet = packet;
2793 /* Helper for handle_flow_miss_without_facet() and
2794 * handle_flow_miss_with_facet(). */
2796 handle_flow_miss_common(struct rule_dpif *rule,
2797 struct ofpbuf *packet, const struct flow *flow)
2799 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2801 ofproto->n_matches++;
2803 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2805 * Extra-special case for fail-open mode.
2807 * We are in fail-open mode and the packet matched the fail-open
2808 * rule, but we are connected to a controller too. We should send
2809 * the packet up to the controller in the hope that it will try to
2810 * set up a flow and thereby allow us to exit fail-open.
2812 * See the top-level comment in fail-open.c for more information.
2814 send_packet_in_miss(ofproto, packet, flow);
2818 /* Figures out whether a flow that missed in 'ofproto', whose details are in
2819 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
2820 * installing a datapath flow. The answer is usually "yes" (a return value of
2821 * true). However, for short flows the cost of bookkeeping is much higher than
2822 * the benefits, so when the datapath holds a large number of flows we impose
2823 * some heuristics to decide which flows are likely to be worth tracking. */
2825 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
2826 struct flow_miss *miss, uint32_t hash)
2828 if (!ofproto->governor) {
2831 n_subfacets = hmap_count(&ofproto->subfacets);
2832 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
2836 ofproto->governor = governor_create(ofproto->up.name);
2839 return governor_should_install_flow(ofproto->governor, hash,
2840 list_size(&miss->packets));
2843 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
2844 * or creating any datapath flow. May add an "execute" operation to 'ops' and
2845 * increment '*n_ops'. */
2847 handle_flow_miss_without_facet(struct flow_miss *miss,
2848 struct rule_dpif *rule,
2849 struct flow_miss_op *ops, size_t *n_ops)
2851 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2852 struct action_xlate_ctx ctx;
2853 struct ofpbuf *packet;
2855 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2856 struct flow_miss_op *op = &ops[*n_ops];
2857 struct dpif_flow_stats stats;
2858 struct ofpbuf odp_actions;
2860 COVERAGE_INC(facet_suppress);
2862 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
2864 dpif_flow_stats_extract(&miss->flow, packet, &stats);
2865 rule_credit_stats(rule, &stats);
2867 action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci,
2869 ctx.resubmit_stats = &stats;
2870 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
2873 if (odp_actions.size) {
2874 struct dpif_execute *execute = &op->dpif_op.u.execute;
2876 init_flow_miss_execute_op(miss, packet, op);
2877 execute->actions = odp_actions.data;
2878 execute->actions_len = odp_actions.size;
2879 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
2883 ofpbuf_uninit(&odp_actions);
2888 /* Handles 'miss', which matches 'facet'. May add any required datapath
2889 * operations to 'ops', incrementing '*n_ops' for each new op. */
2891 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
2892 struct flow_miss_op *ops, size_t *n_ops)
2894 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
2895 enum subfacet_path want_path;
2896 struct subfacet *subfacet;
2897 struct ofpbuf *packet;
2899 subfacet = subfacet_create(facet,
2900 miss->key_fitness, miss->key, miss->key_len,
2903 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2904 struct flow_miss_op *op = &ops[*n_ops];
2905 struct dpif_flow_stats stats;
2906 struct ofpbuf odp_actions;
2908 handle_flow_miss_common(facet->rule, packet, &miss->flow);
2910 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
2911 if (!subfacet->actions || subfacet->slow) {
2912 subfacet_make_actions(subfacet, packet, &odp_actions);
2915 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2916 subfacet_update_stats(subfacet, &stats);
2918 if (subfacet->actions_len) {
2919 struct dpif_execute *execute = &op->dpif_op.u.execute;
2921 init_flow_miss_execute_op(miss, packet, op);
2922 op->subfacet = subfacet;
2923 if (!subfacet->slow) {
2924 execute->actions = subfacet->actions;
2925 execute->actions_len = subfacet->actions_len;
2926 ofpbuf_uninit(&odp_actions);
2928 execute->actions = odp_actions.data;
2929 execute->actions_len = odp_actions.size;
2930 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
2935 ofpbuf_uninit(&odp_actions);
2939 want_path = subfacet_want_path(subfacet->slow);
2940 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
2941 struct flow_miss_op *op = &ops[(*n_ops)++];
2942 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
2944 op->subfacet = subfacet;
2946 op->dpif_op.type = DPIF_OP_FLOW_PUT;
2947 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2948 put->key = miss->key;
2949 put->key_len = miss->key_len;
2950 if (want_path == SF_FAST_PATH) {
2951 put->actions = subfacet->actions;
2952 put->actions_len = subfacet->actions_len;
2954 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
2955 op->stub, sizeof op->stub,
2956 &put->actions, &put->actions_len);
2962 /* Handles flow miss 'miss' on 'ofproto'. May add any required datapath
2963 * operations to 'ops', incrementing '*n_ops' for each new op. */
2965 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2966 struct flow_miss_op *ops, size_t *n_ops)
2968 struct facet *facet;
2971 /* The caller must ensure that miss->hmap_node.hash contains
2972 * flow_hash(miss->flow, 0). */
2973 hash = miss->hmap_node.hash;
2975 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
2977 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
2979 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
2980 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
2984 facet = facet_create(rule, &miss->flow, hash);
2986 handle_flow_miss_with_facet(miss, facet, ops, n_ops);
2989 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
2990 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
2991 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
2992 * what a flow key should contain.
2994 * This function also includes some logic to help make VLAN splinters
2995 * transparent to the rest of the upcall processing logic. In particular, if
2996 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
2997 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
2998 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
3000 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
3001 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
3002 * (This differs from the value returned in flow->vlan_tci only for packets
3003 * received on VLAN splinters.)
3005 static enum odp_key_fitness
3006 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
3007 const struct nlattr *key, size_t key_len,
3008 struct flow *flow, ovs_be16 *initial_tci,
3009 struct ofpbuf *packet)
3011 enum odp_key_fitness fitness;
3013 fitness = odp_flow_key_to_flow(key, key_len, flow);
3014 if (fitness == ODP_FIT_ERROR) {
3017 *initial_tci = flow->vlan_tci;
3019 if (vsp_adjust_flow(ofproto, flow)) {
3021 /* Make the packet resemble the flow, so that it gets sent to an
3022 * OpenFlow controller properly, so that it looks correct for
3023 * sFlow, and so that flow_extract() will get the correct vlan_tci
3024 * if it is called on 'packet'.
3026 * The allocated space inside 'packet' probably also contains
3027 * 'key', that is, both 'packet' and 'key' are probably part of a
3028 * struct dpif_upcall (see the large comment on that structure
3029 * definition), so pushing data on 'packet' is in general not a
3030 * good idea since it could overwrite 'key' or free it as a side
3031 * effect. However, it's OK in this special case because we know
3032 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
3033 * will just overwrite the 4-byte "struct nlattr", which is fine
3034 * since we don't need that header anymore. */
3035 eth_push_vlan(packet, flow->vlan_tci);
3038 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
3039 if (fitness == ODP_FIT_PERFECT) {
3040 fitness = ODP_FIT_TOO_MUCH;
3048 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
3051 struct dpif_upcall *upcall;
3052 struct flow_miss *miss;
3053 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3054 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3055 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3065 /* Construct the to-do list.
3067 * This just amounts to extracting the flow from each packet and sticking
3068 * the packets that have the same flow in the same "flow_miss" structure so
3069 * that we can process them together. */
3072 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3073 struct flow_miss *miss = &misses[n_misses];
3074 struct flow_miss *existing_miss;
3077 /* Obtain metadata and check userspace/kernel agreement on flow match,
3078 * then set 'flow''s header pointers. */
3079 miss->key_fitness = ofproto_dpif_extract_flow_key(
3080 ofproto, upcall->key, upcall->key_len,
3081 &miss->flow, &miss->initial_tci, upcall->packet);
3082 if (miss->key_fitness == ODP_FIT_ERROR) {
3085 flow_extract(upcall->packet, miss->flow.skb_priority,
3086 miss->flow.tun_id, miss->flow.in_port, &miss->flow);
3088 /* Add other packets to a to-do list. */
3089 hash = flow_hash(&miss->flow, 0);
3090 existing_miss = flow_miss_find(&todo, &miss->flow, hash);
3091 if (!existing_miss) {
3092 hmap_insert(&todo, &miss->hmap_node, hash);
3093 miss->key = upcall->key;
3094 miss->key_len = upcall->key_len;
3095 miss->upcall_type = upcall->type;
3096 list_init(&miss->packets);
3100 miss = existing_miss;
3102 list_push_back(&miss->packets, &upcall->packet->list_node);
3105 /* Process each element in the to-do list, constructing the set of
3106 * operations to batch. */
3108 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3109 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
3111 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3113 /* Execute batch. */
3114 for (i = 0; i < n_ops; i++) {
3115 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3117 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
3119 /* Free memory and update facets. */
3120 for (i = 0; i < n_ops; i++) {
3121 struct flow_miss_op *op = &flow_miss_ops[i];
3123 switch (op->dpif_op.type) {
3124 case DPIF_OP_EXECUTE:
3127 case DPIF_OP_FLOW_PUT:
3128 if (!op->dpif_op.error) {
3129 op->subfacet->path = subfacet_want_path(op->subfacet->slow);
3133 case DPIF_OP_FLOW_DEL:
3139 hmap_destroy(&todo);
3142 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
3143 classify_upcall(const struct dpif_upcall *upcall)
3145 union user_action_cookie cookie;
3147 /* First look at the upcall type. */
3148 switch (upcall->type) {
3149 case DPIF_UC_ACTION:
3155 case DPIF_N_UC_TYPES:
3157 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3161 /* "action" upcalls need a closer look. */
3162 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3163 switch (cookie.type) {
3164 case USER_ACTION_COOKIE_SFLOW:
3165 return SFLOW_UPCALL;
3167 case USER_ACTION_COOKIE_SLOW_PATH:
3170 case USER_ACTION_COOKIE_UNSPEC:
3172 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
3178 handle_sflow_upcall(struct ofproto_dpif *ofproto,
3179 const struct dpif_upcall *upcall)
3181 union user_action_cookie cookie;
3182 enum odp_key_fitness fitness;
3183 ovs_be16 initial_tci;
3186 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
3187 upcall->key_len, &flow,
3188 &initial_tci, upcall->packet);
3189 if (fitness == ODP_FIT_ERROR) {
3193 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3194 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, &cookie);
3198 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
3200 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3201 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3202 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3207 assert(max_batch <= FLOW_MISS_MAX_BATCH);
3210 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3211 struct dpif_upcall *upcall = &misses[n_misses];
3212 struct ofpbuf *buf = &miss_bufs[n_misses];
3215 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3216 sizeof miss_buf_stubs[n_misses]);
3217 error = dpif_recv(ofproto->dpif, upcall, buf);
3223 switch (classify_upcall(upcall)) {
3225 /* Handle it later. */
3230 if (ofproto->sflow) {
3231 handle_sflow_upcall(ofproto, upcall);
3242 /* Handle deferred MISS_UPCALL processing. */
3243 handle_miss_upcalls(ofproto, misses, n_misses);
3244 for (i = 0; i < n_misses; i++) {
3245 ofpbuf_uninit(&miss_bufs[i]);
3251 /* Flow expiration. */
3253 static int subfacet_max_idle(const struct ofproto_dpif *);
3254 static void update_stats(struct ofproto_dpif *);
3255 static void rule_expire(struct rule_dpif *);
3256 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
3258 /* This function is called periodically by run(). Its job is to collect
3259 * updates for the flows that have been installed into the datapath, most
3260 * importantly when they last were used, and then use that information to
3261 * expire flows that have not been used recently.
