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 calls this function to report some
247 * significant decision, e.g. to explain why OFPP_NORMAL translation
248 * dropped a packet. */
249 void (*report_hook)(struct action_xlate_ctx *, const char *s);
251 /* If nonnull, flow translation credits the specified statistics to each
252 * rule reached through a resubmit or OFPP_TABLE action.
254 * This is normally null so the client has to set it manually after
255 * calling action_xlate_ctx_init(). */
256 const struct dpif_flow_stats *resubmit_stats;
258 /* xlate_actions() initializes and uses these members. The client might want
259 * to look at them after it returns. */
261 struct ofpbuf *odp_actions; /* Datapath actions. */
262 tag_type tags; /* Tags associated with actions. */
263 enum slow_path_reason slow; /* 0 if fast path may be used. */
264 bool has_learn; /* Actions include NXAST_LEARN? */
265 bool has_normal; /* Actions output to OFPP_NORMAL? */
266 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
267 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
268 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
270 /* xlate_actions() initializes and uses these members, but the client has no
271 * reason to look at them. */
273 int recurse; /* Recursion level, via xlate_table_action. */
274 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
275 struct flow base_flow; /* Flow at the last commit. */
276 uint32_t orig_skb_priority; /* Priority when packet arrived. */
277 uint8_t table_id; /* OpenFlow table ID where flow was found. */
278 uint32_t sflow_n_outputs; /* Number of output ports. */
279 uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
280 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
281 bool exit; /* No further actions should be processed. */
282 struct flow orig_flow; /* Copy of original flow. */
285 static void action_xlate_ctx_init(struct action_xlate_ctx *,
286 struct ofproto_dpif *, const struct flow *,
287 ovs_be16 initial_tci, struct rule_dpif *,
288 uint8_t tcp_flags, const struct ofpbuf *);
289 static void xlate_actions(struct action_xlate_ctx *,
290 const struct ofpact *ofpacts, size_t ofpacts_len,
291 struct ofpbuf *odp_actions);
292 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
293 const struct ofpact *ofpacts,
296 static size_t put_userspace_action(const struct ofproto_dpif *,
297 struct ofpbuf *odp_actions,
299 const union user_action_cookie *);
301 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
302 enum slow_path_reason,
303 uint64_t *stub, size_t stub_size,
304 const struct nlattr **actionsp,
305 size_t *actions_lenp);
307 static void xlate_report(struct action_xlate_ctx *ctx, const char *s);
309 /* A subfacet (see "struct subfacet" below) has three possible installation
312 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
313 * case just after the subfacet is created, just before the subfacet is
314 * destroyed, or if the datapath returns an error when we try to install a
317 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
319 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
320 * ofproto_dpif is installed in the datapath.
323 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
324 SF_FAST_PATH, /* Full actions are installed. */
325 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
328 static const char *subfacet_path_to_string(enum subfacet_path);
330 /* A dpif flow and actions associated with a facet.
332 * See also the large comment on struct facet. */
335 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
336 struct list list_node; /* In struct facet's 'facets' list. */
337 struct facet *facet; /* Owning facet. */
341 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
342 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
343 * regenerate the ODP flow key from ->facet->flow. */
344 enum odp_key_fitness key_fitness;
348 long long int used; /* Time last used; time created if not used. */
350 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
351 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
355 * These should be essentially identical for every subfacet in a facet, but
356 * may differ in trivial ways due to VLAN splinters. */
357 size_t actions_len; /* Number of bytes in actions[]. */
358 struct nlattr *actions; /* Datapath actions. */
360 enum slow_path_reason slow; /* 0 if fast path may be used. */
361 enum subfacet_path path; /* Installed in datapath? */
363 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
364 * splinters can cause it to differ. This value should be removed when
365 * the VLAN splinters feature is no longer needed. */
366 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
369 static struct subfacet *subfacet_create(struct facet *, enum odp_key_fitness,
370 const struct nlattr *key,
371 size_t key_len, ovs_be16 initial_tci);
372 static struct subfacet *subfacet_find(struct ofproto_dpif *,
373 const struct nlattr *key, size_t key_len);
374 static void subfacet_destroy(struct subfacet *);
375 static void subfacet_destroy__(struct subfacet *);
376 static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *,
378 static void subfacet_reset_dp_stats(struct subfacet *,
379 struct dpif_flow_stats *);
380 static void subfacet_update_time(struct subfacet *, long long int used);
381 static void subfacet_update_stats(struct subfacet *,
382 const struct dpif_flow_stats *);
383 static void subfacet_make_actions(struct subfacet *,
384 const struct ofpbuf *packet,
385 struct ofpbuf *odp_actions);
386 static int subfacet_install(struct subfacet *,
387 const struct nlattr *actions, size_t actions_len,
388 struct dpif_flow_stats *, enum slow_path_reason);
389 static void subfacet_uninstall(struct subfacet *);
391 static enum subfacet_path subfacet_want_path(enum slow_path_reason);
393 /* An exact-match instantiation of an OpenFlow flow.
395 * A facet associates a "struct flow", which represents the Open vSwitch
396 * userspace idea of an exact-match flow, with one or more subfacets. Each
397 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
398 * the facet. When the kernel module (or other dpif implementation) and Open
399 * vSwitch userspace agree on the definition of a flow key, there is exactly
400 * one subfacet per facet. If the dpif implementation supports more-specific
401 * flow matching than userspace, however, a facet can have more than one
402 * subfacet, each of which corresponds to some distinction in flow that
403 * userspace simply doesn't understand.
405 * Flow expiration works in terms of subfacets, so a facet must have at least
406 * one subfacet or it will never expire, leaking memory. */
409 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
410 struct list list_node; /* In owning rule's 'facets' list. */
411 struct rule_dpif *rule; /* Owning rule. */
414 struct list subfacets;
415 long long int used; /* Time last used; time created if not used. */
422 * - Do include packets and bytes sent "by hand", e.g. with
425 * - Do include packets and bytes that were obtained from the datapath
426 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
427 * DPIF_FP_ZERO_STATS).
429 * - Do not include packets or bytes that can be obtained from the
430 * datapath for any existing subfacet.
432 uint64_t packet_count; /* Number of packets received. */
433 uint64_t byte_count; /* Number of bytes received. */
435 /* Resubmit statistics. */
436 uint64_t prev_packet_count; /* Number of packets from last stats push. */
437 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
438 long long int prev_used; /* Used time from last stats push. */
441 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
442 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
443 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
445 /* Properties of datapath actions.
447 * Every subfacet has its own actions because actions can differ slightly
448 * between splintered and non-splintered subfacets due to the VLAN tag
449 * being initially different (present vs. absent). All of them have these
450 * properties in common so we just store one copy of them here. */
451 bool has_learn; /* Actions include NXAST_LEARN? */
452 bool has_normal; /* Actions output to OFPP_NORMAL? */
453 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
454 tag_type tags; /* Tags that would require revalidation. */
455 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
457 /* Storage for a single subfacet, to reduce malloc() time and space
458 * overhead. (A facet always has at least one subfacet and in the common
459 * case has exactly one subfacet.) */
460 struct subfacet one_subfacet;
463 static struct facet *facet_create(struct rule_dpif *,
464 const struct flow *, uint32_t hash);
465 static void facet_remove(struct facet *);
466 static void facet_free(struct facet *);
468 static struct facet *facet_find(struct ofproto_dpif *,
469 const struct flow *, uint32_t hash);
470 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
471 const struct flow *, uint32_t hash);
472 static void facet_revalidate(struct facet *);
473 static bool facet_check_consistency(struct facet *);
475 static void facet_flush_stats(struct facet *);
477 static void facet_update_time(struct facet *, long long int used);
478 static void facet_reset_counters(struct facet *);
479 static void facet_push_stats(struct facet *);
480 static void facet_learn(struct facet *);
481 static void facet_account(struct facet *);
483 static bool facet_is_controller_flow(struct facet *);
489 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
490 struct list bundle_node; /* In struct ofbundle's "ports" list. */
491 struct cfm *cfm; /* Connectivity Fault Management, if any. */
492 tag_type tag; /* Tag associated with this port. */
493 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
494 bool may_enable; /* May be enabled in bonds. */
495 long long int carrier_seq; /* Carrier status changes. */
498 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
499 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
500 long long int stp_state_entered;
502 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
504 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
506 * This is deprecated. It is only for compatibility with broken device
507 * drivers in old versions of Linux that do not properly support VLANs when
508 * VLAN devices are not used. When broken device drivers are no longer in
509 * widespread use, we will delete these interfaces. */
510 uint16_t realdev_ofp_port;
514 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
515 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
516 * traffic egressing the 'ofport' with that priority should be marked with. */
517 struct priority_to_dscp {
518 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
519 uint32_t priority; /* Priority of this queue (see struct flow). */
521 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
524 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
526 * This is deprecated. It is only for compatibility with broken device drivers
527 * in old versions of Linux that do not properly support VLANs when VLAN
528 * devices are not used. When broken device drivers are no longer in
529 * widespread use, we will delete these interfaces. */
530 struct vlan_splinter {
531 struct hmap_node realdev_vid_node;
532 struct hmap_node vlandev_node;
533 uint16_t realdev_ofp_port;
534 uint16_t vlandev_ofp_port;
538 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
539 uint32_t realdev, ovs_be16 vlan_tci);
540 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
541 static void vsp_remove(struct ofport_dpif *);
542 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
544 static struct ofport_dpif *
545 ofport_dpif_cast(const struct ofport *ofport)
547 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
548 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
551 static void port_run(struct ofport_dpif *);
552 static void port_run_fast(struct ofport_dpif *);
553 static void port_wait(struct ofport_dpif *);
554 static int set_cfm(struct ofport *, const struct cfm_settings *);
555 static void ofport_clear_priorities(struct ofport_dpif *);
557 struct dpif_completion {
558 struct list list_node;
559 struct ofoperation *op;
562 /* Extra information about a classifier table.
563 * Currently used just for optimized flow revalidation. */
565 /* If either of these is nonnull, then this table has a form that allows
566 * flows to be tagged to avoid revalidating most flows for the most common
567 * kinds of flow table changes. */
568 struct cls_table *catchall_table; /* Table that wildcards all fields. */
569 struct cls_table *other_table; /* Table with any other wildcard set. */
570 uint32_t basis; /* Keeps each table's tags separate. */
573 /* Reasons that we might need to revalidate every facet, and corresponding
576 * A value of 0 means that there is no need to revalidate.
578 * It would be nice to have some cleaner way to integrate with coverage
579 * counters, but with only a few reasons I guess this is good enough for
581 enum revalidate_reason {
582 REV_RECONFIGURE = 1, /* Switch configuration changed. */
583 REV_STP, /* Spanning tree protocol port status change. */
584 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
585 REV_FLOW_TABLE, /* Flow table changed. */
586 REV_INCONSISTENCY /* Facet self-check failed. */
588 COVERAGE_DEFINE(rev_reconfigure);
589 COVERAGE_DEFINE(rev_stp);
590 COVERAGE_DEFINE(rev_port_toggled);
591 COVERAGE_DEFINE(rev_flow_table);
592 COVERAGE_DEFINE(rev_inconsistency);
594 struct ofproto_dpif {
595 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
600 /* Special OpenFlow rules. */
601 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
602 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
608 struct netflow *netflow;
609 struct dpif_sflow *sflow;
610 struct hmap bundles; /* Contains "struct ofbundle"s. */
611 struct mac_learning *ml;
612 struct ofmirror *mirrors[MAX_MIRRORS];
614 bool has_bonded_bundles;
617 struct timer next_expiration;
621 struct hmap subfacets;
622 struct governor *governor;
625 struct table_dpif tables[N_TABLES];
626 enum revalidate_reason need_revalidate;
627 struct tag_set revalidate_set;
629 /* Support for debugging async flow mods. */
630 struct list completions;
632 bool has_bundle_action; /* True when the first bundle action appears. */
633 struct netdev_stats stats; /* To account packets generated and consumed in
638 long long int stp_last_tick;
640 /* VLAN splinters. */
641 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
642 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
645 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
646 * for debugging the asynchronous flow_mod implementation.) */
649 /* All existing ofproto_dpif instances, indexed by ->up.name. */
650 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
652 static void ofproto_dpif_unixctl_init(void);
654 static struct ofproto_dpif *
655 ofproto_dpif_cast(const struct ofproto *ofproto)
657 assert(ofproto->ofproto_class == &ofproto_dpif_class);
658 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
661 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
663 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
665 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
666 const struct ofpbuf *, ovs_be16 initial_tci,
669 /* Packet processing. */
670 static void update_learning_table(struct ofproto_dpif *,
671 const struct flow *, int vlan,
674 #define FLOW_MISS_MAX_BATCH 50
675 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
677 /* Flow expiration. */
678 static int expire(struct ofproto_dpif *);
681 static void send_netflow_active_timeouts(struct ofproto_dpif *);
684 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
685 static size_t compose_sflow_action(const struct ofproto_dpif *,
686 struct ofpbuf *odp_actions,
687 const struct flow *, uint32_t odp_port);
688 static void add_mirror_actions(struct action_xlate_ctx *ctx,
689 const struct flow *flow);
690 /* Global variables. */
691 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
693 /* Factory functions. */
696 enumerate_types(struct sset *types)
698 dp_enumerate_types(types);
702 enumerate_names(const char *type, struct sset *names)
704 return dp_enumerate_names(type, names);
708 del(const char *type, const char *name)
713 error = dpif_open(name, type, &dpif);
715 error = dpif_delete(dpif);
721 /* Basic life-cycle. */
723 static int add_internal_flows(struct ofproto_dpif *);
725 static struct ofproto *
728 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
733 dealloc(struct ofproto *ofproto_)
735 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
740 construct(struct ofproto *ofproto_)
742 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
743 const char *name = ofproto->up.name;
747 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
749 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
753 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
754 ofproto->n_matches = 0;
756 dpif_flow_flush(ofproto->dpif);
757 dpif_recv_purge(ofproto->dpif);
759 error = dpif_recv_set(ofproto->dpif, true);
761 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
762 dpif_close(ofproto->dpif);
766 ofproto->netflow = NULL;
767 ofproto->sflow = NULL;
769 hmap_init(&ofproto->bundles);
770 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
771 for (i = 0; i < MAX_MIRRORS; i++) {
772 ofproto->mirrors[i] = NULL;
774 ofproto->has_bonded_bundles = false;
776 timer_set_duration(&ofproto->next_expiration, 1000);
778 hmap_init(&ofproto->facets);
779 hmap_init(&ofproto->subfacets);
780 ofproto->governor = NULL;
782 for (i = 0; i < N_TABLES; i++) {
783 struct table_dpif *table = &ofproto->tables[i];
785 table->catchall_table = NULL;
786 table->other_table = NULL;
787 table->basis = random_uint32();
789 ofproto->need_revalidate = 0;
790 tag_set_init(&ofproto->revalidate_set);
792 list_init(&ofproto->completions);
794 ofproto_dpif_unixctl_init();
796 ofproto->has_mirrors = false;
797 ofproto->has_bundle_action = false;
799 hmap_init(&ofproto->vlandev_map);
800 hmap_init(&ofproto->realdev_vid_map);
802 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
803 hash_string(ofproto->up.name, 0));
804 memset(&ofproto->stats, 0, sizeof ofproto->stats);
806 ofproto_init_tables(ofproto_, N_TABLES);
807 error = add_internal_flows(ofproto);
808 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
814 add_internal_flow(struct ofproto_dpif *ofproto, int id,
815 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
817 struct ofputil_flow_mod fm;
820 cls_rule_init_catchall(&fm.cr, 0);
821 cls_rule_set_reg(&fm.cr, 0, id);
822 fm.new_cookie = htonll(0);
823 fm.cookie = htonll(0);
824 fm.cookie_mask = htonll(0);
825 fm.table_id = TBL_INTERNAL;
826 fm.command = OFPFC_ADD;
832 fm.ofpacts = ofpacts->data;
833 fm.ofpacts_len = ofpacts->size;
835 error = ofproto_flow_mod(&ofproto->up, &fm);
837 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
838 id, ofperr_to_string(error));
842 *rulep = rule_dpif_lookup__(ofproto, &fm.cr.flow, TBL_INTERNAL);
843 assert(*rulep != NULL);
849 add_internal_flows(struct ofproto_dpif *ofproto)
851 struct ofpact_controller *controller;
852 uint64_t ofpacts_stub[128 / 8];
853 struct ofpbuf ofpacts;
857 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
860 controller = ofpact_put_CONTROLLER(&ofpacts);
861 controller->max_len = UINT16_MAX;
862 controller->controller_id = 0;
863 controller->reason = OFPR_NO_MATCH;
864 ofpact_pad(&ofpacts);
866 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
871 ofpbuf_clear(&ofpacts);
872 error = add_internal_flow(ofproto, id++, &ofpacts,
873 &ofproto->no_packet_in_rule);
878 complete_operations(struct ofproto_dpif *ofproto)
880 struct dpif_completion *c, *next;
882 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
883 ofoperation_complete(c->op, 0);
884 list_remove(&c->list_node);
890 destruct(struct ofproto *ofproto_)
892 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
893 struct rule_dpif *rule, *next_rule;
894 struct oftable *table;
897 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
898 complete_operations(ofproto);
900 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
901 struct cls_cursor cursor;
903 cls_cursor_init(&cursor, &table->cls, NULL);
904 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
905 ofproto_rule_destroy(&rule->up);
909 for (i = 0; i < MAX_MIRRORS; i++) {
910 mirror_destroy(ofproto->mirrors[i]);
913 netflow_destroy(ofproto->netflow);
914 dpif_sflow_destroy(ofproto->sflow);
915 hmap_destroy(&ofproto->bundles);
916 mac_learning_destroy(ofproto->ml);
918 hmap_destroy(&ofproto->facets);
919 hmap_destroy(&ofproto->subfacets);
920 governor_destroy(ofproto->governor);
922 hmap_destroy(&ofproto->vlandev_map);
923 hmap_destroy(&ofproto->realdev_vid_map);
925 dpif_close(ofproto->dpif);
929 run_fast(struct ofproto *ofproto_)
931 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
932 struct ofport_dpif *ofport;
935 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
936 port_run_fast(ofport);
939 /* Handle one or more batches of upcalls, until there's nothing left to do
940 * or until we do a fixed total amount of work.
942 * We do work in batches because it can be much cheaper to set up a number
943 * of flows and fire off their patches all at once. We do multiple batches
944 * because in some cases handling a packet can cause another packet to be
945 * queued almost immediately as part of the return flow. Both
946 * optimizations can make major improvements on some benchmarks and
947 * presumably for real traffic as well. */
949 while (work < FLOW_MISS_MAX_BATCH) {
950 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
960 run(struct ofproto *ofproto_)
962 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
963 struct ofport_dpif *ofport;
964 struct ofbundle *bundle;
968 complete_operations(ofproto);
970 dpif_run(ofproto->dpif);
972 error = run_fast(ofproto_);
977 if (timer_expired(&ofproto->next_expiration)) {
978 int delay = expire(ofproto);
979 timer_set_duration(&ofproto->next_expiration, delay);
982 if (ofproto->netflow) {
983 if (netflow_run(ofproto->netflow)) {
984 send_netflow_active_timeouts(ofproto);
987 if (ofproto->sflow) {
988 dpif_sflow_run(ofproto->sflow);
991 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
994 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
999 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
1001 /* Now revalidate if there's anything to do. */
1002 if (ofproto->need_revalidate
1003 || !tag_set_is_empty(&ofproto->revalidate_set)) {
1004 struct tag_set revalidate_set = ofproto->revalidate_set;
1005 bool revalidate_all = ofproto->need_revalidate;
1006 struct facet *facet;
1008 switch (ofproto->need_revalidate) {
1009 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
1010 case REV_STP: COVERAGE_INC(rev_stp); break;
1011 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
1012 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
1013 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
1016 /* Clear the revalidation flags. */
1017 tag_set_init(&ofproto->revalidate_set);
1018 ofproto->need_revalidate = 0;
1020 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
1022 || tag_set_intersects(&revalidate_set, facet->tags)) {
1023 facet_revalidate(facet);
1028 /* Check the consistency of a random facet, to aid debugging. */
1029 if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) {
1030 struct facet *facet;
1032 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1033 struct facet, hmap_node);
1034 if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) {
1035 if (!facet_check_consistency(facet)) {
1036 ofproto->need_revalidate = REV_INCONSISTENCY;
1041 if (ofproto->governor) {
1044 governor_run(ofproto->governor);
1046 /* If the governor has shrunk to its minimum size and the number of
1047 * subfacets has dwindled, then drop the governor entirely.
