2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-provider.h"
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"
54 #include "unaligned.h"
56 #include "vlan-bitmap.h"
59 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
61 COVERAGE_DEFINE(ofproto_dpif_expired);
62 COVERAGE_DEFINE(ofproto_dpif_xlate);
63 COVERAGE_DEFINE(facet_changed_rule);
64 COVERAGE_DEFINE(facet_revalidate);
65 COVERAGE_DEFINE(facet_unexpected);
66 COVERAGE_DEFINE(facet_suppress);
68 /* Maximum depth of flow table recursion (due to resubmit actions) in a
69 * flow translation. */
70 #define MAX_RESUBMIT_RECURSION 64
72 /* Number of implemented OpenFlow tables. */
73 enum { N_TABLES = 255 };
74 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
75 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
86 * - Do include packets and bytes from facets that have been deleted or
87 * whose own statistics have been folded into the rule.
89 * - Do include packets and bytes sent "by hand" that were accounted to
90 * the rule without any facet being involved (this is a rare corner
91 * case in rule_execute()).
93 * - Do not include packet or bytes that can be obtained from any facet's
94 * packet_count or byte_count member or that can be obtained from the
95 * datapath by, e.g., dpif_flow_get() for any subfacet.
97 uint64_t packet_count; /* Number of packets received. */
98 uint64_t byte_count; /* Number of bytes received. */
100 tag_type tag; /* Caches rule_calculate_tag() result. */
102 struct list facets; /* List of "struct facet"s. */
105 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
107 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
110 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
111 const struct flow *);
112 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
115 static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto,
116 const struct flow *flow);
118 static void rule_credit_stats(struct rule_dpif *,
119 const struct dpif_flow_stats *);
120 static void flow_push_stats(struct rule_dpif *, const struct flow *,
121 const struct dpif_flow_stats *);
122 static tag_type rule_calculate_tag(const struct flow *,
123 const struct minimask *, uint32_t basis);
124 static void rule_invalidate(const struct rule_dpif *);
126 #define MAX_MIRRORS 32
127 typedef uint32_t mirror_mask_t;
128 #define MIRROR_MASK_C(X) UINT32_C(X)
129 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
131 struct ofproto_dpif *ofproto; /* Owning ofproto. */
132 size_t idx; /* In ofproto's "mirrors" array. */
133 void *aux; /* Key supplied by ofproto's client. */
134 char *name; /* Identifier for log messages. */
136 /* Selection criteria. */
137 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
138 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
139 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
141 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
142 struct ofbundle *out; /* Output port or NULL. */
143 int out_vlan; /* Output VLAN or -1. */
144 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
147 int64_t packet_count; /* Number of packets sent. */
148 int64_t byte_count; /* Number of bytes sent. */
151 static void mirror_destroy(struct ofmirror *);
152 static void update_mirror_stats(struct ofproto_dpif *ofproto,
153 mirror_mask_t mirrors,
154 uint64_t packets, uint64_t bytes);
157 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
158 struct ofproto_dpif *ofproto; /* Owning ofproto. */
159 void *aux; /* Key supplied by ofproto's client. */
160 char *name; /* Identifier for log messages. */
163 struct list ports; /* Contains "struct ofport"s. */
164 enum port_vlan_mode vlan_mode; /* VLAN mode */
165 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
166 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
167 * NULL if all VLANs are trunked. */
168 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
169 struct bond *bond; /* Nonnull iff more than one port. */
170 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
173 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
175 /* Port mirroring info. */
176 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
177 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
178 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
181 static void bundle_remove(struct ofport *);
182 static void bundle_update(struct ofbundle *);
183 static void bundle_destroy(struct ofbundle *);
184 static void bundle_del_port(struct ofport_dpif *);
185 static void bundle_run(struct ofbundle *);
186 static void bundle_wait(struct ofbundle *);
187 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
188 uint16_t in_port, bool warn,
189 struct ofport_dpif **in_ofportp);
191 /* A controller may use OFPP_NONE as the ingress port to indicate that
192 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
193 * when an input bundle is needed for validation (e.g., mirroring or
194 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
195 * any 'port' structs, so care must be taken when dealing with it. */
196 static struct ofbundle ofpp_none_bundle = {
198 .vlan_mode = PORT_VLAN_TRUNK
201 static void stp_run(struct ofproto_dpif *ofproto);
202 static void stp_wait(struct ofproto_dpif *ofproto);
203 static int set_stp_port(struct ofport *,
204 const struct ofproto_port_stp_settings *);
206 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
208 struct action_xlate_ctx {
209 /* action_xlate_ctx_init() initializes these members. */
212 struct ofproto_dpif *ofproto;
214 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
215 * this flow when actions change header fields. */
218 /* The packet corresponding to 'flow', or a null pointer if we are
219 * revalidating without a packet to refer to. */
220 const struct ofpbuf *packet;
222 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
223 * actions update the flow table?
225 * We want to update these tables if we are actually processing a packet,
226 * or if we are accounting for packets that the datapath has processed, but
227 * not if we are just revalidating. */
230 /* The rule that we are currently translating, or NULL. */
231 struct rule_dpif *rule;
233 /* Union of the set of TCP flags seen so far in this flow. (Used only by
234 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
238 /* If nonnull, flow translation calls this function just before executing a
239 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
240 * when the recursion depth is exceeded.
242 * 'rule' is the rule being submitted into. It will be null if the
243 * resubmit or OFPP_TABLE action didn't find a matching rule.
245 * This is normally null so the client has to set it manually after
246 * calling action_xlate_ctx_init(). */
247 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule);
249 /* If nonnull, flow translation calls this function to report some
250 * significant decision, e.g. to explain why OFPP_NORMAL translation
251 * dropped a packet. */
252 void (*report_hook)(struct action_xlate_ctx *, const char *s);
254 /* If nonnull, flow translation credits the specified statistics to each
255 * rule reached through a resubmit or OFPP_TABLE action.
257 * This is normally null so the client has to set it manually after
258 * calling action_xlate_ctx_init(). */
259 const struct dpif_flow_stats *resubmit_stats;
261 /* xlate_actions() initializes and uses these members. The client might want
262 * to look at them after it returns. */
264 struct ofpbuf *odp_actions; /* Datapath actions. */
265 tag_type tags; /* Tags associated with actions. */
266 enum slow_path_reason slow; /* 0 if fast path may be used. */
267 bool has_learn; /* Actions include NXAST_LEARN? */
268 bool has_normal; /* Actions output to OFPP_NORMAL? */
269 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
270 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
271 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
273 /* xlate_actions() initializes and uses these members, but the client has no
274 * reason to look at them. */
276 int recurse; /* Recursion level, via xlate_table_action. */
277 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
278 struct flow base_flow; /* Flow at the last commit. */
279 uint32_t orig_skb_priority; /* Priority when packet arrived. */
280 uint8_t table_id; /* OpenFlow table ID where flow was found. */
281 uint32_t sflow_n_outputs; /* Number of output ports. */
282 uint32_t sflow_odp_port; /* Output port for composing sFlow action. */
283 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
284 bool exit; /* No further actions should be processed. */
285 struct flow orig_flow; /* Copy of original flow. */
288 static void action_xlate_ctx_init(struct action_xlate_ctx *,
289 struct ofproto_dpif *, const struct flow *,
290 ovs_be16 initial_tci, struct rule_dpif *,
291 uint8_t tcp_flags, const struct ofpbuf *);
292 static void xlate_actions(struct action_xlate_ctx *,
293 const struct ofpact *ofpacts, size_t ofpacts_len,
294 struct ofpbuf *odp_actions);
295 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
296 const struct ofpact *ofpacts,
299 static size_t put_userspace_action(const struct ofproto_dpif *,
300 struct ofpbuf *odp_actions,
302 const union user_action_cookie *);
304 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
305 enum slow_path_reason,
306 uint64_t *stub, size_t stub_size,
307 const struct nlattr **actionsp,
308 size_t *actions_lenp);
310 static void xlate_report(struct action_xlate_ctx *ctx, const char *s);
312 /* A subfacet (see "struct subfacet" below) has three possible installation
315 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
316 * case just after the subfacet is created, just before the subfacet is
317 * destroyed, or if the datapath returns an error when we try to install a
320 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
322 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
323 * ofproto_dpif is installed in the datapath.
326 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
327 SF_FAST_PATH, /* Full actions are installed. */
328 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
331 static const char *subfacet_path_to_string(enum subfacet_path);
333 /* A dpif flow and actions associated with a facet.
335 * See also the large comment on struct facet. */
338 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
339 struct list list_node; /* In struct facet's 'facets' list. */
340 struct facet *facet; /* Owning facet. */
344 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
345 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
346 * regenerate the ODP flow key from ->facet->flow. */
347 enum odp_key_fitness key_fitness;
351 long long int used; /* Time last used; time created if not used. */
353 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
354 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
358 * These should be essentially identical for every subfacet in a facet, but
359 * may differ in trivial ways due to VLAN splinters. */
360 size_t actions_len; /* Number of bytes in actions[]. */
361 struct nlattr *actions; /* Datapath actions. */
363 enum slow_path_reason slow; /* 0 if fast path may be used. */
364 enum subfacet_path path; /* Installed in datapath? */
366 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
367 * splinters can cause it to differ. This value should be removed when
368 * the VLAN splinters feature is no longer needed. */
369 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
371 /* Datapath port the packet arrived on. This is needed to remove
372 * flows for ports that are no longer part of the bridge. Since the
373 * flow definition only has the OpenFlow port number and the port is
374 * no longer part of the bridge, we can't determine the datapath port
375 * number needed to delete the flow from the datapath. */
376 uint32_t odp_in_port;
379 #define SUBFACET_DESTROY_MAX_BATCH 50
381 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
383 static struct subfacet *subfacet_find(struct ofproto_dpif *,
384 const struct nlattr *key, size_t key_len,
386 const struct flow *flow);
387 static void subfacet_destroy(struct subfacet *);
388 static void subfacet_destroy__(struct subfacet *);
389 static void subfacet_destroy_batch(struct ofproto_dpif *,
390 struct subfacet **, int n);
391 static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *,
393 static void subfacet_reset_dp_stats(struct subfacet *,
394 struct dpif_flow_stats *);
395 static void subfacet_update_time(struct subfacet *, long long int used);
396 static void subfacet_update_stats(struct subfacet *,
397 const struct dpif_flow_stats *);
398 static void subfacet_make_actions(struct subfacet *,
399 const struct ofpbuf *packet,
400 struct ofpbuf *odp_actions);
401 static int subfacet_install(struct subfacet *,
402 const struct nlattr *actions, size_t actions_len,
403 struct dpif_flow_stats *, enum slow_path_reason);
404 static void subfacet_uninstall(struct subfacet *);
406 static enum subfacet_path subfacet_want_path(enum slow_path_reason);
408 /* An exact-match instantiation of an OpenFlow flow.
410 * A facet associates a "struct flow", which represents the Open vSwitch
411 * userspace idea of an exact-match flow, with one or more subfacets. Each
412 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
413 * the facet. When the kernel module (or other dpif implementation) and Open
414 * vSwitch userspace agree on the definition of a flow key, there is exactly
415 * one subfacet per facet. If the dpif implementation supports more-specific
416 * flow matching than userspace, however, a facet can have more than one
417 * subfacet, each of which corresponds to some distinction in flow that
418 * userspace simply doesn't understand.
420 * Flow expiration works in terms of subfacets, so a facet must have at least
421 * one subfacet or it will never expire, leaking memory. */
424 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
425 struct list list_node; /* In owning rule's 'facets' list. */
426 struct rule_dpif *rule; /* Owning rule. */
429 struct list subfacets;
430 long long int used; /* Time last used; time created if not used. */
437 * - Do include packets and bytes sent "by hand", e.g. with
440 * - Do include packets and bytes that were obtained from the datapath
441 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
442 * DPIF_FP_ZERO_STATS).
444 * - Do not include packets or bytes that can be obtained from the
445 * datapath for any existing subfacet.
447 uint64_t packet_count; /* Number of packets received. */
448 uint64_t byte_count; /* Number of bytes received. */
450 /* Resubmit statistics. */
451 uint64_t prev_packet_count; /* Number of packets from last stats push. */
452 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
453 long long int prev_used; /* Used time from last stats push. */
456 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
457 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
458 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
460 /* Properties of datapath actions.
462 * Every subfacet has its own actions because actions can differ slightly
463 * between splintered and non-splintered subfacets due to the VLAN tag
464 * being initially different (present vs. absent). All of them have these
465 * properties in common so we just store one copy of them here. */
466 bool has_learn; /* Actions include NXAST_LEARN? */
467 bool has_normal; /* Actions output to OFPP_NORMAL? */
468 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
469 tag_type tags; /* Tags that would require revalidation. */
470 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
472 /* Storage for a single subfacet, to reduce malloc() time and space
473 * overhead. (A facet always has at least one subfacet and in the common
474 * case has exactly one subfacet.) */
475 struct subfacet one_subfacet;
478 static struct facet *facet_create(struct rule_dpif *,
479 const struct flow *, uint32_t hash);
480 static void facet_remove(struct facet *);
481 static void facet_free(struct facet *);
483 static struct facet *facet_find(struct ofproto_dpif *,
484 const struct flow *, uint32_t hash);
485 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
486 const struct flow *, uint32_t hash);
487 static void facet_revalidate(struct facet *);
488 static bool facet_check_consistency(struct facet *);
490 static void facet_flush_stats(struct facet *);
492 static void facet_update_time(struct facet *, long long int used);
493 static void facet_reset_counters(struct facet *);
494 static void facet_push_stats(struct facet *);
495 static void facet_learn(struct facet *);
496 static void facet_account(struct facet *);
498 static bool facet_is_controller_flow(struct facet *);
501 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
505 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
506 struct list bundle_node; /* In struct ofbundle's "ports" list. */
507 struct cfm *cfm; /* Connectivity Fault Management, if any. */
508 tag_type tag; /* Tag associated with this port. */
509 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
510 bool may_enable; /* May be enabled in bonds. */
511 long long int carrier_seq; /* Carrier status changes. */
514 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
515 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
516 long long int stp_state_entered;
518 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
520 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
522 * This is deprecated. It is only for compatibility with broken device
523 * drivers in old versions of Linux that do not properly support VLANs when
524 * VLAN devices are not used. When broken device drivers are no longer in
525 * widespread use, we will delete these interfaces. */
526 uint16_t realdev_ofp_port;
530 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
531 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
532 * traffic egressing the 'ofport' with that priority should be marked with. */
533 struct priority_to_dscp {
534 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
535 uint32_t priority; /* Priority of this queue (see struct flow). */
537 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
540 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
542 * This is deprecated. It is only for compatibility with broken device drivers
543 * in old versions of Linux that do not properly support VLANs when VLAN
544 * devices are not used. When broken device drivers are no longer in
545 * widespread use, we will delete these interfaces. */
546 struct vlan_splinter {
547 struct hmap_node realdev_vid_node;
548 struct hmap_node vlandev_node;
549 uint16_t realdev_ofp_port;
550 uint16_t vlandev_ofp_port;
554 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
555 uint32_t realdev, ovs_be16 vlan_tci);
556 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
557 static void vsp_remove(struct ofport_dpif *);
558 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
560 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
562 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
565 static struct ofport_dpif *
566 ofport_dpif_cast(const struct ofport *ofport)
568 ovs_assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
569 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
572 static void port_run(struct ofport_dpif *);
573 static void port_run_fast(struct ofport_dpif *);
574 static void port_wait(struct ofport_dpif *);
575 static int set_cfm(struct ofport *, const struct cfm_settings *);
576 static void ofport_clear_priorities(struct ofport_dpif *);
578 struct dpif_completion {
579 struct list list_node;
580 struct ofoperation *op;
583 /* Extra information about a classifier table.