3263 * Returns the number of milliseconds after which it should be called again. */
3265 expire(struct ofproto_dpif *ofproto)
3267 struct rule_dpif *rule, *next_rule;
3268 struct oftable *table;
3271 /* Update stats for each flow in the datapath. */
3272 update_stats(ofproto);
3274 /* Expire subfacets that have been idle too long. */
3275 dp_max_idle = subfacet_max_idle(ofproto);
3276 expire_subfacets(ofproto, dp_max_idle);
3278 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
3279 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
3280 struct cls_cursor cursor;
3282 cls_cursor_init(&cursor, &table->cls, NULL);
3283 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
3288 /* All outstanding data in existing flows has been accounted, so it's a
3289 * good time to do bond rebalancing. */
3290 if (ofproto->has_bonded_bundles) {
3291 struct ofbundle *bundle;
3293 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3295 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
3300 return MIN(dp_max_idle, 1000);
3303 /* Updates flow table statistics given that the datapath just reported 'stats'
3304 * as 'subfacet''s statistics. */
3306 update_subfacet_stats(struct subfacet *subfacet,
3307 const struct dpif_flow_stats *stats)
3309 struct facet *facet = subfacet->facet;
3311 if (stats->n_packets >= subfacet->dp_packet_count) {
3312 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
3313 facet->packet_count += extra;
3315 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3318 if (stats->n_bytes >= subfacet->dp_byte_count) {
3319 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
3321 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3324 subfacet->dp_packet_count = stats->n_packets;
3325 subfacet->dp_byte_count = stats->n_bytes;
3327 facet->tcp_flags |= stats->tcp_flags;
3329 subfacet_update_time(subfacet, stats->used);
3330 if (facet->accounted_bytes < facet->byte_count) {
3332 facet_account(facet);
3333 facet->accounted_bytes = facet->byte_count;
3335 facet_push_stats(facet);
3338 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3339 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3341 delete_unexpected_flow(struct dpif *dpif,
3342 const struct nlattr *key, size_t key_len)
3344 if (!VLOG_DROP_WARN(&rl)) {
3348 odp_flow_key_format(key, key_len, &s);
3349 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
3353 COVERAGE_INC(facet_unexpected);
3354 dpif_flow_del(dpif, key, key_len, NULL);
3357 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3359 * This function also pushes statistics updates to rules which each facet
3360 * resubmits into. Generally these statistics will be accurate. However, if a
3361 * facet changes the rule it resubmits into at some time in between
3362 * update_stats() runs, it is possible that statistics accrued to the
3363 * old rule will be incorrectly attributed to the new rule. This could be
3364 * avoided by calling update_stats() whenever rules are created or
3365 * deleted. However, the performance impact of making so many calls to the
3366 * datapath do not justify the benefit of having perfectly accurate statistics.
3369 update_stats(struct ofproto_dpif *p)
3371 const struct dpif_flow_stats *stats;
3372 struct dpif_flow_dump dump;
3373 const struct nlattr *key;
3376 dpif_flow_dump_start(&dump, p->dpif);
3377 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
3378 struct subfacet *subfacet;
3380 subfacet = subfacet_find(p, key, key_len);
3381 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3383 update_subfacet_stats(subfacet, stats);
3387 /* Stats are updated per-packet. */
3390 case SF_NOT_INSTALLED:
3392 delete_unexpected_flow(p->dpif, key, key_len);
3396 dpif_flow_dump_done(&dump);
3399 /* Calculates and returns the number of milliseconds of idle time after which
3400 * subfacets should expire from the datapath. When a subfacet expires, we fold
3401 * its statistics into its facet, and when a facet's last subfacet expires, we
3402 * fold its statistic into its rule. */
3404 subfacet_max_idle(const struct ofproto_dpif *ofproto)
3407 * Idle time histogram.
3409 * Most of the time a switch has a relatively small number of subfacets.
3410 * When this is the case we might as well keep statistics for all of them
3411 * in userspace and to cache them in the kernel datapath for performance as
3414 * As the number of subfacets increases, the memory required to maintain
3415 * statistics about them in userspace and in the kernel becomes
3416 * significant. However, with a large number of subfacets it is likely
3417 * that only a few of them are "heavy hitters" that consume a large amount
3418 * of bandwidth. At this point, only heavy hitters are worth caching in
3419 * the kernel and maintaining in userspaces; other subfacets we can
3422 * The technique used to compute the idle time is to build a histogram with
3423 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3424 * that is installed in the kernel gets dropped in the appropriate bucket.
3425 * After the histogram has been built, we compute the cutoff so that only
3426 * the most-recently-used 1% of subfacets (but at least
3427 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3428 * the most-recently-used bucket of subfacets is kept, so actually an
3429 * arbitrary number of subfacets can be kept in any given expiration run
3430 * (though the next run will delete most of those unless they receive
3433 * This requires a second pass through the subfacets, in addition to the
3434 * pass made by update_stats(), because the former function never looks at
3435 * uninstallable subfacets.
3437 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3438 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3439 int buckets[N_BUCKETS] = { 0 };
3440 int total, subtotal, bucket;
3441 struct subfacet *subfacet;
3445 total = hmap_count(&ofproto->subfacets);
3446 if (total <= ofproto->up.flow_eviction_threshold) {
3447 return N_BUCKETS * BUCKET_WIDTH;
3450 /* Build histogram. */
3452 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3453 long long int idle = now - subfacet->used;
3454 int bucket = (idle <= 0 ? 0
3455 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3456 : (unsigned int) idle / BUCKET_WIDTH);
3460 /* Find the first bucket whose flows should be expired. */
3461 subtotal = bucket = 0;
3463 subtotal += buckets[bucket++];
3464 } while (bucket < N_BUCKETS &&
3465 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3467 if (VLOG_IS_DBG_ENABLED()) {
3471 ds_put_cstr(&s, "keep");
3472 for (i = 0; i < N_BUCKETS; i++) {
3474 ds_put_cstr(&s, ", drop");
3477 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3480 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3484 return bucket * BUCKET_WIDTH;
3487 enum { EXPIRE_MAX_BATCH = 50 };
3490 expire_batch(struct ofproto_dpif *ofproto, struct subfacet **subfacets, int n)
3492 struct odputil_keybuf keybufs[EXPIRE_MAX_BATCH];
3493 struct dpif_op ops[EXPIRE_MAX_BATCH];
3494 struct dpif_op *opsp[EXPIRE_MAX_BATCH];
3495 struct ofpbuf keys[EXPIRE_MAX_BATCH];
3496 struct dpif_flow_stats stats[EXPIRE_MAX_BATCH];
3499 for (i = 0; i < n; i++) {
3500 ops[i].type = DPIF_OP_FLOW_DEL;
3501 subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]);
3502 ops[i].u.flow_del.key = keys[i].data;
3503 ops[i].u.flow_del.key_len = keys[i].size;
3504 ops[i].u.flow_del.stats = &stats[i];
3508 dpif_operate(ofproto->dpif, opsp, n);
3509 for (i = 0; i < n; i++) {
3510 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
3511 subfacets[i]->path = SF_NOT_INSTALLED;
3512 subfacet_destroy(subfacets[i]);
3517 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3519 /* Cutoff time for most flows. */
3520 long long int normal_cutoff = time_msec() - dp_max_idle;
3522 /* We really want to keep flows for special protocols around, so use a more
3523 * conservative cutoff. */
3524 long long int special_cutoff = time_msec() - 10000;
3526 struct subfacet *subfacet, *next_subfacet;
3527 struct subfacet *batch[EXPIRE_MAX_BATCH];
3531 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3532 &ofproto->subfacets) {
3533 long long int cutoff;
3535 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
3538 if (subfacet->used < cutoff) {
3539 if (subfacet->path != SF_NOT_INSTALLED) {
3540 batch[n_batch++] = subfacet;
3541 if (n_batch >= EXPIRE_MAX_BATCH) {
3542 expire_batch(ofproto, batch, n_batch);
3546 subfacet_destroy(subfacet);
3552 expire_batch(ofproto, batch, n_batch);
3556 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3557 * then delete it entirely. */
3559 rule_expire(struct rule_dpif *rule)
3561 struct facet *facet, *next_facet;
3565 /* Has 'rule' expired? */
3567 if (rule->up.hard_timeout
3568 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3569 reason = OFPRR_HARD_TIMEOUT;
3570 } else if (rule->up.idle_timeout
3571 && now > rule->up.used + rule->up.idle_timeout * 1000) {
3572 reason = OFPRR_IDLE_TIMEOUT;
3577 COVERAGE_INC(ofproto_dpif_expired);
3579 /* Update stats. (This is a no-op if the rule expired due to an idle
3580 * timeout, because that only happens when the rule has no facets left.) */
3581 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3582 facet_remove(facet);
3585 /* Get rid of the rule. */
3586 ofproto_rule_expire(&rule->up, reason);
3591 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3593 * The caller must already have determined that no facet with an identical
3594 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3595 * the ofproto's classifier table.
3597 * 'hash' must be the return value of flow_hash(flow, 0).
3599 * The facet will initially have no subfacets. The caller should create (at
3600 * least) one subfacet with subfacet_create(). */
3601 static struct facet *
3602 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
3604 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3605 struct facet *facet;
3607 facet = xzalloc(sizeof *facet);
3608 facet->used = time_msec();
3609 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
3610 list_push_back(&rule->facets, &facet->list_node);
3612 facet->flow = *flow;
3613 list_init(&facet->subfacets);
3614 netflow_flow_init(&facet->nf_flow);
3615 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3621 facet_free(struct facet *facet)
3626 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3627 * 'packet', which arrived on 'in_port'.
3629 * Takes ownership of 'packet'. */
3631 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3632 const struct nlattr *odp_actions, size_t actions_len,
3633 struct ofpbuf *packet)
3635 struct odputil_keybuf keybuf;
3639 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3640 odp_flow_key_from_flow(&key, flow);
3642 error = dpif_execute(ofproto->dpif, key.data, key.size,
3643 odp_actions, actions_len, packet);
3645 ofpbuf_delete(packet);
3649 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3651 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3652 * rule's statistics, via subfacet_uninstall().
3654 * - Removes 'facet' from its rule and from ofproto->facets.
3657 facet_remove(struct facet *facet)
3659 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3660 struct subfacet *subfacet, *next_subfacet;
3662 assert(!list_is_empty(&facet->subfacets));
3664 /* First uninstall all of the subfacets to get final statistics. */
3665 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3666 subfacet_uninstall(subfacet);
3669 /* Flush the final stats to the rule.
3671 * This might require us to have at least one subfacet around so that we
3672 * can use its actions for accounting in facet_account(), which is why we
3673 * have uninstalled but not yet destroyed the subfacets. */
3674 facet_flush_stats(facet);
3676 /* Now we're really all done so destroy everything. */
3677 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3678 &facet->subfacets) {
3679 subfacet_destroy__(subfacet);
3681 hmap_remove(&ofproto->facets, &facet->hmap_node);
3682 list_remove(&facet->list_node);
3686 /* Feed information from 'facet' back into the learning table to keep it in
3687 * sync with what is actually flowing through the datapath. */
3689 facet_learn(struct facet *facet)
3691 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3692 struct action_xlate_ctx ctx;
3694 if (!facet->has_learn
3695 && !facet->has_normal
3696 && (!facet->has_fin_timeout
3697 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
3701 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3702 facet->flow.vlan_tci,
3703 facet->rule, facet->tcp_flags, NULL);
3704 ctx.may_learn = true;
3705 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
3706 facet->rule->up.ofpacts_len);
3710 facet_account(struct facet *facet)
3712 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3713 struct subfacet *subfacet;
3714 const struct nlattr *a;
3719 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3722 n_bytes = facet->byte_count - facet->accounted_bytes;
3724 /* This loop feeds byte counters to bond_account() for rebalancing to use
3725 * as a basis. We also need to track the actual VLAN on which the packet
3726 * is going to be sent to ensure that it matches the one passed to
3727 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3730 * We use the actions from an arbitrary subfacet because they should all
3731 * be equally valid for our purpose. */
3732 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3733 struct subfacet, list_node);
3734 vlan_tci = facet->flow.vlan_tci;
3735 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3736 subfacet->actions, subfacet->actions_len) {
3737 const struct ovs_action_push_vlan *vlan;
3738 struct ofport_dpif *port;
3740 switch (nl_attr_type(a)) {
3741 case OVS_ACTION_ATTR_OUTPUT:
3742 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3743 if (port && port->bundle && port->bundle->bond) {
3744 bond_account(port->bundle->bond, &facet->flow,
3745 vlan_tci_to_vid(vlan_tci), n_bytes);
3749 case OVS_ACTION_ATTR_POP_VLAN:
3750 vlan_tci = htons(0);
3753 case OVS_ACTION_ATTR_PUSH_VLAN:
3754 vlan = nl_attr_get(a);
3755 vlan_tci = vlan->vlan_tci;
3761 /* Returns true if the only action for 'facet' is to send to the controller.