1049 * For hysteresis, the number of subfacets to drop the governor is
1050 * smaller than the number needed to trigger its creation. */
1051 n_subfacets = hmap_count(&ofproto->subfacets);
1052 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1053 && governor_is_idle(ofproto->governor)) {
1054 governor_destroy(ofproto->governor);
1055 ofproto->governor = NULL;
1063 wait(struct ofproto *ofproto_)
1065 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1066 struct ofport_dpif *ofport;
1067 struct ofbundle *bundle;
1069 if (!clogged && !list_is_empty(&ofproto->completions)) {
1070 poll_immediate_wake();
1073 dpif_wait(ofproto->dpif);
1074 dpif_recv_wait(ofproto->dpif);
1075 if (ofproto->sflow) {
1076 dpif_sflow_wait(ofproto->sflow);
1078 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
1079 poll_immediate_wake();
1081 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1084 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1085 bundle_wait(bundle);
1087 if (ofproto->netflow) {
1088 netflow_wait(ofproto->netflow);
1090 mac_learning_wait(ofproto->ml);
1092 if (ofproto->need_revalidate) {
1093 /* Shouldn't happen, but if it does just go around again. */
1094 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1095 poll_immediate_wake();
1097 timer_wait(&ofproto->next_expiration);
1099 if (ofproto->governor) {
1100 governor_wait(ofproto->governor);
1105 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1107 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1109 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1110 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1114 flush(struct ofproto *ofproto_)
1116 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1117 struct facet *facet, *next_facet;
1119 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1120 /* Mark the facet as not installed so that facet_remove() doesn't
1121 * bother trying to uninstall it. There is no point in uninstalling it
1122 * individually since we are about to blow away all the facets with
1123 * dpif_flow_flush(). */
1124 struct subfacet *subfacet;
1126 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
1127 subfacet->path = SF_NOT_INSTALLED;
1128 subfacet->dp_packet_count = 0;
1129 subfacet->dp_byte_count = 0;
1131 facet_remove(facet);
1133 dpif_flow_flush(ofproto->dpif);
1137 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1138 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1140 *arp_match_ip = true;
1141 *actions = (OFPUTIL_A_OUTPUT |
1142 OFPUTIL_A_SET_VLAN_VID |
1143 OFPUTIL_A_SET_VLAN_PCP |
1144 OFPUTIL_A_STRIP_VLAN |
1145 OFPUTIL_A_SET_DL_SRC |
1146 OFPUTIL_A_SET_DL_DST |
1147 OFPUTIL_A_SET_NW_SRC |
1148 OFPUTIL_A_SET_NW_DST |
1149 OFPUTIL_A_SET_NW_TOS |
1150 OFPUTIL_A_SET_TP_SRC |
1151 OFPUTIL_A_SET_TP_DST |
1156 get_tables(struct ofproto *ofproto_, struct ofp10_table_stats *ots)
1158 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1159 struct dpif_dp_stats s;
1161 strcpy(ots->name, "classifier");
1163 dpif_get_dp_stats(ofproto->dpif, &s);
1164 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
1165 put_32aligned_be64(&ots->matched_count,
1166 htonll(s.n_hit + ofproto->n_matches));
1169 static struct ofport *
1172 struct ofport_dpif *port = xmalloc(sizeof *port);
1177 port_dealloc(struct ofport *port_)
1179 struct ofport_dpif *port = ofport_dpif_cast(port_);
1184 port_construct(struct ofport *port_)
1186 struct ofport_dpif *port = ofport_dpif_cast(port_);
1187 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1189 ofproto->need_revalidate = REV_RECONFIGURE;
1190 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
1191 port->bundle = NULL;
1193 port->tag = tag_create_random();
1194 port->may_enable = true;
1195 port->stp_port = NULL;
1196 port->stp_state = STP_DISABLED;
1197 hmap_init(&port->priorities);
1198 port->realdev_ofp_port = 0;
1199 port->vlandev_vid = 0;
1200 port->carrier_seq = netdev_get_carrier_resets(port->up.netdev);
1202 if (ofproto->sflow) {
1203 dpif_sflow_add_port(ofproto->sflow, port_);
1210 port_destruct(struct ofport *port_)
1212 struct ofport_dpif *port = ofport_dpif_cast(port_);
1213 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1215 ofproto->need_revalidate = REV_RECONFIGURE;
1216 bundle_remove(port_);
1217 set_cfm(port_, NULL);
1218 if (ofproto->sflow) {
1219 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1222 ofport_clear_priorities(port);
1223 hmap_destroy(&port->priorities);
1227 port_modified(struct ofport *port_)
1229 struct ofport_dpif *port = ofport_dpif_cast(port_);
1231 if (port->bundle && port->bundle->bond) {
1232 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1237 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1239 struct ofport_dpif *port = ofport_dpif_cast(port_);
1240 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1241 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1243 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1244 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1245 OFPUTIL_PC_NO_PACKET_IN)) {
1246 ofproto->need_revalidate = REV_RECONFIGURE;
1248 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1249 bundle_update(port->bundle);
1255 set_sflow(struct ofproto *ofproto_,
1256 const struct ofproto_sflow_options *sflow_options)
1258 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1259 struct dpif_sflow *ds = ofproto->sflow;
1261 if (sflow_options) {
1263 struct ofport_dpif *ofport;
1265 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1266 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1267 dpif_sflow_add_port(ds, &ofport->up);
1269 ofproto->need_revalidate = REV_RECONFIGURE;
1271 dpif_sflow_set_options(ds, sflow_options);
1274 dpif_sflow_destroy(ds);
1275 ofproto->need_revalidate = REV_RECONFIGURE;
1276 ofproto->sflow = NULL;
1283 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1285 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1292 struct ofproto_dpif *ofproto;
1294 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1295 ofproto->need_revalidate = REV_RECONFIGURE;
1296 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1299 if (cfm_configure(ofport->cfm, s)) {
1305 cfm_destroy(ofport->cfm);
1311 get_cfm_fault(const struct ofport *ofport_)
1313 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1315 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1319 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1322 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1325 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1333 get_cfm_health(const struct ofport *ofport_)
1335 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1337 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1340 /* Spanning Tree. */
1343 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1345 struct ofproto_dpif *ofproto = ofproto_;
1346 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1347 struct ofport_dpif *ofport;
1349 ofport = stp_port_get_aux(sp);
1351 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1352 ofproto->up.name, port_num);
1354 struct eth_header *eth = pkt->l2;
1356 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1357 if (eth_addr_is_zero(eth->eth_src)) {
1358 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1359 "with unknown MAC", ofproto->up.name, port_num);
1361 send_packet(ofport, pkt);
1367 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1369 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1371 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1373 /* Only revalidate flows if the configuration changed. */
1374 if (!s != !ofproto->stp) {
1375 ofproto->need_revalidate = REV_RECONFIGURE;
1379 if (!ofproto->stp) {
1380 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1381 send_bpdu_cb, ofproto);
1382 ofproto->stp_last_tick = time_msec();
1385 stp_set_bridge_id(ofproto->stp, s->system_id);
1386 stp_set_bridge_priority(ofproto->stp, s->priority);
1387 stp_set_hello_time(ofproto->stp, s->hello_time);
1388 stp_set_max_age(ofproto->stp, s->max_age);
1389 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1391 struct ofport *ofport;
1393 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1394 set_stp_port(ofport, NULL);
1397 stp_destroy(ofproto->stp);
1398 ofproto->stp = NULL;
1405 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1407 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1411 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1412 s->designated_root = stp_get_designated_root(ofproto->stp);
1413 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1422 update_stp_port_state(struct ofport_dpif *ofport)
1424 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1425 enum stp_state state;
1427 /* Figure out new state. */
1428 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1432 if (ofport->stp_state != state) {
1433 enum ofputil_port_state of_state;
1436 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1437 netdev_get_name(ofport->up.netdev),
1438 stp_state_name(ofport->stp_state),
1439 stp_state_name(state));
1440 if (stp_learn_in_state(ofport->stp_state)
1441 != stp_learn_in_state(state)) {
1442 /* xxx Learning action flows should also be flushed. */
1443 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1445 fwd_change = stp_forward_in_state(ofport->stp_state)
1446 != stp_forward_in_state(state);
1448 ofproto->need_revalidate = REV_STP;
1449 ofport->stp_state = state;
1450 ofport->stp_state_entered = time_msec();
1452 if (fwd_change && ofport->bundle) {
1453 bundle_update(ofport->bundle);
1456 /* Update the STP state bits in the OpenFlow port description. */
1457 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1458 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1459 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1460 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1461 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1463 ofproto_port_set_state(&ofport->up, of_state);
1467 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1468 * caller is responsible for assigning STP port numbers and ensuring
1469 * there are no duplicates. */
1471 set_stp_port(struct ofport *ofport_,
1472 const struct ofproto_port_stp_settings *s)
1474 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1475 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1476 struct stp_port *sp = ofport->stp_port;
1478 if (!s || !s->enable) {
1480 ofport->stp_port = NULL;
1481 stp_port_disable(sp);
1482 update_stp_port_state(ofport);
1485 } else if (sp && stp_port_no(sp) != s->port_num
1486 && ofport == stp_port_get_aux(sp)) {
1487 /* The port-id changed, so disable the old one if it's not
1488 * already in use by another port. */
1489 stp_port_disable(sp);
1492 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1493 stp_port_enable(sp);
1495 stp_port_set_aux(sp, ofport);
1496 stp_port_set_priority(sp, s->priority);
1497 stp_port_set_path_cost(sp, s->path_cost);
1499 update_stp_port_state(ofport);
1505 get_stp_port_status(struct ofport *ofport_,
1506 struct ofproto_port_stp_status *s)
1508 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1509 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1510 struct stp_port *sp = ofport->stp_port;
1512 if (!ofproto->stp || !sp) {
1518 s->port_id = stp_port_get_id(sp);
1519 s->state = stp_port_get_state(sp);
1520 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1521 s->role = stp_port_get_role(sp);
1522 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1528 stp_run(struct ofproto_dpif *ofproto)
1531 long long int now = time_msec();
1532 long long int elapsed = now - ofproto->stp_last_tick;
1533 struct stp_port *sp;
1536 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1537 ofproto->stp_last_tick = now;
1539 while (stp_get_changed_port(ofproto->stp, &sp)) {
1540 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1543 update_stp_port_state(ofport);
1547 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1548 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1554 stp_wait(struct ofproto_dpif *ofproto)
1557 poll_timer_wait(1000);
1561 /* Returns true if STP should process 'flow'. */
1563 stp_should_process_flow(const struct flow *flow)
1565 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1569 stp_process_packet(const struct ofport_dpif *ofport,
1570 const struct ofpbuf *packet)
1572 struct ofpbuf payload = *packet;
1573 struct eth_header *eth = payload.data;
1574 struct stp_port *sp = ofport->stp_port;
1576 /* Sink packets on ports that have STP disabled when the bridge has
1578 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1582 /* Trim off padding on payload. */
1583 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1584 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1587 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1588 stp_received_bpdu(sp, payload.data, payload.size);
1592 static struct priority_to_dscp *
1593 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1595 struct priority_to_dscp *pdscp;
1598 hash = hash_int(priority, 0);
1599 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1600 if (pdscp->priority == priority) {
1608 ofport_clear_priorities(struct ofport_dpif *ofport)
1610 struct priority_to_dscp *pdscp, *next;
1612 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1613 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1619 set_queues(struct ofport *ofport_,
1620 const struct ofproto_port_queue *qdscp_list,
1623 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1624 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1625 struct hmap new = HMAP_INITIALIZER(&new);
1628 for (i = 0; i < n_qdscp; i++) {
1629 struct priority_to_dscp *pdscp;
1633 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1634 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1639 pdscp = get_priority(ofport, priority);
1641 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1643 pdscp = xmalloc(sizeof *pdscp);
1644 pdscp->priority = priority;
1646 ofproto->need_revalidate = REV_RECONFIGURE;
1649 if (pdscp->dscp != dscp) {
1651 ofproto->need_revalidate = REV_RECONFIGURE;
1654 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1657 if (!hmap_is_empty(&ofport->priorities)) {
1658 ofport_clear_priorities(ofport);
1659 ofproto->need_revalidate = REV_RECONFIGURE;
1662 hmap_swap(&new, &ofport->priorities);
1670 /* Expires all MAC learning entries associated with 'bundle' and forces its
1671 * ofproto to revalidate every flow.
1673 * Normally MAC learning entries are removed only from the ofproto associated
1674 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1675 * are removed from every ofproto. When patch ports and SLB bonds are in use
1676 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1677 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1678 * with the host from which it migrated. */
1680 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1682 struct ofproto_dpif *ofproto = bundle->ofproto;
1683 struct mac_learning *ml = ofproto->ml;
1684 struct mac_entry *mac, *next_mac;
1686 ofproto->need_revalidate = REV_RECONFIGURE;
1687 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1688 if (mac->port.p == bundle) {
1690 struct ofproto_dpif *o;
1692 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1694 struct mac_entry *e;
1696 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1699 tag_set_add(&o->revalidate_set, e->tag);
1700 mac_learning_expire(o->ml, e);
1706 mac_learning_expire(ml, mac);
1711 static struct ofbundle *
1712 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1714 struct ofbundle *bundle;
1716 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1717 &ofproto->bundles) {
1718 if (bundle->aux == aux) {
1725 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1726 * ones that are found to 'bundles'. */
1728 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1729 void **auxes, size_t n_auxes,
1730 struct hmapx *bundles)
1734 hmapx_init(bundles);
1735 for (i = 0; i < n_auxes; i++) {
1736 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1738 hmapx_add(bundles, bundle);
1744 bundle_update(struct ofbundle *bundle)
1746 struct ofport_dpif *port;
1748 bundle->floodable = true;
1749 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1750 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1751 || !stp_forward_in_state(port->stp_state)) {
1752 bundle->floodable = false;
1759 bundle_del_port(struct ofport_dpif *port)
1761 struct ofbundle *bundle = port->bundle;
1763 bundle->ofproto->need_revalidate = REV_RECONFIGURE;
1765 list_remove(&port->bundle_node);
1766 port->bundle = NULL;
1769 lacp_slave_unregister(bundle->lacp, port);
1772 bond_slave_unregister(bundle->bond, port);
1775 bundle_update(bundle);
1779 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1780 struct lacp_slave_settings *lacp,
1781 uint32_t bond_stable_id)
1783 struct ofport_dpif *port;
1785 port = get_ofp_port(bundle->ofproto, ofp_port);
1790 if (port->bundle != bundle) {
1791 bundle->ofproto->need_revalidate = REV_RECONFIGURE;
1793 bundle_del_port(port);
1796 port->bundle = bundle;
1797 list_push_back(&bundle->ports, &port->bundle_node);
1798 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1799 || !stp_forward_in_state(port->stp_state)) {
1800 bundle->floodable = false;
1804 port->bundle->ofproto->need_revalidate = REV_RECONFIGURE;
1805 lacp_slave_register(bundle->lacp, port, lacp);
1808 port->bond_stable_id = bond_stable_id;
1814 bundle_destroy(struct ofbundle *bundle)
1816 struct ofproto_dpif *ofproto;
1817 struct ofport_dpif *port, *next_port;
1824 ofproto = bundle->ofproto;
1825 for (i = 0; i < MAX_MIRRORS; i++) {
1826 struct ofmirror *m = ofproto->mirrors[i];
1828 if (m->out == bundle) {
1830 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1831 || hmapx_find_and_delete(&m->dsts, bundle)) {
1832 ofproto->need_revalidate = REV_RECONFIGURE;
1837 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1838 bundle_del_port(port);
1841 bundle_flush_macs(bundle, true);
1842 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1844 free(bundle->trunks);
1845 lacp_destroy(bundle->lacp);
1846 bond_destroy(bundle->bond);
1851 bundle_set(struct ofproto *ofproto_, void *aux,
1852 const struct ofproto_bundle_settings *s)
1854 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1855 bool need_flush = false;
1856 struct ofport_dpif *port;
1857 struct ofbundle *bundle;
1858 unsigned long *trunks;
1864 bundle_destroy(bundle_lookup(ofproto, aux));
1868 assert(s->n_slaves == 1 || s->bond != NULL);
1869 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1871 bundle = bundle_lookup(ofproto, aux);
1873 bundle = xmalloc(sizeof *bundle);
1875 bundle->ofproto = ofproto;
1876 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1877 hash_pointer(aux, 0));
1879 bundle->name = NULL;
1881 list_init(&bundle->ports);
1882 bundle->vlan_mode = PORT_VLAN_TRUNK;
1884 bundle->trunks = NULL;
1885 bundle->use_priority_tags = s->use_priority_tags;
1886 bundle->lacp = NULL;
1887 bundle->bond = NULL;
1889 bundle->floodable = true;
1891 bundle->src_mirrors = 0;
1892 bundle->dst_mirrors = 0;
1893 bundle->mirror_out = 0;
1896 if (!bundle->name || strcmp(s->name, bundle->name)) {
1898 bundle->name = xstrdup(s->name);
1903 if (!bundle->lacp) {
1904 ofproto->need_revalidate = REV_RECONFIGURE;
1905 bundle->lacp = lacp_create();
1907 lacp_configure(bundle->lacp, s->lacp);
1909 lacp_destroy(bundle->lacp);
1910 bundle->lacp = NULL;
1913 /* Update set of ports. */
1915 for (i = 0; i < s->n_slaves; i++) {
1916 if (!bundle_add_port(bundle, s->slaves[i],
1917 s->lacp ? &s->lacp_slaves[i] : NULL,
1918 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1922 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1923 struct ofport_dpif *next_port;
1925 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1926 for (i = 0; i < s->n_slaves; i++) {
1927 if (s->slaves[i] == port->up.ofp_port) {
1932 bundle_del_port(port);
1936 assert(list_size(&bundle->ports) <= s->n_slaves);
1938 if (list_is_empty(&bundle->ports)) {
1939 bundle_destroy(bundle);
1943 /* Set VLAN tagging mode */
1944 if (s->vlan_mode != bundle->vlan_mode
1945 || s->use_priority_tags != bundle->use_priority_tags) {
1946 bundle->vlan_mode = s->vlan_mode;
1947 bundle->use_priority_tags = s->use_priority_tags;
1952 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1953 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1955 if (vlan != bundle->vlan) {
1956 bundle->vlan = vlan;
1960 /* Get trunked VLANs. */
1961 switch (s->vlan_mode) {
1962 case PORT_VLAN_ACCESS:
1966 case PORT_VLAN_TRUNK:
1967 trunks = CONST_CAST(unsigned long *, s->trunks);
1970 case PORT_VLAN_NATIVE_UNTAGGED:
1971 case PORT_VLAN_NATIVE_TAGGED:
1972 if (vlan != 0 && (!s->trunks
1973 || !bitmap_is_set(s->trunks, vlan)
1974 || bitmap_is_set(s->trunks, 0))) {
1975 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1977 trunks = bitmap_clone(s->trunks, 4096);
1979 trunks = bitmap_allocate1(4096);
1981 bitmap_set1(trunks, vlan);
1982 bitmap_set0(trunks, 0);
1984 trunks = CONST_CAST(unsigned long *, s->trunks);
1991 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1992 free(bundle->trunks);
1993 if (trunks == s->trunks) {
1994 bundle->trunks = vlan_bitmap_clone(trunks);
1996 bundle->trunks = trunks;
2001 if (trunks != s->trunks) {
2006 if (!list_is_short(&bundle->ports)) {
2007 bundle->ofproto->has_bonded_bundles = true;
2009 if (bond_reconfigure(bundle->bond, s->bond)) {
2010 ofproto->need_revalidate = REV_RECONFIGURE;
2013 bundle->bond = bond_create(s->bond);
2014 ofproto->need_revalidate = REV_RECONFIGURE;
2017 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2018 bond_slave_register(bundle->bond, port, port->bond_stable_id,
2022 bond_destroy(bundle->bond);
2023 bundle->bond = NULL;
2026 /* If we changed something that would affect MAC learning, un-learn
2027 * everything on this port and force flow revalidation. */
2029 bundle_flush_macs(bundle, false);
2036 bundle_remove(struct ofport *port_)
2038 struct ofport_dpif *port = ofport_dpif_cast(port_);
2039 struct ofbundle *bundle = port->bundle;
2042 bundle_del_port(port);
2043 if (list_is_empty(&bundle->ports)) {
2044 bundle_destroy(bundle);
2045 } else if (list_is_short(&bundle->ports)) {
2046 bond_destroy(bundle->bond);
2047 bundle->bond = NULL;
2053 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2055 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2056 struct ofport_dpif *port = port_;
2057 uint8_t ea[ETH_ADDR_LEN];
2060 error = netdev_get_etheraddr(port->up.netdev, ea);