584 * Currently used just for optimized flow revalidation. */
586 /* If either of these is nonnull, then this table has a form that allows
587 * flows to be tagged to avoid revalidating most flows for the most common
588 * kinds of flow table changes. */
589 struct cls_table *catchall_table; /* Table that wildcards all fields. */
590 struct cls_table *other_table; /* Table with any other wildcard set. */
591 uint32_t basis; /* Keeps each table's tags separate. */
594 /* Reasons that we might need to revalidate every facet, and corresponding
597 * A value of 0 means that there is no need to revalidate.
599 * It would be nice to have some cleaner way to integrate with coverage
600 * counters, but with only a few reasons I guess this is good enough for
602 enum revalidate_reason {
603 REV_RECONFIGURE = 1, /* Switch configuration changed. */
604 REV_STP, /* Spanning tree protocol port status change. */
605 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
606 REV_FLOW_TABLE, /* Flow table changed. */
607 REV_INCONSISTENCY /* Facet self-check failed. */
609 COVERAGE_DEFINE(rev_reconfigure);
610 COVERAGE_DEFINE(rev_stp);
611 COVERAGE_DEFINE(rev_port_toggled);
612 COVERAGE_DEFINE(rev_flow_table);
613 COVERAGE_DEFINE(rev_inconsistency);
615 /* All datapaths of a given type share a single dpif backer instance. */
620 struct timer next_expiration;
621 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
623 /* Facet revalidation flags applying to facets which use this backer. */
624 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
625 struct tag_set revalidate_set; /* Revalidate only matching facets. */
628 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
629 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
631 static struct ofport_dpif *
632 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
634 struct ofproto_dpif {
635 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
637 struct dpif_backer *backer;
639 /* Special OpenFlow rules. */
640 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
641 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
647 struct netflow *netflow;
648 struct dpif_sflow *sflow;
649 struct hmap bundles; /* Contains "struct ofbundle"s. */
650 struct mac_learning *ml;
651 struct ofmirror *mirrors[MAX_MIRRORS];
653 bool has_bonded_bundles;
657 struct hmap subfacets;
658 struct governor *governor;
661 struct table_dpif tables[N_TABLES];
663 /* Support for debugging async flow mods. */
664 struct list completions;
666 bool has_bundle_action; /* True when the first bundle action appears. */
667 struct netdev_stats stats; /* To account packets generated and consumed in
672 long long int stp_last_tick;
674 /* VLAN splinters. */
675 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
676 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
679 struct sset ports; /* Set of port names. */
680 struct sset port_poll_set; /* Queued names for port_poll() reply. */
681 int port_poll_errno; /* Last errno for port_poll() reply. */
684 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
685 * for debugging the asynchronous flow_mod implementation.) */
688 /* All existing ofproto_dpif instances, indexed by ->up.name. */
689 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
691 static void ofproto_dpif_unixctl_init(void);
693 static struct ofproto_dpif *
694 ofproto_dpif_cast(const struct ofproto *ofproto)
696 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
697 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
700 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
702 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
704 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
705 const struct ofpbuf *, ovs_be16 initial_tci,
708 /* Packet processing. */
709 static void update_learning_table(struct ofproto_dpif *,
710 const struct flow *, int vlan,
713 #define FLOW_MISS_MAX_BATCH 50
714 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
716 /* Flow expiration. */
717 static int expire(struct dpif_backer *);
720 static void send_netflow_active_timeouts(struct ofproto_dpif *);
723 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
724 static size_t compose_sflow_action(const struct ofproto_dpif *,
725 struct ofpbuf *odp_actions,
726 const struct flow *, uint32_t odp_port);
727 static void add_mirror_actions(struct action_xlate_ctx *ctx,
728 const struct flow *flow);
729 /* Global variables. */
730 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
732 /* Initial mappings of port to bridge mappings. */
733 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
735 /* Factory functions. */
738 init(const struct shash *iface_hints)
740 struct shash_node *node;
742 /* Make a local copy, since we don't own 'iface_hints' elements. */
743 SHASH_FOR_EACH(node, iface_hints) {
744 const struct iface_hint *orig_hint = node->data;
745 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
747 new_hint->br_name = xstrdup(orig_hint->br_name);
748 new_hint->br_type = xstrdup(orig_hint->br_type);
749 new_hint->ofp_port = orig_hint->ofp_port;
751 shash_add(&init_ofp_ports, node->name, new_hint);
756 enumerate_types(struct sset *types)
758 dp_enumerate_types(types);
762 enumerate_names(const char *type, struct sset *names)
764 struct ofproto_dpif *ofproto;
767 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
768 if (strcmp(type, ofproto->up.type)) {
771 sset_add(names, ofproto->up.name);
778 del(const char *type, const char *name)
783 error = dpif_open(name, type, &dpif);
785 error = dpif_delete(dpif);
792 port_open_type(const char *datapath_type, const char *port_type)
794 return dpif_port_open_type(datapath_type, port_type);
797 /* Type functions. */
799 static struct ofproto_dpif *
800 lookup_ofproto_dpif_by_port_name(const char *name)
802 struct ofproto_dpif *ofproto;
804 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
805 if (sset_contains(&ofproto->ports, name)) {
814 type_run(const char *type)
816 struct dpif_backer *backer;
820 backer = shash_find_data(&all_dpif_backers, type);
822 /* This is not necessarily a problem, since backers are only
823 * created on demand. */
827 dpif_run(backer->dpif);
829 if (backer->need_revalidate
830 || !tag_set_is_empty(&backer->revalidate_set)) {
831 struct tag_set revalidate_set = backer->revalidate_set;
832 bool need_revalidate = backer->need_revalidate;
833 struct ofproto_dpif *ofproto;
835 switch (backer->need_revalidate) {
836 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
837 case REV_STP: COVERAGE_INC(rev_stp); break;
838 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
839 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
840 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
843 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
846 if (ofproto->backer != backer) {
850 /* Clear the revalidation flags. */
851 tag_set_init(&backer->revalidate_set);
852 backer->need_revalidate = 0;
854 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
856 || tag_set_intersects(&revalidate_set, facet->tags)) {
857 facet_revalidate(facet);
864 if (timer_expired(&backer->next_expiration)) {
865 int delay = expire(backer);
866 timer_set_duration(&backer->next_expiration, delay);
869 /* Check for port changes in the dpif. */
870 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
871 struct ofproto_dpif *ofproto;
872 struct dpif_port port;
874 /* Don't report on the datapath's device. */
875 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
879 ofproto = lookup_ofproto_dpif_by_port_name(devname);
880 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
881 /* The port was removed. If we know the datapath,
882 * report it through poll_set(). If we don't, it may be
883 * notifying us of a removal we initiated, so ignore it.
884 * If there's a pending ENOBUFS, let it stand, since
885 * everything will be reevaluated. */
886 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
887 sset_add(&ofproto->port_poll_set, devname);
888 ofproto->port_poll_errno = 0;
890 } else if (!ofproto) {
891 /* The port was added, but we don't know with which
892 * ofproto we should associate it. Delete it. */
893 dpif_port_del(backer->dpif, port.port_no);
895 dpif_port_destroy(&port);
901 if (error != EAGAIN) {
902 struct ofproto_dpif *ofproto;
904 /* There was some sort of error, so propagate it to all
905 * ofprotos that use this backer. */
906 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
907 &all_ofproto_dpifs) {
908 if (ofproto->backer == backer) {
909 sset_clear(&ofproto->port_poll_set);
910 ofproto->port_poll_errno = error;
919 type_run_fast(const char *type)
921 struct dpif_backer *backer;
924 backer = shash_find_data(&all_dpif_backers, type);
926 /* This is not necessarily a problem, since backers are only
927 * created on demand. */
931 /* Handle one or more batches of upcalls, until there's nothing left to do
932 * or until we do a fixed total amount of work.
934 * We do work in batches because it can be much cheaper to set up a number
935 * of flows and fire off their patches all at once. We do multiple batches
936 * because in some cases handling a packet can cause another packet to be
937 * queued almost immediately as part of the return flow. Both
938 * optimizations can make major improvements on some benchmarks and
939 * presumably for real traffic as well. */
941 while (work < FLOW_MISS_MAX_BATCH) {
942 int retval = handle_upcalls(backer, FLOW_MISS_MAX_BATCH - work);
953 type_wait(const char *type)
955 struct dpif_backer *backer;
957 backer = shash_find_data(&all_dpif_backers, type);
959 /* This is not necessarily a problem, since backers are only
960 * created on demand. */
964 timer_wait(&backer->next_expiration);
967 /* Basic life-cycle. */
969 static int add_internal_flows(struct ofproto_dpif *);
971 static struct ofproto *
974 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
979 dealloc(struct ofproto *ofproto_)
981 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
986 close_dpif_backer(struct dpif_backer *backer)
988 struct shash_node *node;
990 ovs_assert(backer->refcount > 0);
992 if (--backer->refcount) {
996 hmap_destroy(&backer->odp_to_ofport_map);
997 node = shash_find(&all_dpif_backers, backer->type);
999 shash_delete(&all_dpif_backers, node);
1000 dpif_close(backer->dpif);
1005 /* Datapath port slated for removal from datapath. */
1006 struct odp_garbage {
1007 struct list list_node;
1012 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1014 struct dpif_backer *backer;
1015 struct dpif_port_dump port_dump;
1016 struct dpif_port port;
1017 struct shash_node *node;
1018 struct list garbage_list;
1019 struct odp_garbage *garbage, *next;
1025 backer = shash_find_data(&all_dpif_backers, type);
1032 backer_name = xasprintf("ovs-%s", type);
1034 /* Remove any existing datapaths, since we assume we're the only
1035 * userspace controlling the datapath. */
1037 dp_enumerate_names(type, &names);
1038 SSET_FOR_EACH(name, &names) {
1039 struct dpif *old_dpif;
1041 /* Don't remove our backer if it exists. */
1042 if (!strcmp(name, backer_name)) {
1046 if (dpif_open(name, type, &old_dpif)) {
1047 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1049 dpif_delete(old_dpif);
1050 dpif_close(old_dpif);
1053 sset_destroy(&names);
1055 backer = xmalloc(sizeof *backer);
1057 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1060 VLOG_ERR("failed to open datapath of type %s: %s", type,
1066 backer->type = xstrdup(type);
1067 backer->refcount = 1;
1068 hmap_init(&backer->odp_to_ofport_map);
1069 timer_set_duration(&backer->next_expiration, 1000);
1070 backer->need_revalidate = 0;
1071 tag_set_init(&backer->revalidate_set);
1074 dpif_flow_flush(backer->dpif);
1076 /* Loop through the ports already on the datapath and remove any
1077 * that we don't need anymore. */
1078 list_init(&garbage_list);
1079 dpif_port_dump_start(&port_dump, backer->dpif);
1080 while (dpif_port_dump_next(&port_dump, &port)) {
1081 node = shash_find(&init_ofp_ports, port.name);
1082 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1083 garbage = xmalloc(sizeof *garbage);
1084 garbage->odp_port = port.port_no;
1085 list_push_front(&garbage_list, &garbage->list_node);
1088 dpif_port_dump_done(&port_dump);
1090 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1091 dpif_port_del(backer->dpif, garbage->odp_port);
1092 list_remove(&garbage->list_node);
1096 shash_add(&all_dpif_backers, type, backer);
1098 error = dpif_recv_set(backer->dpif, true);
1100 VLOG_ERR("failed to listen on datapath of type %s: %s",
1101 type, strerror(error));
1102 close_dpif_backer(backer);
1110 construct(struct ofproto *ofproto_)
1112 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1113 struct shash_node *node, *next;
1118 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1123 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1124 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1126 ofproto->n_matches = 0;
1128 ofproto->netflow = NULL;
1129 ofproto->sflow = NULL;
1130 ofproto->stp = NULL;
1131 hmap_init(&ofproto->bundles);
1132 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1133 for (i = 0; i < MAX_MIRRORS; i++) {
1134 ofproto->mirrors[i] = NULL;
1136 ofproto->has_bonded_bundles = false;
1138 hmap_init(&ofproto->facets);
1139 hmap_init(&ofproto->subfacets);
1140 ofproto->governor = NULL;
1142 for (i = 0; i < N_TABLES; i++) {
1143 struct table_dpif *table = &ofproto->tables[i];
1145 table->catchall_table = NULL;
1146 table->other_table = NULL;
1147 table->basis = random_uint32();
1150 list_init(&ofproto->completions);
1152 ofproto_dpif_unixctl_init();
1154 ofproto->has_mirrors = false;
1155 ofproto->has_bundle_action = false;
1157 hmap_init(&ofproto->vlandev_map);
1158 hmap_init(&ofproto->realdev_vid_map);
1160 sset_init(&ofproto->ports);
1161 sset_init(&ofproto->port_poll_set);
1162 ofproto->port_poll_errno = 0;
1164 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1165 struct iface_hint *iface_hint = node->data;
1167 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1168 /* Check if the datapath already has this port. */
1169 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1170 sset_add(&ofproto->ports, node->name);
1173 free(iface_hint->br_name);
1174 free(iface_hint->br_type);
1176 shash_delete(&init_ofp_ports, node);
1180 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1181 hash_string(ofproto->up.name, 0));
1182 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1184 ofproto_init_tables(ofproto_, N_TABLES);
1185 error = add_internal_flows(ofproto);
1186 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1192 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1193 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1195 struct ofputil_flow_mod fm;
1198 match_init_catchall(&fm.match);
1200 match_set_reg(&fm.match, 0, id);
1201 fm.new_cookie = htonll(0);
1202 fm.cookie = htonll(0);
1203 fm.cookie_mask = htonll(0);
1204 fm.table_id = TBL_INTERNAL;
1205 fm.command = OFPFC_ADD;
1206 fm.idle_timeout = 0;
1207 fm.hard_timeout = 0;
1211 fm.ofpacts = ofpacts->data;
1212 fm.ofpacts_len = ofpacts->size;
1214 error = ofproto_flow_mod(&ofproto->up, &fm);
1216 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1217 id, ofperr_to_string(error));
1221 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1222 ovs_assert(*rulep != NULL);
1228 add_internal_flows(struct ofproto_dpif *ofproto)
1230 struct ofpact_controller *controller;
1231 uint64_t ofpacts_stub[128 / 8];
1232 struct ofpbuf ofpacts;
1236 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1239 controller = ofpact_put_CONTROLLER(&ofpacts);
1240 controller->max_len = UINT16_MAX;
1241 controller->controller_id = 0;
1242 controller->reason = OFPR_NO_MATCH;
1243 ofpact_pad(&ofpacts);
1245 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1250 ofpbuf_clear(&ofpacts);
1251 error = add_internal_flow(ofproto, id++, &ofpacts,
1252 &ofproto->no_packet_in_rule);
1257 complete_operations(struct ofproto_dpif *ofproto)
1259 struct dpif_completion *c, *next;
1261 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1262 ofoperation_complete(c->op, 0);
1263 list_remove(&c->list_node);
1269 destruct(struct ofproto *ofproto_)
1271 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1272 struct rule_dpif *rule, *next_rule;
1273 struct oftable *table;
1276 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1277 complete_operations(ofproto);
1279 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1280 struct cls_cursor cursor;
1282 cls_cursor_init(&cursor, &table->cls, NULL);
1283 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1284 ofproto_rule_destroy(&rule->up);
1288 for (i = 0; i < MAX_MIRRORS; i++) {
1289 mirror_destroy(ofproto->mirrors[i]);
1292 netflow_destroy(ofproto->netflow);
1293 dpif_sflow_destroy(ofproto->sflow);
1294 hmap_destroy(&ofproto->bundles);
1295 mac_learning_destroy(ofproto->ml);
1297 hmap_destroy(&ofproto->facets);
1298 hmap_destroy(&ofproto->subfacets);
1299 governor_destroy(ofproto->governor);
1301 hmap_destroy(&ofproto->vlandev_map);
1302 hmap_destroy(&ofproto->realdev_vid_map);
1304 sset_destroy(&ofproto->ports);
1305 sset_destroy(&ofproto->port_poll_set);
1307 close_dpif_backer(ofproto->backer);
1311 run_fast(struct ofproto *ofproto_)
1313 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1314 struct ofport_dpif *ofport;
1316 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1317 port_run_fast(ofport);
1324 run(struct ofproto *ofproto_)
1326 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1327 struct ofport_dpif *ofport;
1328 struct ofbundle *bundle;
1332 complete_operations(ofproto);
1335 error = run_fast(ofproto_);
1340 if (ofproto->netflow) {
1341 if (netflow_run(ofproto->netflow)) {
1342 send_netflow_active_timeouts(ofproto);
1345 if (ofproto->sflow) {
1346 dpif_sflow_run(ofproto->sflow);
1349 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1352 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1357 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1359 /* Check the consistency of a random facet, to aid debugging. */
1360 if (!hmap_is_empty(&ofproto->facets)
1361 && !ofproto->backer->need_revalidate) {
1362 struct facet *facet;
1364 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1365 struct facet, hmap_node);
1366 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1368 if (!facet_check_consistency(facet)) {
1369 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1374 if (ofproto->governor) {
1377 governor_run(ofproto->governor);
1379 /* If the governor has shrunk to its minimum size and the number of
1380 * subfacets has dwindled, then drop the governor entirely.