3762 * (We don't report NetFlow expiration messages for such facets because they
3763 * are just part of the control logic for the network, not real traffic). */
3765 facet_is_controller_flow(struct facet *facet)
3768 const struct rule *rule = &facet->rule->up;
3769 const struct ofpact *ofpacts = rule->ofpacts;
3770 size_t ofpacts_len = rule->ofpacts_len;
3772 if (ofpacts->type == OFPACT_CONTROLLER &&
3773 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
3780 /* Folds all of 'facet''s statistics into its rule. Also updates the
3781 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3782 * 'facet''s statistics in the datapath should have been zeroed and folded into
3783 * its packet and byte counts before this function is called. */
3785 facet_flush_stats(struct facet *facet)
3787 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3788 struct subfacet *subfacet;
3790 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3791 assert(!subfacet->dp_byte_count);
3792 assert(!subfacet->dp_packet_count);
3795 facet_push_stats(facet);
3796 if (facet->accounted_bytes < facet->byte_count) {
3797 facet_account(facet);
3798 facet->accounted_bytes = facet->byte_count;
3801 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3802 struct ofexpired expired;
3803 expired.flow = facet->flow;
3804 expired.packet_count = facet->packet_count;
3805 expired.byte_count = facet->byte_count;
3806 expired.used = facet->used;
3807 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3810 facet->rule->packet_count += facet->packet_count;
3811 facet->rule->byte_count += facet->byte_count;
3813 /* Reset counters to prevent double counting if 'facet' ever gets
3815 facet_reset_counters(facet);
3817 netflow_flow_clear(&facet->nf_flow);
3818 facet->tcp_flags = 0;
3821 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3822 * Returns it if found, otherwise a null pointer.
3824 * 'hash' must be the return value of flow_hash(flow, 0).
3826 * The returned facet might need revalidation; use facet_lookup_valid()
3827 * instead if that is important. */
3828 static struct facet *
3829 facet_find(struct ofproto_dpif *ofproto,
3830 const struct flow *flow, uint32_t hash)
3832 struct facet *facet;
3834 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
3835 if (flow_equal(flow, &facet->flow)) {
3843 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3844 * Returns it if found, otherwise a null pointer.
3846 * 'hash' must be the return value of flow_hash(flow, 0).
3848 * The returned facet is guaranteed to be valid. */
3849 static struct facet *
3850 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
3853 struct facet *facet;
3855 facet = facet_find(ofproto, flow, hash);
3857 && (ofproto->need_revalidate
3858 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))) {
3859 facet_revalidate(facet);
3866 subfacet_path_to_string(enum subfacet_path path)
3869 case SF_NOT_INSTALLED:
3870 return "not installed";
3872 return "in fast path";
3874 return "in slow path";
3880 /* Returns the path in which a subfacet should be installed if its 'slow'
3881 * member has the specified value. */
3882 static enum subfacet_path
3883 subfacet_want_path(enum slow_path_reason slow)
3885 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
3888 /* Returns true if 'subfacet' needs to have its datapath flow updated,
3889 * supposing that its actions have been recalculated as 'want_actions' and that
3890 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
3892 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
3893 const struct ofpbuf *want_actions)
3895 enum subfacet_path want_path = subfacet_want_path(slow);
3896 return (want_path != subfacet->path
3897 || (want_path == SF_FAST_PATH
3898 && (subfacet->actions_len != want_actions->size
3899 || memcmp(subfacet->actions, want_actions->data,
3900 subfacet->actions_len))));
3904 facet_check_consistency(struct facet *facet)
3906 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3908 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3910 uint64_t odp_actions_stub[1024 / 8];
3911 struct ofpbuf odp_actions;
3913 struct rule_dpif *rule;
3914 struct subfacet *subfacet;
3915 bool may_log = false;
3918 /* Check the rule for consistency. */
3919 rule = rule_dpif_lookup(ofproto, &facet->flow);
3920 ok = rule == facet->rule;
3922 may_log = !VLOG_DROP_WARN(&rl);
3927 flow_format(&s, &facet->flow);
3928 ds_put_format(&s, ": facet associated with wrong rule (was "
3929 "table=%"PRIu8",", facet->rule->up.table_id);
3930 cls_rule_format(&facet->rule->up.cr, &s);
3931 ds_put_format(&s, ") (should have been table=%"PRIu8",",
3933 cls_rule_format(&rule->up.cr, &s);
3934 ds_put_char(&s, ')');
3936 VLOG_WARN("%s", ds_cstr(&s));
3941 /* Check the datapath actions for consistency. */
3942 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
3943 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3944 enum subfacet_path want_path;
3945 struct odputil_keybuf keybuf;
3946 struct action_xlate_ctx ctx;
3950 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3951 subfacet->initial_tci, rule, 0, NULL);
3952 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3955 if (subfacet->path == SF_NOT_INSTALLED) {
3956 /* This only happens if the datapath reported an error when we
3957 * tried to install the flow. Don't flag another error here. */
3961 want_path = subfacet_want_path(subfacet->slow);
3962 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
3963 /* The actions for slow-path flows may legitimately vary from one
3964 * packet to the next. We're done. */
3968 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
3972 /* Inconsistency! */
3974 may_log = !VLOG_DROP_WARN(&rl);
3978 /* Rate-limited, skip reporting. */
3983 subfacet_get_key(subfacet, &keybuf, &key);
3984 odp_flow_key_format(key.data, key.size, &s);
3986 ds_put_cstr(&s, ": inconsistency in subfacet");
3987 if (want_path != subfacet->path) {
3988 enum odp_key_fitness fitness = subfacet->key_fitness;
3990 ds_put_format(&s, " (%s, fitness=%s)",
3991 subfacet_path_to_string(subfacet->path),
3992 odp_key_fitness_to_string(fitness));
3993 ds_put_format(&s, " (should have been %s)",
3994 subfacet_path_to_string(want_path));
3995 } else if (want_path == SF_FAST_PATH) {
3996 ds_put_cstr(&s, " (actions were: ");
3997 format_odp_actions(&s, subfacet->actions,
3998 subfacet->actions_len);
3999 ds_put_cstr(&s, ") (correct actions: ");
4000 format_odp_actions(&s, odp_actions.data, odp_actions.size);
4001 ds_put_char(&s, ')');
4003 ds_put_cstr(&s, " (actions: ");
4004 format_odp_actions(&s, subfacet->actions,
4005 subfacet->actions_len);
4006 ds_put_char(&s, ')');
4008 VLOG_WARN("%s", ds_cstr(&s));
4011 ofpbuf_uninit(&odp_actions);
4016 /* Re-searches the classifier for 'facet':
4018 * - If the rule found is different from 'facet''s current rule, moves
4019 * 'facet' to the new rule and recompiles its actions.
4021 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4022 * where it is and recompiles its actions anyway. */
4024 facet_revalidate(struct facet *facet)
4026 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4028 struct nlattr *odp_actions;
4031 struct actions *new_actions;
4033 struct action_xlate_ctx ctx;
4034 uint64_t odp_actions_stub[1024 / 8];
4035 struct ofpbuf odp_actions;
4037 struct rule_dpif *new_rule;
4038 struct subfacet *subfacet;
4041 COVERAGE_INC(facet_revalidate);
4043 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
4045 /* Calculate new datapath actions.
4047 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4048 * emit a NetFlow expiration and, if so, we need to have the old state
4049 * around to properly compose it. */
4051 /* If the datapath actions changed or the installability changed,
4052 * then we need to talk to the datapath. */
4055 memset(&ctx, 0, sizeof ctx);
4056 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4057 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4058 enum slow_path_reason slow;
4060 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4061 subfacet->initial_tci, new_rule, 0, NULL);
4062 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
4065 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4066 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
4067 struct dpif_flow_stats stats;
4069 subfacet_install(subfacet,
4070 odp_actions.data, odp_actions.size, &stats, slow);
4071 subfacet_update_stats(subfacet, &stats);
4074 new_actions = xcalloc(list_size(&facet->subfacets),
4075 sizeof *new_actions);
4077 new_actions[i].odp_actions = xmemdup(odp_actions.data,
4079 new_actions[i].actions_len = odp_actions.size;
4084 ofpbuf_uninit(&odp_actions);
4087 facet_flush_stats(facet);
4090 /* Update 'facet' now that we've taken care of all the old state. */
4091 facet->tags = ctx.tags;
4092 facet->nf_flow.output_iface = ctx.nf_output_iface;
4093 facet->has_learn = ctx.has_learn;
4094 facet->has_normal = ctx.has_normal;
4095 facet->has_fin_timeout = ctx.has_fin_timeout;
4096 facet->mirrors = ctx.mirrors;
4099 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4100 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4102 if (new_actions && new_actions[i].odp_actions) {
4103 free(subfacet->actions);
4104 subfacet->actions = new_actions[i].odp_actions;
4105 subfacet->actions_len = new_actions[i].actions_len;
4111 if (facet->rule != new_rule) {
4112 COVERAGE_INC(facet_changed_rule);
4113 list_remove(&facet->list_node);
4114 list_push_back(&new_rule->facets, &facet->list_node);
4115 facet->rule = new_rule;
4116 facet->used = new_rule->up.created;
4117 facet->prev_used = facet->used;
4121 /* Updates 'facet''s used time. Caller is responsible for calling
4122 * facet_push_stats() to update the flows which 'facet' resubmits into. */
4124 facet_update_time(struct facet *facet, long long int used)
4126 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4127 if (used > facet->used) {
4129 ofproto_rule_update_used(&facet->rule->up, used);
4130 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
4135 facet_reset_counters(struct facet *facet)
4137 facet->packet_count = 0;
4138 facet->byte_count = 0;
4139 facet->prev_packet_count = 0;
4140 facet->prev_byte_count = 0;
4141 facet->accounted_bytes = 0;
4145 facet_push_stats(struct facet *facet)
4147 struct dpif_flow_stats stats;
4149 assert(facet->packet_count >= facet->prev_packet_count);
4150 assert(facet->byte_count >= facet->prev_byte_count);
4151 assert(facet->used >= facet->prev_used);
4153 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4154 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4155 stats.used = facet->used;
4156 stats.tcp_flags = 0;
4158 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
4159 facet->prev_packet_count = facet->packet_count;
4160 facet->prev_byte_count = facet->byte_count;
4161 facet->prev_used = facet->used;
4163 flow_push_stats(facet->rule, &facet->flow, &stats);
4165 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
4166 facet->mirrors, stats.n_packets, stats.n_bytes);
4171 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4173 rule->packet_count += stats->n_packets;
4174 rule->byte_count += stats->n_bytes;
4175 ofproto_rule_update_used(&rule->up, stats->used);
4178 /* Pushes flow statistics to the rules which 'flow' resubmits into given
4179 * 'rule''s actions and mirrors. */
4181 flow_push_stats(struct rule_dpif *rule,
4182 const struct flow *flow, const struct dpif_flow_stats *stats)
4184 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4185 struct action_xlate_ctx ctx;
4187 ofproto_rule_update_used(&rule->up, stats->used);
4189 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule,
4191 ctx.resubmit_stats = stats;
4192 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
4193 rule->up.ofpacts_len);
4198 static struct subfacet *
4199 subfacet_find__(struct ofproto_dpif *ofproto,
4200 const struct nlattr *key, size_t key_len, uint32_t key_hash,
4201 const struct flow *flow)
4203 struct subfacet *subfacet;
4205 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4206 &ofproto->subfacets) {
4208 ? (subfacet->key_len == key_len
4209 && !memcmp(key, subfacet->key, key_len))
4210 : flow_equal(flow, &subfacet->facet->flow)) {
4218 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4219 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
4220 * there is one, otherwise creates and returns a new subfacet.