2062 struct ofpbuf packet;
2065 ofpbuf_init(&packet, 0);
2066 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2068 memcpy(packet_pdu, pdu, pdu_size);
2070 send_packet(port, &packet);
2071 ofpbuf_uninit(&packet);
2073 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2074 "%s (%s)", port->bundle->name,
2075 netdev_get_name(port->up.netdev), strerror(error));
2080 bundle_send_learning_packets(struct ofbundle *bundle)
2082 struct ofproto_dpif *ofproto = bundle->ofproto;
2083 int error, n_packets, n_errors;
2084 struct mac_entry *e;
2086 error = n_packets = n_errors = 0;
2087 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2088 if (e->port.p != bundle) {
2089 struct ofpbuf *learning_packet;
2090 struct ofport_dpif *port;
2094 /* The assignment to "port" is unnecessary but makes "grep"ing for
2095 * struct ofport_dpif more effective. */
2096 learning_packet = bond_compose_learning_packet(bundle->bond,
2100 ret = send_packet(port, learning_packet);
2101 ofpbuf_delete(learning_packet);
2111 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2112 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2113 "packets, last error was: %s",
2114 bundle->name, n_errors, n_packets, strerror(error));
2116 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2117 bundle->name, n_packets);
2122 bundle_run(struct ofbundle *bundle)
2125 lacp_run(bundle->lacp, send_pdu_cb);
2128 struct ofport_dpif *port;
2130 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2131 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2134 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
2135 lacp_status(bundle->lacp));
2136 if (bond_should_send_learning_packets(bundle->bond)) {
2137 bundle_send_learning_packets(bundle);
2143 bundle_wait(struct ofbundle *bundle)
2146 lacp_wait(bundle->lacp);
2149 bond_wait(bundle->bond);
2156 mirror_scan(struct ofproto_dpif *ofproto)
2160 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2161 if (!ofproto->mirrors[idx]) {
2168 static struct ofmirror *
2169 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2173 for (i = 0; i < MAX_MIRRORS; i++) {
2174 struct ofmirror *mirror = ofproto->mirrors[i];
2175 if (mirror && mirror->aux == aux) {
2183 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2185 mirror_update_dups(struct ofproto_dpif *ofproto)
2189 for (i = 0; i < MAX_MIRRORS; i++) {
2190 struct ofmirror *m = ofproto->mirrors[i];
2193 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2197 for (i = 0; i < MAX_MIRRORS; i++) {
2198 struct ofmirror *m1 = ofproto->mirrors[i];
2205 for (j = i + 1; j < MAX_MIRRORS; j++) {
2206 struct ofmirror *m2 = ofproto->mirrors[j];
2208 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2209 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2210 m2->dup_mirrors |= m1->dup_mirrors;
2217 mirror_set(struct ofproto *ofproto_, void *aux,
2218 const struct ofproto_mirror_settings *s)
2220 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2221 mirror_mask_t mirror_bit;
2222 struct ofbundle *bundle;
2223 struct ofmirror *mirror;
2224 struct ofbundle *out;
2225 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2226 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2229 mirror = mirror_lookup(ofproto, aux);
2231 mirror_destroy(mirror);
2237 idx = mirror_scan(ofproto);
2239 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2241 ofproto->up.name, MAX_MIRRORS, s->name);
2245 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2246 mirror->ofproto = ofproto;
2249 mirror->out_vlan = -1;
2250 mirror->name = NULL;
2253 if (!mirror->name || strcmp(s->name, mirror->name)) {
2255 mirror->name = xstrdup(s->name);
2258 /* Get the new configuration. */
2259 if (s->out_bundle) {
2260 out = bundle_lookup(ofproto, s->out_bundle);
2262 mirror_destroy(mirror);
2268 out_vlan = s->out_vlan;
2270 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2271 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2273 /* If the configuration has not changed, do nothing. */
2274 if (hmapx_equals(&srcs, &mirror->srcs)
2275 && hmapx_equals(&dsts, &mirror->dsts)
2276 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2277 && mirror->out == out
2278 && mirror->out_vlan == out_vlan)
2280 hmapx_destroy(&srcs);
2281 hmapx_destroy(&dsts);
2285 hmapx_swap(&srcs, &mirror->srcs);
2286 hmapx_destroy(&srcs);
2288 hmapx_swap(&dsts, &mirror->dsts);
2289 hmapx_destroy(&dsts);
2291 free(mirror->vlans);
2292 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2295 mirror->out_vlan = out_vlan;
2297 /* Update bundles. */
2298 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2299 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2300 if (hmapx_contains(&mirror->srcs, bundle)) {
2301 bundle->src_mirrors |= mirror_bit;
2303 bundle->src_mirrors &= ~mirror_bit;
2306 if (hmapx_contains(&mirror->dsts, bundle)) {
2307 bundle->dst_mirrors |= mirror_bit;
2309 bundle->dst_mirrors &= ~mirror_bit;
2312 if (mirror->out == bundle) {
2313 bundle->mirror_out |= mirror_bit;
2315 bundle->mirror_out &= ~mirror_bit;
2319 ofproto->need_revalidate = REV_RECONFIGURE;
2320 ofproto->has_mirrors = true;
2321 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2322 mirror_update_dups(ofproto);
2328 mirror_destroy(struct ofmirror *mirror)
2330 struct ofproto_dpif *ofproto;
2331 mirror_mask_t mirror_bit;
2332 struct ofbundle *bundle;
2339 ofproto = mirror->ofproto;
2340 ofproto->need_revalidate = REV_RECONFIGURE;
2341 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2343 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2344 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2345 bundle->src_mirrors &= ~mirror_bit;
2346 bundle->dst_mirrors &= ~mirror_bit;
2347 bundle->mirror_out &= ~mirror_bit;
2350 hmapx_destroy(&mirror->srcs);
2351 hmapx_destroy(&mirror->dsts);
2352 free(mirror->vlans);
2354 ofproto->mirrors[mirror->idx] = NULL;
2358 mirror_update_dups(ofproto);
2360 ofproto->has_mirrors = false;
2361 for (i = 0; i < MAX_MIRRORS; i++) {
2362 if (ofproto->mirrors[i]) {
2363 ofproto->has_mirrors = true;
2370 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2371 uint64_t *packets, uint64_t *bytes)
2373 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2374 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2377 *packets = *bytes = UINT64_MAX;
2381 *packets = mirror->packet_count;
2382 *bytes = mirror->byte_count;
2388 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2390 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2391 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2392 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2398 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2400 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2401 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2402 return bundle && bundle->mirror_out != 0;
2406 forward_bpdu_changed(struct ofproto *ofproto_)
2408 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2409 ofproto->need_revalidate = REV_RECONFIGURE;
2413 set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time)
2415 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2416 mac_learning_set_idle_time(ofproto->ml, idle_time);
2421 static struct ofport_dpif *
2422 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2424 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2425 return ofport ? ofport_dpif_cast(ofport) : NULL;
2428 static struct ofport_dpif *
2429 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2431 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2435 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2436 struct dpif_port *dpif_port)
2438 ofproto_port->name = dpif_port->name;
2439 ofproto_port->type = dpif_port->type;
2440 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2444 port_run_fast(struct ofport_dpif *ofport)
2446 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2447 struct ofpbuf packet;
2449 ofpbuf_init(&packet, 0);
2450 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2451 send_packet(ofport, &packet);
2452 ofpbuf_uninit(&packet);
2457 port_run(struct ofport_dpif *ofport)
2459 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2460 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2461 bool enable = netdev_get_carrier(ofport->up.netdev);
2463 ofport->carrier_seq = carrier_seq;
2465 port_run_fast(ofport);
2467 int cfm_opup = cfm_get_opup(ofport->cfm);
2469 cfm_run(ofport->cfm);
2470 enable = enable && !cfm_get_fault(ofport->cfm);
2472 if (cfm_opup >= 0) {
2473 enable = enable && cfm_opup;
2477 if (ofport->bundle) {
2478 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2479 if (carrier_changed) {
2480 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2484 if (ofport->may_enable != enable) {
2485 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2487 if (ofproto->has_bundle_action) {
2488 ofproto->need_revalidate = REV_PORT_TOGGLED;
2492 ofport->may_enable = enable;
2496 port_wait(struct ofport_dpif *ofport)
2499 cfm_wait(ofport->cfm);
2504 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2505 struct ofproto_port *ofproto_port)
2507 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2508 struct dpif_port dpif_port;
2511 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2513 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2519 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2521 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2522 uint16_t odp_port = UINT16_MAX;
2525 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2527 *ofp_portp = odp_port_to_ofp_port(odp_port);
2533 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2535 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2538 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2540 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2542 /* The caller is going to close ofport->up.netdev. If this is a
2543 * bonded port, then the bond is using that netdev, so remove it
2544 * from the bond. The client will need to reconfigure everything
2545 * after deleting ports, so then the slave will get re-added. */
2546 bundle_remove(&ofport->up);
2553 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2555 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2558 error = netdev_get_stats(ofport->up.netdev, stats);
2560 if (!error && ofport->odp_port == OVSP_LOCAL) {
2561 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2563 /* ofproto->stats.tx_packets represents packets that we created
2564 * internally and sent to some port (e.g. packets sent with
2565 * send_packet()). Account for them as if they had come from
2566 * OFPP_LOCAL and got forwarded. */
2568 if (stats->rx_packets != UINT64_MAX) {
2569 stats->rx_packets += ofproto->stats.tx_packets;
2572 if (stats->rx_bytes != UINT64_MAX) {
2573 stats->rx_bytes += ofproto->stats.tx_bytes;
2576 /* ofproto->stats.rx_packets represents packets that were received on
2577 * some port and we processed internally and dropped (e.g. STP).
2578 * Account fro them as if they had been forwarded to OFPP_LOCAL. */
2580 if (stats->tx_packets != UINT64_MAX) {
2581 stats->tx_packets += ofproto->stats.rx_packets;
2584 if (stats->tx_bytes != UINT64_MAX) {
2585 stats->tx_bytes += ofproto->stats.rx_bytes;
2592 /* Account packets for LOCAL port. */
2594 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2595 size_t tx_size, size_t rx_size)
2597 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2600 ofproto->stats.rx_packets++;
2601 ofproto->stats.rx_bytes += rx_size;
2604 ofproto->stats.tx_packets++;
2605 ofproto->stats.tx_bytes += tx_size;
2609 struct port_dump_state {
2610 struct dpif_port_dump dump;
2615 port_dump_start(const struct ofproto *ofproto_, void **statep)
2617 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2618 struct port_dump_state *state;
2620 *statep = state = xmalloc(sizeof *state);
2621 dpif_port_dump_start(&state->dump, ofproto->dpif);
2622 state->done = false;
2627 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2628 struct ofproto_port *port)
2630 struct port_dump_state *state = state_;
2631 struct dpif_port dpif_port;
2633 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2634 ofproto_port_from_dpif_port(port, &dpif_port);
2637 int error = dpif_port_dump_done(&state->dump);
2639 return error ? error : EOF;
2644 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2646 struct port_dump_state *state = state_;
2649 dpif_port_dump_done(&state->dump);
2656 port_poll(const struct ofproto *ofproto_, char **devnamep)
2658 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2659 return dpif_port_poll(ofproto->dpif, devnamep);
2663 port_poll_wait(const struct ofproto *ofproto_)
2665 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2666 dpif_port_poll_wait(ofproto->dpif);
2670 port_is_lacp_current(const struct ofport *ofport_)
2672 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2673 return (ofport->bundle && ofport->bundle->lacp
2674 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2678 /* Upcall handling. */
2680 /* Flow miss batching.
2682 * Some dpifs implement operations faster when you hand them off in a batch.
2683 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2684 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2685 * more packets, plus possibly installing the flow in the dpif.
2687 * So far we only batch the operations that affect flow setup time the most.
2688 * It's possible to batch more than that, but the benefit might be minimal. */
2690 struct hmap_node hmap_node;
2692 enum odp_key_fitness key_fitness;
2693 const struct nlattr *key;
2695 ovs_be16 initial_tci;
2696 struct list packets;
2697 enum dpif_upcall_type upcall_type;
2700 struct flow_miss_op {
2701 struct dpif_op dpif_op;
2702 struct subfacet *subfacet; /* Subfacet */
2703 void *garbage; /* Pointer to pass to free(), NULL if none. */
2704 uint64_t stub[1024 / 8]; /* Temporary buffer. */
2707 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2708 * OpenFlow controller as necessary according to their individual
2709 * configurations. */
2711 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
2712 const struct flow *flow)
2714 struct ofputil_packet_in pin;
2716 pin.packet = packet->data;
2717 pin.packet_len = packet->size;
2718 pin.reason = OFPR_NO_MATCH;
2719 pin.controller_id = 0;
2724 pin.send_len = 0; /* not used for flow table misses */
2726 flow_get_metadata(flow, &pin.fmd);
2728 /* Registers aren't meaningful on a miss. */
2729 memset(pin.fmd.reg_masks, 0, sizeof pin.fmd.reg_masks);
2731 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
2734 static enum slow_path_reason
2735 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2736 const struct ofpbuf *packet)
2738 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2744 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2746 cfm_process_heartbeat(ofport->cfm, packet);
2749 } else if (ofport->bundle && ofport->bundle->lacp
2750 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2752 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2755 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2757 stp_process_packet(ofport, packet);
2764 static struct flow_miss *
2765 flow_miss_find(struct hmap *todo, const struct flow *flow, uint32_t hash)
2767 struct flow_miss *miss;
2769 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2770 if (flow_equal(&miss->flow, flow)) {
2778 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
2779 * 'packet'. The caller must initialize op->actions and op->actions_len. If
2780 * 'miss' is associated with a subfacet the caller must also initialize the
2781 * returned op->subfacet, and if anything needs to be freed after processing
2782 * the op, the caller must initialize op->garbage also. */
2784 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
2785 struct flow_miss_op *op)
2787 if (miss->flow.vlan_tci != miss->initial_tci) {
2788 /* This packet was received on a VLAN splinter port. We
2789 * added a VLAN to the packet to make the packet resemble
2790 * the flow, but the actions were composed assuming that
2791 * the packet contained no VLAN. So, we must remove the
2792 * VLAN header from the packet before trying to execute the
2794 eth_pop_vlan(packet);
2797 op->subfacet = NULL;
2799 op->dpif_op.type = DPIF_OP_EXECUTE;
2800 op->dpif_op.u.execute.key = miss->key;
2801 op->dpif_op.u.execute.key_len = miss->key_len;
2802 op->dpif_op.u.execute.packet = packet;
2805 /* Helper for handle_flow_miss_without_facet() and
2806 * handle_flow_miss_with_facet(). */
2808 handle_flow_miss_common(struct rule_dpif *rule,
2809 struct ofpbuf *packet, const struct flow *flow)
2811 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2813 ofproto->n_matches++;
2815 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2817 * Extra-special case for fail-open mode.
2819 * We are in fail-open mode and the packet matched the fail-open
2820 * rule, but we are connected to a controller too. We should send
2821 * the packet up to the controller in the hope that it will try to
2822 * set up a flow and thereby allow us to exit fail-open.
2824 * See the top-level comment in fail-open.c for more information.
2826 send_packet_in_miss(ofproto, packet, flow);
2830 /* Figures out whether a flow that missed in 'ofproto', whose details are in
2831 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
2832 * installing a datapath flow. The answer is usually "yes" (a return value of
2833 * true). However, for short flows the cost of bookkeeping is much higher than
2834 * the benefits, so when the datapath holds a large number of flows we impose
2835 * some heuristics to decide which flows are likely to be worth tracking. */
2837 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
2838 struct flow_miss *miss, uint32_t hash)
2840 if (!ofproto->governor) {
2843 n_subfacets = hmap_count(&ofproto->subfacets);
2844 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
2848 ofproto->governor = governor_create(ofproto->up.name);
2851 return governor_should_install_flow(ofproto->governor, hash,
2852 list_size(&miss->packets));
2855 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
2856 * or creating any datapath flow. May add an "execute" operation to 'ops' and
2857 * increment '*n_ops'. */
2859 handle_flow_miss_without_facet(struct flow_miss *miss,
2860 struct rule_dpif *rule,
2861 struct flow_miss_op *ops, size_t *n_ops)
2863 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2864 struct action_xlate_ctx ctx;
2865 struct ofpbuf *packet;
2867 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2868 struct flow_miss_op *op = &ops[*n_ops];
2869 struct dpif_flow_stats stats;
2870 struct ofpbuf odp_actions;
2872 COVERAGE_INC(facet_suppress);
2874 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
2876 dpif_flow_stats_extract(&miss->flow, packet, &stats);
2877 rule_credit_stats(rule, &stats);
2879 action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci,
2881 ctx.resubmit_stats = &stats;
2882 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
2885 if (odp_actions.size) {
2886 struct dpif_execute *execute = &op->dpif_op.u.execute;
2888 init_flow_miss_execute_op(miss, packet, op);
2889 execute->actions = odp_actions.data;
2890 execute->actions_len = odp_actions.size;
2891 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
2895 ofpbuf_uninit(&odp_actions);
2900 /* Handles 'miss', which matches 'facet'. May add any required datapath
2901 * operations to 'ops', incrementing '*n_ops' for each new op. */
2903 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
2904 struct flow_miss_op *ops, size_t *n_ops)
2906 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
2907 enum subfacet_path want_path;
2908 struct subfacet *subfacet;
2909 struct ofpbuf *packet;
2911 subfacet = subfacet_create(facet,
2912 miss->key_fitness, miss->key, miss->key_len,
2915 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2916 struct flow_miss_op *op = &ops[*n_ops];
2917 struct dpif_flow_stats stats;
2918 struct ofpbuf odp_actions;
2920 handle_flow_miss_common(facet->rule, packet, &miss->flow);
2922 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
2923 if (!subfacet->actions || subfacet->slow) {
2924 subfacet_make_actions(subfacet, packet, &odp_actions);
2927 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2928 subfacet_update_stats(subfacet, &stats);
2930 if (subfacet->actions_len) {
2931 struct dpif_execute *execute = &op->dpif_op.u.execute;
2933 init_flow_miss_execute_op(miss, packet, op);
2934 op->subfacet = subfacet;
2935 if (!subfacet->slow) {
2936 execute->actions = subfacet->actions;
2937 execute->actions_len = subfacet->actions_len;
2938 ofpbuf_uninit(&odp_actions);
2940 execute->actions = odp_actions.data;
2941 execute->actions_len = odp_actions.size;
2942 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
2947 ofpbuf_uninit(&odp_actions);
2951 want_path = subfacet_want_path(subfacet->slow);
2952 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
2953 struct flow_miss_op *op = &ops[(*n_ops)++];
2954 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
2956 op->subfacet = subfacet;
2958 op->dpif_op.type = DPIF_OP_FLOW_PUT;
2959 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2960 put->key = miss->key;
2961 put->key_len = miss->key_len;
2962 if (want_path == SF_FAST_PATH) {
2963 put->actions = subfacet->actions;
2964 put->actions_len = subfacet->actions_len;
2966 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
2967 op->stub, sizeof op->stub,
2968 &put->actions, &put->actions_len);
2974 /* Handles flow miss 'miss' on 'ofproto'. May add any required datapath
2975 * operations to 'ops', incrementing '*n_ops' for each new op. */
2977 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2978 struct flow_miss_op *ops, size_t *n_ops)
2980 struct facet *facet;
2983 /* The caller must ensure that miss->hmap_node.hash contains
2984 * flow_hash(miss->flow, 0). */
2985 hash = miss->hmap_node.hash;
2987 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
2989 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
2991 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
2992 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
2996 facet = facet_create(rule, &miss->flow, hash);
2998 handle_flow_miss_with_facet(miss, facet, ops, n_ops);
3001 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
3002 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
3003 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
3004 * what a flow key should contain.