1382 * For hysteresis, the number of subfacets to drop the governor is
1383 * smaller than the number needed to trigger its creation. */
1384 n_subfacets = hmap_count(&ofproto->subfacets);
1385 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1386 && governor_is_idle(ofproto->governor)) {
1387 governor_destroy(ofproto->governor);
1388 ofproto->governor = NULL;
1396 wait(struct ofproto *ofproto_)
1398 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1399 struct ofport_dpif *ofport;
1400 struct ofbundle *bundle;
1402 if (!clogged && !list_is_empty(&ofproto->completions)) {
1403 poll_immediate_wake();
1406 dpif_wait(ofproto->backer->dpif);
1407 dpif_recv_wait(ofproto->backer->dpif);
1408 if (ofproto->sflow) {
1409 dpif_sflow_wait(ofproto->sflow);
1411 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1412 poll_immediate_wake();
1414 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1417 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1418 bundle_wait(bundle);
1420 if (ofproto->netflow) {
1421 netflow_wait(ofproto->netflow);
1423 mac_learning_wait(ofproto->ml);
1425 if (ofproto->backer->need_revalidate) {
1426 /* Shouldn't happen, but if it does just go around again. */
1427 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1428 poll_immediate_wake();
1430 if (ofproto->governor) {
1431 governor_wait(ofproto->governor);
1436 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1438 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1440 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1441 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1445 flush(struct ofproto *ofproto_)
1447 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1448 struct subfacet *subfacet, *next_subfacet;
1449 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1453 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1454 &ofproto->subfacets) {
1455 if (subfacet->path != SF_NOT_INSTALLED) {
1456 batch[n_batch++] = subfacet;
1457 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1458 subfacet_destroy_batch(ofproto, batch, n_batch);
1462 subfacet_destroy(subfacet);
1467 subfacet_destroy_batch(ofproto, batch, n_batch);
1472 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1473 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1475 *arp_match_ip = true;
1476 *actions = (OFPUTIL_A_OUTPUT |
1477 OFPUTIL_A_SET_VLAN_VID |
1478 OFPUTIL_A_SET_VLAN_PCP |
1479 OFPUTIL_A_STRIP_VLAN |
1480 OFPUTIL_A_SET_DL_SRC |
1481 OFPUTIL_A_SET_DL_DST |
1482 OFPUTIL_A_SET_NW_SRC |
1483 OFPUTIL_A_SET_NW_DST |
1484 OFPUTIL_A_SET_NW_TOS |
1485 OFPUTIL_A_SET_TP_SRC |
1486 OFPUTIL_A_SET_TP_DST |
1491 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1493 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1494 struct dpif_dp_stats s;
1496 strcpy(ots->name, "classifier");
1498 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1500 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1501 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1504 static struct ofport *
1507 struct ofport_dpif *port = xmalloc(sizeof *port);
1512 port_dealloc(struct ofport *port_)
1514 struct ofport_dpif *port = ofport_dpif_cast(port_);
1519 port_construct(struct ofport *port_)
1521 struct ofport_dpif *port = ofport_dpif_cast(port_);
1522 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1523 struct dpif_port dpif_port;
1526 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1527 port->bundle = NULL;
1529 port->tag = tag_create_random();
1530 port->may_enable = true;
1531 port->stp_port = NULL;
1532 port->stp_state = STP_DISABLED;
1533 hmap_init(&port->priorities);
1534 port->realdev_ofp_port = 0;
1535 port->vlandev_vid = 0;
1536 port->carrier_seq = netdev_get_carrier_resets(port->up.netdev);
1538 error = dpif_port_query_by_name(ofproto->backer->dpif,
1539 netdev_get_name(port->up.netdev),
1545 port->odp_port = dpif_port.port_no;
1547 /* Sanity-check that a mapping doesn't already exist. This
1548 * shouldn't happen. */
1549 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1550 VLOG_ERR("port %s already has an OpenFlow port number\n",
1555 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1556 hash_int(port->odp_port, 0));
1558 if (ofproto->sflow) {
1559 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1566 port_destruct(struct ofport *port_)
1568 struct ofport_dpif *port = ofport_dpif_cast(port_);
1569 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1570 const char *devname = netdev_get_name(port->up.netdev);
1572 if (dpif_port_exists(ofproto->backer->dpif, devname)) {
1573 /* The underlying device is still there, so delete it. This
1574 * happens when the ofproto is being destroyed, since the caller
1575 * assumes that removal of attached ports will happen as part of
1577 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1580 sset_find_and_delete(&ofproto->ports, devname);
1581 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1582 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1583 bundle_remove(port_);
1584 set_cfm(port_, NULL);
1585 if (ofproto->sflow) {
1586 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1589 ofport_clear_priorities(port);
1590 hmap_destroy(&port->priorities);
1594 port_modified(struct ofport *port_)
1596 struct ofport_dpif *port = ofport_dpif_cast(port_);
1598 if (port->bundle && port->bundle->bond) {
1599 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1604 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1606 struct ofport_dpif *port = ofport_dpif_cast(port_);
1607 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1608 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1610 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1611 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1612 OFPUTIL_PC_NO_PACKET_IN)) {
1613 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1615 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1616 bundle_update(port->bundle);
1622 set_sflow(struct ofproto *ofproto_,
1623 const struct ofproto_sflow_options *sflow_options)
1625 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1626 struct dpif_sflow *ds = ofproto->sflow;
1628 if (sflow_options) {
1630 struct ofport_dpif *ofport;
1632 ds = ofproto->sflow = dpif_sflow_create();
1633 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1634 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1636 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1638 dpif_sflow_set_options(ds, sflow_options);
1641 dpif_sflow_destroy(ds);
1642 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1643 ofproto->sflow = NULL;
1650 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1652 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1659 struct ofproto_dpif *ofproto;
1661 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1662 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1663 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1666 if (cfm_configure(ofport->cfm, s)) {
1672 cfm_destroy(ofport->cfm);
1678 get_cfm_fault(const struct ofport *ofport_)
1680 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1682 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1686 get_cfm_opup(const struct ofport *ofport_)
1688 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1690 return ofport->cfm ? cfm_get_opup(ofport->cfm) : -1;
1694 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1697 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1700 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1708 get_cfm_health(const struct ofport *ofport_)
1710 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1712 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1715 /* Spanning Tree. */
1718 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1720 struct ofproto_dpif *ofproto = ofproto_;
1721 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1722 struct ofport_dpif *ofport;
1724 ofport = stp_port_get_aux(sp);
1726 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1727 ofproto->up.name, port_num);
1729 struct eth_header *eth = pkt->l2;
1731 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1732 if (eth_addr_is_zero(eth->eth_src)) {
1733 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1734 "with unknown MAC", ofproto->up.name, port_num);
1736 send_packet(ofport, pkt);
1742 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1744 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1746 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1748 /* Only revalidate flows if the configuration changed. */
1749 if (!s != !ofproto->stp) {
1750 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1754 if (!ofproto->stp) {
1755 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1756 send_bpdu_cb, ofproto);
1757 ofproto->stp_last_tick = time_msec();
1760 stp_set_bridge_id(ofproto->stp, s->system_id);
1761 stp_set_bridge_priority(ofproto->stp, s->priority);
1762 stp_set_hello_time(ofproto->stp, s->hello_time);
1763 stp_set_max_age(ofproto->stp, s->max_age);
1764 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1766 struct ofport *ofport;
1768 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1769 set_stp_port(ofport, NULL);
1772 stp_destroy(ofproto->stp);
1773 ofproto->stp = NULL;
1780 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1782 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1786 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1787 s->designated_root = stp_get_designated_root(ofproto->stp);
1788 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1797 update_stp_port_state(struct ofport_dpif *ofport)
1799 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1800 enum stp_state state;
1802 /* Figure out new state. */
1803 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1807 if (ofport->stp_state != state) {
1808 enum ofputil_port_state of_state;
1811 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1812 netdev_get_name(ofport->up.netdev),
1813 stp_state_name(ofport->stp_state),
1814 stp_state_name(state));
1815 if (stp_learn_in_state(ofport->stp_state)
1816 != stp_learn_in_state(state)) {
1817 /* xxx Learning action flows should also be flushed. */
1818 mac_learning_flush(ofproto->ml,
1819 &ofproto->backer->revalidate_set);
1821 fwd_change = stp_forward_in_state(ofport->stp_state)
1822 != stp_forward_in_state(state);
1824 ofproto->backer->need_revalidate = REV_STP;
1825 ofport->stp_state = state;
1826 ofport->stp_state_entered = time_msec();
1828 if (fwd_change && ofport->bundle) {
1829 bundle_update(ofport->bundle);
1832 /* Update the STP state bits in the OpenFlow port description. */
1833 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1834 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1835 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1836 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1837 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1839 ofproto_port_set_state(&ofport->up, of_state);
1843 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1844 * caller is responsible for assigning STP port numbers and ensuring
1845 * there are no duplicates. */
1847 set_stp_port(struct ofport *ofport_,
1848 const struct ofproto_port_stp_settings *s)
1850 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1851 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1852 struct stp_port *sp = ofport->stp_port;
1854 if (!s || !s->enable) {
1856 ofport->stp_port = NULL;
1857 stp_port_disable(sp);
1858 update_stp_port_state(ofport);
1861 } else if (sp && stp_port_no(sp) != s->port_num
1862 && ofport == stp_port_get_aux(sp)) {
1863 /* The port-id changed, so disable the old one if it's not
1864 * already in use by another port. */
1865 stp_port_disable(sp);
1868 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1869 stp_port_enable(sp);
1871 stp_port_set_aux(sp, ofport);
1872 stp_port_set_priority(sp, s->priority);
1873 stp_port_set_path_cost(sp, s->path_cost);
1875 update_stp_port_state(ofport);
1881 get_stp_port_status(struct ofport *ofport_,
1882 struct ofproto_port_stp_status *s)
1884 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1885 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1886 struct stp_port *sp = ofport->stp_port;
1888 if (!ofproto->stp || !sp) {
1894 s->port_id = stp_port_get_id(sp);
1895 s->state = stp_port_get_state(sp);
1896 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1897 s->role = stp_port_get_role(sp);
1898 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1904 stp_run(struct ofproto_dpif *ofproto)
1907 long long int now = time_msec();
1908 long long int elapsed = now - ofproto->stp_last_tick;
1909 struct stp_port *sp;
1912 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1913 ofproto->stp_last_tick = now;
1915 while (stp_get_changed_port(ofproto->stp, &sp)) {
1916 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1919 update_stp_port_state(ofport);
1923 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1924 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
1930 stp_wait(struct ofproto_dpif *ofproto)
1933 poll_timer_wait(1000);
1937 /* Returns true if STP should process 'flow'. */
1939 stp_should_process_flow(const struct flow *flow)
1941 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1945 stp_process_packet(const struct ofport_dpif *ofport,
1946 const struct ofpbuf *packet)
1948 struct ofpbuf payload = *packet;
1949 struct eth_header *eth = payload.data;
1950 struct stp_port *sp = ofport->stp_port;
1952 /* Sink packets on ports that have STP disabled when the bridge has
1954 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1958 /* Trim off padding on payload. */
1959 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1960 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1963 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1964 stp_received_bpdu(sp, payload.data, payload.size);
1968 static struct priority_to_dscp *
1969 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1971 struct priority_to_dscp *pdscp;
1974 hash = hash_int(priority, 0);
1975 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1976 if (pdscp->priority == priority) {
1984 ofport_clear_priorities(struct ofport_dpif *ofport)
1986 struct priority_to_dscp *pdscp, *next;
1988 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1989 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1995 set_queues(struct ofport *ofport_,
1996 const struct ofproto_port_queue *qdscp_list,
1999 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2000 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2001 struct hmap new = HMAP_INITIALIZER(&new);
2004 for (i = 0; i < n_qdscp; i++) {
2005 struct priority_to_dscp *pdscp;
2009 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2010 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2015 pdscp = get_priority(ofport, priority);
2017 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2019 pdscp = xmalloc(sizeof *pdscp);
2020 pdscp->priority = priority;
2022 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2025 if (pdscp->dscp != dscp) {
2027 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2030 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2033 if (!hmap_is_empty(&ofport->priorities)) {
2034 ofport_clear_priorities(ofport);
2035 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2038 hmap_swap(&new, &ofport->priorities);
2046 /* Expires all MAC learning entries associated with 'bundle' and forces its
2047 * ofproto to revalidate every flow.