4222 * If the returned subfacet is new, then subfacet->actions will be NULL, in
4223 * which case the caller must populate the actions with
4224 * subfacet_make_actions(). */
4225 static struct subfacet *
4226 subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness,
4227 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
4229 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4230 uint32_t key_hash = odp_flow_key_hash(key, key_len);
4231 struct subfacet *subfacet;
4233 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
4235 if (subfacet->facet == facet) {
4239 /* This shouldn't happen. */
4240 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4241 subfacet_destroy(subfacet);
4244 subfacet = (list_is_empty(&facet->subfacets)
4245 ? &facet->one_subfacet
4246 : xmalloc(sizeof *subfacet));
4247 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
4248 list_push_back(&facet->subfacets, &subfacet->list_node);
4249 subfacet->facet = facet;
4250 subfacet->key_fitness = key_fitness;
4251 if (key_fitness != ODP_FIT_PERFECT) {
4252 subfacet->key = xmemdup(key, key_len);
4253 subfacet->key_len = key_len;
4255 subfacet->key = NULL;
4256 subfacet->key_len = 0;
4258 subfacet->used = time_msec();
4259 subfacet->dp_packet_count = 0;
4260 subfacet->dp_byte_count = 0;
4261 subfacet->actions_len = 0;
4262 subfacet->actions = NULL;
4263 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
4266 subfacet->path = SF_NOT_INSTALLED;
4267 subfacet->initial_tci = initial_tci;
4272 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
4273 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
4274 static struct subfacet *
4275 subfacet_find(struct ofproto_dpif *ofproto,
4276 const struct nlattr *key, size_t key_len)
4278 uint32_t key_hash = odp_flow_key_hash(key, key_len);
4279 enum odp_key_fitness fitness;
4282 fitness = odp_flow_key_to_flow(key, key_len, &flow);
4283 if (fitness == ODP_FIT_ERROR) {
4287 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
4290 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4291 * its facet within 'ofproto', and frees it. */
4293 subfacet_destroy__(struct subfacet *subfacet)
4295 struct facet *facet = subfacet->facet;
4296 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4298 subfacet_uninstall(subfacet);
4299 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
4300 list_remove(&subfacet->list_node);
4301 free(subfacet->key);
4302 free(subfacet->actions);
4303 if (subfacet != &facet->one_subfacet) {
4308 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4309 * last remaining subfacet in its facet destroys the facet too. */
4311 subfacet_destroy(struct subfacet *subfacet)
4313 struct facet *facet = subfacet->facet;
4315 if (list_is_singleton(&facet->subfacets)) {
4316 /* facet_remove() needs at least one subfacet (it will remove it). */
4317 facet_remove(facet);
4319 subfacet_destroy__(subfacet);
4323 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
4324 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
4325 * for use as temporary storage. */
4327 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
4330 if (!subfacet->key) {
4331 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
4332 odp_flow_key_from_flow(key, &subfacet->facet->flow);
4334 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
4338 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
4339 * Translates the actions into 'odp_actions', which the caller must have
4340 * initialized and is responsible for uninitializing. */
4342 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
4343 struct ofpbuf *odp_actions)
4345 struct facet *facet = subfacet->facet;
4346 struct rule_dpif *rule = facet->rule;
4347 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4349 struct action_xlate_ctx ctx;
4351 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
4353 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
4354 facet->tags = ctx.tags;
4355 facet->has_learn = ctx.has_learn;
4356 facet->has_normal = ctx.has_normal;
4357 facet->has_fin_timeout = ctx.has_fin_timeout;
4358 facet->nf_flow.output_iface = ctx.nf_output_iface;
4359 facet->mirrors = ctx.mirrors;
4361 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4362 if (subfacet->actions_len != odp_actions->size
4363 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
4364 free(subfacet->actions);
4365 subfacet->actions_len = odp_actions->size;
4366 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
4370 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4371 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4372 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4373 * since 'subfacet' was last updated.
4375 * Returns 0 if successful, otherwise a positive errno value. */
4377 subfacet_install(struct subfacet *subfacet,
4378 const struct nlattr *actions, size_t actions_len,
4379 struct dpif_flow_stats *stats,
4380 enum slow_path_reason slow)
4382 struct facet *facet = subfacet->facet;
4383 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4384 enum subfacet_path path = subfacet_want_path(slow);
4385 uint64_t slow_path_stub[128 / 8];
4386 struct odputil_keybuf keybuf;
4387 enum dpif_flow_put_flags flags;
4391 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
4393 flags |= DPIF_FP_ZERO_STATS;
4396 if (path == SF_SLOW_PATH) {
4397 compose_slow_path(ofproto, &facet->flow, slow,
4398 slow_path_stub, sizeof slow_path_stub,
4399 &actions, &actions_len);
4402 subfacet_get_key(subfacet, &keybuf, &key);
4403 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
4404 actions, actions_len, stats);
4407 subfacet_reset_dp_stats(subfacet, stats);
4411 subfacet->path = path;
4417 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
4419 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
4420 stats, subfacet->slow);
4423 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4425 subfacet_uninstall(struct subfacet *subfacet)
4427 if (subfacet->path != SF_NOT_INSTALLED) {
4428 struct rule_dpif *rule = subfacet->facet->rule;
4429 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4430 struct odputil_keybuf keybuf;
4431 struct dpif_flow_stats stats;
4435 subfacet_get_key(subfacet, &keybuf, &key);
4436 error = dpif_flow_del(ofproto->dpif, key.data, key.size, &stats);
4437 subfacet_reset_dp_stats(subfacet, &stats);
4439 subfacet_update_stats(subfacet, &stats);
4441 subfacet->path = SF_NOT_INSTALLED;
4443 assert(subfacet->dp_packet_count == 0);
4444 assert(subfacet->dp_byte_count == 0);
4448 /* Resets 'subfacet''s datapath statistics counters. This should be called
4449 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4450 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4451 * was reset in the datapath. 'stats' will be modified to include only
4452 * statistics new since 'subfacet' was last updated. */
4454 subfacet_reset_dp_stats(struct subfacet *subfacet,
4455 struct dpif_flow_stats *stats)
4458 && subfacet->dp_packet_count <= stats->n_packets
4459 && subfacet->dp_byte_count <= stats->n_bytes) {
4460 stats->n_packets -= subfacet->dp_packet_count;
4461 stats->n_bytes -= subfacet->dp_byte_count;
4464 subfacet->dp_packet_count = 0;
4465 subfacet->dp_byte_count = 0;
4468 /* Updates 'subfacet''s used time. The caller is responsible for calling
4469 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
4471 subfacet_update_time(struct subfacet *subfacet, long long int used)
4473 if (used > subfacet->used) {
4474 subfacet->used = used;
4475 facet_update_time(subfacet->facet, used);
4479 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4481 * Because of the meaning of a subfacet's counters, it only makes sense to do
4482 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4483 * represents a packet that was sent by hand or if it represents statistics
4484 * that have been cleared out of the datapath. */
4486 subfacet_update_stats(struct subfacet *subfacet,
4487 const struct dpif_flow_stats *stats)
4489 if (stats->n_packets || stats->used > subfacet->used) {
4490 struct facet *facet = subfacet->facet;
4492 subfacet_update_time(subfacet, stats->used);
4493 facet->packet_count += stats->n_packets;
4494 facet->byte_count += stats->n_bytes;
4495 facet->tcp_flags |= stats->tcp_flags;
4496 facet_push_stats(facet);
4497 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
4503 static struct rule_dpif *
4504 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
4506 struct ofport_dpif *port;
4507 struct rule_dpif *rule;
4509 rule = rule_dpif_lookup__(ofproto, flow, 0);
4514 port = get_ofp_port(ofproto, flow->in_port);
4516 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
4517 return ofproto->miss_rule;
4520 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
4521 return ofproto->no_packet_in_rule;
4523 return ofproto->miss_rule;
4526 static struct rule_dpif *
4527 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
4530 struct cls_rule *cls_rule;
4531 struct classifier *cls;
4533 if (table_id >= N_TABLES) {
4537 cls = &ofproto->up.tables[table_id].cls;
4538 if (flow->nw_frag & FLOW_NW_FRAG_ANY
4539 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4540 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
4541 * are unavailable. */
4542 struct flow ofpc_normal_flow = *flow;
4543 ofpc_normal_flow.tp_src = htons(0);
4544 ofpc_normal_flow.tp_dst = htons(0);
4545 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
4547 cls_rule = classifier_lookup(cls, flow);
4549 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
4553 complete_operation(struct rule_dpif *rule)
4555 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4557 rule_invalidate(rule);
4559 struct dpif_completion *c = xmalloc(sizeof *c);
4560 c->op = rule->up.pending;
4561 list_push_back(&ofproto->completions, &c->list_node);
4563 ofoperation_complete(rule->up.pending, 0);
4567 static struct rule *
4570 struct rule_dpif *rule = xmalloc(sizeof *rule);
4575 rule_dealloc(struct rule *rule_)
4577 struct rule_dpif *rule = rule_dpif_cast(rule_);
4582 rule_construct(struct rule *rule_)
4584 struct rule_dpif *rule = rule_dpif_cast(rule_);
4585 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4586 struct rule_dpif *victim;
4590 error = ofpacts_check(rule->up.ofpacts, rule->up.ofpacts_len,
4591 &rule->up.cr.flow, ofproto->max_ports);
4596 rule->packet_count = 0;
4597 rule->byte_count = 0;
4599 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
4600 if (victim && !list_is_empty(&victim->facets)) {
4601 struct facet *facet;
4603 rule->facets = victim->facets;
4604 list_moved(&rule->facets);
4605 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4606 /* XXX: We're only clearing our local counters here. It's possible
4607 * that quite a few packets are unaccounted for in the datapath
4608 * statistics. These will be accounted to the new rule instead of
4609 * cleared as required. This could be fixed by clearing out the
4610 * datapath statistics for this facet, but currently it doesn't
4612 facet_reset_counters(facet);
4616 /* Must avoid list_moved() in this case. */
4617 list_init(&rule->facets);
4620 table_id = rule->up.table_id;
4621 rule->tag = (victim ? victim->tag
4623 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
4624 ofproto->tables[table_id].basis));
4626 complete_operation(rule);
4631 rule_destruct(struct rule *rule_)
4633 struct rule_dpif *rule = rule_dpif_cast(rule_);
4634 struct facet *facet, *next_facet;
4636 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4637 facet_revalidate(facet);
4640 complete_operation(rule);
4644 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4646 struct rule_dpif *rule = rule_dpif_cast(rule_);
4647 struct facet *facet;
4649 /* Start from historical data for 'rule' itself that are no longer tracked
4650 * in facets. This counts, for example, facets that have expired. */
4651 *packets = rule->packet_count;
4652 *bytes = rule->byte_count;
4654 /* Add any statistics that are tracked by facets. This includes
4655 * statistical data recently updated by ofproto_update_stats() as well as
4656 * stats for packets that were executed "by hand" via dpif_execute(). */
4657 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4658 *packets += facet->packet_count;
4659 *bytes += facet->byte_count;
4664 rule_execute(struct rule *rule_, const struct flow *flow,
4665 struct ofpbuf *packet)
4667 struct rule_dpif *rule = rule_dpif_cast(rule_);
4668 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4670 struct dpif_flow_stats stats;
4672 struct action_xlate_ctx ctx;
4673 uint64_t odp_actions_stub[1024 / 8];
4674 struct ofpbuf odp_actions;
4676 dpif_flow_stats_extract(flow, packet, &stats);
4677 rule_credit_stats(rule, &stats);
4679 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4680 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
4681 rule, stats.tcp_flags, packet);
4682 ctx.resubmit_stats = &stats;
4683 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
4685 execute_odp_actions(ofproto, flow, odp_actions.data,
4686 odp_actions.size, packet);
4688 ofpbuf_uninit(&odp_actions);
4694 rule_modify_actions(struct rule *rule_)
4696 struct rule_dpif *rule = rule_dpif_cast(rule_);
4697 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4700 error = ofpacts_check(rule->up.ofpacts, rule->up.ofpacts_len,
4701 &rule->up.cr.flow, ofproto->max_ports);
4703 ofoperation_complete(rule->up.pending, error);
4707 complete_operation(rule);
4710 /* Sends 'packet' out 'ofport'.
4711 * May modify 'packet'.
4712 * Returns 0 if successful, otherwise a positive errno value. */
4714 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4716 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4717 struct ofpbuf key, odp_actions;
4718 struct odputil_keybuf keybuf;
4723 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
4724 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4726 if (odp_port != ofport->odp_port) {
4727 eth_pop_vlan(packet);
4728 flow.vlan_tci = htons(0);
4731 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4732 odp_flow_key_from_flow(&key, &flow);
4734 ofpbuf_init(&odp_actions, 32);
4735 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4737 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4738 error = dpif_execute(ofproto->dpif,
4740 odp_actions.data, odp_actions.size,
4742 ofpbuf_uninit(&odp_actions);
4745 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4746 ofproto->up.name, odp_port, strerror(error));
4748 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
4752 /* OpenFlow to datapath action translation. */
4754 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
4755 struct action_xlate_ctx *);
4756 static void xlate_normal(struct action_xlate_ctx *);
4758 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
4759 * The action will state 'slow' as the reason that the action is in the slow
4760 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
4761 * dump-flows" output to see why a flow is in the slow path.)
4763 * The 'stub_size' bytes in 'stub' will be used to store the action.
4764 * 'stub_size' must be large enough for the action.