3006 * This function also includes some logic to help make VLAN splinters
3007 * transparent to the rest of the upcall processing logic. In particular, if
3008 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
3009 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
3010 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
3012 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
3013 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
3014 * (This differs from the value returned in flow->vlan_tci only for packets
3015 * received on VLAN splinters.)
3017 static enum odp_key_fitness
3018 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
3019 const struct nlattr *key, size_t key_len,
3020 struct flow *flow, ovs_be16 *initial_tci,
3021 struct ofpbuf *packet)
3023 enum odp_key_fitness fitness;
3025 fitness = odp_flow_key_to_flow(key, key_len, flow);
3026 if (fitness == ODP_FIT_ERROR) {
3029 *initial_tci = flow->vlan_tci;
3031 if (vsp_adjust_flow(ofproto, flow)) {
3033 /* Make the packet resemble the flow, so that it gets sent to an
3034 * OpenFlow controller properly, so that it looks correct for
3035 * sFlow, and so that flow_extract() will get the correct vlan_tci
3036 * if it is called on 'packet'.
3038 * The allocated space inside 'packet' probably also contains
3039 * 'key', that is, both 'packet' and 'key' are probably part of a
3040 * struct dpif_upcall (see the large comment on that structure
3041 * definition), so pushing data on 'packet' is in general not a
3042 * good idea since it could overwrite 'key' or free it as a side
3043 * effect. However, it's OK in this special case because we know
3044 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
3045 * will just overwrite the 4-byte "struct nlattr", which is fine
3046 * since we don't need that header anymore. */
3047 eth_push_vlan(packet, flow->vlan_tci);
3050 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
3051 if (fitness == ODP_FIT_PERFECT) {
3052 fitness = ODP_FIT_TOO_MUCH;
3060 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
3063 struct dpif_upcall *upcall;
3064 struct flow_miss *miss;
3065 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3066 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3067 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3077 /* Construct the to-do list.
3079 * This just amounts to extracting the flow from each packet and sticking
3080 * the packets that have the same flow in the same "flow_miss" structure so
3081 * that we can process them together. */
3084 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3085 struct flow_miss *miss = &misses[n_misses];
3086 struct flow_miss *existing_miss;
3089 /* Obtain metadata and check userspace/kernel agreement on flow match,
3090 * then set 'flow''s header pointers. */
3091 miss->key_fitness = ofproto_dpif_extract_flow_key(
3092 ofproto, upcall->key, upcall->key_len,
3093 &miss->flow, &miss->initial_tci, upcall->packet);
3094 if (miss->key_fitness == ODP_FIT_ERROR) {
3097 flow_extract(upcall->packet, miss->flow.skb_priority,
3098 miss->flow.tun_id, miss->flow.in_port, &miss->flow);
3100 /* Add other packets to a to-do list. */
3101 hash = flow_hash(&miss->flow, 0);
3102 existing_miss = flow_miss_find(&todo, &miss->flow, hash);
3103 if (!existing_miss) {
3104 hmap_insert(&todo, &miss->hmap_node, hash);
3105 miss->key = upcall->key;
3106 miss->key_len = upcall->key_len;
3107 miss->upcall_type = upcall->type;
3108 list_init(&miss->packets);
3112 miss = existing_miss;
3114 list_push_back(&miss->packets, &upcall->packet->list_node);
3117 /* Process each element in the to-do list, constructing the set of
3118 * operations to batch. */
3120 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3121 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
3123 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3125 /* Execute batch. */
3126 for (i = 0; i < n_ops; i++) {
3127 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3129 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
3131 /* Free memory and update facets. */
3132 for (i = 0; i < n_ops; i++) {
3133 struct flow_miss_op *op = &flow_miss_ops[i];
3135 switch (op->dpif_op.type) {
3136 case DPIF_OP_EXECUTE:
3139 case DPIF_OP_FLOW_PUT:
3140 if (!op->dpif_op.error) {
3141 op->subfacet->path = subfacet_want_path(op->subfacet->slow);
3145 case DPIF_OP_FLOW_DEL:
3151 hmap_destroy(&todo);
3154 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
3155 classify_upcall(const struct dpif_upcall *upcall)
3157 union user_action_cookie cookie;
3159 /* First look at the upcall type. */
3160 switch (upcall->type) {
3161 case DPIF_UC_ACTION:
3167 case DPIF_N_UC_TYPES:
3169 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3173 /* "action" upcalls need a closer look. */
3174 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3175 switch (cookie.type) {
3176 case USER_ACTION_COOKIE_SFLOW:
3177 return SFLOW_UPCALL;
3179 case USER_ACTION_COOKIE_SLOW_PATH:
3182 case USER_ACTION_COOKIE_UNSPEC:
3184 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
3190 handle_sflow_upcall(struct ofproto_dpif *ofproto,
3191 const struct dpif_upcall *upcall)
3193 union user_action_cookie cookie;
3194 enum odp_key_fitness fitness;
3195 ovs_be16 initial_tci;
3198 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
3199 upcall->key_len, &flow,
3200 &initial_tci, upcall->packet);
3201 if (fitness == ODP_FIT_ERROR) {
3205 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3206 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, &cookie);
3210 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
3212 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3213 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3214 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3219 assert(max_batch <= FLOW_MISS_MAX_BATCH);
3222 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3223 struct dpif_upcall *upcall = &misses[n_misses];
3224 struct ofpbuf *buf = &miss_bufs[n_misses];
3227 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3228 sizeof miss_buf_stubs[n_misses]);
3229 error = dpif_recv(ofproto->dpif, upcall, buf);
3235 switch (classify_upcall(upcall)) {
3237 /* Handle it later. */
3242 if (ofproto->sflow) {
3243 handle_sflow_upcall(ofproto, upcall);
3254 /* Handle deferred MISS_UPCALL processing. */
3255 handle_miss_upcalls(ofproto, misses, n_misses);
3256 for (i = 0; i < n_misses; i++) {
3257 ofpbuf_uninit(&miss_bufs[i]);
3263 /* Flow expiration. */
3265 static int subfacet_max_idle(const struct ofproto_dpif *);
3266 static void update_stats(struct ofproto_dpif *);
3267 static void rule_expire(struct rule_dpif *);
3268 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
3270 /* This function is called periodically by run(). Its job is to collect
3271 * updates for the flows that have been installed into the datapath, most
3272 * importantly when they last were used, and then use that information to
3273 * expire flows that have not been used recently.
3275 * Returns the number of milliseconds after which it should be called again. */
3277 expire(struct ofproto_dpif *ofproto)
3279 struct rule_dpif *rule, *next_rule;
3280 struct oftable *table;
3283 /* Update stats for each flow in the datapath. */
3284 update_stats(ofproto);
3286 /* Expire subfacets that have been idle too long. */
3287 dp_max_idle = subfacet_max_idle(ofproto);
3288 expire_subfacets(ofproto, dp_max_idle);
3290 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
3291 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
3292 struct cls_cursor cursor;
3294 cls_cursor_init(&cursor, &table->cls, NULL);
3295 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
3300 /* All outstanding data in existing flows has been accounted, so it's a
3301 * good time to do bond rebalancing. */
3302 if (ofproto->has_bonded_bundles) {
3303 struct ofbundle *bundle;
3305 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3307 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
3312 return MIN(dp_max_idle, 1000);
3315 /* Updates flow table statistics given that the datapath just reported 'stats'
3316 * as 'subfacet''s statistics. */
3318 update_subfacet_stats(struct subfacet *subfacet,
3319 const struct dpif_flow_stats *stats)
3321 struct facet *facet = subfacet->facet;
3323 if (stats->n_packets >= subfacet->dp_packet_count) {
3324 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
3325 facet->packet_count += extra;
3327 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3330 if (stats->n_bytes >= subfacet->dp_byte_count) {
3331 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
3333 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3336 subfacet->dp_packet_count = stats->n_packets;
3337 subfacet->dp_byte_count = stats->n_bytes;
3339 facet->tcp_flags |= stats->tcp_flags;
3341 subfacet_update_time(subfacet, stats->used);
3342 if (facet->accounted_bytes < facet->byte_count) {
3344 facet_account(facet);
3345 facet->accounted_bytes = facet->byte_count;
3347 facet_push_stats(facet);
3350 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3351 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3353 delete_unexpected_flow(struct dpif *dpif,
3354 const struct nlattr *key, size_t key_len)
3356 if (!VLOG_DROP_WARN(&rl)) {
3360 odp_flow_key_format(key, key_len, &s);
3361 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
3365 COVERAGE_INC(facet_unexpected);
3366 dpif_flow_del(dpif, key, key_len, NULL);
3369 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3371 * This function also pushes statistics updates to rules which each facet
3372 * resubmits into. Generally these statistics will be accurate. However, if a
3373 * facet changes the rule it resubmits into at some time in between
3374 * update_stats() runs, it is possible that statistics accrued to the
3375 * old rule will be incorrectly attributed to the new rule. This could be
3376 * avoided by calling update_stats() whenever rules are created or
3377 * deleted. However, the performance impact of making so many calls to the
3378 * datapath do not justify the benefit of having perfectly accurate statistics.
3381 update_stats(struct ofproto_dpif *p)
3383 const struct dpif_flow_stats *stats;
3384 struct dpif_flow_dump dump;
3385 const struct nlattr *key;
3388 dpif_flow_dump_start(&dump, p->dpif);
3389 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
3390 struct subfacet *subfacet;
3392 subfacet = subfacet_find(p, key, key_len);
3393 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3395 update_subfacet_stats(subfacet, stats);
3399 /* Stats are updated per-packet. */
3402 case SF_NOT_INSTALLED:
3404 delete_unexpected_flow(p->dpif, key, key_len);
3408 dpif_flow_dump_done(&dump);
3411 /* Calculates and returns the number of milliseconds of idle time after which
3412 * subfacets should expire from the datapath. When a subfacet expires, we fold
3413 * its statistics into its facet, and when a facet's last subfacet expires, we
3414 * fold its statistic into its rule. */
3416 subfacet_max_idle(const struct ofproto_dpif *ofproto)
3419 * Idle time histogram.
3421 * Most of the time a switch has a relatively small number of subfacets.
3422 * When this is the case we might as well keep statistics for all of them
3423 * in userspace and to cache them in the kernel datapath for performance as
3426 * As the number of subfacets increases, the memory required to maintain
3427 * statistics about them in userspace and in the kernel becomes
3428 * significant. However, with a large number of subfacets it is likely
3429 * that only a few of them are "heavy hitters" that consume a large amount
3430 * of bandwidth. At this point, only heavy hitters are worth caching in
3431 * the kernel and maintaining in userspaces; other subfacets we can
3434 * The technique used to compute the idle time is to build a histogram with
3435 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3436 * that is installed in the kernel gets dropped in the appropriate bucket.
3437 * After the histogram has been built, we compute the cutoff so that only
3438 * the most-recently-used 1% of subfacets (but at least
3439 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3440 * the most-recently-used bucket of subfacets is kept, so actually an
3441 * arbitrary number of subfacets can be kept in any given expiration run
3442 * (though the next run will delete most of those unless they receive
3445 * This requires a second pass through the subfacets, in addition to the
3446 * pass made by update_stats(), because the former function never looks at
3447 * uninstallable subfacets.
3449 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3450 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3451 int buckets[N_BUCKETS] = { 0 };
3452 int total, subtotal, bucket;
3453 struct subfacet *subfacet;
3457 total = hmap_count(&ofproto->subfacets);
3458 if (total <= ofproto->up.flow_eviction_threshold) {
3459 return N_BUCKETS * BUCKET_WIDTH;
3462 /* Build histogram. */
3464 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3465 long long int idle = now - subfacet->used;
3466 int bucket = (idle <= 0 ? 0
3467 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3468 : (unsigned int) idle / BUCKET_WIDTH);
3472 /* Find the first bucket whose flows should be expired. */
3473 subtotal = bucket = 0;
3475 subtotal += buckets[bucket++];
3476 } while (bucket < N_BUCKETS &&
3477 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3479 if (VLOG_IS_DBG_ENABLED()) {
3483 ds_put_cstr(&s, "keep");
3484 for (i = 0; i < N_BUCKETS; i++) {
3486 ds_put_cstr(&s, ", drop");
3489 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3492 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3496 return bucket * BUCKET_WIDTH;
3499 enum { EXPIRE_MAX_BATCH = 50 };
3502 expire_batch(struct ofproto_dpif *ofproto, struct subfacet **subfacets, int n)
3504 struct odputil_keybuf keybufs[EXPIRE_MAX_BATCH];
3505 struct dpif_op ops[EXPIRE_MAX_BATCH];
3506 struct dpif_op *opsp[EXPIRE_MAX_BATCH];
3507 struct ofpbuf keys[EXPIRE_MAX_BATCH];
3508 struct dpif_flow_stats stats[EXPIRE_MAX_BATCH];
3511 for (i = 0; i < n; i++) {
3512 ops[i].type = DPIF_OP_FLOW_DEL;
3513 subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]);
3514 ops[i].u.flow_del.key = keys[i].data;
3515 ops[i].u.flow_del.key_len = keys[i].size;
3516 ops[i].u.flow_del.stats = &stats[i];
3520 dpif_operate(ofproto->dpif, opsp, n);
3521 for (i = 0; i < n; i++) {
3522 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
3523 subfacets[i]->path = SF_NOT_INSTALLED;
3524 subfacet_destroy(subfacets[i]);
3529 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3531 /* Cutoff time for most flows. */
3532 long long int normal_cutoff = time_msec() - dp_max_idle;
3534 /* We really want to keep flows for special protocols around, so use a more
3535 * conservative cutoff. */
3536 long long int special_cutoff = time_msec() - 10000;
3538 struct subfacet *subfacet, *next_subfacet;
3539 struct subfacet *batch[EXPIRE_MAX_BATCH];
3543 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3544 &ofproto->subfacets) {
3545 long long int cutoff;
3547 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
3550 if (subfacet->used < cutoff) {
3551 if (subfacet->path != SF_NOT_INSTALLED) {
3552 batch[n_batch++] = subfacet;
3553 if (n_batch >= EXPIRE_MAX_BATCH) {
3554 expire_batch(ofproto, batch, n_batch);
3558 subfacet_destroy(subfacet);
3564 expire_batch(ofproto, batch, n_batch);
3568 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3569 * then delete it entirely. */
3571 rule_expire(struct rule_dpif *rule)
3573 struct facet *facet, *next_facet;
3577 if (rule->up.pending) {
3578 /* We'll have to expire it later. */
3582 /* Has 'rule' expired? */
3584 if (rule->up.hard_timeout
3585 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3586 reason = OFPRR_HARD_TIMEOUT;
3587 } else if (rule->up.idle_timeout
3588 && now > rule->up.used + rule->up.idle_timeout * 1000) {
3589 reason = OFPRR_IDLE_TIMEOUT;
3594 COVERAGE_INC(ofproto_dpif_expired);
3596 /* Update stats. (This is a no-op if the rule expired due to an idle
3597 * timeout, because that only happens when the rule has no facets left.) */
3598 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3599 facet_remove(facet);
3602 /* Get rid of the rule. */
3603 ofproto_rule_expire(&rule->up, reason);
3608 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3610 * The caller must already have determined that no facet with an identical
3611 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3612 * the ofproto's classifier table.
3614 * 'hash' must be the return value of flow_hash(flow, 0).
3616 * The facet will initially have no subfacets. The caller should create (at
3617 * least) one subfacet with subfacet_create(). */
3618 static struct facet *
3619 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
3621 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3622 struct facet *facet;
3624 facet = xzalloc(sizeof *facet);
3625 facet->used = time_msec();
3626 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
3627 list_push_back(&rule->facets, &facet->list_node);
3629 facet->flow = *flow;
3630 list_init(&facet->subfacets);
3631 netflow_flow_init(&facet->nf_flow);
3632 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3638 facet_free(struct facet *facet)
3643 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3644 * 'packet', which arrived on 'in_port'.
3646 * Takes ownership of 'packet'. */
3648 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3649 const struct nlattr *odp_actions, size_t actions_len,
3650 struct ofpbuf *packet)
3652 struct odputil_keybuf keybuf;
3656 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3657 odp_flow_key_from_flow(&key, flow);
3659 error = dpif_execute(ofproto->dpif, key.data, key.size,
3660 odp_actions, actions_len, packet);
3662 ofpbuf_delete(packet);
3666 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3668 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3669 * rule's statistics, via subfacet_uninstall().
3671 * - Removes 'facet' from its rule and from ofproto->facets.
3674 facet_remove(struct facet *facet)
3676 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3677 struct subfacet *subfacet, *next_subfacet;
3679 assert(!list_is_empty(&facet->subfacets));
3681 /* First uninstall all of the subfacets to get final statistics. */
3682 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3683 subfacet_uninstall(subfacet);
3686 /* Flush the final stats to the rule.
3688 * This might require us to have at least one subfacet around so that we
3689 * can use its actions for accounting in facet_account(), which is why we
3690 * have uninstalled but not yet destroyed the subfacets. */
3691 facet_flush_stats(facet);
3693 /* Now we're really all done so destroy everything. */
3694 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3695 &facet->subfacets) {
3696 subfacet_destroy__(subfacet);
3698 hmap_remove(&ofproto->facets, &facet->hmap_node);
3699 list_remove(&facet->list_node);
3703 /* Feed information from 'facet' back into the learning table to keep it in
3704 * sync with what is actually flowing through the datapath. */
3706 facet_learn(struct facet *facet)
3708 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3709 struct action_xlate_ctx ctx;
3711 if (!facet->has_learn
3712 && !facet->has_normal
3713 && (!facet->has_fin_timeout
3714 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
3718 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3719 facet->flow.vlan_tci,
3720 facet->rule, facet->tcp_flags, NULL);
3721 ctx.may_learn = true;
3722 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
3723 facet->rule->up.ofpacts_len);
3727 facet_account(struct facet *facet)
3729 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3730 struct subfacet *subfacet;
3731 const struct nlattr *a;
3736 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3739 n_bytes = facet->byte_count - facet->accounted_bytes;
3741 /* This loop feeds byte counters to bond_account() for rebalancing to use
3742 * as a basis. We also need to track the actual VLAN on which the packet
3743 * is going to be sent to ensure that it matches the one passed to
3744 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3747 * We use the actions from an arbitrary subfacet because they should all
3748 * be equally valid for our purpose. */
3749 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3750 struct subfacet, list_node);
3751 vlan_tci = facet->flow.vlan_tci;
3752 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3753 subfacet->actions, subfacet->actions_len) {
3754 const struct ovs_action_push_vlan *vlan;
3755 struct ofport_dpif *port;
3757 switch (nl_attr_type(a)) {
3758 case OVS_ACTION_ATTR_OUTPUT:
3759 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3760 if (port && port->bundle && port->bundle->bond) {
3761 bond_account(port->bundle->bond, &facet->flow,
3762 vlan_tci_to_vid(vlan_tci), n_bytes);
3766 case OVS_ACTION_ATTR_POP_VLAN:
3767 vlan_tci = htons(0);
3770 case OVS_ACTION_ATTR_PUSH_VLAN:
3771 vlan = nl_attr_get(a);
3772 vlan_tci = vlan->vlan_tci;
3778 /* Returns true if the only action for 'facet' is to send to the controller.