2049 * Normally MAC learning entries are removed only from the ofproto associated
2050 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2051 * are removed from every ofproto. When patch ports and SLB bonds are in use
2052 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2053 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2054 * with the host from which it migrated. */
2056 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2058 struct ofproto_dpif *ofproto = bundle->ofproto;
2059 struct mac_learning *ml = ofproto->ml;
2060 struct mac_entry *mac, *next_mac;
2062 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2063 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2064 if (mac->port.p == bundle) {
2066 struct ofproto_dpif *o;
2068 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2070 struct mac_entry *e;
2072 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2075 mac_learning_expire(o->ml, e);
2081 mac_learning_expire(ml, mac);
2086 static struct ofbundle *
2087 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2089 struct ofbundle *bundle;
2091 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2092 &ofproto->bundles) {
2093 if (bundle->aux == aux) {
2100 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2101 * ones that are found to 'bundles'. */
2103 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2104 void **auxes, size_t n_auxes,
2105 struct hmapx *bundles)
2109 hmapx_init(bundles);
2110 for (i = 0; i < n_auxes; i++) {
2111 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2113 hmapx_add(bundles, bundle);
2119 bundle_update(struct ofbundle *bundle)
2121 struct ofport_dpif *port;
2123 bundle->floodable = true;
2124 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2125 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2126 || !stp_forward_in_state(port->stp_state)) {
2127 bundle->floodable = false;
2134 bundle_del_port(struct ofport_dpif *port)
2136 struct ofbundle *bundle = port->bundle;
2138 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2140 list_remove(&port->bundle_node);
2141 port->bundle = NULL;
2144 lacp_slave_unregister(bundle->lacp, port);
2147 bond_slave_unregister(bundle->bond, port);
2150 bundle_update(bundle);
2154 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2155 struct lacp_slave_settings *lacp,
2156 uint32_t bond_stable_id)
2158 struct ofport_dpif *port;
2160 port = get_ofp_port(bundle->ofproto, ofp_port);
2165 if (port->bundle != bundle) {
2166 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2168 bundle_del_port(port);
2171 port->bundle = bundle;
2172 list_push_back(&bundle->ports, &port->bundle_node);
2173 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2174 || !stp_forward_in_state(port->stp_state)) {
2175 bundle->floodable = false;
2179 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2180 lacp_slave_register(bundle->lacp, port, lacp);
2183 port->bond_stable_id = bond_stable_id;
2189 bundle_destroy(struct ofbundle *bundle)
2191 struct ofproto_dpif *ofproto;
2192 struct ofport_dpif *port, *next_port;
2199 ofproto = bundle->ofproto;
2200 for (i = 0; i < MAX_MIRRORS; i++) {
2201 struct ofmirror *m = ofproto->mirrors[i];
2203 if (m->out == bundle) {
2205 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2206 || hmapx_find_and_delete(&m->dsts, bundle)) {
2207 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2212 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2213 bundle_del_port(port);
2216 bundle_flush_macs(bundle, true);
2217 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2219 free(bundle->trunks);
2220 lacp_destroy(bundle->lacp);
2221 bond_destroy(bundle->bond);
2226 bundle_set(struct ofproto *ofproto_, void *aux,
2227 const struct ofproto_bundle_settings *s)
2229 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2230 bool need_flush = false;
2231 struct ofport_dpif *port;
2232 struct ofbundle *bundle;
2233 unsigned long *trunks;
2239 bundle_destroy(bundle_lookup(ofproto, aux));
2243 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2244 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2246 bundle = bundle_lookup(ofproto, aux);
2248 bundle = xmalloc(sizeof *bundle);
2250 bundle->ofproto = ofproto;
2251 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2252 hash_pointer(aux, 0));
2254 bundle->name = NULL;
2256 list_init(&bundle->ports);
2257 bundle->vlan_mode = PORT_VLAN_TRUNK;
2259 bundle->trunks = NULL;
2260 bundle->use_priority_tags = s->use_priority_tags;
2261 bundle->lacp = NULL;
2262 bundle->bond = NULL;
2264 bundle->floodable = true;
2266 bundle->src_mirrors = 0;
2267 bundle->dst_mirrors = 0;
2268 bundle->mirror_out = 0;
2271 if (!bundle->name || strcmp(s->name, bundle->name)) {
2273 bundle->name = xstrdup(s->name);
2278 if (!bundle->lacp) {
2279 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2280 bundle->lacp = lacp_create();
2282 lacp_configure(bundle->lacp, s->lacp);
2284 lacp_destroy(bundle->lacp);
2285 bundle->lacp = NULL;
2288 /* Update set of ports. */
2290 for (i = 0; i < s->n_slaves; i++) {
2291 if (!bundle_add_port(bundle, s->slaves[i],
2292 s->lacp ? &s->lacp_slaves[i] : NULL,
2293 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
2297 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2298 struct ofport_dpif *next_port;
2300 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2301 for (i = 0; i < s->n_slaves; i++) {
2302 if (s->slaves[i] == port->up.ofp_port) {
2307 bundle_del_port(port);
2311 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2313 if (list_is_empty(&bundle->ports)) {
2314 bundle_destroy(bundle);
2318 /* Set VLAN tagging mode */
2319 if (s->vlan_mode != bundle->vlan_mode
2320 || s->use_priority_tags != bundle->use_priority_tags) {
2321 bundle->vlan_mode = s->vlan_mode;
2322 bundle->use_priority_tags = s->use_priority_tags;
2327 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2328 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2330 if (vlan != bundle->vlan) {
2331 bundle->vlan = vlan;
2335 /* Get trunked VLANs. */
2336 switch (s->vlan_mode) {
2337 case PORT_VLAN_ACCESS:
2341 case PORT_VLAN_TRUNK:
2342 trunks = CONST_CAST(unsigned long *, s->trunks);
2345 case PORT_VLAN_NATIVE_UNTAGGED:
2346 case PORT_VLAN_NATIVE_TAGGED:
2347 if (vlan != 0 && (!s->trunks
2348 || !bitmap_is_set(s->trunks, vlan)
2349 || bitmap_is_set(s->trunks, 0))) {
2350 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2352 trunks = bitmap_clone(s->trunks, 4096);
2354 trunks = bitmap_allocate1(4096);
2356 bitmap_set1(trunks, vlan);
2357 bitmap_set0(trunks, 0);
2359 trunks = CONST_CAST(unsigned long *, s->trunks);
2366 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2367 free(bundle->trunks);
2368 if (trunks == s->trunks) {
2369 bundle->trunks = vlan_bitmap_clone(trunks);
2371 bundle->trunks = trunks;
2376 if (trunks != s->trunks) {
2381 if (!list_is_short(&bundle->ports)) {
2382 bundle->ofproto->has_bonded_bundles = true;
2384 if (bond_reconfigure(bundle->bond, s->bond)) {
2385 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2388 bundle->bond = bond_create(s->bond);
2389 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2392 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2393 bond_slave_register(bundle->bond, port, port->bond_stable_id,
2397 bond_destroy(bundle->bond);
2398 bundle->bond = NULL;
2401 /* If we changed something that would affect MAC learning, un-learn
2402 * everything on this port and force flow revalidation. */
2404 bundle_flush_macs(bundle, false);
2411 bundle_remove(struct ofport *port_)
2413 struct ofport_dpif *port = ofport_dpif_cast(port_);
2414 struct ofbundle *bundle = port->bundle;
2417 bundle_del_port(port);
2418 if (list_is_empty(&bundle->ports)) {
2419 bundle_destroy(bundle);
2420 } else if (list_is_short(&bundle->ports)) {
2421 bond_destroy(bundle->bond);
2422 bundle->bond = NULL;
2428 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2430 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2431 struct ofport_dpif *port = port_;
2432 uint8_t ea[ETH_ADDR_LEN];
2435 error = netdev_get_etheraddr(port->up.netdev, ea);
2437 struct ofpbuf packet;
2440 ofpbuf_init(&packet, 0);
2441 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2443 memcpy(packet_pdu, pdu, pdu_size);
2445 send_packet(port, &packet);
2446 ofpbuf_uninit(&packet);
2448 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2449 "%s (%s)", port->bundle->name,
2450 netdev_get_name(port->up.netdev), strerror(error));
2455 bundle_send_learning_packets(struct ofbundle *bundle)
2457 struct ofproto_dpif *ofproto = bundle->ofproto;
2458 int error, n_packets, n_errors;
2459 struct mac_entry *e;
2461 error = n_packets = n_errors = 0;
2462 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2463 if (e->port.p != bundle) {
2464 struct ofpbuf *learning_packet;
2465 struct ofport_dpif *port;
2469 /* The assignment to "port" is unnecessary but makes "grep"ing for
2470 * struct ofport_dpif more effective. */
2471 learning_packet = bond_compose_learning_packet(bundle->bond,
2475 ret = send_packet(port, learning_packet);
2476 ofpbuf_delete(learning_packet);
2486 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2487 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2488 "packets, last error was: %s",
2489 bundle->name, n_errors, n_packets, strerror(error));
2491 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2492 bundle->name, n_packets);
2497 bundle_run(struct ofbundle *bundle)
2500 lacp_run(bundle->lacp, send_pdu_cb);
2503 struct ofport_dpif *port;
2505 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2506 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2509 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2510 lacp_status(bundle->lacp));
2511 if (bond_should_send_learning_packets(bundle->bond)) {
2512 bundle_send_learning_packets(bundle);
2518 bundle_wait(struct ofbundle *bundle)
2521 lacp_wait(bundle->lacp);
2524 bond_wait(bundle->bond);
2531 mirror_scan(struct ofproto_dpif *ofproto)
2535 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2536 if (!ofproto->mirrors[idx]) {
2543 static struct ofmirror *
2544 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2548 for (i = 0; i < MAX_MIRRORS; i++) {
2549 struct ofmirror *mirror = ofproto->mirrors[i];
2550 if (mirror && mirror->aux == aux) {
2558 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2560 mirror_update_dups(struct ofproto_dpif *ofproto)
2564 for (i = 0; i < MAX_MIRRORS; i++) {
2565 struct ofmirror *m = ofproto->mirrors[i];
2568 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2572 for (i = 0; i < MAX_MIRRORS; i++) {
2573 struct ofmirror *m1 = ofproto->mirrors[i];
2580 for (j = i + 1; j < MAX_MIRRORS; j++) {
2581 struct ofmirror *m2 = ofproto->mirrors[j];
2583 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2584 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2585 m2->dup_mirrors |= m1->dup_mirrors;
2592 mirror_set(struct ofproto *ofproto_, void *aux,
2593 const struct ofproto_mirror_settings *s)
2595 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2596 mirror_mask_t mirror_bit;
2597 struct ofbundle *bundle;
2598 struct ofmirror *mirror;
2599 struct ofbundle *out;
2600 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2601 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2604 mirror = mirror_lookup(ofproto, aux);
2606 mirror_destroy(mirror);
2612 idx = mirror_scan(ofproto);
2614 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2616 ofproto->up.name, MAX_MIRRORS, s->name);
2620 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2621 mirror->ofproto = ofproto;
2624 mirror->out_vlan = -1;
2625 mirror->name = NULL;
2628 if (!mirror->name || strcmp(s->name, mirror->name)) {
2630 mirror->name = xstrdup(s->name);
2633 /* Get the new configuration. */
2634 if (s->out_bundle) {
2635 out = bundle_lookup(ofproto, s->out_bundle);
2637 mirror_destroy(mirror);
2643 out_vlan = s->out_vlan;
2645 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2646 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2648 /* If the configuration has not changed, do nothing. */
2649 if (hmapx_equals(&srcs, &mirror->srcs)
2650 && hmapx_equals(&dsts, &mirror->dsts)
2651 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2652 && mirror->out == out
2653 && mirror->out_vlan == out_vlan)
2655 hmapx_destroy(&srcs);
2656 hmapx_destroy(&dsts);
2660 hmapx_swap(&srcs, &mirror->srcs);
2661 hmapx_destroy(&srcs);
2663 hmapx_swap(&dsts, &mirror->dsts);
2664 hmapx_destroy(&dsts);
2666 free(mirror->vlans);
2667 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2670 mirror->out_vlan = out_vlan;
2672 /* Update bundles. */
2673 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2674 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2675 if (hmapx_contains(&mirror->srcs, bundle)) {
2676 bundle->src_mirrors |= mirror_bit;
2678 bundle->src_mirrors &= ~mirror_bit;
2681 if (hmapx_contains(&mirror->dsts, bundle)) {
2682 bundle->dst_mirrors |= mirror_bit;
2684 bundle->dst_mirrors &= ~mirror_bit;
2687 if (mirror->out == bundle) {
2688 bundle->mirror_out |= mirror_bit;
2690 bundle->mirror_out &= ~mirror_bit;
2694 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2695 ofproto->has_mirrors = true;
2696 mac_learning_flush(ofproto->ml,
2697 &ofproto->backer->revalidate_set);
2698 mirror_update_dups(ofproto);
2704 mirror_destroy(struct ofmirror *mirror)
2706 struct ofproto_dpif *ofproto;
2707 mirror_mask_t mirror_bit;
2708 struct ofbundle *bundle;
2715 ofproto = mirror->ofproto;
2716 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2717 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2719 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2720 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2721 bundle->src_mirrors &= ~mirror_bit;
2722 bundle->dst_mirrors &= ~mirror_bit;
2723 bundle->mirror_out &= ~mirror_bit;
2726 hmapx_destroy(&mirror->srcs);
2727 hmapx_destroy(&mirror->dsts);
2728 free(mirror->vlans);
2730 ofproto->mirrors[mirror->idx] = NULL;
2734 mirror_update_dups(ofproto);
2736 ofproto->has_mirrors = false;
2737 for (i = 0; i < MAX_MIRRORS; i++) {
2738 if (ofproto->mirrors[i]) {
2739 ofproto->has_mirrors = true;
2746 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2747 uint64_t *packets, uint64_t *bytes)
2749 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2750 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2753 *packets = *bytes = UINT64_MAX;
2757 *packets = mirror->packet_count;
2758 *bytes = mirror->byte_count;
2764 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2766 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2767 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2768 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2774 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2776 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2777 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2778 return bundle && bundle->mirror_out != 0;
2782 forward_bpdu_changed(struct ofproto *ofproto_)
2784 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2785 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2789 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2792 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2793 mac_learning_set_idle_time(ofproto->ml, idle_time);
2794 mac_learning_set_max_entries(ofproto->ml, max_entries);
2799 static struct ofport_dpif *
2800 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2802 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2803 return ofport ? ofport_dpif_cast(ofport) : NULL;
2806 static struct ofport_dpif *
2807 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2809 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2810 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2814 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2815 struct ofproto_port *ofproto_port,
2816 struct dpif_port *dpif_port)
2818 ofproto_port->name = dpif_port->name;
2819 ofproto_port->type = dpif_port->type;
2820 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2824 port_run_fast(struct ofport_dpif *ofport)
2826 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2827 struct ofpbuf packet;
2829 ofpbuf_init(&packet, 0);
2830 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2831 send_packet(ofport, &packet);
2832 ofpbuf_uninit(&packet);
2837 port_run(struct ofport_dpif *ofport)
2839 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2840 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2841 bool enable = netdev_get_carrier(ofport->up.netdev);
2843 ofport->carrier_seq = carrier_seq;
2845 port_run_fast(ofport);
2847 int cfm_opup = cfm_get_opup(ofport->cfm);
2849 cfm_run(ofport->cfm);
2850 enable = enable && !cfm_get_fault(ofport->cfm);
2852 if (cfm_opup >= 0) {
2853 enable = enable && cfm_opup;
2857 if (ofport->bundle) {
2858 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2859 if (carrier_changed) {
2860 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2864 if (ofport->may_enable != enable) {
2865 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2867 if (ofproto->has_bundle_action) {
2868 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2872 ofport->may_enable = enable;
2876 port_wait(struct ofport_dpif *ofport)
2879 cfm_wait(ofport->cfm);
2884 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2885 struct ofproto_port *ofproto_port)
2887 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2888 struct dpif_port dpif_port;
2891 if (!sset_contains(&ofproto->ports, devname)) {
2894 error = dpif_port_query_by_name(ofproto->backer->dpif,
2895 devname, &dpif_port);
2897 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2903 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2905 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2906 uint32_t odp_port = UINT32_MAX;
2909 error = dpif_port_add(ofproto->backer->dpif, netdev, &odp_port);
2911 sset_add(&ofproto->ports, netdev_get_name(netdev));
2917 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2919 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2920 uint32_t odp_port = ofp_port_to_odp_port(ofproto, ofp_port);
2923 if (odp_port != OFPP_NONE) {
2924 error = dpif_port_del(ofproto->backer->dpif, odp_port);
2927 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2929 /* The caller is going to close ofport->up.netdev. If this is a
2930 * bonded port, then the bond is using that netdev, so remove it
2931 * from the bond. The client will need to reconfigure everything
2932 * after deleting ports, so then the slave will get re-added. */
2933 bundle_remove(&ofport->up);
2940 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2942 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2945 error = netdev_get_stats(ofport->up.netdev, stats);
2947 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
2948 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2950 /* ofproto->stats.tx_packets represents packets that we created
2951 * internally and sent to some port (e.g. packets sent with
2952 * send_packet()). Account for them as if they had come from
2953 * OFPP_LOCAL and got forwarded. */
2955 if (stats->rx_packets != UINT64_MAX) {
2956 stats->rx_packets += ofproto->stats.tx_packets;
2959 if (stats->rx_bytes != UINT64_MAX) {
2960 stats->rx_bytes += ofproto->stats.tx_bytes;
2963 /* ofproto->stats.rx_packets represents packets that were received on
2964 * some port and we processed internally and dropped (e.g. STP).