4766 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
4769 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
4770 enum slow_path_reason slow,
4771 uint64_t *stub, size_t stub_size,
4772 const struct nlattr **actionsp, size_t *actions_lenp)
4774 union user_action_cookie cookie;
4777 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
4778 cookie.slow_path.unused = 0;
4779 cookie.slow_path.reason = slow;
4781 ofpbuf_use_stack(&buf, stub, stub_size);
4782 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
4783 uint32_t pid = dpif_port_get_pid(ofproto->dpif, UINT16_MAX);
4784 odp_put_userspace_action(pid, &cookie, &buf);
4786 put_userspace_action(ofproto, &buf, flow, &cookie);
4788 *actionsp = buf.data;
4789 *actions_lenp = buf.size;
4793 put_userspace_action(const struct ofproto_dpif *ofproto,
4794 struct ofpbuf *odp_actions,
4795 const struct flow *flow,
4796 const union user_action_cookie *cookie)
4800 pid = dpif_port_get_pid(ofproto->dpif,
4801 ofp_port_to_odp_port(flow->in_port));
4803 return odp_put_userspace_action(pid, cookie, odp_actions);
4807 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
4808 ovs_be16 vlan_tci, uint32_t odp_port,
4809 unsigned int n_outputs, union user_action_cookie *cookie)
4813 cookie->type = USER_ACTION_COOKIE_SFLOW;
4814 cookie->sflow.vlan_tci = vlan_tci;
4816 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
4817 * port information") for the interpretation of cookie->output. */
4818 switch (n_outputs) {
4820 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
4821 cookie->sflow.output = 0x40000000 | 256;
4825 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4827 cookie->sflow.output = ifindex;
4832 /* 0x80000000 means "multiple output ports. */
4833 cookie->sflow.output = 0x80000000 | n_outputs;
4838 /* Compose SAMPLE action for sFlow. */
4840 compose_sflow_action(const struct ofproto_dpif *ofproto,
4841 struct ofpbuf *odp_actions,
4842 const struct flow *flow,
4845 uint32_t probability;
4846 union user_action_cookie cookie;
4847 size_t sample_offset, actions_offset;
4850 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4854 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4856 /* Number of packets out of UINT_MAX to sample. */
4857 probability = dpif_sflow_get_probability(ofproto->sflow);
4858 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4860 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4861 compose_sflow_cookie(ofproto, htons(0), odp_port,
4862 odp_port == OVSP_NONE ? 0 : 1, &cookie);
4863 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4865 nl_msg_end_nested(odp_actions, actions_offset);
4866 nl_msg_end_nested(odp_actions, sample_offset);
4867 return cookie_offset;
4870 /* SAMPLE action must be first action in any given list of actions.
4871 * At this point we do not have all information required to build it. So try to
4872 * build sample action as complete as possible. */
4874 add_sflow_action(struct action_xlate_ctx *ctx)
4876 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4878 &ctx->flow, OVSP_NONE);
4879 ctx->sflow_odp_port = 0;
4880 ctx->sflow_n_outputs = 0;
4883 /* Fix SAMPLE action according to data collected while composing ODP actions.
4884 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4885 * USERSPACE action's user-cookie which is required for sflow. */
4887 fix_sflow_action(struct action_xlate_ctx *ctx)
4889 const struct flow *base = &ctx->base_flow;
4890 union user_action_cookie *cookie;
4892 if (!ctx->user_cookie_offset) {
4896 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4898 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4900 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
4901 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
4905 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4908 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4909 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4910 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4911 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4915 struct priority_to_dscp *pdscp;
4917 if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD
4918 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4922 pdscp = get_priority(ofport, ctx->flow.skb_priority);
4924 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4925 ctx->flow.nw_tos |= pdscp->dscp;
4928 /* We may not have an ofport record for this port, but it doesn't hurt
4929 * to allow forwarding to it anyhow. Maybe such a port will appear
4930 * later and we're pre-populating the flow table. */
4933 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4934 ctx->flow.vlan_tci);
4935 if (out_port != odp_port) {
4936 ctx->flow.vlan_tci = htons(0);
4938 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4939 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4941 ctx->sflow_odp_port = odp_port;
4942 ctx->sflow_n_outputs++;
4943 ctx->nf_output_iface = ofp_port;
4944 ctx->flow.vlan_tci = flow_vlan_tci;
4945 ctx->flow.nw_tos = flow_nw_tos;
4949 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4951 compose_output_action__(ctx, ofp_port, true);
4955 xlate_table_action(struct action_xlate_ctx *ctx,
4956 uint16_t in_port, uint8_t table_id)
4958 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4959 struct ofproto_dpif *ofproto = ctx->ofproto;
4960 struct rule_dpif *rule;
4961 uint16_t old_in_port;
4962 uint8_t old_table_id;
4964 old_table_id = ctx->table_id;
4965 ctx->table_id = table_id;
4967 /* Look up a flow with 'in_port' as the input port. */
4968 old_in_port = ctx->flow.in_port;
4969 ctx->flow.in_port = in_port;
4970 rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id);
4973 if (table_id > 0 && table_id < N_TABLES) {
4974 struct table_dpif *table = &ofproto->tables[table_id];
4975 if (table->other_table) {
4976 ctx->tags |= (rule && rule->tag
4978 : rule_calculate_tag(&ctx->flow,
4979 &table->other_table->wc,
4984 /* Restore the original input port. Otherwise OFPP_NORMAL and
4985 * OFPP_IN_PORT will have surprising behavior. */
4986 ctx->flow.in_port = old_in_port;
4988 if (ctx->resubmit_hook) {
4989 ctx->resubmit_hook(ctx, rule);
4993 struct rule_dpif *old_rule = ctx->rule;
4995 if (ctx->resubmit_stats) {
4996 rule_credit_stats(rule, ctx->resubmit_stats);
5001 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
5002 ctx->rule = old_rule;
5006 ctx->table_id = old_table_id;
5008 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5010 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
5011 MAX_RESUBMIT_RECURSION);
5012 ctx->max_resubmit_trigger = true;
5017 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
5018 const struct ofpact_resubmit *resubmit)
5023 in_port = resubmit->in_port;
5024 if (in_port == OFPP_IN_PORT) {
5025 in_port = ctx->flow.in_port;
5028 table_id = resubmit->table_id;
5029 if (table_id == 255) {
5030 table_id = ctx->table_id;
5033 xlate_table_action(ctx, in_port, table_id);
5037 flood_packets(struct action_xlate_ctx *ctx, bool all)
5039 struct ofport_dpif *ofport;
5041 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
5042 uint16_t ofp_port = ofport->up.ofp_port;
5044 if (ofp_port == ctx->flow.in_port) {
5049 compose_output_action__(ctx, ofp_port, false);
5050 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
5051 compose_output_action(ctx, ofp_port);
5055 ctx->nf_output_iface = NF_OUT_FLOOD;
5059 execute_controller_action(struct action_xlate_ctx *ctx, int len,
5060 enum ofp_packet_in_reason reason,
5061 uint16_t controller_id)
5063 struct ofputil_packet_in pin;
5064 struct ofpbuf *packet;
5066 ctx->slow |= SLOW_CONTROLLER;
5071 packet = ofpbuf_clone(ctx->packet);
5073 if (packet->l2 && packet->l3) {
5074 struct eth_header *eh;
5076 eth_pop_vlan(packet);
5079 /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
5080 * LLC frame. Calculating the Ethernet type of these frames is more
5081 * trouble than seems appropriate for a simple assertion. */
5082 assert(ntohs(eh->eth_type) < ETH_TYPE_MIN
5083 || eh->eth_type == ctx->flow.dl_type);
5085 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
5086 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
5088 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
5089 eth_push_vlan(packet, ctx->flow.vlan_tci);
5093 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5094 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
5095 ctx->flow.nw_tos, ctx->flow.nw_ttl);
5099 if (ctx->flow.nw_proto == IPPROTO_TCP) {
5100 packet_set_tcp_port(packet, ctx->flow.tp_src,
5102 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
5103 packet_set_udp_port(packet, ctx->flow.tp_src,
5110 pin.packet = packet->data;
5111 pin.packet_len = packet->size;
5112 pin.reason = reason;
5113 pin.controller_id = controller_id;
5114 pin.table_id = ctx->table_id;
5115 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
5118 flow_get_metadata(&ctx->flow, &pin.fmd);
5120 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
5121 ofpbuf_delete(packet);
5125 compose_dec_ttl(struct action_xlate_ctx *ctx)
5127 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
5128 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
5132 if (ctx->flow.nw_ttl > 1) {
5136 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
5138 /* Stop processing for current table. */
5144 xlate_output_action(struct action_xlate_ctx *ctx,
5145 uint16_t port, uint16_t max_len)
5147 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
5149 ctx->nf_output_iface = NF_OUT_DROP;
5153 compose_output_action(ctx, ctx->flow.in_port);
5156 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
5162 flood_packets(ctx, false);
5165 flood_packets(ctx, true);
5167 case OFPP_CONTROLLER:
5168 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
5174 if (port != ctx->flow.in_port) {
5175 compose_output_action(ctx, port);
5180 if (prev_nf_output_iface == NF_OUT_FLOOD) {
5181 ctx->nf_output_iface = NF_OUT_FLOOD;
5182 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
5183 ctx->nf_output_iface = prev_nf_output_iface;
5184 } else if (prev_nf_output_iface != NF_OUT_DROP &&
5185 ctx->nf_output_iface != NF_OUT_FLOOD) {
5186 ctx->nf_output_iface = NF_OUT_MULTI;
5191 xlate_output_reg_action(struct action_xlate_ctx *ctx,
5192 const struct ofpact_output_reg *or)
5194 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
5195 if (port <= UINT16_MAX) {
5196 xlate_output_action(ctx, port, or->max_len);
5201 xlate_enqueue_action(struct action_xlate_ctx *ctx,
5202 const struct ofpact_enqueue *enqueue)
5204 uint16_t ofp_port = enqueue->port;
5205 uint32_t queue_id = enqueue->queue;
5206 uint32_t flow_priority, priority;
5209 /* Translate queue to priority. */
5210 error = dpif_queue_to_priority(ctx->ofproto->dpif, queue_id, &priority);
5212 /* Fall back to ordinary output action. */
5213 xlate_output_action(ctx, enqueue->port, 0);
5217 /* Check output port. */
5218 if (ofp_port == OFPP_IN_PORT) {
5219 ofp_port = ctx->flow.in_port;
5220 } else if (ofp_port == ctx->flow.in_port) {
5224 /* Add datapath actions. */
5225 flow_priority = ctx->flow.skb_priority;
5226 ctx->flow.skb_priority = priority;
5227 compose_output_action(ctx, ofp_port);
5228 ctx->flow.skb_priority = flow_priority;
5230 /* Update NetFlow output port. */
5231 if (ctx->nf_output_iface == NF_OUT_DROP) {
5232 ctx->nf_output_iface = ofp_port;
5233 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
5234 ctx->nf_output_iface = NF_OUT_MULTI;
5239 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
5241 uint32_t skb_priority;
5243 if (!dpif_queue_to_priority(ctx->ofproto->dpif, queue_id, &skb_priority)) {
5244 ctx->flow.skb_priority = skb_priority;
5246 /* Couldn't translate queue to a priority. Nothing to do. A warning
5247 * has already been logged. */
5251 struct xlate_reg_state {
5257 xlate_autopath(struct action_xlate_ctx *ctx,
5258 const struct ofpact_autopath *ap)
5260 uint16_t ofp_port = ap->port;
5261 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
5263 if (!port || !port->bundle) {
5264 ofp_port = OFPP_NONE;
5265 } else if (port->bundle->bond) {
5266 /* Autopath does not support VLAN hashing. */
5267 struct ofport_dpif *slave = bond_choose_output_slave(
5268 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
5270 ofp_port = slave->up.ofp_port;
5273 nxm_reg_load(&ap->dst, ofp_port, &ctx->flow);
5277 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
5279 struct ofproto_dpif *ofproto = ofproto_;
5280 struct ofport_dpif *port;
5290 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
5293 port = get_ofp_port(ofproto, ofp_port);
5294 return port ? port->may_enable : false;
5299 xlate_bundle_action(struct action_xlate_ctx *ctx,
5300 const struct ofpact_bundle *bundle)
5304 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
5305 if (bundle->dst.field) {
5306 nxm_reg_load(&bundle->dst, port, &ctx->flow);
5308 xlate_output_action(ctx, port, 0);
5313 xlate_learn_action(struct action_xlate_ctx *ctx,
5314 const struct ofpact_learn *learn)
5316 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
5317 struct ofputil_flow_mod fm;
5318 uint64_t ofpacts_stub[1024 / 8];
5319 struct ofpbuf ofpacts;
5322 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
5323 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
5325 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
5326 if (error && !VLOG_DROP_WARN(&rl)) {
5327 VLOG_WARN("learning action failed to modify flow table (%s)",
5328 ofperr_get_name(error));
5331 ofpbuf_uninit(&ofpacts);
5334 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
5335 * means "infinite". */
5337 reduce_timeout(uint16_t max, uint16_t *timeout)
5339 if (max && (!*timeout || *timeout > max)) {
5345 xlate_fin_timeout(struct action_xlate_ctx *ctx,
5346 const struct ofpact_fin_timeout *oft)
5348 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
5349 struct rule_dpif *rule = ctx->rule;
5351 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
5352 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
5357 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
5359 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
5360 ? OFPUTIL_PC_NO_RECV_STP
5361 : OFPUTIL_PC_NO_RECV)) {
5365 /* Only drop packets here if both forwarding and learning are
5366 * disabled. If just learning is enabled, we need to have
5367 * OFPP_NORMAL and the learning action have a look at the packet
5368 * before we can drop it. */
5369 if (!stp_forward_in_state(port->stp_state)
5370 && !stp_learn_in_state(port->stp_state)) {
5378 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
5379 struct action_xlate_ctx *ctx)
5381 const struct ofport_dpif *port;
5382 bool was_evictable = true;
5383 const struct ofpact *a;
5385 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5386 if (port && !may_receive(port, ctx)) {
5387 /* Drop this flow. */
5392 /* Don't let the rule we're working on get evicted underneath us. */
5393 was_evictable = ctx->rule->up.evictable;
5394 ctx->rule->up.evictable = false;
5396 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
5397 struct ofpact_controller *controller;
5405 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
5406 ofpact_get_OUTPUT(a)->max_len);
5409 case OFPACT_CONTROLLER:
5410 controller = ofpact_get_CONTROLLER(a);
5411 execute_controller_action(ctx, controller->max_len,
5413 controller->controller_id);
5416 case OFPACT_ENQUEUE:
5417 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
5420 case OFPACT_SET_VLAN_VID:
5421 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
5422 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
5426 case OFPACT_SET_VLAN_PCP:
5427 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
5428 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
5433 case OFPACT_STRIP_VLAN:
5434 ctx->flow.vlan_tci = htons(0);
5437 case OFPACT_SET_ETH_SRC:
5438 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
5442 case OFPACT_SET_ETH_DST:
5443 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
5447 case OFPACT_SET_IPV4_SRC:
5448 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
5451 case OFPACT_SET_IPV4_DST:
5452 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
5455 case OFPACT_SET_IPV4_DSCP:
5456 /* OpenFlow 1.0 only supports IPv4. */
5457 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5458 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5459 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
5463 case OFPACT_SET_L4_SRC_PORT:
5464 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
5467 case OFPACT_SET_L4_DST_PORT:
5468 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
5471 case OFPACT_RESUBMIT:
5472 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
5475 case OFPACT_SET_TUNNEL:
5476 ctx->flow.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
5479 case OFPACT_SET_QUEUE:
5480 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
5483 case OFPACT_POP_QUEUE:
5484 ctx->flow.skb_priority = ctx->orig_skb_priority;
5487 case OFPACT_REG_MOVE:
5488 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
5491 case OFPACT_REG_LOAD:
5492 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
5495 case OFPACT_DEC_TTL:
5496 if (compose_dec_ttl(ctx)) {
5502 /* Nothing to do. */
5505 case OFPACT_MULTIPATH:
5506 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
5509 case OFPACT_AUTOPATH:
5510 xlate_autopath(ctx, ofpact_get_AUTOPATH(a));
5514 ctx->ofproto->has_bundle_action = true;
5515 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
5518 case OFPACT_OUTPUT_REG:
5519 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
5523 ctx->has_learn = true;
5524 if (ctx->may_learn) {
5525 xlate_learn_action(ctx, ofpact_get_LEARN(a));
5533 case OFPACT_FIN_TIMEOUT:
5534 ctx->has_fin_timeout = true;
5535 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
5541 /* We've let OFPP_NORMAL and the learning action look at the packet,
5542 * so drop it now if forwarding is disabled. */
5543 if (port && !stp_forward_in_state(port->stp_state)) {
5544 ofpbuf_clear(ctx->odp_actions);
5545 add_sflow_action(ctx);
5548 ctx->rule->up.evictable = was_evictable;
5553 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
5554 struct ofproto_dpif *ofproto, const struct flow *flow,
5555 ovs_be16 initial_tci, struct rule_dpif *rule,
5556 uint8_t tcp_flags, const struct ofpbuf *packet)
5558 ctx->ofproto = ofproto;
5560 ctx->base_flow = ctx->flow;
5561 ctx->base_flow.tun_id = 0;
5562 ctx->base_flow.vlan_tci = initial_tci;
5564 ctx->packet = packet;
5565 ctx->may_learn = packet != NULL;
5566 ctx->tcp_flags = tcp_flags;
5567 ctx->resubmit_hook = NULL;
5568 ctx->resubmit_stats = NULL;
5571 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
5572 * into datapath actions in 'odp_actions', using 'ctx'. */
5574 xlate_actions(struct action_xlate_ctx *ctx,
5575 const struct ofpact *ofpacts, size_t ofpacts_len,
5576 struct ofpbuf *odp_actions)
5578 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
5579 * that in the future we always keep a copy of the original flow for
5580 * tracing purposes. */
5581 static bool hit_resubmit_limit;
5583 enum slow_path_reason special;
5585 COVERAGE_INC(ofproto_dpif_xlate);
5587 ofpbuf_clear(odp_actions);
5588 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
5590 ctx->odp_actions = odp_actions;
5593 ctx->has_learn = false;
5594 ctx->has_normal = false;
5595 ctx->has_fin_timeout = false;
5596 ctx->nf_output_iface = NF_OUT_DROP;
5599 ctx->max_resubmit_trigger = false;
5600 ctx->orig_skb_priority = ctx->flow.skb_priority;
5604 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
5605 /* Do this conditionally because the copy is expensive enough that it
5606 * shows up in profiles.
5608 * We keep orig_flow in 'ctx' only because I couldn't make GCC 4.4
5609 * believe that I wasn't using it without initializing it if I kept it
5610 * in a local variable. */
5611 ctx->orig_flow = ctx->flow;
5614 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
5615 switch (ctx->ofproto->up.frag_handling) {
5616 case OFPC_FRAG_NORMAL:
5617 /* We must pretend that transport ports are unavailable. */
5618 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
5619 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
5622 case OFPC_FRAG_DROP:
5625 case OFPC_FRAG_REASM:
5628 case OFPC_FRAG_NX_MATCH:
5629 /* Nothing to do. */
5632 case OFPC_INVALID_TTL_TO_CONTROLLER:
5637 special = process_special(ctx->ofproto, &ctx->flow, ctx->packet);
5639 ctx->slow |= special;
5641 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5642 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
5644 add_sflow_action(ctx);
5645 do_xlate_actions(ofpacts, ofpacts_len, ctx);
5647 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
5648 if (!hit_resubmit_limit) {
5649 /* We didn't record the original flow. Make sure we do from
5651 hit_resubmit_limit = true;
5652 } else if (!VLOG_DROP_ERR(&trace_rl)) {
5653 struct ds ds = DS_EMPTY_INITIALIZER;
5655 ofproto_trace(ctx->ofproto, &ctx->orig_flow, ctx->packet,
5657 VLOG_ERR("Trace triggered by excessive resubmit "
5658 "recursion:\n%s", ds_cstr(&ds));
5663 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
5664 ctx->odp_actions->data,
5665 ctx->odp_actions->size)) {
5666 ctx->slow |= SLOW_IN_BAND;
5668 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
5670 compose_output_action(ctx, OFPP_LOCAL);
5673 if (ctx->ofproto->has_mirrors) {
5674 add_mirror_actions(ctx, &ctx->orig_flow);
5676 fix_sflow_action(ctx);
5680 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
5681 * into datapath actions, using 'ctx', and discards the datapath actions. */
5683 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
5684 const struct ofpact *ofpacts,
5687 uint64_t odp_actions_stub[1024 / 8];
5688 struct ofpbuf odp_actions;
5690 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5691 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
5692 ofpbuf_uninit(&odp_actions);
5695 /* OFPP_NORMAL implementation. */
5697 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
5699 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
5700 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
5701 * the bundle on which the packet was received, returns the VLAN to which the
5704 * Both 'vid' and the return value are in the range 0...4095. */
5706 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
5708 switch (in_bundle->vlan_mode) {
5709 case PORT_VLAN_ACCESS:
5710 return in_bundle->vlan;
5713 case PORT_VLAN_TRUNK:
5716 case PORT_VLAN_NATIVE_UNTAGGED:
5717 case PORT_VLAN_NATIVE_TAGGED:
5718 return vid ? vid : in_bundle->vlan;
5725 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
5726 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
5729 * 'vid' should be the VID obtained from the 802.1Q header that was received as
5730 * part of a packet (specify 0 if there was no 802.1Q header), in the range
5733 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
5735 /* Allow any VID on the OFPP_NONE port. */
5736 if (in_bundle == &ofpp_none_bundle) {
5740 switch (in_bundle->vlan_mode) {
5741 case PORT_VLAN_ACCESS:
5744 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5745 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
5746 "packet received on port %s configured as VLAN "
5747 "%"PRIu16" access port",
5748 in_bundle->ofproto->up.name, vid,
5749 in_bundle->name, in_bundle->vlan);
5755 case PORT_VLAN_NATIVE_UNTAGGED:
5756 case PORT_VLAN_NATIVE_TAGGED:
5758 /* Port must always carry its native VLAN. */
5762 case PORT_VLAN_TRUNK:
5763 if (!ofbundle_includes_vlan(in_bundle, vid)) {
5765 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5766 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
5767 "received on port %s not configured for trunking "
5769 in_bundle->ofproto->up.name, vid,
5770 in_bundle->name, vid);
5782 /* Given 'vlan', the VLAN that a packet belongs to, and
5783 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
5784 * that should be included in the 802.1Q header. (If the return value is 0,
5785 * then the 802.1Q header should only be included in the packet if there is a
5788 * Both 'vlan' and the return value are in the range 0...4095. */
5790 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
5792 switch (out_bundle->vlan_mode) {
5793 case PORT_VLAN_ACCESS:
5796 case PORT_VLAN_TRUNK:
5797 case PORT_VLAN_NATIVE_TAGGED:
5800 case PORT_VLAN_NATIVE_UNTAGGED:
5801 return vlan == out_bundle->vlan ? 0 : vlan;
5809 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
5812 struct ofport_dpif *port;
5814 ovs_be16 tci, old_tci;
5816 vid = output_vlan_to_vid(out_bundle, vlan);
5817 if (!out_bundle->bond) {
5818 port = ofbundle_get_a_port(out_bundle);
5820 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
5823 /* No slaves enabled, so drop packet. */
5828 old_tci = ctx->flow.vlan_tci;
5830 if (tci || out_bundle->use_priority_tags) {
5831 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
5833 tci |= htons(VLAN_CFI);
5836 ctx->flow.vlan_tci = tci;
5838 compose_output_action(ctx, port->up.ofp_port);
5839 ctx->flow.vlan_tci = old_tci;
5843 mirror_mask_ffs(mirror_mask_t mask)
5845 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
5850 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
5852 return (bundle->vlan_mode != PORT_VLAN_ACCESS
5853 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
5857 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
5859 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
5862 /* Returns an arbitrary interface within 'bundle'. */
5863 static struct ofport_dpif *
5864 ofbundle_get_a_port(const struct ofbundle *bundle)
5866 return CONTAINER_OF(list_front(&bundle->ports),
5867 struct ofport_dpif, bundle_node);
5871 vlan_is_mirrored(const struct ofmirror *m, int vlan)
5873 return !m->vlans || bitmap_is_set(m->vlans, vlan);
5877 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5879 struct ofproto_dpif *ofproto = ctx->ofproto;
5880 mirror_mask_t mirrors;
5881 struct ofbundle *in_bundle;
5884 const struct nlattr *a;
5887 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5888 ctx->packet != NULL, NULL);
5892 mirrors = in_bundle->src_mirrors;
5894 /* Drop frames on bundles reserved for mirroring. */
5895 if (in_bundle->mirror_out) {
5896 if (ctx->packet != NULL) {
5897 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5898 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5899 "%s, which is reserved exclusively for mirroring",
5900 ctx->ofproto->up.name, in_bundle->name);
5906 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5907 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5910 vlan = input_vid_to_vlan(in_bundle, vid);
5912 /* Look at the output ports to check for destination selections. */
5914 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5915 ctx->odp_actions->size) {
5916 enum ovs_action_attr type = nl_attr_type(a);
5917 struct ofport_dpif *ofport;
5919 if (type != OVS_ACTION_ATTR_OUTPUT) {
5923 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5924 if (ofport && ofport->bundle) {
5925 mirrors |= ofport->bundle->dst_mirrors;
5933 /* Restore the original packet before adding the mirror actions. */
5934 ctx->flow = *orig_flow;
5939 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5941 if (!vlan_is_mirrored(m, vlan)) {
5942 mirrors &= mirrors - 1;
5946 mirrors &= ~m->dup_mirrors;
5947 ctx->mirrors |= m->dup_mirrors;
5949 output_normal(ctx, m->out, vlan);
5950 } else if (vlan != m->out_vlan
5951 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
5952 struct ofbundle *bundle;
5954 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5955 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5956 && !bundle->mirror_out) {
5957 output_normal(ctx, bundle, m->out_vlan);
5965 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5966 uint64_t packets, uint64_t bytes)
5972 for (; mirrors; mirrors &= mirrors - 1) {
5975 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5978 /* In normal circumstances 'm' will not be NULL. However,
5979 * if mirrors are reconfigured, we can temporarily get out
5980 * of sync in facet_revalidate(). We could "correct" the
5981 * mirror list before reaching here, but doing that would
5982 * not properly account the traffic stats we've currently
5983 * accumulated for previous mirror configuration. */
5987 m->packet_count += packets;
5988 m->byte_count += bytes;
5992 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
5993 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
5994 * indicate this; newer upstream kernels use gratuitous ARP requests. */
5996 is_gratuitous_arp(const struct flow *flow)
5998 return (flow->dl_type == htons(ETH_TYPE_ARP)
5999 && eth_addr_is_broadcast(flow->dl_dst)
6000 && (flow->nw_proto == ARP_OP_REPLY
6001 || (flow->nw_proto == ARP_OP_REQUEST
6002 && flow->nw_src == flow->nw_dst)));
6006 update_learning_table(struct ofproto_dpif *ofproto,
6007 const struct flow *flow, int vlan,
6008 struct ofbundle *in_bundle)
6010 struct mac_entry *mac;
6012 /* Don't learn the OFPP_NONE port. */
6013 if (in_bundle == &ofpp_none_bundle) {
6017 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
6021 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
6022 if (is_gratuitous_arp(flow)) {
6023 /* We don't want to learn from gratuitous ARP packets that are
6024 * reflected back over bond slaves so we lock the learning table. */
6025 if (!in_bundle->bond) {
6026 mac_entry_set_grat_arp_lock(mac);
6027 } else if (mac_entry_is_grat_arp_locked(mac)) {
6032 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
6033 /* The log messages here could actually be useful in debugging,
6034 * so keep the rate limit relatively high. */
6035 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
6036 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
6037 "on port %s in VLAN %d",
6038 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
6039 in_bundle->name, vlan);
6041 mac->port.p = in_bundle;
6042 tag_set_add(&ofproto->revalidate_set,
6043 mac_learning_changed(ofproto->ml, mac));
6047 static struct ofbundle *
6048 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
6049 bool warn, struct ofport_dpif **in_ofportp)
6051 struct ofport_dpif *ofport;
6053 /* Find the port and bundle for the received packet. */
6054 ofport = get_ofp_port(ofproto, in_port);
6056 *in_ofportp = ofport;
6058 if (ofport && ofport->bundle) {
6059 return ofport->bundle;
6062 /* Special-case OFPP_NONE, which a controller may use as the ingress
6063 * port for traffic that it is sourcing. */
6064 if (in_port == OFPP_NONE) {
6065 return &ofpp_none_bundle;
6068 /* Odd. A few possible reasons here:
6070 * - We deleted a port but there are still a few packets queued up
6073 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
6074 * we don't know about.