3779 * (We don't report NetFlow expiration messages for such facets because they
3780 * are just part of the control logic for the network, not real traffic). */
3782 facet_is_controller_flow(struct facet *facet)
3785 const struct rule *rule = &facet->rule->up;
3786 const struct ofpact *ofpacts = rule->ofpacts;
3787 size_t ofpacts_len = rule->ofpacts_len;
3789 if (ofpacts->type == OFPACT_CONTROLLER &&
3790 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
3797 /* Folds all of 'facet''s statistics into its rule. Also updates the
3798 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3799 * 'facet''s statistics in the datapath should have been zeroed and folded into
3800 * its packet and byte counts before this function is called. */
3802 facet_flush_stats(struct facet *facet)
3804 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3805 struct subfacet *subfacet;
3807 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3808 assert(!subfacet->dp_byte_count);
3809 assert(!subfacet->dp_packet_count);
3812 facet_push_stats(facet);
3813 if (facet->accounted_bytes < facet->byte_count) {
3814 facet_account(facet);
3815 facet->accounted_bytes = facet->byte_count;
3818 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3819 struct ofexpired expired;
3820 expired.flow = facet->flow;
3821 expired.packet_count = facet->packet_count;
3822 expired.byte_count = facet->byte_count;
3823 expired.used = facet->used;
3824 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3827 facet->rule->packet_count += facet->packet_count;
3828 facet->rule->byte_count += facet->byte_count;
3830 /* Reset counters to prevent double counting if 'facet' ever gets
3832 facet_reset_counters(facet);
3834 netflow_flow_clear(&facet->nf_flow);
3835 facet->tcp_flags = 0;
3838 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3839 * Returns it if found, otherwise a null pointer.
3841 * 'hash' must be the return value of flow_hash(flow, 0).
3843 * The returned facet might need revalidation; use facet_lookup_valid()
3844 * instead if that is important. */
3845 static struct facet *
3846 facet_find(struct ofproto_dpif *ofproto,
3847 const struct flow *flow, uint32_t hash)
3849 struct facet *facet;
3851 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
3852 if (flow_equal(flow, &facet->flow)) {
3860 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3861 * Returns it if found, otherwise a null pointer.
3863 * 'hash' must be the return value of flow_hash(flow, 0).
3865 * The returned facet is guaranteed to be valid. */
3866 static struct facet *
3867 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
3870 struct facet *facet;
3872 facet = facet_find(ofproto, flow, hash);
3874 && (ofproto->need_revalidate
3875 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))) {
3876 facet_revalidate(facet);
3883 subfacet_path_to_string(enum subfacet_path path)
3886 case SF_NOT_INSTALLED:
3887 return "not installed";
3889 return "in fast path";
3891 return "in slow path";
3897 /* Returns the path in which a subfacet should be installed if its 'slow'
3898 * member has the specified value. */
3899 static enum subfacet_path
3900 subfacet_want_path(enum slow_path_reason slow)
3902 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
3905 /* Returns true if 'subfacet' needs to have its datapath flow updated,
3906 * supposing that its actions have been recalculated as 'want_actions' and that
3907 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
3909 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
3910 const struct ofpbuf *want_actions)
3912 enum subfacet_path want_path = subfacet_want_path(slow);
3913 return (want_path != subfacet->path
3914 || (want_path == SF_FAST_PATH
3915 && (subfacet->actions_len != want_actions->size
3916 || memcmp(subfacet->actions, want_actions->data,
3917 subfacet->actions_len))));
3921 facet_check_consistency(struct facet *facet)
3923 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3925 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3927 uint64_t odp_actions_stub[1024 / 8];
3928 struct ofpbuf odp_actions;
3930 struct rule_dpif *rule;
3931 struct subfacet *subfacet;
3932 bool may_log = false;
3935 /* Check the rule for consistency. */
3936 rule = rule_dpif_lookup(ofproto, &facet->flow);
3937 ok = rule == facet->rule;
3939 may_log = !VLOG_DROP_WARN(&rl);
3944 flow_format(&s, &facet->flow);
3945 ds_put_format(&s, ": facet associated with wrong rule (was "
3946 "table=%"PRIu8",", facet->rule->up.table_id);
3947 cls_rule_format(&facet->rule->up.cr, &s);
3948 ds_put_format(&s, ") (should have been table=%"PRIu8",",
3950 cls_rule_format(&rule->up.cr, &s);
3951 ds_put_char(&s, ')');
3953 VLOG_WARN("%s", ds_cstr(&s));
3958 /* Check the datapath actions for consistency. */
3959 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
3960 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3961 enum subfacet_path want_path;
3962 struct odputil_keybuf keybuf;
3963 struct action_xlate_ctx ctx;
3967 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3968 subfacet->initial_tci, rule, 0, NULL);
3969 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3972 if (subfacet->path == SF_NOT_INSTALLED) {
3973 /* This only happens if the datapath reported an error when we
3974 * tried to install the flow. Don't flag another error here. */
3978 want_path = subfacet_want_path(subfacet->slow);
3979 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
3980 /* The actions for slow-path flows may legitimately vary from one
3981 * packet to the next. We're done. */
3985 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
3989 /* Inconsistency! */
3991 may_log = !VLOG_DROP_WARN(&rl);
3995 /* Rate-limited, skip reporting. */
4000 subfacet_get_key(subfacet, &keybuf, &key);
4001 odp_flow_key_format(key.data, key.size, &s);
4003 ds_put_cstr(&s, ": inconsistency in subfacet");
4004 if (want_path != subfacet->path) {
4005 enum odp_key_fitness fitness = subfacet->key_fitness;
4007 ds_put_format(&s, " (%s, fitness=%s)",
4008 subfacet_path_to_string(subfacet->path),
4009 odp_key_fitness_to_string(fitness));
4010 ds_put_format(&s, " (should have been %s)",
4011 subfacet_path_to_string(want_path));
4012 } else if (want_path == SF_FAST_PATH) {
4013 ds_put_cstr(&s, " (actions were: ");
4014 format_odp_actions(&s, subfacet->actions,
4015 subfacet->actions_len);
4016 ds_put_cstr(&s, ") (correct actions: ");
4017 format_odp_actions(&s, odp_actions.data, odp_actions.size);
4018 ds_put_char(&s, ')');
4020 ds_put_cstr(&s, " (actions: ");
4021 format_odp_actions(&s, subfacet->actions,
4022 subfacet->actions_len);
4023 ds_put_char(&s, ')');
4025 VLOG_WARN("%s", ds_cstr(&s));
4028 ofpbuf_uninit(&odp_actions);
4033 /* Re-searches the classifier for 'facet':
4035 * - If the rule found is different from 'facet''s current rule, moves
4036 * 'facet' to the new rule and recompiles its actions.
4038 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4039 * where it is and recompiles its actions anyway. */
4041 facet_revalidate(struct facet *facet)
4043 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4045 struct nlattr *odp_actions;
4048 struct actions *new_actions;
4050 struct action_xlate_ctx ctx;
4051 uint64_t odp_actions_stub[1024 / 8];
4052 struct ofpbuf odp_actions;
4054 struct rule_dpif *new_rule;
4055 struct subfacet *subfacet;
4058 COVERAGE_INC(facet_revalidate);
4060 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
4062 /* Calculate new datapath actions.
4064 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4065 * emit a NetFlow expiration and, if so, we need to have the old state
4066 * around to properly compose it. */
4068 /* If the datapath actions changed or the installability changed,
4069 * then we need to talk to the datapath. */
4072 memset(&ctx, 0, sizeof ctx);
4073 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4074 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4075 enum slow_path_reason slow;
4077 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4078 subfacet->initial_tci, new_rule, 0, NULL);
4079 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
4082 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4083 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
4084 struct dpif_flow_stats stats;
4086 subfacet_install(subfacet,
4087 odp_actions.data, odp_actions.size, &stats, slow);
4088 subfacet_update_stats(subfacet, &stats);
4091 new_actions = xcalloc(list_size(&facet->subfacets),
4092 sizeof *new_actions);
4094 new_actions[i].odp_actions = xmemdup(odp_actions.data,
4096 new_actions[i].actions_len = odp_actions.size;
4101 ofpbuf_uninit(&odp_actions);
4104 facet_flush_stats(facet);
4107 /* Update 'facet' now that we've taken care of all the old state. */
4108 facet->tags = ctx.tags;
4109 facet->nf_flow.output_iface = ctx.nf_output_iface;
4110 facet->has_learn = ctx.has_learn;
4111 facet->has_normal = ctx.has_normal;
4112 facet->has_fin_timeout = ctx.has_fin_timeout;
4113 facet->mirrors = ctx.mirrors;
4116 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4117 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4119 if (new_actions && new_actions[i].odp_actions) {
4120 free(subfacet->actions);
4121 subfacet->actions = new_actions[i].odp_actions;
4122 subfacet->actions_len = new_actions[i].actions_len;
4128 if (facet->rule != new_rule) {
4129 COVERAGE_INC(facet_changed_rule);
4130 list_remove(&facet->list_node);
4131 list_push_back(&new_rule->facets, &facet->list_node);
4132 facet->rule = new_rule;
4133 facet->used = new_rule->up.created;
4134 facet->prev_used = facet->used;
4138 /* Updates 'facet''s used time. Caller is responsible for calling
4139 * facet_push_stats() to update the flows which 'facet' resubmits into. */
4141 facet_update_time(struct facet *facet, long long int used)
4143 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4144 if (used > facet->used) {
4146 ofproto_rule_update_used(&facet->rule->up, used);
4147 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
4152 facet_reset_counters(struct facet *facet)
4154 facet->packet_count = 0;
4155 facet->byte_count = 0;
4156 facet->prev_packet_count = 0;
4157 facet->prev_byte_count = 0;
4158 facet->accounted_bytes = 0;
4162 facet_push_stats(struct facet *facet)
4164 struct dpif_flow_stats stats;
4166 assert(facet->packet_count >= facet->prev_packet_count);
4167 assert(facet->byte_count >= facet->prev_byte_count);
4168 assert(facet->used >= facet->prev_used);
4170 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4171 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4172 stats.used = facet->used;
4173 stats.tcp_flags = 0;
4175 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
4176 facet->prev_packet_count = facet->packet_count;
4177 facet->prev_byte_count = facet->byte_count;
4178 facet->prev_used = facet->used;
4180 flow_push_stats(facet->rule, &facet->flow, &stats);
4182 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
4183 facet->mirrors, stats.n_packets, stats.n_bytes);
4188 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4190 rule->packet_count += stats->n_packets;
4191 rule->byte_count += stats->n_bytes;
4192 ofproto_rule_update_used(&rule->up, stats->used);
4195 /* Pushes flow statistics to the rules which 'flow' resubmits into given
4196 * 'rule''s actions and mirrors. */
4198 flow_push_stats(struct rule_dpif *rule,
4199 const struct flow *flow, const struct dpif_flow_stats *stats)
4201 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4202 struct action_xlate_ctx ctx;
4204 ofproto_rule_update_used(&rule->up, stats->used);
4206 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule,
4208 ctx.resubmit_stats = stats;
4209 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
4210 rule->up.ofpacts_len);
4215 static struct subfacet *
4216 subfacet_find__(struct ofproto_dpif *ofproto,
4217 const struct nlattr *key, size_t key_len, uint32_t key_hash,
4218 const struct flow *flow)
4220 struct subfacet *subfacet;
4222 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4223 &ofproto->subfacets) {
4225 ? (subfacet->key_len == key_len
4226 && !memcmp(key, subfacet->key, key_len))
4227 : flow_equal(flow, &subfacet->facet->flow)) {
4235 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4236 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
4237 * there is one, otherwise creates and returns a new subfacet.
4239 * If the returned subfacet is new, then subfacet->actions will be NULL, in
4240 * which case the caller must populate the actions with
4241 * subfacet_make_actions(). */
4242 static struct subfacet *
4243 subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness,
4244 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
4246 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4247 uint32_t key_hash = odp_flow_key_hash(key, key_len);
4248 struct subfacet *subfacet;
4250 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
4252 if (subfacet->facet == facet) {
4256 /* This shouldn't happen. */
4257 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4258 subfacet_destroy(subfacet);
4261 subfacet = (list_is_empty(&facet->subfacets)
4262 ? &facet->one_subfacet
4263 : xmalloc(sizeof *subfacet));
4264 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
4265 list_push_back(&facet->subfacets, &subfacet->list_node);
4266 subfacet->facet = facet;
4267 subfacet->key_fitness = key_fitness;
4268 if (key_fitness != ODP_FIT_PERFECT) {
4269 subfacet->key = xmemdup(key, key_len);
4270 subfacet->key_len = key_len;
4272 subfacet->key = NULL;
4273 subfacet->key_len = 0;
4275 subfacet->used = time_msec();
4276 subfacet->dp_packet_count = 0;
4277 subfacet->dp_byte_count = 0;
4278 subfacet->actions_len = 0;
4279 subfacet->actions = NULL;
4280 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
4283 subfacet->path = SF_NOT_INSTALLED;
4284 subfacet->initial_tci = initial_tci;
4289 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
4290 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
4291 static struct subfacet *
4292 subfacet_find(struct ofproto_dpif *ofproto,
4293 const struct nlattr *key, size_t key_len)
4295 uint32_t key_hash = odp_flow_key_hash(key, key_len);
4296 enum odp_key_fitness fitness;
4299 fitness = odp_flow_key_to_flow(key, key_len, &flow);
4300 if (fitness == ODP_FIT_ERROR) {
4304 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
4307 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4308 * its facet within 'ofproto', and frees it. */
4310 subfacet_destroy__(struct subfacet *subfacet)
4312 struct facet *facet = subfacet->facet;
4313 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4315 subfacet_uninstall(subfacet);
4316 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
4317 list_remove(&subfacet->list_node);
4318 free(subfacet->key);
4319 free(subfacet->actions);
4320 if (subfacet != &facet->one_subfacet) {
4325 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4326 * last remaining subfacet in its facet destroys the facet too. */
4328 subfacet_destroy(struct subfacet *subfacet)
4330 struct facet *facet = subfacet->facet;
4332 if (list_is_singleton(&facet->subfacets)) {
4333 /* facet_remove() needs at least one subfacet (it will remove it). */
4334 facet_remove(facet);
4336 subfacet_destroy__(subfacet);
4340 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
4341 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
4342 * for use as temporary storage. */
4344 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
4347 if (!subfacet->key) {
4348 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
4349 odp_flow_key_from_flow(key, &subfacet->facet->flow);
4351 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
4355 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
4356 * Translates the actions into 'odp_actions', which the caller must have
4357 * initialized and is responsible for uninitializing. */
4359 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
4360 struct ofpbuf *odp_actions)
4362 struct facet *facet = subfacet->facet;
4363 struct rule_dpif *rule = facet->rule;
4364 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4366 struct action_xlate_ctx ctx;
4368 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
4370 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
4371 facet->tags = ctx.tags;
4372 facet->has_learn = ctx.has_learn;
4373 facet->has_normal = ctx.has_normal;
4374 facet->has_fin_timeout = ctx.has_fin_timeout;
4375 facet->nf_flow.output_iface = ctx.nf_output_iface;
4376 facet->mirrors = ctx.mirrors;
4378 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4379 if (subfacet->actions_len != odp_actions->size
4380 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
4381 free(subfacet->actions);
4382 subfacet->actions_len = odp_actions->size;
4383 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
4387 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4388 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4389 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4390 * since 'subfacet' was last updated.