2965 * Account for them as if they had been forwarded to OFPP_LOCAL. */
2967 if (stats->tx_packets != UINT64_MAX) {
2968 stats->tx_packets += ofproto->stats.rx_packets;
2971 if (stats->tx_bytes != UINT64_MAX) {
2972 stats->tx_bytes += ofproto->stats.rx_bytes;
2979 /* Account packets for LOCAL port. */
2981 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2982 size_t tx_size, size_t rx_size)
2984 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2987 ofproto->stats.rx_packets++;
2988 ofproto->stats.rx_bytes += rx_size;
2991 ofproto->stats.tx_packets++;
2992 ofproto->stats.tx_bytes += tx_size;
2996 struct port_dump_state {
3002 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3004 struct port_dump_state *state;
3006 *statep = state = xmalloc(sizeof *state);
3013 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
3014 struct ofproto_port *port)
3016 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3017 struct port_dump_state *state = state_;
3018 struct sset_node *node;
3020 while ((node = sset_at_position(&ofproto->ports, &state->bucket,
3024 error = port_query_by_name(ofproto_, node->name, port);
3025 if (error != ENODEV) {
3034 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3036 struct port_dump_state *state = state_;
3043 port_poll(const struct ofproto *ofproto_, char **devnamep)
3045 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3047 if (ofproto->port_poll_errno) {
3048 int error = ofproto->port_poll_errno;
3049 ofproto->port_poll_errno = 0;
3053 if (sset_is_empty(&ofproto->port_poll_set)) {
3057 *devnamep = sset_pop(&ofproto->port_poll_set);
3062 port_poll_wait(const struct ofproto *ofproto_)
3064 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3065 dpif_port_poll_wait(ofproto->backer->dpif);
3069 port_is_lacp_current(const struct ofport *ofport_)
3071 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3072 return (ofport->bundle && ofport->bundle->lacp
3073 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3077 /* Upcall handling. */
3079 /* Flow miss batching.
3081 * Some dpifs implement operations faster when you hand them off in a batch.
3082 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3083 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3084 * more packets, plus possibly installing the flow in the dpif.
3086 * So far we only batch the operations that affect flow setup time the most.
3087 * It's possible to batch more than that, but the benefit might be minimal. */
3089 struct hmap_node hmap_node;
3090 struct ofproto_dpif *ofproto;
3092 enum odp_key_fitness key_fitness;
3093 const struct nlattr *key;
3095 ovs_be16 initial_tci;
3096 struct list packets;
3097 enum dpif_upcall_type upcall_type;
3098 uint32_t odp_in_port;
3101 struct flow_miss_op {
3102 struct dpif_op dpif_op;
3103 struct subfacet *subfacet; /* Subfacet */
3104 void *garbage; /* Pointer to pass to free(), NULL if none. */
3105 uint64_t stub[1024 / 8]; /* Temporary buffer. */
3108 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3109 * OpenFlow controller as necessary according to their individual
3110 * configurations. */
3112 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3113 const struct flow *flow)
3115 struct ofputil_packet_in pin;
3117 pin.packet = packet->data;
3118 pin.packet_len = packet->size;
3119 pin.reason = OFPR_NO_MATCH;
3120 pin.controller_id = 0;
3125 pin.send_len = 0; /* not used for flow table misses */
3127 flow_get_metadata(flow, &pin.fmd);
3129 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3132 static enum slow_path_reason
3133 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3134 const struct ofpbuf *packet)
3136 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
3142 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3144 cfm_process_heartbeat(ofport->cfm, packet);
3147 } else if (ofport->bundle && ofport->bundle->lacp
3148 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3150 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3153 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3155 stp_process_packet(ofport, packet);
3162 static struct flow_miss *
3163 flow_miss_find(struct hmap *todo, const struct flow *flow, uint32_t hash)
3165 struct flow_miss *miss;
3167 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3168 if (flow_equal(&miss->flow, flow)) {
3176 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3177 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3178 * 'miss' is associated with a subfacet the caller must also initialize the
3179 * returned op->subfacet, and if anything needs to be freed after processing
3180 * the op, the caller must initialize op->garbage also. */
3182 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3183 struct flow_miss_op *op)
3185 if (miss->flow.vlan_tci != miss->initial_tci) {
3186 /* This packet was received on a VLAN splinter port. We
3187 * added a VLAN to the packet to make the packet resemble
3188 * the flow, but the actions were composed assuming that
3189 * the packet contained no VLAN. So, we must remove the
3190 * VLAN header from the packet before trying to execute the
3192 eth_pop_vlan(packet);
3195 op->subfacet = NULL;
3197 op->dpif_op.type = DPIF_OP_EXECUTE;
3198 op->dpif_op.u.execute.key = miss->key;
3199 op->dpif_op.u.execute.key_len = miss->key_len;
3200 op->dpif_op.u.execute.packet = packet;
3203 /* Helper for handle_flow_miss_without_facet() and
3204 * handle_flow_miss_with_facet(). */
3206 handle_flow_miss_common(struct rule_dpif *rule,
3207 struct ofpbuf *packet, const struct flow *flow)
3209 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3211 ofproto->n_matches++;
3213 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3215 * Extra-special case for fail-open mode.
3217 * We are in fail-open mode and the packet matched the fail-open
3218 * rule, but we are connected to a controller too. We should send
3219 * the packet up to the controller in the hope that it will try to
3220 * set up a flow and thereby allow us to exit fail-open.
3222 * See the top-level comment in fail-open.c for more information.
3224 send_packet_in_miss(ofproto, packet, flow);
3228 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3229 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3230 * installing a datapath flow. The answer is usually "yes" (a return value of
3231 * true). However, for short flows the cost of bookkeeping is much higher than
3232 * the benefits, so when the datapath holds a large number of flows we impose
3233 * some heuristics to decide which flows are likely to be worth tracking. */
3235 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3236 struct flow_miss *miss, uint32_t hash)
3238 if (!ofproto->governor) {
3241 n_subfacets = hmap_count(&ofproto->subfacets);
3242 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3246 ofproto->governor = governor_create(ofproto->up.name);
3249 return governor_should_install_flow(ofproto->governor, hash,
3250 list_size(&miss->packets));
3253 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3254 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3255 * increment '*n_ops'. */
3257 handle_flow_miss_without_facet(struct flow_miss *miss,
3258 struct rule_dpif *rule,
3259 struct flow_miss_op *ops, size_t *n_ops)
3261 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3262 long long int now = time_msec();
3263 struct action_xlate_ctx ctx;
3264 struct ofpbuf *packet;
3266 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3267 struct flow_miss_op *op = &ops[*n_ops];
3268 struct dpif_flow_stats stats;
3269 struct ofpbuf odp_actions;
3271 COVERAGE_INC(facet_suppress);
3273 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3275 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3276 rule_credit_stats(rule, &stats);
3278 action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci,
3280 ctx.resubmit_stats = &stats;
3281 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3284 if (odp_actions.size) {
3285 struct dpif_execute *execute = &op->dpif_op.u.execute;
3287 init_flow_miss_execute_op(miss, packet, op);
3288 execute->actions = odp_actions.data;
3289 execute->actions_len = odp_actions.size;
3290 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3294 ofpbuf_uninit(&odp_actions);
3299 /* Handles 'miss', which matches 'facet'. May add any required datapath
3300 * operations to 'ops', incrementing '*n_ops' for each new op.
3302 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3303 * This is really important only for new facets: if we just called time_msec()
3304 * here, then the new subfacet or its packets could look (occasionally) as
3305 * though it was used some time after the facet was used. That can make a
3306 * one-packet flow look like it has a nonzero duration, which looks odd in
3307 * e.g. NetFlow statistics. */
3309 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3311 struct flow_miss_op *ops, size_t *n_ops)
3313 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3314 enum subfacet_path want_path;
3315 struct subfacet *subfacet;
3316 struct ofpbuf *packet;
3318 subfacet = subfacet_create(facet, miss, now);
3320 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3321 struct flow_miss_op *op = &ops[*n_ops];
3322 struct dpif_flow_stats stats;
3323 struct ofpbuf odp_actions;
3325 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3327 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3328 if (!subfacet->actions || subfacet->slow) {
3329 subfacet_make_actions(subfacet, packet, &odp_actions);
3332 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3333 subfacet_update_stats(subfacet, &stats);
3335 if (subfacet->actions_len) {
3336 struct dpif_execute *execute = &op->dpif_op.u.execute;
3338 init_flow_miss_execute_op(miss, packet, op);
3339 op->subfacet = subfacet;
3340 if (!subfacet->slow) {
3341 execute->actions = subfacet->actions;
3342 execute->actions_len = subfacet->actions_len;
3343 ofpbuf_uninit(&odp_actions);
3345 execute->actions = odp_actions.data;
3346 execute->actions_len = odp_actions.size;
3347 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3352 ofpbuf_uninit(&odp_actions);
3356 want_path = subfacet_want_path(subfacet->slow);
3357 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3358 struct flow_miss_op *op = &ops[(*n_ops)++];
3359 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3361 op->subfacet = subfacet;
3363 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3364 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3365 put->key = miss->key;
3366 put->key_len = miss->key_len;
3367 if (want_path == SF_FAST_PATH) {
3368 put->actions = subfacet->actions;
3369 put->actions_len = subfacet->actions_len;
3371 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
3372 op->stub, sizeof op->stub,
3373 &put->actions, &put->actions_len);
3379 /* Handles flow miss 'miss'. May add any required datapath operations
3380 * to 'ops', incrementing '*n_ops' for each new op. */
3382 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3385 struct ofproto_dpif *ofproto = miss->ofproto;
3386 struct facet *facet;
3390 /* The caller must ensure that miss->hmap_node.hash contains
3391 * flow_hash(miss->flow, 0). */
3392 hash = miss->hmap_node.hash;
3394 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3396 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
3398 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
3399 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
3403 facet = facet_create(rule, &miss->flow, hash);
3408 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3411 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3412 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3413 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3414 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3415 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3416 * 'packet' ingressed.
3418 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3419 * 'flow''s in_port to OFPP_NONE.
3421 * This function does post-processing on data returned from
3422 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3423 * of the upcall processing logic. In particular, if the extracted in_port is
3424 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3425 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3426 * a VLAN header onto 'packet' (if it is nonnull).
3428 * Optionally, if nonnull, sets '*initial_tci' to the VLAN TCI with which the
3429 * packet was really received, that is, the actual VLAN TCI extracted by
3430 * odp_flow_key_to_flow(). (This differs from the value returned in
3431 * flow->vlan_tci only for packets received on VLAN splinters.)
3433 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3434 * or some other positive errno if there are other problems. */
3436 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3437 const struct nlattr *key, size_t key_len,
3438 struct flow *flow, enum odp_key_fitness *fitnessp,
3439 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3440 ovs_be16 *initial_tci)
3442 const struct ofport_dpif *port;
3443 enum odp_key_fitness fitness;
3446 fitness = odp_flow_key_to_flow(key, key_len, flow);
3447 if (fitness == ODP_FIT_ERROR) {
3453 *initial_tci = flow->vlan_tci;
3457 *odp_in_port = flow->in_port;
3460 port = odp_port_to_ofport(backer, flow->in_port);
3462 flow->in_port = OFPP_NONE;
3463 error = ofproto ? ENODEV : 0;
3468 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3471 flow->in_port = port->up.ofp_port;
3472 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3474 /* Make the packet resemble the flow, so that it gets sent to an
3475 * OpenFlow controller properly, so that it looks correct for
3476 * sFlow, and so that flow_extract() will get the correct vlan_tci
3477 * if it is called on 'packet'.
3479 * The allocated space inside 'packet' probably also contains
3480 * 'key', that is, both 'packet' and 'key' are probably part of a
3481 * struct dpif_upcall (see the large comment on that structure
3482 * definition), so pushing data on 'packet' is in general not a
3483 * good idea since it could overwrite 'key' or free it as a side
3484 * effect. However, it's OK in this special case because we know
3485 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
3486 * will just overwrite the 4-byte "struct nlattr", which is fine
3487 * since we don't need that header anymore. */
3488 eth_push_vlan(packet, flow->vlan_tci);
3491 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
3492 if (fitness == ODP_FIT_PERFECT) {
3493 fitness = ODP_FIT_TOO_MUCH;
3500 *fitnessp = fitness;
3506 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3509 struct dpif_upcall *upcall;
3510 struct flow_miss *miss;
3511 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3512 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3513 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3523 /* Construct the to-do list.