6076 * - The ofproto client didn't configure the port as part of a bundle.
6077 * This is particularly likely to happen if a packet was received on the
6078 * port after it was created, but before the client had a chance to
6079 * configure its bundle.
6082 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6084 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
6085 "port %"PRIu16, ofproto->up.name, in_port);
6090 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
6091 * dropped. Returns true if they may be forwarded, false if they should be
6094 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
6095 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
6097 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
6098 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
6099 * checked by input_vid_is_valid().
6101 * May also add tags to '*tags', although the current implementation only does
6102 * so in one special case.
6105 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
6106 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
6108 struct ofbundle *in_bundle = in_port->bundle;
6110 /* Drop frames for reserved multicast addresses
6111 * only if forward_bpdu option is absent. */
6112 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
6116 if (in_bundle->bond) {
6117 struct mac_entry *mac;
6119 switch (bond_check_admissibility(in_bundle->bond, in_port,
6120 flow->dl_dst, tags)) {
6127 case BV_DROP_IF_MOVED:
6128 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
6129 if (mac && mac->port.p != in_bundle &&
6130 (!is_gratuitous_arp(flow)
6131 || mac_entry_is_grat_arp_locked(mac))) {
6142 xlate_normal(struct action_xlate_ctx *ctx)
6144 struct ofport_dpif *in_port;
6145 struct ofbundle *in_bundle;
6146 struct mac_entry *mac;
6150 ctx->has_normal = true;
6152 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
6153 ctx->packet != NULL, &in_port);
6158 /* Drop malformed frames. */
6159 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
6160 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
6161 if (ctx->packet != NULL) {
6162 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6163 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
6164 "VLAN tag received on port %s",
6165 ctx->ofproto->up.name, in_bundle->name);
6170 /* Drop frames on bundles reserved for mirroring. */
6171 if (in_bundle->mirror_out) {
6172 if (ctx->packet != NULL) {
6173 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6174 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6175 "%s, which is reserved exclusively for mirroring",
6176 ctx->ofproto->up.name, in_bundle->name);
6182 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
6183 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6186 vlan = input_vid_to_vlan(in_bundle, vid);
6188 /* Check other admissibility requirements. */
6190 !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
6194 /* Learn source MAC. */
6195 if (ctx->may_learn) {
6196 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
6199 /* Determine output bundle. */
6200 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
6203 if (mac->port.p != in_bundle) {
6204 output_normal(ctx, mac->port.p, vlan);
6207 struct ofbundle *bundle;
6209 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
6210 if (bundle != in_bundle
6211 && ofbundle_includes_vlan(bundle, vlan)
6212 && bundle->floodable
6213 && !bundle->mirror_out) {
6214 output_normal(ctx, bundle, vlan);
6217 ctx->nf_output_iface = NF_OUT_FLOOD;
6221 /* Optimized flow revalidation.
6223 * It's a difficult problem, in general, to tell which facets need to have
6224 * their actions recalculated whenever the OpenFlow flow table changes. We
6225 * don't try to solve that general problem: for most kinds of OpenFlow flow
6226 * table changes, we recalculate the actions for every facet. This is
6227 * relatively expensive, but it's good enough if the OpenFlow flow table
6228 * doesn't change very often.
6230 * However, we can expect one particular kind of OpenFlow flow table change to
6231 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
6232 * of CPU on revalidating every facet whenever MAC learning modifies the flow
6233 * table, we add a special case that applies to flow tables in which every rule
6234 * has the same form (that is, the same wildcards), except that the table is
6235 * also allowed to have a single "catch-all" flow that matches all packets. We
6236 * optimize this case by tagging all of the facets that resubmit into the table
6237 * and invalidating the same tag whenever a flow changes in that table. The
6238 * end result is that we revalidate just the facets that need it (and sometimes
6239 * a few more, but not all of the facets or even all of the facets that
6240 * resubmit to the table modified by MAC learning). */
6242 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
6243 * into an OpenFlow table with the given 'basis'. */
6245 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
6248 if (flow_wildcards_is_catchall(wc)) {
6251 struct flow tag_flow = *flow;
6252 flow_zero_wildcards(&tag_flow, wc);
6253 return tag_create_deterministic(flow_hash(&tag_flow, secret));
6257 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
6258 * taggability of that table.
6260 * This function must be called after *each* change to a flow table. If you
6261 * skip calling it on some changes then the pointer comparisons at the end can
6262 * be invalid if you get unlucky. For example, if a flow removal causes a
6263 * cls_table to be destroyed and then a flow insertion causes a cls_table with
6264 * different wildcards to be created with the same address, then this function
6265 * will incorrectly skip revalidation. */
6267 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
6269 struct table_dpif *table = &ofproto->tables[table_id];
6270 const struct oftable *oftable = &ofproto->up.tables[table_id];
6271 struct cls_table *catchall, *other;
6272 struct cls_table *t;
6274 catchall = other = NULL;
6276 switch (hmap_count(&oftable->cls.tables)) {
6278 /* We could tag this OpenFlow table but it would make the logic a
6279 * little harder and it's a corner case that doesn't seem worth it
6285 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
6286 if (cls_table_is_catchall(t)) {
6288 } else if (!other) {
6291 /* Indicate that we can't tag this by setting both tables to
6292 * NULL. (We know that 'catchall' is already NULL.) */
6299 /* Can't tag this table. */
6303 if (table->catchall_table != catchall || table->other_table != other) {
6304 table->catchall_table = catchall;
6305 table->other_table = other;
6306 ofproto->need_revalidate = REV_FLOW_TABLE;
6310 /* Given 'rule' that has changed in some way (either it is a rule being
6311 * inserted, a rule being deleted, or a rule whose actions are being
6312 * modified), marks facets for revalidation to ensure that packets will be
6313 * forwarded correctly according to the new state of the flow table.
6315 * This function must be called after *each* change to a flow table. See
6316 * the comment on table_update_taggable() for more information. */
6318 rule_invalidate(const struct rule_dpif *rule)
6320 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
6322 table_update_taggable(ofproto, rule->up.table_id);
6324 if (!ofproto->need_revalidate) {
6325 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
6327 if (table->other_table && rule->tag) {
6328 tag_set_add(&ofproto->revalidate_set, rule->tag);
6330 ofproto->need_revalidate = REV_FLOW_TABLE;
6336 set_frag_handling(struct ofproto *ofproto_,
6337 enum ofp_config_flags frag_handling)
6339 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6341 if (frag_handling != OFPC_FRAG_REASM) {
6342 ofproto->need_revalidate = REV_RECONFIGURE;
6350 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
6351 const struct flow *flow,
6352 const struct ofpact *ofpacts, size_t ofpacts_len)
6354 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6357 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
6358 return OFPERR_NXBRC_BAD_IN_PORT;
6361 error = ofpacts_check(ofpacts, ofpacts_len, flow, ofproto->max_ports);
6363 struct odputil_keybuf keybuf;
6364 struct dpif_flow_stats stats;
6368 struct action_xlate_ctx ctx;
6369 uint64_t odp_actions_stub[1024 / 8];
6370 struct ofpbuf odp_actions;
6372 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
6373 odp_flow_key_from_flow(&key, flow);
6375 dpif_flow_stats_extract(flow, packet, &stats);
6377 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL,
6378 packet_get_tcp_flags(packet, flow), packet);
6379 ctx.resubmit_stats = &stats;
6381 ofpbuf_use_stub(&odp_actions,
6382 odp_actions_stub, sizeof odp_actions_stub);
6383 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
6384 dpif_execute(ofproto->dpif, key.data, key.size,
6385 odp_actions.data, odp_actions.size, packet);
6386 ofpbuf_uninit(&odp_actions);
6394 set_netflow(struct ofproto *ofproto_,
6395 const struct netflow_options *netflow_options)
6397 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6399 if (netflow_options) {
6400 if (!ofproto->netflow) {
6401 ofproto->netflow = netflow_create();
6403 return netflow_set_options(ofproto->netflow, netflow_options);
6405 netflow_destroy(ofproto->netflow);
6406 ofproto->netflow = NULL;
6412 get_netflow_ids(const struct ofproto *ofproto_,
6413 uint8_t *engine_type, uint8_t *engine_id)
6415 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6417 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
6421 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
6423 if (!facet_is_controller_flow(facet) &&
6424 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
6425 struct subfacet *subfacet;
6426 struct ofexpired expired;
6428 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
6429 if (subfacet->path == SF_FAST_PATH) {
6430 struct dpif_flow_stats stats;
6432 subfacet_reinstall(subfacet, &stats);
6433 subfacet_update_stats(subfacet, &stats);
6437 expired.flow = facet->flow;
6438 expired.packet_count = facet->packet_count;
6439 expired.byte_count = facet->byte_count;
6440 expired.used = facet->used;
6441 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
6446 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
6448 struct facet *facet;
6450 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6451 send_active_timeout(ofproto, facet);
6455 static struct ofproto_dpif *
6456 ofproto_dpif_lookup(const char *name)
6458 struct ofproto_dpif *ofproto;
6460 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
6461 hash_string(name, 0), &all_ofproto_dpifs) {
6462 if (!strcmp(ofproto->up.name, name)) {
6470 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
6471 const char *argv[], void *aux OVS_UNUSED)
6473 struct ofproto_dpif *ofproto;
6476 ofproto = ofproto_dpif_lookup(argv[1]);
6478 unixctl_command_reply_error(conn, "no such bridge");
6481 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6483 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6484 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6488 unixctl_command_reply(conn, "table successfully flushed");
6492 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6493 const char *argv[], void *aux OVS_UNUSED)
6495 struct ds ds = DS_EMPTY_INITIALIZER;
6496 const struct ofproto_dpif *ofproto;
6497 const struct mac_entry *e;
6499 ofproto = ofproto_dpif_lookup(argv[1]);
6501 unixctl_command_reply_error(conn, "no such bridge");
6505 ds_put_cstr(&ds, " port VLAN MAC Age\n");
6506 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
6507 struct ofbundle *bundle = e->port.p;
6508 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
6509 ofbundle_get_a_port(bundle)->odp_port,
6510 e->vlan, ETH_ADDR_ARGS(e->mac),
6511 mac_entry_age(ofproto->ml, e));
6513 unixctl_command_reply(conn, ds_cstr(&ds));
6518 struct action_xlate_ctx ctx;
6524 trace_format_rule(struct ds *result, uint8_t table_id, int level,
6525 const struct rule_dpif *rule)
6527 ds_put_char_multiple(result, '\t', level);
6529 ds_put_cstr(result, "No match\n");
6533 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
6534 table_id, ntohll(rule->up.flow_cookie));
6535 cls_rule_format(&rule->up.cr, result);
6536 ds_put_char(result, '\n');
6538 ds_put_char_multiple(result, '\t', level);
6539 ds_put_cstr(result, "OpenFlow ");
6540 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
6541 ds_put_char(result, '\n');
6545 trace_format_flow(struct ds *result, int level, const char *title,
6546 struct trace_ctx *trace)
6548 ds_put_char_multiple(result, '\t', level);
6549 ds_put_format(result, "%s: ", title);
6550 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
6551 ds_put_cstr(result, "unchanged");
6553 flow_format(result, &trace->ctx.flow);
6554 trace->flow = trace->ctx.flow;
6556 ds_put_char(result, '\n');
6560 trace_format_regs(struct ds *result, int level, const char *title,
6561 struct trace_ctx *trace)
6565 ds_put_char_multiple(result, '\t', level);
6566 ds_put_format(result, "%s:", title);
6567 for (i = 0; i < FLOW_N_REGS; i++) {
6568 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
6570 ds_put_char(result, '\n');
6574 trace_format_odp(struct ds *result, int level, const char *title,
6575 struct trace_ctx *trace)
6577 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
6579 ds_put_char_multiple(result, '\t', level);
6580 ds_put_format(result, "%s: ", title);
6581 format_odp_actions(result, odp_actions->data, odp_actions->size);
6582 ds_put_char(result, '\n');
6586 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
6588 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
6589 struct ds *result = trace->result;
6591 ds_put_char(result, '\n');
6592 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
6593 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
6594 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
6595 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
6599 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
6600 void *aux OVS_UNUSED)
6602 const char *dpname = argv[1];
6603 struct ofproto_dpif *ofproto;
6604 struct ofpbuf odp_key;
6605 struct ofpbuf *packet;
6606 ovs_be16 initial_tci;
6612 ofpbuf_init(&odp_key, 0);
6615 ofproto = ofproto_dpif_lookup(dpname);
6617 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
6621 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
6622 /* ofproto/trace dpname flow [-generate] */
6623 const char *flow_s = argv[2];
6624 const char *generate_s = argv[3];
6626 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
6627 * flow. We guess which type it is based on whether 'flow_s' contains
6628 * an '(', since a datapath flow always contains '(') but an
6629 * OpenFlow-like flow should not (in fact it's allowed but I believe
6630 * that's not documented anywhere).