4392 * Returns 0 if successful, otherwise a positive errno value. */
4394 subfacet_install(struct subfacet *subfacet,
4395 const struct nlattr *actions, size_t actions_len,
4396 struct dpif_flow_stats *stats,
4397 enum slow_path_reason slow)
4399 struct facet *facet = subfacet->facet;
4400 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4401 enum subfacet_path path = subfacet_want_path(slow);
4402 uint64_t slow_path_stub[128 / 8];
4403 struct odputil_keybuf keybuf;
4404 enum dpif_flow_put_flags flags;
4408 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
4410 flags |= DPIF_FP_ZERO_STATS;
4413 if (path == SF_SLOW_PATH) {
4414 compose_slow_path(ofproto, &facet->flow, slow,
4415 slow_path_stub, sizeof slow_path_stub,
4416 &actions, &actions_len);
4419 subfacet_get_key(subfacet, &keybuf, &key);
4420 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
4421 actions, actions_len, stats);
4424 subfacet_reset_dp_stats(subfacet, stats);
4428 subfacet->path = path;
4434 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
4436 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
4437 stats, subfacet->slow);
4440 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4442 subfacet_uninstall(struct subfacet *subfacet)
4444 if (subfacet->path != SF_NOT_INSTALLED) {
4445 struct rule_dpif *rule = subfacet->facet->rule;
4446 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4447 struct odputil_keybuf keybuf;
4448 struct dpif_flow_stats stats;
4452 subfacet_get_key(subfacet, &keybuf, &key);
4453 error = dpif_flow_del(ofproto->dpif, key.data, key.size, &stats);
4454 subfacet_reset_dp_stats(subfacet, &stats);
4456 subfacet_update_stats(subfacet, &stats);
4458 subfacet->path = SF_NOT_INSTALLED;
4460 assert(subfacet->dp_packet_count == 0);
4461 assert(subfacet->dp_byte_count == 0);
4465 /* Resets 'subfacet''s datapath statistics counters. This should be called
4466 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4467 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4468 * was reset in the datapath. 'stats' will be modified to include only
4469 * statistics new since 'subfacet' was last updated. */
4471 subfacet_reset_dp_stats(struct subfacet *subfacet,
4472 struct dpif_flow_stats *stats)
4475 && subfacet->dp_packet_count <= stats->n_packets
4476 && subfacet->dp_byte_count <= stats->n_bytes) {
4477 stats->n_packets -= subfacet->dp_packet_count;
4478 stats->n_bytes -= subfacet->dp_byte_count;
4481 subfacet->dp_packet_count = 0;
4482 subfacet->dp_byte_count = 0;
4485 /* Updates 'subfacet''s used time. The caller is responsible for calling
4486 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
4488 subfacet_update_time(struct subfacet *subfacet, long long int used)
4490 if (used > subfacet->used) {
4491 subfacet->used = used;
4492 facet_update_time(subfacet->facet, used);
4496 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4498 * Because of the meaning of a subfacet's counters, it only makes sense to do
4499 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4500 * represents a packet that was sent by hand or if it represents statistics
4501 * that have been cleared out of the datapath. */
4503 subfacet_update_stats(struct subfacet *subfacet,
4504 const struct dpif_flow_stats *stats)
4506 if (stats->n_packets || stats->used > subfacet->used) {
4507 struct facet *facet = subfacet->facet;
4509 subfacet_update_time(subfacet, stats->used);
4510 facet->packet_count += stats->n_packets;
4511 facet->byte_count += stats->n_bytes;
4512 facet->tcp_flags |= stats->tcp_flags;
4513 facet_push_stats(facet);
4514 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
4520 static struct rule_dpif *
4521 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
4523 struct ofport_dpif *port;
4524 struct rule_dpif *rule;
4526 rule = rule_dpif_lookup__(ofproto, flow, 0);
4531 port = get_ofp_port(ofproto, flow->in_port);
4533 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
4534 return ofproto->miss_rule;
4537 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
4538 return ofproto->no_packet_in_rule;
4540 return ofproto->miss_rule;
4543 static struct rule_dpif *
4544 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
4547 struct cls_rule *cls_rule;
4548 struct classifier *cls;
4550 if (table_id >= N_TABLES) {
4554 cls = &ofproto->up.tables[table_id].cls;
4555 if (flow->nw_frag & FLOW_NW_FRAG_ANY
4556 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4557 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
4558 * are unavailable. */
4559 struct flow ofpc_normal_flow = *flow;
4560 ofpc_normal_flow.tp_src = htons(0);
4561 ofpc_normal_flow.tp_dst = htons(0);
4562 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
4564 cls_rule = classifier_lookup(cls, flow);
4566 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
4570 complete_operation(struct rule_dpif *rule)
4572 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4574 rule_invalidate(rule);
4576 struct dpif_completion *c = xmalloc(sizeof *c);
4577 c->op = rule->up.pending;
4578 list_push_back(&ofproto->completions, &c->list_node);
4580 ofoperation_complete(rule->up.pending, 0);
4584 static struct rule *
4587 struct rule_dpif *rule = xmalloc(sizeof *rule);
4592 rule_dealloc(struct rule *rule_)
4594 struct rule_dpif *rule = rule_dpif_cast(rule_);
4599 rule_construct(struct rule *rule_)
4601 struct rule_dpif *rule = rule_dpif_cast(rule_);
4602 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4603 struct rule_dpif *victim;
4607 error = ofpacts_check(rule->up.ofpacts, rule->up.ofpacts_len,
4608 &rule->up.cr.flow, ofproto->max_ports);
4613 rule->packet_count = 0;
4614 rule->byte_count = 0;
4616 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
4617 if (victim && !list_is_empty(&victim->facets)) {
4618 struct facet *facet;
4620 rule->facets = victim->facets;
4621 list_moved(&rule->facets);
4622 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4623 /* XXX: We're only clearing our local counters here. It's possible
4624 * that quite a few packets are unaccounted for in the datapath
4625 * statistics. These will be accounted to the new rule instead of
4626 * cleared as required. This could be fixed by clearing out the
4627 * datapath statistics for this facet, but currently it doesn't
4629 facet_reset_counters(facet);
4633 /* Must avoid list_moved() in this case. */
4634 list_init(&rule->facets);
4637 table_id = rule->up.table_id;
4638 rule->tag = (victim ? victim->tag
4640 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
4641 ofproto->tables[table_id].basis));
4643 complete_operation(rule);
4648 rule_destruct(struct rule *rule_)
4650 struct rule_dpif *rule = rule_dpif_cast(rule_);
4651 struct facet *facet, *next_facet;
4653 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4654 facet_revalidate(facet);
4657 complete_operation(rule);
4661 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4663 struct rule_dpif *rule = rule_dpif_cast(rule_);
4664 struct facet *facet;
4666 /* Start from historical data for 'rule' itself that are no longer tracked
4667 * in facets. This counts, for example, facets that have expired. */
4668 *packets = rule->packet_count;
4669 *bytes = rule->byte_count;
4671 /* Add any statistics that are tracked by facets. This includes
4672 * statistical data recently updated by ofproto_update_stats() as well as
4673 * stats for packets that were executed "by hand" via dpif_execute(). */
4674 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4675 *packets += facet->packet_count;
4676 *bytes += facet->byte_count;
4681 rule_execute(struct rule *rule_, const struct flow *flow,
4682 struct ofpbuf *packet)
4684 struct rule_dpif *rule = rule_dpif_cast(rule_);
4685 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4687 struct dpif_flow_stats stats;
4689 struct action_xlate_ctx ctx;
4690 uint64_t odp_actions_stub[1024 / 8];
4691 struct ofpbuf odp_actions;
4693 dpif_flow_stats_extract(flow, packet, &stats);
4694 rule_credit_stats(rule, &stats);
4696 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4697 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
4698 rule, stats.tcp_flags, packet);
4699 ctx.resubmit_stats = &stats;
4700 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
4702 execute_odp_actions(ofproto, flow, odp_actions.data,
4703 odp_actions.size, packet);
4705 ofpbuf_uninit(&odp_actions);
4711 rule_modify_actions(struct rule *rule_)
4713 struct rule_dpif *rule = rule_dpif_cast(rule_);
4714 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4717 error = ofpacts_check(rule->up.ofpacts, rule->up.ofpacts_len,
4718 &rule->up.cr.flow, ofproto->max_ports);
4720 ofoperation_complete(rule->up.pending, error);
4724 complete_operation(rule);
4727 /* Sends 'packet' out 'ofport'.
4728 * May modify 'packet'.
4729 * Returns 0 if successful, otherwise a positive errno value. */
4731 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4733 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4734 struct ofpbuf key, odp_actions;
4735 struct odputil_keybuf keybuf;
4740 flow_extract(packet, 0, 0, 0, &flow);
4741 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4743 if (odp_port != ofport->odp_port) {
4744 eth_pop_vlan(packet);
4745 flow.vlan_tci = htons(0);
4748 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4749 odp_flow_key_from_flow(&key, &flow);
4751 ofpbuf_init(&odp_actions, 32);
4752 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4754 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4755 error = dpif_execute(ofproto->dpif,
4757 odp_actions.data, odp_actions.size,
4759 ofpbuf_uninit(&odp_actions);
4762 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4763 ofproto->up.name, odp_port, strerror(error));
4765 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
4769 /* OpenFlow to datapath action translation. */
4771 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
4772 struct action_xlate_ctx *);
4773 static void xlate_normal(struct action_xlate_ctx *);
4775 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
4776 * The action will state 'slow' as the reason that the action is in the slow
4777 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
4778 * dump-flows" output to see why a flow is in the slow path.)
4780 * The 'stub_size' bytes in 'stub' will be used to store the action.
4781 * 'stub_size' must be large enough for the action.
4783 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
4786 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
4787 enum slow_path_reason slow,
4788 uint64_t *stub, size_t stub_size,
4789 const struct nlattr **actionsp, size_t *actions_lenp)
4791 union user_action_cookie cookie;
4794 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
4795 cookie.slow_path.unused = 0;
4796 cookie.slow_path.reason = slow;
4798 ofpbuf_use_stack(&buf, stub, stub_size);
4799 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
4800 uint32_t pid = dpif_port_get_pid(ofproto->dpif, UINT16_MAX);
4801 odp_put_userspace_action(pid, &cookie, &buf);
4803 put_userspace_action(ofproto, &buf, flow, &cookie);
4805 *actionsp = buf.data;
4806 *actions_lenp = buf.size;
4810 put_userspace_action(const struct ofproto_dpif *ofproto,
4811 struct ofpbuf *odp_actions,
4812 const struct flow *flow,
4813 const union user_action_cookie *cookie)
4817 pid = dpif_port_get_pid(ofproto->dpif,
4818 ofp_port_to_odp_port(flow->in_port));
4820 return odp_put_userspace_action(pid, cookie, odp_actions);
4824 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
4825 ovs_be16 vlan_tci, uint32_t odp_port,
4826 unsigned int n_outputs, union user_action_cookie *cookie)
4830 cookie->type = USER_ACTION_COOKIE_SFLOW;
4831 cookie->sflow.vlan_tci = vlan_tci;
4833 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
4834 * port information") for the interpretation of cookie->output. */
4835 switch (n_outputs) {
4837 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
4838 cookie->sflow.output = 0x40000000 | 256;
4842 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4844 cookie->sflow.output = ifindex;
4849 /* 0x80000000 means "multiple output ports. */
4850 cookie->sflow.output = 0x80000000 | n_outputs;
4855 /* Compose SAMPLE action for sFlow. */
4857 compose_sflow_action(const struct ofproto_dpif *ofproto,
4858 struct ofpbuf *odp_actions,
4859 const struct flow *flow,
4862 uint32_t probability;
4863 union user_action_cookie cookie;
4864 size_t sample_offset, actions_offset;
4867 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4871 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4873 /* Number of packets out of UINT_MAX to sample. */
4874 probability = dpif_sflow_get_probability(ofproto->sflow);
4875 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4877 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4878 compose_sflow_cookie(ofproto, htons(0), odp_port,
4879 odp_port == OVSP_NONE ? 0 : 1, &cookie);
4880 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4882 nl_msg_end_nested(odp_actions, actions_offset);
4883 nl_msg_end_nested(odp_actions, sample_offset);
4884 return cookie_offset;
4887 /* SAMPLE action must be first action in any given list of actions.
4888 * At this point we do not have all information required to build it. So try to
4889 * build sample action as complete as possible. */
4891 add_sflow_action(struct action_xlate_ctx *ctx)
4893 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4895 &ctx->flow, OVSP_NONE);
4896 ctx->sflow_odp_port = 0;
4897 ctx->sflow_n_outputs = 0;
4900 /* Fix SAMPLE action according to data collected while composing ODP actions.
4901 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4902 * USERSPACE action's user-cookie which is required for sflow. */
4904 fix_sflow_action(struct action_xlate_ctx *ctx)
4906 const struct flow *base = &ctx->base_flow;
4907 union user_action_cookie *cookie;
4909 if (!ctx->user_cookie_offset) {
4913 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4915 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4917 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
4918 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
4922 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4925 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4926 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4927 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4928 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4932 struct priority_to_dscp *pdscp;
4934 if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD
4935 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4939 pdscp = get_priority(ofport, ctx->flow.skb_priority);
4941 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4942 ctx->flow.nw_tos |= pdscp->dscp;
4945 /* We may not have an ofport record for this port, but it doesn't hurt
4946 * to allow forwarding to it anyhow. Maybe such a port will appear
4947 * later and we're pre-populating the flow table. */
4950 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4951 ctx->flow.vlan_tci);
4952 if (out_port != odp_port) {
4953 ctx->flow.vlan_tci = htons(0);
4955 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4956 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4958 ctx->sflow_odp_port = odp_port;
4959 ctx->sflow_n_outputs++;
4960 ctx->nf_output_iface = ofp_port;
4961 ctx->flow.vlan_tci = flow_vlan_tci;
4962 ctx->flow.nw_tos = flow_nw_tos;
4966 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4968 compose_output_action__(ctx, ofp_port, true);
4972 xlate_table_action(struct action_xlate_ctx *ctx,
4973 uint16_t in_port, uint8_t table_id)
4975 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4976 struct ofproto_dpif *ofproto = ctx->ofproto;
4977 struct rule_dpif *rule;
4978 uint16_t old_in_port;
4979 uint8_t old_table_id;
4981 old_table_id = ctx->table_id;
4982 ctx->table_id = table_id;
4984 /* Look up a flow with 'in_port' as the input port. */
4985 old_in_port = ctx->flow.in_port;
4986 ctx->flow.in_port = in_port;
4987 rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id);
4990 if (table_id > 0 && table_id < N_TABLES) {
4991 struct table_dpif *table = &ofproto->tables[table_id];
4992 if (table->other_table) {
4993 ctx->tags |= (rule && rule->tag
4995 : rule_calculate_tag(&ctx->flow,
4996 &table->other_table->wc,
5001 /* Restore the original input port. Otherwise OFPP_NORMAL and
5002 * OFPP_IN_PORT will have surprising behavior. */
5003 ctx->flow.in_port = old_in_port;
5005 if (ctx->resubmit_hook) {
5006 ctx->resubmit_hook(ctx, rule);
5010 struct rule_dpif *old_rule = ctx->rule;
5012 if (ctx->resubmit_stats) {
5013 rule_credit_stats(rule, ctx->resubmit_stats);
5018 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
5019 ctx->rule = old_rule;
5023 ctx->table_id = old_table_id;
5025 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5027 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
5028 MAX_RESUBMIT_RECURSION);
5029 ctx->max_resubmit_trigger = true;
5034 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
5035 const struct ofpact_resubmit *resubmit)
5040 in_port = resubmit->in_port;
5041 if (in_port == OFPP_IN_PORT) {
5042 in_port = ctx->flow.in_port;
5045 table_id = resubmit->table_id;
5046 if (table_id == 255) {
5047 table_id = ctx->table_id;
5050 xlate_table_action(ctx, in_port, table_id);
5054 flood_packets(struct action_xlate_ctx *ctx, bool all)
5056 struct ofport_dpif *ofport;
5058 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
5059 uint16_t ofp_port = ofport->up.ofp_port;
5061 if (ofp_port == ctx->flow.in_port) {
5066 compose_output_action__(ctx, ofp_port, false);
5067 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
5068 compose_output_action(ctx, ofp_port);
5072 ctx->nf_output_iface = NF_OUT_FLOOD;
5076 execute_controller_action(struct action_xlate_ctx *ctx, int len,
5077 enum ofp_packet_in_reason reason,
5078 uint16_t controller_id)
5080 struct ofputil_packet_in pin;
5081 struct ofpbuf *packet;
5083 ctx->slow |= SLOW_CONTROLLER;
5088 packet = ofpbuf_clone(ctx->packet);
5090 if (packet->l2 && packet->l3) {
5091 struct eth_header *eh;
5093 eth_pop_vlan(packet);
5096 /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
5097 * LLC frame. Calculating the Ethernet type of these frames is more
5098 * trouble than seems appropriate for a simple assertion. */
5099 assert(ntohs(eh->eth_type) < ETH_TYPE_MIN
5100 || eh->eth_type == ctx->flow.dl_type);
5102 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
5103 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
5105 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
5106 eth_push_vlan(packet, ctx->flow.vlan_tci);
5110 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5111 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
5112 ctx->flow.nw_tos, ctx->flow.nw_ttl);
5116 if (ctx->flow.nw_proto == IPPROTO_TCP) {
5117 packet_set_tcp_port(packet, ctx->flow.tp_src,
5119 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
5120 packet_set_udp_port(packet, ctx->flow.tp_src,
5127 pin.packet = packet->data;
5128 pin.packet_len = packet->size;
5129 pin.reason = reason;
5130 pin.controller_id = controller_id;
5131 pin.table_id = ctx->table_id;
5132 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
5135 flow_get_metadata(&ctx->flow, &pin.fmd);
5137 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
5138 ofpbuf_delete(packet);
5142 compose_dec_ttl(struct action_xlate_ctx *ctx)
5144 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
5145 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
5149 if (ctx->flow.nw_ttl > 1) {
5153 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
5155 /* Stop processing for current table. */
5161 xlate_output_action(struct action_xlate_ctx *ctx,
5162 uint16_t port, uint16_t max_len)
5164 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
5166 ctx->nf_output_iface = NF_OUT_DROP;
5170 compose_output_action(ctx, ctx->flow.in_port);
5173 xlate_table_action(ctx, ctx->flow.in_port, 0);
5179 flood_packets(ctx, false);
5182 flood_packets(ctx, true);
5184 case OFPP_CONTROLLER:
5185 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
5191 if (port != ctx->flow.in_port) {
5192 compose_output_action(ctx, port);
5197 if (prev_nf_output_iface == NF_OUT_FLOOD) {
5198 ctx->nf_output_iface = NF_OUT_FLOOD;
5199 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
5200 ctx->nf_output_iface = prev_nf_output_iface;
5201 } else if (prev_nf_output_iface != NF_OUT_DROP &&
5202 ctx->nf_output_iface != NF_OUT_FLOOD) {
5203 ctx->nf_output_iface = NF_OUT_MULTI;
5208 xlate_output_reg_action(struct action_xlate_ctx *ctx,
5209 const struct ofpact_output_reg *or)
5211 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
5212 if (port <= UINT16_MAX) {
5213 xlate_output_action(ctx, port, or->max_len);
5218 xlate_enqueue_action(struct action_xlate_ctx *ctx,
5219 const struct ofpact_enqueue *enqueue)
5221 uint16_t ofp_port = enqueue->port;
5222 uint32_t queue_id = enqueue->queue;
5223 uint32_t flow_priority, priority;
5226 /* Translate queue to priority. */
5227 error = dpif_queue_to_priority(ctx->ofproto->dpif, queue_id, &priority);
5229 /* Fall back to ordinary output action. */
5230 xlate_output_action(ctx, enqueue->port, 0);
5234 /* Check output port. */
5235 if (ofp_port == OFPP_IN_PORT) {
5236 ofp_port = ctx->flow.in_port;
5237 } else if (ofp_port == ctx->flow.in_port) {
5241 /* Add datapath actions. */
5242 flow_priority = ctx->flow.skb_priority;
5243 ctx->flow.skb_priority = priority;
5244 compose_output_action(ctx, ofp_port);
5245 ctx->flow.skb_priority = flow_priority;
5247 /* Update NetFlow output port. */
5248 if (ctx->nf_output_iface == NF_OUT_DROP) {
5249 ctx->nf_output_iface = ofp_port;
5250 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
5251 ctx->nf_output_iface = NF_OUT_MULTI;
5256 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
5258 uint32_t skb_priority;
5260 if (!dpif_queue_to_priority(ctx->ofproto->dpif, queue_id, &skb_priority)) {
5261 ctx->flow.skb_priority = skb_priority;
5263 /* Couldn't translate queue to a priority. Nothing to do. A warning
5264 * has already been logged. */
5268 struct xlate_reg_state {
5274 xlate_autopath(struct action_xlate_ctx *ctx,
5275 const struct ofpact_autopath *ap)
5277 uint16_t ofp_port = ap->port;
5278 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
5280 if (!port || !port->bundle) {
5281 ofp_port = OFPP_NONE;
5282 } else if (port->bundle->bond) {
5283 /* Autopath does not support VLAN hashing. */
5284 struct ofport_dpif *slave = bond_choose_output_slave(
5285 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
5287 ofp_port = slave->up.ofp_port;
5290 nxm_reg_load(&ap->dst, ofp_port, &ctx->flow);
5294 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
5296 struct ofproto_dpif *ofproto = ofproto_;
5297 struct ofport_dpif *port;
5307 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
5310 port = get_ofp_port(ofproto, ofp_port);
5311 return port ? port->may_enable : false;
5316 xlate_bundle_action(struct action_xlate_ctx *ctx,
5317 const struct ofpact_bundle *bundle)
5321 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
5322 if (bundle->dst.field) {
5323 nxm_reg_load(&bundle->dst, port, &ctx->flow);
5325 xlate_output_action(ctx, port, 0);
5330 xlate_learn_action(struct action_xlate_ctx *ctx,
5331 const struct ofpact_learn *learn)
5333 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
5334 struct ofputil_flow_mod fm;
5335 uint64_t ofpacts_stub[1024 / 8];
5336 struct ofpbuf ofpacts;
5339 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
5340 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
5342 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
5343 if (error && !VLOG_DROP_WARN(&rl)) {
5344 VLOG_WARN("learning action failed to modify flow table (%s)",
5345 ofperr_get_name(error));
5348 ofpbuf_uninit(&ofpacts);
5351 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
5352 * means "infinite". */
5354 reduce_timeout(uint16_t max, uint16_t *timeout)
5356 if (max && (!*timeout || *timeout > max)) {
5362 xlate_fin_timeout(struct action_xlate_ctx *ctx,
5363 const struct ofpact_fin_timeout *oft)
5365 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
5366 struct rule_dpif *rule = ctx->rule;
5368 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
5369 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
5374 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
5376 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
5377 ? OFPUTIL_PC_NO_RECV_STP
5378 : OFPUTIL_PC_NO_RECV)) {
5382 /* Only drop packets here if both forwarding and learning are
5383 * disabled. If just learning is enabled, we need to have
5384 * OFPP_NORMAL and the learning action have a look at the packet
5385 * before we can drop it. */
5386 if (!stp_forward_in_state(port->stp_state)
5387 && !stp_learn_in_state(port->stp_state)) {
5395 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
5396 struct action_xlate_ctx *ctx)
5398 const struct ofport_dpif *port;
5399 bool was_evictable = true;
5400 const struct ofpact *a;
5402 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5403 if (port && !may_receive(port, ctx)) {
5404 /* Drop this flow. */
5409 /* Don't let the rule we're working on get evicted underneath us. */
5410 was_evictable = ctx->rule->up.evictable;
5411 ctx->rule->up.evictable = false;
5413 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
5414 struct ofpact_controller *controller;
5422 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
5423 ofpact_get_OUTPUT(a)->max_len);
5426 case OFPACT_CONTROLLER:
5427 controller = ofpact_get_CONTROLLER(a);
5428 execute_controller_action(ctx, controller->max_len,
5430 controller->controller_id);
5433 case OFPACT_ENQUEUE:
5434 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
5437 case OFPACT_SET_VLAN_VID:
5438 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
5439 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
5443 case OFPACT_SET_VLAN_PCP:
5444 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
5445 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
5450 case OFPACT_STRIP_VLAN:
5451 ctx->flow.vlan_tci = htons(0);
5454 case OFPACT_SET_ETH_SRC:
5455 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
5459 case OFPACT_SET_ETH_DST:
5460 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
5464 case OFPACT_SET_IPV4_SRC:
5465 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
5468 case OFPACT_SET_IPV4_DST:
5469 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
5472 case OFPACT_SET_IPV4_DSCP:
5473 /* OpenFlow 1.0 only supports IPv4. */
5474 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5475 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5476 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
5480 case OFPACT_SET_L4_SRC_PORT:
5481 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
5484 case OFPACT_SET_L4_DST_PORT:
5485 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
5488 case OFPACT_RESUBMIT:
5489 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
5492 case OFPACT_SET_TUNNEL:
5493 ctx->flow.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
5496 case OFPACT_SET_QUEUE:
5497 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
5500 case OFPACT_POP_QUEUE:
5501 ctx->flow.skb_priority = ctx->orig_skb_priority;
5504 case OFPACT_REG_MOVE:
5505 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
5508 case OFPACT_REG_LOAD:
5509 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
5512 case OFPACT_DEC_TTL:
5513 if (compose_dec_ttl(ctx)) {
5519 /* Nothing to do. */
5522 case OFPACT_MULTIPATH:
5523 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
5526 case OFPACT_AUTOPATH:
5527 xlate_autopath(ctx, ofpact_get_AUTOPATH(a));
5531 ctx->ofproto->has_bundle_action = true;
5532 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
5535 case OFPACT_OUTPUT_REG:
5536 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
5540 ctx->has_learn = true;
5541 if (ctx->may_learn) {
5542 xlate_learn_action(ctx, ofpact_get_LEARN(a));
5550 case OFPACT_FIN_TIMEOUT:
5551 ctx->has_fin_timeout = true;
5552 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
5558 /* We've let OFPP_NORMAL and the learning action look at the packet,
5559 * so drop it now if forwarding is disabled. */
5560 if (port && !stp_forward_in_state(port->stp_state)) {
5561 ofpbuf_clear(ctx->odp_actions);
5562 add_sflow_action(ctx);
5565 ctx->rule->up.evictable = was_evictable;
5570 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
5571 struct ofproto_dpif *ofproto, const struct flow *flow,
5572 ovs_be16 initial_tci, struct rule_dpif *rule,
5573 uint8_t tcp_flags, const struct ofpbuf *packet)
5575 ctx->ofproto = ofproto;
5577 ctx->base_flow = ctx->flow;
5578 ctx->base_flow.tun_id = 0;
5579 ctx->base_flow.vlan_tci = initial_tci;
5581 ctx->packet = packet;
5582 ctx->may_learn = packet != NULL;
5583 ctx->tcp_flags = tcp_flags;
5584 ctx->resubmit_hook = NULL;
5585 ctx->report_hook = NULL;
5586 ctx->resubmit_stats = NULL;
5589 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
5590 * into datapath actions in 'odp_actions', using 'ctx'. */
5592 xlate_actions(struct action_xlate_ctx *ctx,
5593 const struct ofpact *ofpacts, size_t ofpacts_len,
5594 struct ofpbuf *odp_actions)
5596 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
5597 * that in the future we always keep a copy of the original flow for
5598 * tracing purposes. */
5599 static bool hit_resubmit_limit;
5601 enum slow_path_reason special;
5603 COVERAGE_INC(ofproto_dpif_xlate);
5605 ofpbuf_clear(odp_actions);
5606 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
5608 ctx->odp_actions = odp_actions;
5611 ctx->has_learn = false;
5612 ctx->has_normal = false;
5613 ctx->has_fin_timeout = false;
5614 ctx->nf_output_iface = NF_OUT_DROP;
5617 ctx->max_resubmit_trigger = false;
5618 ctx->orig_skb_priority = ctx->flow.skb_priority;
5622 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
5623 /* Do this conditionally because the copy is expensive enough that it
5624 * shows up in profiles.