3525 * This just amounts to extracting the flow from each packet and sticking
3526 * the packets that have the same flow in the same "flow_miss" structure so
3527 * that we can process them together. */
3530 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3531 struct flow_miss *miss = &misses[n_misses];
3532 struct flow_miss *existing_miss;
3533 struct ofproto_dpif *ofproto;
3534 uint32_t odp_in_port;
3539 error = ofproto_receive(backer, upcall->packet, upcall->key,
3540 upcall->key_len, &flow, &miss->key_fitness,
3541 &ofproto, &odp_in_port, &miss->initial_tci);
3542 if (error == ENODEV) {
3543 /* Received packet on port for which we couldn't associate
3544 * an ofproto. This can happen if a port is removed while
3545 * traffic is being received. Print a rate-limited message
3546 * in case it happens frequently. */
3547 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
3553 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3554 &flow.tunnel, flow.in_port, &miss->flow);
3556 /* Add other packets to a to-do list. */
3557 hash = flow_hash(&miss->flow, 0);
3558 existing_miss = flow_miss_find(&todo, &miss->flow, hash);
3559 if (!existing_miss) {
3560 hmap_insert(&todo, &miss->hmap_node, hash);
3561 miss->ofproto = ofproto;
3562 miss->key = upcall->key;
3563 miss->key_len = upcall->key_len;
3564 miss->upcall_type = upcall->type;
3565 miss->odp_in_port = odp_in_port;
3566 list_init(&miss->packets);
3570 miss = existing_miss;
3572 list_push_back(&miss->packets, &upcall->packet->list_node);
3575 /* Process each element in the to-do list, constructing the set of
3576 * operations to batch. */
3578 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3579 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3581 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3583 /* Execute batch. */
3584 for (i = 0; i < n_ops; i++) {
3585 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3587 dpif_operate(backer->dpif, dpif_ops, n_ops);
3589 /* Free memory and update facets. */
3590 for (i = 0; i < n_ops; i++) {
3591 struct flow_miss_op *op = &flow_miss_ops[i];
3593 switch (op->dpif_op.type) {
3594 case DPIF_OP_EXECUTE:
3597 case DPIF_OP_FLOW_PUT:
3598 if (!op->dpif_op.error) {
3599 op->subfacet->path = subfacet_want_path(op->subfacet->slow);
3603 case DPIF_OP_FLOW_DEL:
3609 hmap_destroy(&todo);
3612 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
3613 classify_upcall(const struct dpif_upcall *upcall)
3615 union user_action_cookie cookie;
3617 /* First look at the upcall type. */
3618 switch (upcall->type) {
3619 case DPIF_UC_ACTION:
3625 case DPIF_N_UC_TYPES:
3627 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3631 /* "action" upcalls need a closer look. */
3632 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3633 switch (cookie.type) {
3634 case USER_ACTION_COOKIE_SFLOW:
3635 return SFLOW_UPCALL;
3637 case USER_ACTION_COOKIE_SLOW_PATH:
3640 case USER_ACTION_COOKIE_UNSPEC:
3642 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
3648 handle_sflow_upcall(struct dpif_backer *backer,
3649 const struct dpif_upcall *upcall)
3651 struct ofproto_dpif *ofproto;
3652 union user_action_cookie cookie;
3654 uint32_t odp_in_port;
3656 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3657 &flow, NULL, &ofproto, &odp_in_port, NULL)
3658 || !ofproto->sflow) {
3662 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3663 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3664 odp_in_port, &cookie);
3668 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3670 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3671 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3672 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3677 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
3680 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3681 struct dpif_upcall *upcall = &misses[n_misses];
3682 struct ofpbuf *buf = &miss_bufs[n_misses];
3685 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3686 sizeof miss_buf_stubs[n_misses]);
3687 error = dpif_recv(backer->dpif, upcall, buf);
3693 switch (classify_upcall(upcall)) {
3695 /* Handle it later. */
3700 handle_sflow_upcall(backer, upcall);
3710 /* Handle deferred MISS_UPCALL processing. */
3711 handle_miss_upcalls(backer, misses, n_misses);
3712 for (i = 0; i < n_misses; i++) {
3713 ofpbuf_uninit(&miss_bufs[i]);
3719 /* Flow expiration. */
3721 static int subfacet_max_idle(const struct ofproto_dpif *);
3722 static void update_stats(struct dpif_backer *);
3723 static void rule_expire(struct rule_dpif *);
3724 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
3726 /* This function is called periodically by run(). Its job is to collect
3727 * updates for the flows that have been installed into the datapath, most
3728 * importantly when they last were used, and then use that information to
3729 * expire flows that have not been used recently.
3731 * Returns the number of milliseconds after which it should be called again. */
3733 expire(struct dpif_backer *backer)
3735 struct ofproto_dpif *ofproto;
3736 int max_idle = INT32_MAX;
3738 /* Update stats for each flow in the backer. */
3739 update_stats(backer);
3741 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3742 struct rule_dpif *rule, *next_rule;
3743 struct oftable *table;
3746 if (ofproto->backer != backer) {
3750 /* Expire subfacets that have been idle too long. */
3751 dp_max_idle = subfacet_max_idle(ofproto);
3752 expire_subfacets(ofproto, dp_max_idle);
3754 max_idle = MIN(max_idle, dp_max_idle);
3756 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3758 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
3759 struct cls_cursor cursor;
3761 cls_cursor_init(&cursor, &table->cls, NULL);
3762 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
3767 /* All outstanding data in existing flows has been accounted, so it's a
3768 * good time to do bond rebalancing. */
3769 if (ofproto->has_bonded_bundles) {
3770 struct ofbundle *bundle;
3772 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3774 bond_rebalance(bundle->bond, &backer->revalidate_set);
3780 return MIN(max_idle, 1000);
3783 /* Updates flow table statistics given that the datapath just reported 'stats'
3784 * as 'subfacet''s statistics. */
3786 update_subfacet_stats(struct subfacet *subfacet,
3787 const struct dpif_flow_stats *stats)
3789 struct facet *facet = subfacet->facet;
3791 if (stats->n_packets >= subfacet->dp_packet_count) {
3792 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
3793 facet->packet_count += extra;
3795 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3798 if (stats->n_bytes >= subfacet->dp_byte_count) {
3799 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
3801 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3804 subfacet->dp_packet_count = stats->n_packets;
3805 subfacet->dp_byte_count = stats->n_bytes;
3807 facet->tcp_flags |= stats->tcp_flags;
3809 subfacet_update_time(subfacet, stats->used);
3810 if (facet->accounted_bytes < facet->byte_count) {
3812 facet_account(facet);
3813 facet->accounted_bytes = facet->byte_count;
3815 facet_push_stats(facet);
3818 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3819 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3821 delete_unexpected_flow(struct ofproto_dpif *ofproto,
3822 const struct nlattr *key, size_t key_len)
3824 if (!VLOG_DROP_WARN(&rl)) {
3828 odp_flow_key_format(key, key_len, &s);
3829 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
3833 COVERAGE_INC(facet_unexpected);
3834 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
3837 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3839 * This function also pushes statistics updates to rules which each facet
3840 * resubmits into. Generally these statistics will be accurate. However, if a
3841 * facet changes the rule it resubmits into at some time in between
3842 * update_stats() runs, it is possible that statistics accrued to the
3843 * old rule will be incorrectly attributed to the new rule. This could be
3844 * avoided by calling update_stats() whenever rules are created or
3845 * deleted. However, the performance impact of making so many calls to the
3846 * datapath do not justify the benefit of having perfectly accurate statistics.
3849 update_stats(struct dpif_backer *backer)
3851 const struct dpif_flow_stats *stats;
3852 struct dpif_flow_dump dump;
3853 const struct nlattr *key;
3856 dpif_flow_dump_start(&dump, backer->dpif);
3857 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
3859 struct subfacet *subfacet;
3860 enum odp_key_fitness fitness;
3861 struct ofproto_dpif *ofproto;
3862 struct ofport_dpif *port;
3865 fitness = odp_flow_key_to_flow(key, key_len, &flow);
3866 if (fitness == ODP_FIT_ERROR) {
3870 port = odp_port_to_ofport(backer, flow.in_port);
3872 /* This flow is for a port for which we couldn't associate an
3873 * ofproto. This can happen if a port is removed while
3874 * traffic is being received. Ignore this flow, since it
3875 * will get timed out. */
3879 ofproto = ofproto_dpif_cast(port->up.ofproto);
3880 flow.in_port = port->up.ofp_port;
3881 key_hash = odp_flow_key_hash(key, key_len);
3883 subfacet = subfacet_find(ofproto, key, key_len, key_hash, &flow);
3884 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3886 update_subfacet_stats(subfacet, stats);
3890 /* Stats are updated per-packet. */
3893 case SF_NOT_INSTALLED:
3895 delete_unexpected_flow(ofproto, key, key_len);
3899 dpif_flow_dump_done(&dump);
3902 /* Calculates and returns the number of milliseconds of idle time after which
3903 * subfacets should expire from the datapath. When a subfacet expires, we fold
3904 * its statistics into its facet, and when a facet's last subfacet expires, we
3905 * fold its statistic into its rule. */
3907 subfacet_max_idle(const struct ofproto_dpif *ofproto)
3910 * Idle time histogram.
3912 * Most of the time a switch has a relatively small number of subfacets.
3913 * When this is the case we might as well keep statistics for all of them
3914 * in userspace and to cache them in the kernel datapath for performance as
3917 * As the number of subfacets increases, the memory required to maintain
3918 * statistics about them in userspace and in the kernel becomes
3919 * significant. However, with a large number of subfacets it is likely
3920 * that only a few of them are "heavy hitters" that consume a large amount
3921 * of bandwidth. At this point, only heavy hitters are worth caching in
3922 * the kernel and maintaining in userspaces; other subfacets we can
3925 * The technique used to compute the idle time is to build a histogram with
3926 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3927 * that is installed in the kernel gets dropped in the appropriate bucket.
3928 * After the histogram has been built, we compute the cutoff so that only
3929 * the most-recently-used 1% of subfacets (but at least
3930 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3931 * the most-recently-used bucket of subfacets is kept, so actually an
3932 * arbitrary number of subfacets can be kept in any given expiration run
3933 * (though the next run will delete most of those unless they receive
3936 * This requires a second pass through the subfacets, in addition to the
3937 * pass made by update_stats(), because the former function never looks at
3938 * uninstallable subfacets.
3940 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3941 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3942 int buckets[N_BUCKETS] = { 0 };
3943 int total, subtotal, bucket;
3944 struct subfacet *subfacet;
3948 total = hmap_count(&ofproto->subfacets);
3949 if (total <= ofproto->up.flow_eviction_threshold) {
3950 return N_BUCKETS * BUCKET_WIDTH;
3953 /* Build histogram. */
3955 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3956 long long int idle = now - subfacet->used;
3957 int bucket = (idle <= 0 ? 0
3958 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3959 : (unsigned int) idle / BUCKET_WIDTH);
3963 /* Find the first bucket whose flows should be expired. */
3964 subtotal = bucket = 0;
3966 subtotal += buckets[bucket++];
3967 } while (bucket < N_BUCKETS &&
3968 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3970 if (VLOG_IS_DBG_ENABLED()) {
3974 ds_put_cstr(&s, "keep");
3975 for (i = 0; i < N_BUCKETS; i++) {
3977 ds_put_cstr(&s, ", drop");
3980 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3983 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3987 return bucket * BUCKET_WIDTH;
3991 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3993 /* Cutoff time for most flows. */
3994 long long int normal_cutoff = time_msec() - dp_max_idle;
3996 /* We really want to keep flows for special protocols around, so use a more
3997 * conservative cutoff. */
3998 long long int special_cutoff = time_msec() - 10000;
4000 struct subfacet *subfacet, *next_subfacet;
4001 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4005 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4006 &ofproto->subfacets) {
4007 long long int cutoff;
4009 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
4012 if (subfacet->used < cutoff) {
4013 if (subfacet->path != SF_NOT_INSTALLED) {
4014 batch[n_batch++] = subfacet;
4015 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4016 subfacet_destroy_batch(ofproto, batch, n_batch);
4020 subfacet_destroy(subfacet);
4026 subfacet_destroy_batch(ofproto, batch, n_batch);
4030 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4031 * then delete it entirely. */
4033 rule_expire(struct rule_dpif *rule)
4035 struct facet *facet, *next_facet;
4039 if (rule->up.pending) {
4040 /* We'll have to expire it later. */
4044 /* Has 'rule' expired? */
4046 if (rule->up.hard_timeout
4047 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4048 reason = OFPRR_HARD_TIMEOUT;
4049 } else if (rule->up.idle_timeout
4050 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4051 reason = OFPRR_IDLE_TIMEOUT;
4056 COVERAGE_INC(ofproto_dpif_expired);
4058 /* Update stats. (This is a no-op if the rule expired due to an idle
4059 * timeout, because that only happens when the rule has no facets left.) */
4060 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4061 facet_remove(facet);
4064 /* Get rid of the rule. */
4065 ofproto_rule_expire(&rule->up, reason);
4070 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
4072 * The caller must already have determined that no facet with an identical
4073 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
4074 * the ofproto's classifier table.
4076 * 'hash' must be the return value of flow_hash(flow, 0).
4078 * The facet will initially have no subfacets. The caller should create (at
4079 * least) one subfacet with subfacet_create(). */
4080 static struct facet *
4081 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
4083 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4084 struct facet *facet;
4086 facet = xzalloc(sizeof *facet);
4087 facet->used = time_msec();
4088 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4089 list_push_back(&rule->facets, &facet->list_node);
4091 facet->flow = *flow;
4092 list_init(&facet->subfacets);
4093 netflow_flow_init(&facet->nf_flow);
4094 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4100 facet_free(struct facet *facet)
4105 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4106 * 'packet', which arrived on 'in_port'.
4108 * Takes ownership of 'packet'. */
4110 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4111 const struct nlattr *odp_actions, size_t actions_len,
4112 struct ofpbuf *packet)
4114 struct odputil_keybuf keybuf;
4118 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4119 odp_flow_key_from_flow(&key, flow,
4120 ofp_port_to_odp_port(ofproto, flow->in_port));
4122 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4123 odp_actions, actions_len, packet);
4125 ofpbuf_delete(packet);
4129 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4131 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4132 * rule's statistics, via subfacet_uninstall().
4134 * - Removes 'facet' from its rule and from ofproto->facets.
4137 facet_remove(struct facet *facet)
4139 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4140 struct subfacet *subfacet, *next_subfacet;
4142 ovs_assert(!list_is_empty(&facet->subfacets));
4144 /* First uninstall all of the subfacets to get final statistics. */
4145 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4146 subfacet_uninstall(subfacet);
4149 /* Flush the final stats to the rule.
4151 * This might require us to have at least one subfacet around so that we
4152 * can use its actions for accounting in facet_account(), which is why we
4153 * have uninstalled but not yet destroyed the subfacets. */
4154 facet_flush_stats(facet);
4156 /* Now we're really all done so destroy everything. */
4157 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4158 &facet->subfacets) {
4159 subfacet_destroy__(subfacet);
4161 hmap_remove(&ofproto->facets, &facet->hmap_node);
4162 list_remove(&facet->list_node);
4166 /* Feed information from 'facet' back into the learning table to keep it in
4167 * sync with what is actually flowing through the datapath. */
4169 facet_learn(struct facet *facet)
4171 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4172 struct action_xlate_ctx ctx;
4174 if (!facet->has_learn
4175 && !facet->has_normal
4176 && (!facet->has_fin_timeout
4177 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4181 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4182 facet->flow.vlan_tci,
4183 facet->rule, facet->tcp_flags, NULL);
4184 ctx.may_learn = true;
4185 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
4186 facet->rule->up.ofpacts_len);
4190 facet_account(struct facet *facet)
4192 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4193 struct subfacet *subfacet;
4194 const struct nlattr *a;
4199 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
4202 n_bytes = facet->byte_count - facet->accounted_bytes;
4204 /* This loop feeds byte counters to bond_account() for rebalancing to use
4205 * as a basis. We also need to track the actual VLAN on which the packet
4206 * is going to be sent to ensure that it matches the one passed to
4207 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4210 * We use the actions from an arbitrary subfacet because they should all
4211 * be equally valid for our purpose. */
4212 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
4213 struct subfacet, list_node);
4214 vlan_tci = facet->flow.vlan_tci;
4215 NL_ATTR_FOR_EACH_UNSAFE (a, left,
4216 subfacet->actions, subfacet->actions_len) {
4217 const struct ovs_action_push_vlan *vlan;
4218 struct ofport_dpif *port;
4220 switch (nl_attr_type(a)) {
4221 case OVS_ACTION_ATTR_OUTPUT:
4222 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4223 if (port && port->bundle && port->bundle->bond) {
4224 bond_account(port->bundle->bond, &facet->flow,
4225 vlan_tci_to_vid(vlan_tci), n_bytes);
4229 case OVS_ACTION_ATTR_POP_VLAN:
4230 vlan_tci = htons(0);
4233 case OVS_ACTION_ATTR_PUSH_VLAN:
4234 vlan = nl_attr_get(a);
4235 vlan_tci = vlan->vlan_tci;
4241 /* Returns true if the only action for 'facet' is to send to the controller.