6632 * An alternative would be to try to parse 'flow_s' both ways, but then
6633 * it would be tricky giving a sensible error message. After all, do
6634 * you just say "syntax error" or do you present both error messages?
6635 * Both choices seem lousy. */
6636 if (strchr(flow_s, '(')) {
6639 /* Convert string to datapath key. */
6640 ofpbuf_init(&odp_key, 0);
6641 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
6643 unixctl_command_reply_error(conn, "Bad flow syntax");
6647 /* Convert odp_key to flow. */
6648 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
6649 odp_key.size, &flow,
6650 &initial_tci, NULL);
6651 if (error == ODP_FIT_ERROR) {
6652 unixctl_command_reply_error(conn, "Invalid flow");
6658 error_s = parse_ofp_exact_flow(&flow, argv[2]);
6660 unixctl_command_reply_error(conn, error_s);
6665 initial_tci = flow.vlan_tci;
6666 vsp_adjust_flow(ofproto, &flow);
6669 /* Generate a packet, if requested. */
6671 packet = ofpbuf_new(0);
6672 flow_compose(packet, &flow);
6674 } else if (argc == 6) {
6675 /* ofproto/trace dpname priority tun_id in_port packet */
6676 const char *priority_s = argv[2];
6677 const char *tun_id_s = argv[3];
6678 const char *in_port_s = argv[4];
6679 const char *packet_s = argv[5];
6680 uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s));
6681 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
6682 uint32_t priority = atoi(priority_s);
6685 msg = eth_from_hex(packet_s, &packet);
6687 unixctl_command_reply_error(conn, msg);
6691 ds_put_cstr(&result, "Packet: ");
6692 s = ofp_packet_to_string(packet->data, packet->size);
6693 ds_put_cstr(&result, s);
6696 flow_extract(packet, priority, tun_id, in_port, &flow);
6697 initial_tci = flow.vlan_tci;
6699 unixctl_command_reply_error(conn, "Bad command syntax");
6703 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
6704 unixctl_command_reply(conn, ds_cstr(&result));
6707 ds_destroy(&result);
6708 ofpbuf_delete(packet);
6709 ofpbuf_uninit(&odp_key);
6713 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
6714 const struct ofpbuf *packet, ovs_be16 initial_tci,
6717 struct rule_dpif *rule;
6719 ds_put_cstr(ds, "Flow: ");
6720 flow_format(ds, flow);
6721 ds_put_char(ds, '\n');
6723 rule = rule_dpif_lookup(ofproto, flow);
6725 trace_format_rule(ds, 0, 0, rule);
6726 if (rule == ofproto->miss_rule) {
6727 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
6728 } else if (rule == ofproto->no_packet_in_rule) {
6729 ds_put_cstr(ds, "\nNo match, packets dropped because "
6730 "OFPPC_NO_PACKET_IN is set on in_port.\n");
6734 uint64_t odp_actions_stub[1024 / 8];
6735 struct ofpbuf odp_actions;
6737 struct trace_ctx trace;
6740 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
6743 ofpbuf_use_stub(&odp_actions,
6744 odp_actions_stub, sizeof odp_actions_stub);
6745 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
6746 rule, tcp_flags, packet);
6747 trace.ctx.resubmit_hook = trace_resubmit;
6748 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
6751 ds_put_char(ds, '\n');
6752 trace_format_flow(ds, 0, "Final flow", &trace);
6753 ds_put_cstr(ds, "Datapath actions: ");
6754 format_odp_actions(ds, odp_actions.data, odp_actions.size);
6755 ofpbuf_uninit(&odp_actions);
6757 if (trace.ctx.slow) {
6758 enum slow_path_reason slow;
6760 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
6761 "slow path because it:");
6762 for (slow = trace.ctx.slow; slow; ) {
6763 enum slow_path_reason bit = rightmost_1bit(slow);
6767 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
6770 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
6773 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
6776 ds_put_cstr(ds, "\n\t- Needs in-band special case "
6779 ds_put_cstr(ds, "\n\t (The datapath actions are "
6780 "incomplete--for complete actions, "
6781 "please supply a packet.)");
6784 case SLOW_CONTROLLER:
6785 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
6786 "to the OpenFlow controller.");
6789 ds_put_cstr(ds, "\n\t- Needs more specific matching "
6790 "than the datapath supports.");
6797 if (slow & ~SLOW_MATCH) {
6798 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
6799 "the special slow-path processing.");
6806 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6807 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6810 unixctl_command_reply(conn, NULL);
6814 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6815 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6818 unixctl_command_reply(conn, NULL);
6821 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
6822 * 'reply' describing the results. */
6824 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
6826 struct facet *facet;
6830 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6831 if (!facet_check_consistency(facet)) {
6836 ofproto->need_revalidate = REV_INCONSISTENCY;
6840 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
6841 ofproto->up.name, errors);
6843 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
6848 ofproto_dpif_self_check(struct unixctl_conn *conn,
6849 int argc, const char *argv[], void *aux OVS_UNUSED)
6851 struct ds reply = DS_EMPTY_INITIALIZER;
6852 struct ofproto_dpif *ofproto;
6855 ofproto = ofproto_dpif_lookup(argv[1]);
6857 unixctl_command_reply_error(conn, "Unknown ofproto (use "
6858 "ofproto/list for help)");
6861 ofproto_dpif_self_check__(ofproto, &reply);
6863 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6864 ofproto_dpif_self_check__(ofproto, &reply);
6868 unixctl_command_reply(conn, ds_cstr(&reply));
6873 ofproto_dpif_unixctl_init(void)
6875 static bool registered;
6881 unixctl_command_register(
6883 "bridge {tun_id in_port packet | odp_flow [-generate]}",
6884 2, 5, ofproto_unixctl_trace, NULL);
6885 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6886 ofproto_unixctl_fdb_flush, NULL);
6887 unixctl_command_register("fdb/show", "bridge", 1, 1,
6888 ofproto_unixctl_fdb_show, NULL);
6889 unixctl_command_register("ofproto/clog", "", 0, 0,
6890 ofproto_dpif_clog, NULL);
6891 unixctl_command_register("ofproto/unclog", "", 0, 0,
6892 ofproto_dpif_unclog, NULL);
6893 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6894 ofproto_dpif_self_check, NULL);
6897 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6899 * This is deprecated. It is only for compatibility with broken device drivers
6900 * in old versions of Linux that do not properly support VLANs when VLAN
6901 * devices are not used. When broken device drivers are no longer in
6902 * widespread use, we will delete these interfaces. */
6905 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
6907 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6908 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6910 if (realdev_ofp_port == ofport->realdev_ofp_port
6911 && vid == ofport->vlandev_vid) {
6915 ofproto->need_revalidate = REV_RECONFIGURE;
6917 if (ofport->realdev_ofp_port) {
6920 if (realdev_ofp_port && ofport->bundle) {
6921 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6922 * themselves be part of a bundle. */
6923 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6926 ofport->realdev_ofp_port = realdev_ofp_port;
6927 ofport->vlandev_vid = vid;
6929 if (realdev_ofp_port) {
6930 vsp_add(ofport, realdev_ofp_port, vid);
6937 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
6939 return hash_2words(realdev_ofp_port, vid);
6942 /* Returns the ODP port number of the Linux VLAN device that corresponds to
6943 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
6944 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
6945 * it would return the port number of eth0.9.
6947 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
6948 * function just returns its 'realdev_odp_port' argument. */
6950 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6951 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
6953 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6954 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
6955 int vid = vlan_tci_to_vid(vlan_tci);
6956 const struct vlan_splinter *vsp;
6958 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6959 hash_realdev_vid(realdev_ofp_port, vid),
6960 &ofproto->realdev_vid_map) {
6961 if (vsp->realdev_ofp_port == realdev_ofp_port
6962 && vsp->vid == vid) {
6963 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
6967 return realdev_odp_port;
6970 static struct vlan_splinter *
6971 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
6973 struct vlan_splinter *vsp;
6975 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
6976 &ofproto->vlandev_map) {
6977 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6985 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6986 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6987 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6988 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6989 * eth0 and store 9 in '*vid'.
6991 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6992 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6995 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6996 uint16_t vlandev_ofp_port, int *vid)
6998 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6999 const struct vlan_splinter *vsp;
7001 vsp = vlandev_find(ofproto, vlandev_ofp_port);
7006 return vsp->realdev_ofp_port;
7012 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
7013 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
7014 * 'flow->in_port' to the "real" device backing the VLAN device, sets
7015 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
7016 * always the case unless VLAN splinters are enabled), returns false without
7017 * making any changes. */
7019 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
7024 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
7029 /* Cause the flow to be processed as if it came in on the real device with
7030 * the VLAN device's VLAN ID. */
7031 flow->in_port = realdev;
7032 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
7037 vsp_remove(struct ofport_dpif *port)
7039 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7040 struct vlan_splinter *vsp;
7042 vsp = vlandev_find(ofproto, port->up.ofp_port);
7044 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
7045 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
7048 port->realdev_ofp_port = 0;
7050 VLOG_ERR("missing vlan device record");
7055 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
7057 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7059 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
7060 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
7061 == realdev_ofp_port)) {
7062 struct vlan_splinter *vsp;
7064 vsp = xmalloc(sizeof *vsp);
7065 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
7066 hash_int(port->up.ofp_port, 0));
7067 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
7068 hash_realdev_vid(realdev_ofp_port, vid));
7069 vsp->realdev_ofp_port = realdev_ofp_port;
7070 vsp->vlandev_ofp_port = port->up.ofp_port;
7073 port->realdev_ofp_port = realdev_ofp_port;
7075 VLOG_ERR("duplicate vlan device record");
7079 const struct ofproto_class ofproto_dpif_class = {
7109 port_is_lacp_current,
7110 NULL, /* rule_choose_table */
7117 rule_modify_actions,
7125 get_cfm_remote_mpids,
7130 get_stp_port_status,
7137 is_mirror_output_bundle,
7138 forward_bpdu_changed,