5626 * We keep orig_flow in 'ctx' only because I couldn't make GCC 4.4
5627 * believe that I wasn't using it without initializing it if I kept it
5628 * in a local variable. */
5629 ctx->orig_flow = ctx->flow;
5632 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
5633 switch (ctx->ofproto->up.frag_handling) {
5634 case OFPC_FRAG_NORMAL:
5635 /* We must pretend that transport ports are unavailable. */
5636 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
5637 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
5640 case OFPC_FRAG_DROP:
5643 case OFPC_FRAG_REASM:
5646 case OFPC_FRAG_NX_MATCH:
5647 /* Nothing to do. */
5650 case OFPC_INVALID_TTL_TO_CONTROLLER:
5655 special = process_special(ctx->ofproto, &ctx->flow, ctx->packet);
5657 ctx->slow |= special;
5659 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5660 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
5662 add_sflow_action(ctx);
5663 do_xlate_actions(ofpacts, ofpacts_len, ctx);
5665 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
5666 if (!hit_resubmit_limit) {
5667 /* We didn't record the original flow. Make sure we do from
5669 hit_resubmit_limit = true;
5670 } else if (!VLOG_DROP_ERR(&trace_rl)) {
5671 struct ds ds = DS_EMPTY_INITIALIZER;
5673 ofproto_trace(ctx->ofproto, &ctx->orig_flow, ctx->packet,
5675 VLOG_ERR("Trace triggered by excessive resubmit "
5676 "recursion:\n%s", ds_cstr(&ds));
5681 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
5682 ctx->odp_actions->data,
5683 ctx->odp_actions->size)) {
5684 ctx->slow |= SLOW_IN_BAND;
5686 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
5688 compose_output_action(ctx, OFPP_LOCAL);
5691 if (ctx->ofproto->has_mirrors) {
5692 add_mirror_actions(ctx, &ctx->orig_flow);
5694 fix_sflow_action(ctx);
5698 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
5699 * into datapath actions, using 'ctx', and discards the datapath actions. */
5701 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
5702 const struct ofpact *ofpacts,
5705 uint64_t odp_actions_stub[1024 / 8];
5706 struct ofpbuf odp_actions;
5708 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5709 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
5710 ofpbuf_uninit(&odp_actions);
5714 xlate_report(struct action_xlate_ctx *ctx, const char *s)
5716 if (ctx->report_hook) {
5717 ctx->report_hook(ctx, s);
5721 /* OFPP_NORMAL implementation. */
5723 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
5725 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
5726 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
5727 * the bundle on which the packet was received, returns the VLAN to which the
5730 * Both 'vid' and the return value are in the range 0...4095. */
5732 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
5734 switch (in_bundle->vlan_mode) {
5735 case PORT_VLAN_ACCESS:
5736 return in_bundle->vlan;
5739 case PORT_VLAN_TRUNK:
5742 case PORT_VLAN_NATIVE_UNTAGGED:
5743 case PORT_VLAN_NATIVE_TAGGED:
5744 return vid ? vid : in_bundle->vlan;
5751 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
5752 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
5755 * 'vid' should be the VID obtained from the 802.1Q header that was received as
5756 * part of a packet (specify 0 if there was no 802.1Q header), in the range
5759 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
5761 /* Allow any VID on the OFPP_NONE port. */
5762 if (in_bundle == &ofpp_none_bundle) {
5766 switch (in_bundle->vlan_mode) {
5767 case PORT_VLAN_ACCESS:
5770 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5771 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
5772 "packet received on port %s configured as VLAN "
5773 "%"PRIu16" access port",
5774 in_bundle->ofproto->up.name, vid,
5775 in_bundle->name, in_bundle->vlan);
5781 case PORT_VLAN_NATIVE_UNTAGGED:
5782 case PORT_VLAN_NATIVE_TAGGED:
5784 /* Port must always carry its native VLAN. */
5788 case PORT_VLAN_TRUNK:
5789 if (!ofbundle_includes_vlan(in_bundle, vid)) {
5791 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5792 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
5793 "received on port %s not configured for trunking "
5795 in_bundle->ofproto->up.name, vid,
5796 in_bundle->name, vid);
5808 /* Given 'vlan', the VLAN that a packet belongs to, and
5809 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
5810 * that should be included in the 802.1Q header. (If the return value is 0,
5811 * then the 802.1Q header should only be included in the packet if there is a
5814 * Both 'vlan' and the return value are in the range 0...4095. */
5816 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
5818 switch (out_bundle->vlan_mode) {
5819 case PORT_VLAN_ACCESS:
5822 case PORT_VLAN_TRUNK:
5823 case PORT_VLAN_NATIVE_TAGGED:
5826 case PORT_VLAN_NATIVE_UNTAGGED:
5827 return vlan == out_bundle->vlan ? 0 : vlan;
5835 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
5838 struct ofport_dpif *port;
5840 ovs_be16 tci, old_tci;
5842 vid = output_vlan_to_vid(out_bundle, vlan);
5843 if (!out_bundle->bond) {
5844 port = ofbundle_get_a_port(out_bundle);
5846 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
5849 /* No slaves enabled, so drop packet. */
5854 old_tci = ctx->flow.vlan_tci;
5856 if (tci || out_bundle->use_priority_tags) {
5857 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
5859 tci |= htons(VLAN_CFI);
5862 ctx->flow.vlan_tci = tci;
5864 compose_output_action(ctx, port->up.ofp_port);
5865 ctx->flow.vlan_tci = old_tci;
5869 mirror_mask_ffs(mirror_mask_t mask)
5871 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
5876 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
5878 return (bundle->vlan_mode != PORT_VLAN_ACCESS
5879 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
5883 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
5885 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
5888 /* Returns an arbitrary interface within 'bundle'. */
5889 static struct ofport_dpif *
5890 ofbundle_get_a_port(const struct ofbundle *bundle)
5892 return CONTAINER_OF(list_front(&bundle->ports),
5893 struct ofport_dpif, bundle_node);
5897 vlan_is_mirrored(const struct ofmirror *m, int vlan)
5899 return !m->vlans || bitmap_is_set(m->vlans, vlan);
5903 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5905 struct ofproto_dpif *ofproto = ctx->ofproto;
5906 mirror_mask_t mirrors;
5907 struct ofbundle *in_bundle;
5910 const struct nlattr *a;
5913 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5914 ctx->packet != NULL, NULL);
5918 mirrors = in_bundle->src_mirrors;
5920 /* Drop frames on bundles reserved for mirroring. */
5921 if (in_bundle->mirror_out) {
5922 if (ctx->packet != NULL) {
5923 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5924 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5925 "%s, which is reserved exclusively for mirroring",
5926 ctx->ofproto->up.name, in_bundle->name);
5932 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5933 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5936 vlan = input_vid_to_vlan(in_bundle, vid);
5938 /* Look at the output ports to check for destination selections. */
5940 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5941 ctx->odp_actions->size) {
5942 enum ovs_action_attr type = nl_attr_type(a);
5943 struct ofport_dpif *ofport;
5945 if (type != OVS_ACTION_ATTR_OUTPUT) {
5949 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5950 if (ofport && ofport->bundle) {
5951 mirrors |= ofport->bundle->dst_mirrors;
5959 /* Restore the original packet before adding the mirror actions. */
5960 ctx->flow = *orig_flow;
5965 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5967 if (!vlan_is_mirrored(m, vlan)) {
5968 mirrors &= mirrors - 1;
5972 mirrors &= ~m->dup_mirrors;
5973 ctx->mirrors |= m->dup_mirrors;
5975 output_normal(ctx, m->out, vlan);
5976 } else if (vlan != m->out_vlan
5977 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
5978 struct ofbundle *bundle;
5980 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5981 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5982 && !bundle->mirror_out) {
5983 output_normal(ctx, bundle, m->out_vlan);
5991 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5992 uint64_t packets, uint64_t bytes)
5998 for (; mirrors; mirrors &= mirrors - 1) {
6001 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6004 /* In normal circumstances 'm' will not be NULL. However,
6005 * if mirrors are reconfigured, we can temporarily get out
6006 * of sync in facet_revalidate(). We could "correct" the
6007 * mirror list before reaching here, but doing that would
6008 * not properly account the traffic stats we've currently
6009 * accumulated for previous mirror configuration. */
6013 m->packet_count += packets;
6014 m->byte_count += bytes;
6018 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
6019 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
6020 * indicate this; newer upstream kernels use gratuitous ARP requests. */
6022 is_gratuitous_arp(const struct flow *flow)
6024 return (flow->dl_type == htons(ETH_TYPE_ARP)
6025 && eth_addr_is_broadcast(flow->dl_dst)
6026 && (flow->nw_proto == ARP_OP_REPLY
6027 || (flow->nw_proto == ARP_OP_REQUEST
6028 && flow->nw_src == flow->nw_dst)));
6032 update_learning_table(struct ofproto_dpif *ofproto,
6033 const struct flow *flow, int vlan,
6034 struct ofbundle *in_bundle)
6036 struct mac_entry *mac;
6038 /* Don't learn the OFPP_NONE port. */
6039 if (in_bundle == &ofpp_none_bundle) {
6043 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
6047 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
6048 if (is_gratuitous_arp(flow)) {
6049 /* We don't want to learn from gratuitous ARP packets that are
6050 * reflected back over bond slaves so we lock the learning table. */
6051 if (!in_bundle->bond) {
6052 mac_entry_set_grat_arp_lock(mac);
6053 } else if (mac_entry_is_grat_arp_locked(mac)) {
6058 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
6059 /* The log messages here could actually be useful in debugging,
6060 * so keep the rate limit relatively high. */
6061 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
6062 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
6063 "on port %s in VLAN %d",
6064 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
6065 in_bundle->name, vlan);
6067 mac->port.p = in_bundle;
6068 tag_set_add(&ofproto->revalidate_set,
6069 mac_learning_changed(ofproto->ml, mac));
6073 static struct ofbundle *
6074 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
6075 bool warn, struct ofport_dpif **in_ofportp)
6077 struct ofport_dpif *ofport;
6079 /* Find the port and bundle for the received packet. */
6080 ofport = get_ofp_port(ofproto, in_port);
6082 *in_ofportp = ofport;
6084 if (ofport && ofport->bundle) {
6085 return ofport->bundle;
6088 /* Special-case OFPP_NONE, which a controller may use as the ingress
6089 * port for traffic that it is sourcing. */
6090 if (in_port == OFPP_NONE) {
6091 return &ofpp_none_bundle;
6094 /* Odd. A few possible reasons here:
6096 * - We deleted a port but there are still a few packets queued up
6099 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
6100 * we don't know about.
6102 * - The ofproto client didn't configure the port as part of a bundle.
6103 * This is particularly likely to happen if a packet was received on the
6104 * port after it was created, but before the client had a chance to
6105 * configure its bundle.
6108 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6110 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
6111 "port %"PRIu16, ofproto->up.name, in_port);
6116 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
6117 * dropped. Returns true if they may be forwarded, false if they should be
6120 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
6121 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
6123 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
6124 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
6125 * checked by input_vid_is_valid().
6127 * May also add tags to '*tags', although the current implementation only does
6128 * so in one special case.
6131 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
6134 struct ofproto_dpif *ofproto = ctx->ofproto;
6135 struct flow *flow = &ctx->flow;
6136 struct ofbundle *in_bundle = in_port->bundle;
6138 /* Drop frames for reserved multicast addresses
6139 * only if forward_bpdu option is absent. */
6140 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
6141 xlate_report(ctx, "packet has reserved destination MAC, dropping");
6145 if (in_bundle->bond) {
6146 struct mac_entry *mac;
6148 switch (bond_check_admissibility(in_bundle->bond, in_port,
6149 flow->dl_dst, &ctx->tags)) {
6154 xlate_report(ctx, "bonding refused admissibility, dropping");
6157 case BV_DROP_IF_MOVED:
6158 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
6159 if (mac && mac->port.p != in_bundle &&
6160 (!is_gratuitous_arp(flow)
6161 || mac_entry_is_grat_arp_locked(mac))) {
6162 xlate_report(ctx, "SLB bond thinks this packet looped back, "
6174 xlate_normal(struct action_xlate_ctx *ctx)
6176 struct ofport_dpif *in_port;
6177 struct ofbundle *in_bundle;
6178 struct mac_entry *mac;
6182 ctx->has_normal = true;
6184 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
6185 ctx->packet != NULL, &in_port);
6187 xlate_report(ctx, "no input bundle, dropping");
6191 /* Drop malformed frames. */
6192 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
6193 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
6194 if (ctx->packet != NULL) {
6195 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6196 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
6197 "VLAN tag received on port %s",
6198 ctx->ofproto->up.name, in_bundle->name);
6200 xlate_report(ctx, "partial VLAN tag, dropping");
6204 /* Drop frames on bundles reserved for mirroring. */
6205 if (in_bundle->mirror_out) {
6206 if (ctx->packet != NULL) {
6207 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6208 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6209 "%s, which is reserved exclusively for mirroring",
6210 ctx->ofproto->up.name, in_bundle->name);
6212 xlate_report(ctx, "input port is mirror output port, dropping");
6217 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
6218 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6219 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
6222 vlan = input_vid_to_vlan(in_bundle, vid);
6224 /* Check other admissibility requirements. */
6225 if (in_port && !is_admissible(ctx, in_port, vlan)) {
6229 /* Learn source MAC. */
6230 if (ctx->may_learn) {
6231 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
6234 /* Determine output bundle. */
6235 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
6238 if (mac->port.p != in_bundle) {
6239 xlate_report(ctx, "forwarding to learned port");
6240 output_normal(ctx, mac->port.p, vlan);
6242 xlate_report(ctx, "learned port is input port, dropping");
6245 struct ofbundle *bundle;
6247 xlate_report(ctx, "no learned MAC for destination, flooding");
6248 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
6249 if (bundle != in_bundle
6250 && ofbundle_includes_vlan(bundle, vlan)
6251 && bundle->floodable
6252 && !bundle->mirror_out) {
6253 output_normal(ctx, bundle, vlan);
6256 ctx->nf_output_iface = NF_OUT_FLOOD;
6260 /* Optimized flow revalidation.
6262 * It's a difficult problem, in general, to tell which facets need to have
6263 * their actions recalculated whenever the OpenFlow flow table changes. We
6264 * don't try to solve that general problem: for most kinds of OpenFlow flow
6265 * table changes, we recalculate the actions for every facet. This is
6266 * relatively expensive, but it's good enough if the OpenFlow flow table
6267 * doesn't change very often.