4242 * (We don't report NetFlow expiration messages for such facets because they
4243 * are just part of the control logic for the network, not real traffic). */
4245 facet_is_controller_flow(struct facet *facet)
4248 const struct rule *rule = &facet->rule->up;
4249 const struct ofpact *ofpacts = rule->ofpacts;
4250 size_t ofpacts_len = rule->ofpacts_len;
4252 if (ofpacts_len > 0 &&
4253 ofpacts->type == OFPACT_CONTROLLER &&
4254 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4261 /* Folds all of 'facet''s statistics into its rule. Also updates the
4262 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4263 * 'facet''s statistics in the datapath should have been zeroed and folded into
4264 * its packet and byte counts before this function is called. */
4266 facet_flush_stats(struct facet *facet)
4268 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4269 struct subfacet *subfacet;
4271 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4272 ovs_assert(!subfacet->dp_byte_count);
4273 ovs_assert(!subfacet->dp_packet_count);
4276 facet_push_stats(facet);
4277 if (facet->accounted_bytes < facet->byte_count) {
4278 facet_account(facet);
4279 facet->accounted_bytes = facet->byte_count;
4282 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4283 struct ofexpired expired;
4284 expired.flow = facet->flow;
4285 expired.packet_count = facet->packet_count;
4286 expired.byte_count = facet->byte_count;
4287 expired.used = facet->used;
4288 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4291 facet->rule->packet_count += facet->packet_count;
4292 facet->rule->byte_count += facet->byte_count;
4294 /* Reset counters to prevent double counting if 'facet' ever gets
4296 facet_reset_counters(facet);
4298 netflow_flow_clear(&facet->nf_flow);
4299 facet->tcp_flags = 0;
4302 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4303 * Returns it if found, otherwise a null pointer.
4305 * 'hash' must be the return value of flow_hash(flow, 0).
4307 * The returned facet might need revalidation; use facet_lookup_valid()
4308 * instead if that is important. */
4309 static struct facet *
4310 facet_find(struct ofproto_dpif *ofproto,
4311 const struct flow *flow, uint32_t hash)
4313 struct facet *facet;
4315 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4316 if (flow_equal(flow, &facet->flow)) {
4324 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4325 * Returns it if found, otherwise a null pointer.
4327 * 'hash' must be the return value of flow_hash(flow, 0).
4329 * The returned facet is guaranteed to be valid. */
4330 static struct facet *
4331 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4334 struct facet *facet;
4336 facet = facet_find(ofproto, flow, hash);
4338 && (ofproto->backer->need_revalidate
4339 || tag_set_intersects(&ofproto->backer->revalidate_set,
4341 facet_revalidate(facet);
4348 subfacet_path_to_string(enum subfacet_path path)
4351 case SF_NOT_INSTALLED:
4352 return "not installed";
4354 return "in fast path";
4356 return "in slow path";
4362 /* Returns the path in which a subfacet should be installed if its 'slow'
4363 * member has the specified value. */
4364 static enum subfacet_path
4365 subfacet_want_path(enum slow_path_reason slow)
4367 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
4370 /* Returns true if 'subfacet' needs to have its datapath flow updated,
4371 * supposing that its actions have been recalculated as 'want_actions' and that
4372 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
4374 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
4375 const struct ofpbuf *want_actions)
4377 enum subfacet_path want_path = subfacet_want_path(slow);
4378 return (want_path != subfacet->path
4379 || (want_path == SF_FAST_PATH
4380 && (subfacet->actions_len != want_actions->size
4381 || memcmp(subfacet->actions, want_actions->data,
4382 subfacet->actions_len))));
4386 facet_check_consistency(struct facet *facet)
4388 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4390 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4392 uint64_t odp_actions_stub[1024 / 8];
4393 struct ofpbuf odp_actions;
4395 struct rule_dpif *rule;
4396 struct subfacet *subfacet;
4397 bool may_log = false;
4400 /* Check the rule for consistency. */
4401 rule = rule_dpif_lookup(ofproto, &facet->flow);
4402 ok = rule == facet->rule;
4404 may_log = !VLOG_DROP_WARN(&rl);
4409 flow_format(&s, &facet->flow);
4410 ds_put_format(&s, ": facet associated with wrong rule (was "
4411 "table=%"PRIu8",", facet->rule->up.table_id);
4412 cls_rule_format(&facet->rule->up.cr, &s);
4413 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4415 cls_rule_format(&rule->up.cr, &s);
4416 ds_put_char(&s, ')');
4418 VLOG_WARN("%s", ds_cstr(&s));
4423 /* Check the datapath actions for consistency. */
4424 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4425 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4426 enum subfacet_path want_path;
4427 struct odputil_keybuf keybuf;
4428 struct action_xlate_ctx ctx;
4432 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4433 subfacet->initial_tci, rule, 0, NULL);
4434 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
4437 if (subfacet->path == SF_NOT_INSTALLED) {
4438 /* This only happens if the datapath reported an error when we
4439 * tried to install the flow. Don't flag another error here. */
4443 want_path = subfacet_want_path(subfacet->slow);
4444 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
4445 /* The actions for slow-path flows may legitimately vary from one
4446 * packet to the next. We're done. */
4450 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
4454 /* Inconsistency! */
4456 may_log = !VLOG_DROP_WARN(&rl);
4460 /* Rate-limited, skip reporting. */
4465 subfacet_get_key(subfacet, &keybuf, &key);
4466 odp_flow_key_format(key.data, key.size, &s);
4468 ds_put_cstr(&s, ": inconsistency in subfacet");
4469 if (want_path != subfacet->path) {
4470 enum odp_key_fitness fitness = subfacet->key_fitness;
4472 ds_put_format(&s, " (%s, fitness=%s)",
4473 subfacet_path_to_string(subfacet->path),
4474 odp_key_fitness_to_string(fitness));
4475 ds_put_format(&s, " (should have been %s)",
4476 subfacet_path_to_string(want_path));
4477 } else if (want_path == SF_FAST_PATH) {
4478 ds_put_cstr(&s, " (actions were: ");
4479 format_odp_actions(&s, subfacet->actions,
4480 subfacet->actions_len);
4481 ds_put_cstr(&s, ") (correct actions: ");
4482 format_odp_actions(&s, odp_actions.data, odp_actions.size);
4483 ds_put_char(&s, ')');
4485 ds_put_cstr(&s, " (actions: ");
4486 format_odp_actions(&s, subfacet->actions,
4487 subfacet->actions_len);
4488 ds_put_char(&s, ')');
4490 VLOG_WARN("%s", ds_cstr(&s));
4493 ofpbuf_uninit(&odp_actions);
4498 /* Re-searches the classifier for 'facet':
4500 * - If the rule found is different from 'facet''s current rule, moves
4501 * 'facet' to the new rule and recompiles its actions.
4503 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4504 * where it is and recompiles its actions anyway. */
4506 facet_revalidate(struct facet *facet)
4508 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4510 struct nlattr *odp_actions;
4513 struct actions *new_actions;
4515 struct action_xlate_ctx ctx;
4516 uint64_t odp_actions_stub[1024 / 8];
4517 struct ofpbuf odp_actions;
4519 struct rule_dpif *new_rule;
4520 struct subfacet *subfacet;
4523 COVERAGE_INC(facet_revalidate);
4525 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
4527 /* Calculate new datapath actions.
4529 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4530 * emit a NetFlow expiration and, if so, we need to have the old state
4531 * around to properly compose it. */
4533 /* If the datapath actions changed or the installability changed,
4534 * then we need to talk to the datapath. */
4537 memset(&ctx, 0, sizeof ctx);
4538 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4539 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4540 enum slow_path_reason slow;
4542 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4543 subfacet->initial_tci, new_rule, 0, NULL);
4544 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
4547 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4548 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
4549 struct dpif_flow_stats stats;
4551 subfacet_install(subfacet,
4552 odp_actions.data, odp_actions.size, &stats, slow);
4553 subfacet_update_stats(subfacet, &stats);
4556 new_actions = xcalloc(list_size(&facet->subfacets),
4557 sizeof *new_actions);
4559 new_actions[i].odp_actions = xmemdup(odp_actions.data,
4561 new_actions[i].actions_len = odp_actions.size;
4566 ofpbuf_uninit(&odp_actions);
4569 facet_flush_stats(facet);
4572 /* Update 'facet' now that we've taken care of all the old state. */
4573 facet->tags = ctx.tags;
4574 facet->nf_flow.output_iface = ctx.nf_output_iface;
4575 facet->has_learn = ctx.has_learn;
4576 facet->has_normal = ctx.has_normal;
4577 facet->has_fin_timeout = ctx.has_fin_timeout;
4578 facet->mirrors = ctx.mirrors;
4581 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4582 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4584 if (new_actions && new_actions[i].odp_actions) {
4585 free(subfacet->actions);
4586 subfacet->actions = new_actions[i].odp_actions;
4587 subfacet->actions_len = new_actions[i].actions_len;
4593 if (facet->rule != new_rule) {
4594 COVERAGE_INC(facet_changed_rule);
4595 list_remove(&facet->list_node);
4596 list_push_back(&new_rule->facets, &facet->list_node);
4597 facet->rule = new_rule;
4598 facet->used = new_rule->up.created;
4599 facet->prev_used = facet->used;
4603 /* Updates 'facet''s used time. Caller is responsible for calling
4604 * facet_push_stats() to update the flows which 'facet' resubmits into. */
4606 facet_update_time(struct facet *facet, long long int used)
4608 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4609 if (used > facet->used) {
4611 ofproto_rule_update_used(&facet->rule->up, used);
4612 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
4617 facet_reset_counters(struct facet *facet)
4619 facet->packet_count = 0;
4620 facet->byte_count = 0;
4621 facet->prev_packet_count = 0;
4622 facet->prev_byte_count = 0;
4623 facet->accounted_bytes = 0;
4627 facet_push_stats(struct facet *facet)
4629 struct dpif_flow_stats stats;
4631 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4632 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4633 ovs_assert(facet->used >= facet->prev_used);
4635 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4636 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4637 stats.used = facet->used;
4638 stats.tcp_flags = 0;
4640 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
4641 facet->prev_packet_count = facet->packet_count;
4642 facet->prev_byte_count = facet->byte_count;
4643 facet->prev_used = facet->used;
4645 flow_push_stats(facet->rule, &facet->flow, &stats);
4647 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
4648 facet->mirrors, stats.n_packets, stats.n_bytes);
4653 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4655 rule->packet_count += stats->n_packets;
4656 rule->byte_count += stats->n_bytes;
4657 ofproto_rule_update_used(&rule->up, stats->used);
4660 /* Pushes flow statistics to the rules which 'flow' resubmits into given
4661 * 'rule''s actions and mirrors. */
4663 flow_push_stats(struct rule_dpif *rule,
4664 const struct flow *flow, const struct dpif_flow_stats *stats)
4666 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4667 struct action_xlate_ctx ctx;
4669 ofproto_rule_update_used(&rule->up, stats->used);
4671 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule,
4673 ctx.resubmit_stats = stats;
4674 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
4675 rule->up.ofpacts_len);
4680 static struct subfacet *
4681 subfacet_find(struct ofproto_dpif *ofproto,
4682 const struct nlattr *key, size_t key_len, uint32_t key_hash,
4683 const struct flow *flow)
4685 struct subfacet *subfacet;
4687 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4688 &ofproto->subfacets) {
4690 ? (subfacet->key_len == key_len
4691 && !memcmp(key, subfacet->key, key_len))
4692 : flow_equal(flow, &subfacet->facet->flow)) {
4700 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4701 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4702 * existing subfacet if there is one, otherwise creates and returns a
4705 * If the returned subfacet is new, then subfacet->actions will be NULL, in
4706 * which case the caller must populate the actions with
4707 * subfacet_make_actions(). */
4708 static struct subfacet *
4709 subfacet_create(struct facet *facet, struct flow_miss *miss,
4712 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4713 enum odp_key_fitness key_fitness = miss->key_fitness;
4714 const struct nlattr *key = miss->key;
4715 size_t key_len = miss->key_len;
4717 struct subfacet *subfacet;
4719 key_hash = odp_flow_key_hash(key, key_len);
4721 if (list_is_empty(&facet->subfacets)) {
4722 subfacet = &facet->one_subfacet;
4724 subfacet = subfacet_find(ofproto, key, key_len, key_hash,
4727 if (subfacet->facet == facet) {
4731 /* This shouldn't happen. */
4732 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4733 subfacet_destroy(subfacet);
4736 subfacet = xmalloc(sizeof *subfacet);
4739 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
4740 list_push_back(&facet->subfacets, &subfacet->list_node);
4741 subfacet->facet = facet;
4742 subfacet->key_fitness = key_fitness;
4743 if (key_fitness != ODP_FIT_PERFECT) {
4744 subfacet->key = xmemdup(key, key_len);
4745 subfacet->key_len = key_len;
4747 subfacet->key = NULL;
4748 subfacet->key_len = 0;
4750 subfacet->used = now;
4751 subfacet->dp_packet_count = 0;
4752 subfacet->dp_byte_count = 0;
4753 subfacet->actions_len = 0;
4754 subfacet->actions = NULL;
4755 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
4758 subfacet->path = SF_NOT_INSTALLED;
4759 subfacet->initial_tci = miss->initial_tci;
4760 subfacet->odp_in_port = miss->odp_in_port;
4765 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4766 * its facet within 'ofproto', and frees it. */
4768 subfacet_destroy__(struct subfacet *subfacet)
4770 struct facet *facet = subfacet->facet;
4771 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4773 subfacet_uninstall(subfacet);
4774 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
4775 list_remove(&subfacet->list_node);
4776 free(subfacet->key);
4777 free(subfacet->actions);
4778 if (subfacet != &facet->one_subfacet) {
4783 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4784 * last remaining subfacet in its facet destroys the facet too. */
4786 subfacet_destroy(struct subfacet *subfacet)
4788 struct facet *facet = subfacet->facet;
4790 if (list_is_singleton(&facet->subfacets)) {
4791 /* facet_remove() needs at least one subfacet (it will remove it). */
4792 facet_remove(facet);
4794 subfacet_destroy__(subfacet);
4799 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
4800 struct subfacet **subfacets, int n)
4802 struct odputil_keybuf keybufs[SUBFACET_DESTROY_MAX_BATCH];
4803 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4804 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4805 struct ofpbuf keys[SUBFACET_DESTROY_MAX_BATCH];
4806 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4809 for (i = 0; i < n; i++) {
4810 ops[i].type = DPIF_OP_FLOW_DEL;
4811 subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]);
4812 ops[i].u.flow_del.key = keys[i].data;
4813 ops[i].u.flow_del.key_len = keys[i].size;
4814 ops[i].u.flow_del.stats = &stats[i];
4818 dpif_operate(ofproto->backer->dpif, opsp, n);
4819 for (i = 0; i < n; i++) {
4820 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4821 subfacets[i]->path = SF_NOT_INSTALLED;
4822 subfacet_destroy(subfacets[i]);
4826 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
4827 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
4828 * for use as temporary storage. */
4830 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
4834 if (!subfacet->key) {
4835 struct flow *flow = &subfacet->facet->flow;
4837 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
4838 odp_flow_key_from_flow(key, flow, subfacet->odp_in_port);
4840 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
4844 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
4845 * Translates the actions into 'odp_actions', which the caller must have
4846 * initialized and is responsible for uninitializing. */
4848 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
4849 struct ofpbuf *odp_actions)
4851 struct facet *facet = subfacet->facet;
4852 struct rule_dpif *rule = facet->rule;
4853 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4855 struct action_xlate_ctx ctx;
4857 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
4859 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
4860 facet->tags = ctx.tags;
4861 facet->has_learn = ctx.has_learn;
4862 facet->has_normal = ctx.has_normal;
4863 facet->has_fin_timeout = ctx.has_fin_timeout;
4864 facet->nf_flow.output_iface = ctx.nf_output_iface;
4865 facet->mirrors = ctx.mirrors;
4867 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4868 if (subfacet->actions_len != odp_actions->size
4869 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
4870 free(subfacet->actions);
4871 subfacet->actions_len = odp_actions->size;
4872 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
4876 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4877 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4878 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4879 * since 'subfacet' was last updated.