6269 * However, we can expect one particular kind of OpenFlow flow table change to
6270 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
6271 * of CPU on revalidating every facet whenever MAC learning modifies the flow
6272 * table, we add a special case that applies to flow tables in which every rule
6273 * has the same form (that is, the same wildcards), except that the table is
6274 * also allowed to have a single "catch-all" flow that matches all packets. We
6275 * optimize this case by tagging all of the facets that resubmit into the table
6276 * and invalidating the same tag whenever a flow changes in that table. The
6277 * end result is that we revalidate just the facets that need it (and sometimes
6278 * a few more, but not all of the facets or even all of the facets that
6279 * resubmit to the table modified by MAC learning). */
6281 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
6282 * into an OpenFlow table with the given 'basis'. */
6284 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
6287 if (flow_wildcards_is_catchall(wc)) {
6290 struct flow tag_flow = *flow;
6291 flow_zero_wildcards(&tag_flow, wc);
6292 return tag_create_deterministic(flow_hash(&tag_flow, secret));
6296 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
6297 * taggability of that table.
6299 * This function must be called after *each* change to a flow table. If you
6300 * skip calling it on some changes then the pointer comparisons at the end can
6301 * be invalid if you get unlucky. For example, if a flow removal causes a
6302 * cls_table to be destroyed and then a flow insertion causes a cls_table with
6303 * different wildcards to be created with the same address, then this function
6304 * will incorrectly skip revalidation. */
6306 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
6308 struct table_dpif *table = &ofproto->tables[table_id];
6309 const struct oftable *oftable = &ofproto->up.tables[table_id];
6310 struct cls_table *catchall, *other;
6311 struct cls_table *t;
6313 catchall = other = NULL;
6315 switch (hmap_count(&oftable->cls.tables)) {
6317 /* We could tag this OpenFlow table but it would make the logic a
6318 * little harder and it's a corner case that doesn't seem worth it
6324 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
6325 if (cls_table_is_catchall(t)) {
6327 } else if (!other) {
6330 /* Indicate that we can't tag this by setting both tables to
6331 * NULL. (We know that 'catchall' is already NULL.) */
6338 /* Can't tag this table. */
6342 if (table->catchall_table != catchall || table->other_table != other) {
6343 table->catchall_table = catchall;
6344 table->other_table = other;
6345 ofproto->need_revalidate = REV_FLOW_TABLE;
6349 /* Given 'rule' that has changed in some way (either it is a rule being
6350 * inserted, a rule being deleted, or a rule whose actions are being
6351 * modified), marks facets for revalidation to ensure that packets will be
6352 * forwarded correctly according to the new state of the flow table.
6354 * This function must be called after *each* change to a flow table. See
6355 * the comment on table_update_taggable() for more information. */
6357 rule_invalidate(const struct rule_dpif *rule)
6359 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
6361 table_update_taggable(ofproto, rule->up.table_id);
6363 if (!ofproto->need_revalidate) {
6364 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
6366 if (table->other_table && rule->tag) {
6367 tag_set_add(&ofproto->revalidate_set, rule->tag);
6369 ofproto->need_revalidate = REV_FLOW_TABLE;
6375 set_frag_handling(struct ofproto *ofproto_,
6376 enum ofp_config_flags frag_handling)
6378 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6380 if (frag_handling != OFPC_FRAG_REASM) {
6381 ofproto->need_revalidate = REV_RECONFIGURE;
6389 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
6390 const struct flow *flow,
6391 const struct ofpact *ofpacts, size_t ofpacts_len)
6393 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6396 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
6397 return OFPERR_NXBRC_BAD_IN_PORT;
6400 error = ofpacts_check(ofpacts, ofpacts_len, flow, ofproto->max_ports);
6402 struct odputil_keybuf keybuf;
6403 struct dpif_flow_stats stats;
6407 struct action_xlate_ctx ctx;
6408 uint64_t odp_actions_stub[1024 / 8];
6409 struct ofpbuf odp_actions;
6411 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
6412 odp_flow_key_from_flow(&key, flow);
6414 dpif_flow_stats_extract(flow, packet, &stats);
6416 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL,
6417 packet_get_tcp_flags(packet, flow), packet);
6418 ctx.resubmit_stats = &stats;
6420 ofpbuf_use_stub(&odp_actions,
6421 odp_actions_stub, sizeof odp_actions_stub);
6422 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
6423 dpif_execute(ofproto->dpif, key.data, key.size,
6424 odp_actions.data, odp_actions.size, packet);
6425 ofpbuf_uninit(&odp_actions);
6433 set_netflow(struct ofproto *ofproto_,
6434 const struct netflow_options *netflow_options)
6436 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6438 if (netflow_options) {
6439 if (!ofproto->netflow) {
6440 ofproto->netflow = netflow_create();
6442 return netflow_set_options(ofproto->netflow, netflow_options);
6444 netflow_destroy(ofproto->netflow);
6445 ofproto->netflow = NULL;
6451 get_netflow_ids(const struct ofproto *ofproto_,
6452 uint8_t *engine_type, uint8_t *engine_id)
6454 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6456 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
6460 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
6462 if (!facet_is_controller_flow(facet) &&
6463 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
6464 struct subfacet *subfacet;
6465 struct ofexpired expired;
6467 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
6468 if (subfacet->path == SF_FAST_PATH) {
6469 struct dpif_flow_stats stats;
6471 subfacet_reinstall(subfacet, &stats);
6472 subfacet_update_stats(subfacet, &stats);
6476 expired.flow = facet->flow;
6477 expired.packet_count = facet->packet_count;
6478 expired.byte_count = facet->byte_count;
6479 expired.used = facet->used;
6480 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
6485 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
6487 struct facet *facet;
6489 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6490 send_active_timeout(ofproto, facet);
6494 static struct ofproto_dpif *
6495 ofproto_dpif_lookup(const char *name)
6497 struct ofproto_dpif *ofproto;
6499 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
6500 hash_string(name, 0), &all_ofproto_dpifs) {
6501 if (!strcmp(ofproto->up.name, name)) {
6509 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
6510 const char *argv[], void *aux OVS_UNUSED)
6512 struct ofproto_dpif *ofproto;
6515 ofproto = ofproto_dpif_lookup(argv[1]);
6517 unixctl_command_reply_error(conn, "no such bridge");
6520 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6522 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6523 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6527 unixctl_command_reply(conn, "table successfully flushed");
6531 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6532 const char *argv[], void *aux OVS_UNUSED)
6534 struct ds ds = DS_EMPTY_INITIALIZER;
6535 const struct ofproto_dpif *ofproto;
6536 const struct mac_entry *e;
6538 ofproto = ofproto_dpif_lookup(argv[1]);
6540 unixctl_command_reply_error(conn, "no such bridge");
6544 ds_put_cstr(&ds, " port VLAN MAC Age\n");
6545 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
6546 struct ofbundle *bundle = e->port.p;
6547 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
6548 ofbundle_get_a_port(bundle)->odp_port,
6549 e->vlan, ETH_ADDR_ARGS(e->mac),
6550 mac_entry_age(ofproto->ml, e));
6552 unixctl_command_reply(conn, ds_cstr(&ds));
6557 struct action_xlate_ctx ctx;
6563 trace_format_rule(struct ds *result, uint8_t table_id, int level,
6564 const struct rule_dpif *rule)
6566 ds_put_char_multiple(result, '\t', level);
6568 ds_put_cstr(result, "No match\n");
6572 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
6573 table_id, ntohll(rule->up.flow_cookie));
6574 cls_rule_format(&rule->up.cr, result);
6575 ds_put_char(result, '\n');
6577 ds_put_char_multiple(result, '\t', level);
6578 ds_put_cstr(result, "OpenFlow ");
6579 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
6580 ds_put_char(result, '\n');
6584 trace_format_flow(struct ds *result, int level, const char *title,
6585 struct trace_ctx *trace)
6587 ds_put_char_multiple(result, '\t', level);
6588 ds_put_format(result, "%s: ", title);
6589 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
6590 ds_put_cstr(result, "unchanged");
6592 flow_format(result, &trace->ctx.flow);
6593 trace->flow = trace->ctx.flow;
6595 ds_put_char(result, '\n');
6599 trace_format_regs(struct ds *result, int level, const char *title,
6600 struct trace_ctx *trace)
6604 ds_put_char_multiple(result, '\t', level);
6605 ds_put_format(result, "%s:", title);
6606 for (i = 0; i < FLOW_N_REGS; i++) {
6607 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
6609 ds_put_char(result, '\n');
6613 trace_format_odp(struct ds *result, int level, const char *title,
6614 struct trace_ctx *trace)
6616 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
6618 ds_put_char_multiple(result, '\t', level);
6619 ds_put_format(result, "%s: ", title);
6620 format_odp_actions(result, odp_actions->data, odp_actions->size);
6621 ds_put_char(result, '\n');
6625 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
6627 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
6628 struct ds *result = trace->result;
6630 ds_put_char(result, '\n');
6631 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
6632 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
6633 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
6634 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
6638 trace_report(struct action_xlate_ctx *ctx, const char *s)
6640 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
6641 struct ds *result = trace->result;
6643 ds_put_char_multiple(result, '\t', ctx->recurse);
6644 ds_put_cstr(result, s);
6645 ds_put_char(result, '\n');
6649 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
6650 void *aux OVS_UNUSED)
6652 const char *dpname = argv[1];
6653 struct ofproto_dpif *ofproto;
6654 struct ofpbuf odp_key;
6655 struct ofpbuf *packet;
6656 ovs_be16 initial_tci;
6662 ofpbuf_init(&odp_key, 0);
6665 ofproto = ofproto_dpif_lookup(dpname);
6667 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
6671 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
6672 /* ofproto/trace dpname flow [-generate] */
6673 const char *flow_s = argv[2];
6674 const char *generate_s = argv[3];
6676 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
6677 * flow. We guess which type it is based on whether 'flow_s' contains
6678 * an '(', since a datapath flow always contains '(') but an
6679 * OpenFlow-like flow should not (in fact it's allowed but I believe
6680 * that's not documented anywhere).
6682 * An alternative would be to try to parse 'flow_s' both ways, but then
6683 * it would be tricky giving a sensible error message. After all, do
6684 * you just say "syntax error" or do you present both error messages?
6685 * Both choices seem lousy. */
6686 if (strchr(flow_s, '(')) {
6689 /* Convert string to datapath key. */
6690 ofpbuf_init(&odp_key, 0);
6691 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
6693 unixctl_command_reply_error(conn, "Bad flow syntax");
6697 /* Convert odp_key to flow. */
6698 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
6699 odp_key.size, &flow,
6700 &initial_tci, NULL);
6701 if (error == ODP_FIT_ERROR) {
6702 unixctl_command_reply_error(conn, "Invalid flow");
6708 error_s = parse_ofp_exact_flow(&flow, argv[2]);
6710 unixctl_command_reply_error(conn, error_s);
6715 initial_tci = flow.vlan_tci;
6716 vsp_adjust_flow(ofproto, &flow);
6719 /* Generate a packet, if requested. */
6721 packet = ofpbuf_new(0);
6722 flow_compose(packet, &flow);
6724 } else if (argc == 6) {
6725 /* ofproto/trace dpname priority tun_id in_port packet */
6726 const char *priority_s = argv[2];
6727 const char *tun_id_s = argv[3];
6728 const char *in_port_s = argv[4];
6729 const char *packet_s = argv[5];
6730 uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s));
6731 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
6732 uint32_t priority = atoi(priority_s);
6735 msg = eth_from_hex(packet_s, &packet);
6737 unixctl_command_reply_error(conn, msg);
6741 ds_put_cstr(&result, "Packet: ");
6742 s = ofp_packet_to_string(packet->data, packet->size);
6743 ds_put_cstr(&result, s);
6746 flow_extract(packet, priority, tun_id, in_port, &flow);
6747 initial_tci = flow.vlan_tci;
6749 unixctl_command_reply_error(conn, "Bad command syntax");
6753 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
6754 unixctl_command_reply(conn, ds_cstr(&result));
6757 ds_destroy(&result);
6758 ofpbuf_delete(packet);
6759 ofpbuf_uninit(&odp_key);
6763 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
6764 const struct ofpbuf *packet, ovs_be16 initial_tci,
6767 struct rule_dpif *rule;
6769 ds_put_cstr(ds, "Flow: ");
6770 flow_format(ds, flow);
6771 ds_put_char(ds, '\n');
6773 rule = rule_dpif_lookup(ofproto, flow);
6775 trace_format_rule(ds, 0, 0, rule);
6776 if (rule == ofproto->miss_rule) {
6777 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
6778 } else if (rule == ofproto->no_packet_in_rule) {
6779 ds_put_cstr(ds, "\nNo match, packets dropped because "
6780 "OFPPC_NO_PACKET_IN is set on in_port.\n");
6784 uint64_t odp_actions_stub[1024 / 8];
6785 struct ofpbuf odp_actions;
6787 struct trace_ctx trace;
6790 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
6793 ofpbuf_use_stub(&odp_actions,
6794 odp_actions_stub, sizeof odp_actions_stub);
6795 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
6796 rule, tcp_flags, packet);
6797 trace.ctx.resubmit_hook = trace_resubmit;
6798 trace.ctx.report_hook = trace_report;
6799 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
6802 ds_put_char(ds, '\n');
6803 trace_format_flow(ds, 0, "Final flow", &trace);
6804 ds_put_cstr(ds, "Datapath actions: ");
6805 format_odp_actions(ds, odp_actions.data, odp_actions.size);
6806 ofpbuf_uninit(&odp_actions);
6808 if (trace.ctx.slow) {
6809 enum slow_path_reason slow;
6811 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
6812 "slow path because it:");
6813 for (slow = trace.ctx.slow; slow; ) {
6814 enum slow_path_reason bit = rightmost_1bit(slow);
6818 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
6821 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
6824 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
6827 ds_put_cstr(ds, "\n\t- Needs in-band special case "
6830 ds_put_cstr(ds, "\n\t (The datapath actions are "
6831 "incomplete--for complete actions, "
6832 "please supply a packet.)");
6835 case SLOW_CONTROLLER:
6836 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
6837 "to the OpenFlow controller.");
6840 ds_put_cstr(ds, "\n\t- Needs more specific matching "
6841 "than the datapath supports.");
6848 if (slow & ~SLOW_MATCH) {
6849 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
6850 "the special slow-path processing.");
6857 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6858 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6861 unixctl_command_reply(conn, NULL);
6865 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6866 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6869 unixctl_command_reply(conn, NULL);
6872 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
6873 * 'reply' describing the results. */
6875 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
6877 struct facet *facet;
6881 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6882 if (!facet_check_consistency(facet)) {
6887 ofproto->need_revalidate = REV_INCONSISTENCY;
6891 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
6892 ofproto->up.name, errors);
6894 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
6899 ofproto_dpif_self_check(struct unixctl_conn *conn,
6900 int argc, const char *argv[], void *aux OVS_UNUSED)
6902 struct ds reply = DS_EMPTY_INITIALIZER;
6903 struct ofproto_dpif *ofproto;
6906 ofproto = ofproto_dpif_lookup(argv[1]);
6908 unixctl_command_reply_error(conn, "Unknown ofproto (use "
6909 "ofproto/list for help)");
6912 ofproto_dpif_self_check__(ofproto, &reply);
6914 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6915 ofproto_dpif_self_check__(ofproto, &reply);
6919 unixctl_command_reply(conn, ds_cstr(&reply));
6924 ofproto_dpif_unixctl_init(void)
6926 static bool registered;
6932 unixctl_command_register(
6934 "bridge {tun_id in_port packet | odp_flow [-generate]}",
6935 2, 5, ofproto_unixctl_trace, NULL);
6936 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6937 ofproto_unixctl_fdb_flush, NULL);
6938 unixctl_command_register("fdb/show", "bridge", 1, 1,
6939 ofproto_unixctl_fdb_show, NULL);
6940 unixctl_command_register("ofproto/clog", "", 0, 0,
6941 ofproto_dpif_clog, NULL);
6942 unixctl_command_register("ofproto/unclog", "", 0, 0,
6943 ofproto_dpif_unclog, NULL);
6944 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6945 ofproto_dpif_self_check, NULL);
6948 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6950 * This is deprecated. It is only for compatibility with broken device drivers
6951 * in old versions of Linux that do not properly support VLANs when VLAN
6952 * devices are not used. When broken device drivers are no longer in
6953 * widespread use, we will delete these interfaces. */
6956 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
6958 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6959 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6961 if (realdev_ofp_port == ofport->realdev_ofp_port
6962 && vid == ofport->vlandev_vid) {
6966 ofproto->need_revalidate = REV_RECONFIGURE;
6968 if (ofport->realdev_ofp_port) {
6971 if (realdev_ofp_port && ofport->bundle) {
6972 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6973 * themselves be part of a bundle. */
6974 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6977 ofport->realdev_ofp_port = realdev_ofp_port;
6978 ofport->vlandev_vid = vid;
6980 if (realdev_ofp_port) {
6981 vsp_add(ofport, realdev_ofp_port, vid);
6988 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
6990 return hash_2words(realdev_ofp_port, vid);
6993 /* Returns the ODP port number of the Linux VLAN device that corresponds to
6994 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
6995 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
6996 * it would return the port number of eth0.9.
6998 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
6999 * function just returns its 'realdev_odp_port' argument. */
7001 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
7002 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
7004 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
7005 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
7006 int vid = vlan_tci_to_vid(vlan_tci);
7007 const struct vlan_splinter *vsp;
7009 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
7010 hash_realdev_vid(realdev_ofp_port, vid),
7011 &ofproto->realdev_vid_map) {
7012 if (vsp->realdev_ofp_port == realdev_ofp_port
7013 && vsp->vid == vid) {
7014 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
7018 return realdev_odp_port;
7021 static struct vlan_splinter *
7022 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
7024 struct vlan_splinter *vsp;
7026 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
7027 &ofproto->vlandev_map) {
7028 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
7036 /* Returns the OpenFlow port number of the "real" device underlying the Linux
7037 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
7038 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
7039 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
7040 * eth0 and store 9 in '*vid'.
7042 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
7043 * VLAN device. Unless VLAN splinters are enabled, this is what this function
7046 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
7047 uint16_t vlandev_ofp_port, int *vid)
7049 if (!hmap_is_empty(&ofproto->vlandev_map)) {
7050 const struct vlan_splinter *vsp;
7052 vsp = vlandev_find(ofproto, vlandev_ofp_port);
7057 return vsp->realdev_ofp_port;
7063 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
7064 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
7065 * 'flow->in_port' to the "real" device backing the VLAN device, sets
7066 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
7067 * always the case unless VLAN splinters are enabled), returns false without
7068 * making any changes. */
7070 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
7075 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
7080 /* Cause the flow to be processed as if it came in on the real device with
7081 * the VLAN device's VLAN ID. */
7082 flow->in_port = realdev;
7083 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
7088 vsp_remove(struct ofport_dpif *port)
7090 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7091 struct vlan_splinter *vsp;
7093 vsp = vlandev_find(ofproto, port->up.ofp_port);
7095 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
7096 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
7099 port->realdev_ofp_port = 0;
7101 VLOG_ERR("missing vlan device record");
7106 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
7108 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7110 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
7111 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
7112 == realdev_ofp_port)) {
7113 struct vlan_splinter *vsp;
7115 vsp = xmalloc(sizeof *vsp);
7116 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
7117 hash_int(port->up.ofp_port, 0));
7118 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
7119 hash_realdev_vid(realdev_ofp_port, vid));
7120 vsp->realdev_ofp_port = realdev_ofp_port;
7121 vsp->vlandev_ofp_port = port->up.ofp_port;
7124 port->realdev_ofp_port = realdev_ofp_port;
7126 VLOG_ERR("duplicate vlan device record");
7130 const struct ofproto_class ofproto_dpif_class = {
7160 port_is_lacp_current,
7161 NULL, /* rule_choose_table */
7168 rule_modify_actions,
7176 get_cfm_remote_mpids,
7181 get_stp_port_status,
7188 is_mirror_output_bundle,
7189 forward_bpdu_changed,