4881 * Returns 0 if successful, otherwise a positive errno value. */
4883 subfacet_install(struct subfacet *subfacet,
4884 const struct nlattr *actions, size_t actions_len,
4885 struct dpif_flow_stats *stats,
4886 enum slow_path_reason slow)
4888 struct facet *facet = subfacet->facet;
4889 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4890 enum subfacet_path path = subfacet_want_path(slow);
4891 uint64_t slow_path_stub[128 / 8];
4892 struct odputil_keybuf keybuf;
4893 enum dpif_flow_put_flags flags;
4897 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
4899 flags |= DPIF_FP_ZERO_STATS;
4902 if (path == SF_SLOW_PATH) {
4903 compose_slow_path(ofproto, &facet->flow, slow,
4904 slow_path_stub, sizeof slow_path_stub,
4905 &actions, &actions_len);
4908 subfacet_get_key(subfacet, &keybuf, &key);
4909 ret = dpif_flow_put(ofproto->backer->dpif, flags, key.data, key.size,
4910 actions, actions_len, stats);
4913 subfacet_reset_dp_stats(subfacet, stats);
4917 subfacet->path = path;
4923 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
4925 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
4926 stats, subfacet->slow);
4929 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4931 subfacet_uninstall(struct subfacet *subfacet)
4933 if (subfacet->path != SF_NOT_INSTALLED) {
4934 struct rule_dpif *rule = subfacet->facet->rule;
4935 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4936 struct odputil_keybuf keybuf;
4937 struct dpif_flow_stats stats;
4941 subfacet_get_key(subfacet, &keybuf, &key);
4942 error = dpif_flow_del(ofproto->backer->dpif,
4943 key.data, key.size, &stats);
4944 subfacet_reset_dp_stats(subfacet, &stats);
4946 subfacet_update_stats(subfacet, &stats);
4948 subfacet->path = SF_NOT_INSTALLED;
4950 ovs_assert(subfacet->dp_packet_count == 0);
4951 ovs_assert(subfacet->dp_byte_count == 0);
4955 /* Resets 'subfacet''s datapath statistics counters. This should be called
4956 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4957 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4958 * was reset in the datapath. 'stats' will be modified to include only
4959 * statistics new since 'subfacet' was last updated. */
4961 subfacet_reset_dp_stats(struct subfacet *subfacet,
4962 struct dpif_flow_stats *stats)
4965 && subfacet->dp_packet_count <= stats->n_packets
4966 && subfacet->dp_byte_count <= stats->n_bytes) {
4967 stats->n_packets -= subfacet->dp_packet_count;
4968 stats->n_bytes -= subfacet->dp_byte_count;
4971 subfacet->dp_packet_count = 0;
4972 subfacet->dp_byte_count = 0;
4975 /* Updates 'subfacet''s used time. The caller is responsible for calling
4976 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
4978 subfacet_update_time(struct subfacet *subfacet, long long int used)
4980 if (used > subfacet->used) {
4981 subfacet->used = used;
4982 facet_update_time(subfacet->facet, used);
4986 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4988 * Because of the meaning of a subfacet's counters, it only makes sense to do
4989 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4990 * represents a packet that was sent by hand or if it represents statistics
4991 * that have been cleared out of the datapath. */
4993 subfacet_update_stats(struct subfacet *subfacet,
4994 const struct dpif_flow_stats *stats)
4996 if (stats->n_packets || stats->used > subfacet->used) {
4997 struct facet *facet = subfacet->facet;
4999 subfacet_update_time(subfacet, stats->used);
5000 facet->packet_count += stats->n_packets;
5001 facet->byte_count += stats->n_bytes;
5002 facet->tcp_flags |= stats->tcp_flags;
5003 facet_push_stats(facet);
5004 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
5010 static struct rule_dpif *
5011 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5013 struct rule_dpif *rule;
5015 rule = rule_dpif_lookup__(ofproto, flow, 0);
5020 return rule_dpif_miss_rule(ofproto, flow);
5023 static struct rule_dpif *
5024 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5027 struct cls_rule *cls_rule;
5028 struct classifier *cls;
5030 if (table_id >= N_TABLES) {
5034 cls = &ofproto->up.tables[table_id].cls;
5035 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5036 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5037 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5038 * are unavailable. */
5039 struct flow ofpc_normal_flow = *flow;
5040 ofpc_normal_flow.tp_src = htons(0);
5041 ofpc_normal_flow.tp_dst = htons(0);
5042 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5044 cls_rule = classifier_lookup(cls, flow);
5046 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5049 static struct rule_dpif *
5050 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5052 struct ofport_dpif *port;
5054 port = get_ofp_port(ofproto, flow->in_port);
5056 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5057 return ofproto->miss_rule;
5060 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5061 return ofproto->no_packet_in_rule;
5063 return ofproto->miss_rule;
5067 complete_operation(struct rule_dpif *rule)
5069 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5071 rule_invalidate(rule);
5073 struct dpif_completion *c = xmalloc(sizeof *c);
5074 c->op = rule->up.pending;
5075 list_push_back(&ofproto->completions, &c->list_node);
5077 ofoperation_complete(rule->up.pending, 0);
5081 static struct rule *
5084 struct rule_dpif *rule = xmalloc(sizeof *rule);
5089 rule_dealloc(struct rule *rule_)
5091 struct rule_dpif *rule = rule_dpif_cast(rule_);
5096 rule_construct(struct rule *rule_)
5098 struct rule_dpif *rule = rule_dpif_cast(rule_);
5099 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5100 struct rule_dpif *victim;
5103 rule->packet_count = 0;
5104 rule->byte_count = 0;
5106 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5107 if (victim && !list_is_empty(&victim->facets)) {
5108 struct facet *facet;
5110 rule->facets = victim->facets;
5111 list_moved(&rule->facets);
5112 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5113 /* XXX: We're only clearing our local counters here. It's possible
5114 * that quite a few packets are unaccounted for in the datapath
5115 * statistics. These will be accounted to the new rule instead of
5116 * cleared as required. This could be fixed by clearing out the
5117 * datapath statistics for this facet, but currently it doesn't
5119 facet_reset_counters(facet);
5123 /* Must avoid list_moved() in this case. */
5124 list_init(&rule->facets);
5127 table_id = rule->up.table_id;
5129 rule->tag = victim->tag;
5130 } else if (table_id == 0) {
5135 miniflow_expand(&rule->up.cr.match.flow, &flow);
5136 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5137 ofproto->tables[table_id].basis);
5140 complete_operation(rule);
5145 rule_destruct(struct rule *rule_)
5147 struct rule_dpif *rule = rule_dpif_cast(rule_);
5148 struct facet *facet, *next_facet;
5150 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5151 facet_revalidate(facet);
5154 complete_operation(rule);
5158 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5160 struct rule_dpif *rule = rule_dpif_cast(rule_);
5161 struct facet *facet;
5163 /* Start from historical data for 'rule' itself that are no longer tracked
5164 * in facets. This counts, for example, facets that have expired. */
5165 *packets = rule->packet_count;
5166 *bytes = rule->byte_count;
5168 /* Add any statistics that are tracked by facets. This includes
5169 * statistical data recently updated by ofproto_update_stats() as well as
5170 * stats for packets that were executed "by hand" via dpif_execute(). */
5171 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5172 *packets += facet->packet_count;
5173 *bytes += facet->byte_count;
5178 rule_execute(struct rule *rule_, const struct flow *flow,
5179 struct ofpbuf *packet)
5181 struct rule_dpif *rule = rule_dpif_cast(rule_);
5182 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5184 struct dpif_flow_stats stats;
5186 struct action_xlate_ctx ctx;
5187 uint64_t odp_actions_stub[1024 / 8];
5188 struct ofpbuf odp_actions;
5190 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5191 rule_credit_stats(rule, &stats);
5193 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5194 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
5195 rule, stats.tcp_flags, packet);
5196 ctx.resubmit_stats = &stats;
5197 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
5199 execute_odp_actions(ofproto, flow, odp_actions.data,
5200 odp_actions.size, packet);
5202 ofpbuf_uninit(&odp_actions);
5208 rule_modify_actions(struct rule *rule_)
5210 struct rule_dpif *rule = rule_dpif_cast(rule_);
5212 complete_operation(rule);
5215 /* Sends 'packet' out 'ofport'.
5216 * May modify 'packet'.
5217 * Returns 0 if successful, otherwise a positive errno value. */
5219 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5221 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5222 struct ofpbuf key, odp_actions;
5223 struct odputil_keybuf keybuf;
5228 flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow);
5229 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5231 if (odp_port != ofport->odp_port) {
5232 eth_pop_vlan(packet);
5233 flow.vlan_tci = htons(0);
5236 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5237 odp_flow_key_from_flow(&key, &flow,
5238 ofp_port_to_odp_port(ofproto, flow.in_port));
5240 ofpbuf_init(&odp_actions, 32);
5241 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5243 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5244 error = dpif_execute(ofproto->backer->dpif,
5246 odp_actions.data, odp_actions.size,
5248 ofpbuf_uninit(&odp_actions);
5251 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5252 ofproto->up.name, odp_port, strerror(error));
5254 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5258 /* OpenFlow to datapath action translation. */
5260 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5261 struct action_xlate_ctx *);
5262 static void xlate_normal(struct action_xlate_ctx *);
5264 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5265 * The action will state 'slow' as the reason that the action is in the slow
5266 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5267 * dump-flows" output to see why a flow is in the slow path.)
5269 * The 'stub_size' bytes in 'stub' will be used to store the action.
5270 * 'stub_size' must be large enough for the action.
5272 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5275 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5276 enum slow_path_reason slow,
5277 uint64_t *stub, size_t stub_size,
5278 const struct nlattr **actionsp, size_t *actions_lenp)
5280 union user_action_cookie cookie;
5283 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5284 cookie.slow_path.unused = 0;
5285 cookie.slow_path.reason = slow;
5287 ofpbuf_use_stack(&buf, stub, stub_size);
5288 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
5289 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5290 odp_put_userspace_action(pid, &cookie, &buf);
5292 put_userspace_action(ofproto, &buf, flow, &cookie);
5294 *actionsp = buf.data;
5295 *actions_lenp = buf.size;
5299 put_userspace_action(const struct ofproto_dpif *ofproto,
5300 struct ofpbuf *odp_actions,
5301 const struct flow *flow,
5302 const union user_action_cookie *cookie)
5306 pid = dpif_port_get_pid(ofproto->backer->dpif,
5307 ofp_port_to_odp_port(ofproto, flow->in_port));
5309 return odp_put_userspace_action(pid, cookie, odp_actions);
5313 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5314 ovs_be16 vlan_tci, uint32_t odp_port,
5315 unsigned int n_outputs, union user_action_cookie *cookie)
5319 cookie->type = USER_ACTION_COOKIE_SFLOW;
5320 cookie->sflow.vlan_tci = vlan_tci;
5322 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5323 * port information") for the interpretation of cookie->output. */
5324 switch (n_outputs) {
5326 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5327 cookie->sflow.output = 0x40000000 | 256;
5331 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5333 cookie->sflow.output = ifindex;
5338 /* 0x80000000 means "multiple output ports. */
5339 cookie->sflow.output = 0x80000000 | n_outputs;
5344 /* Compose SAMPLE action for sFlow. */
5346 compose_sflow_action(const struct ofproto_dpif *ofproto,
5347 struct ofpbuf *odp_actions,
5348 const struct flow *flow,
5351 uint32_t probability;
5352 union user_action_cookie cookie;
5353 size_t sample_offset, actions_offset;
5356 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5360 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5362 /* Number of packets out of UINT_MAX to sample. */
5363 probability = dpif_sflow_get_probability(ofproto->sflow);
5364 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5366 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5367 compose_sflow_cookie(ofproto, htons(0), odp_port,
5368 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5369 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
5371 nl_msg_end_nested(odp_actions, actions_offset);
5372 nl_msg_end_nested(odp_actions, sample_offset);
5373 return cookie_offset;
5376 /* SAMPLE action must be first action in any given list of actions.
5377 * At this point we do not have all information required to build it. So try to
5378 * build sample action as complete as possible. */
5380 add_sflow_action(struct action_xlate_ctx *ctx)
5382 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5384 &ctx->flow, OVSP_NONE);
5385 ctx->sflow_odp_port = 0;
5386 ctx->sflow_n_outputs = 0;
5389 /* Fix SAMPLE action according to data collected while composing ODP actions.
5390 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5391 * USERSPACE action's user-cookie which is required for sflow. */
5393 fix_sflow_action(struct action_xlate_ctx *ctx)
5395 const struct flow *base = &ctx->base_flow;
5396 union user_action_cookie *cookie;
5398 if (!ctx->user_cookie_offset) {
5402 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
5404 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5406 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5407 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5411 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
5414 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5415 uint32_t odp_port = ofp_port_to_odp_port(ctx->ofproto, ofp_port);
5416 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
5417 uint8_t flow_nw_tos = ctx->flow.nw_tos;
5418 struct priority_to_dscp *pdscp;
5422 xlate_report(ctx, "Nonexistent output port");
5424 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5425 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5427 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5428 xlate_report(ctx, "STP not in forwarding state, skipping output");
5432 pdscp = get_priority(ofport, ctx->flow.skb_priority);
5434 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5435 ctx->flow.nw_tos |= pdscp->dscp;
5438 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
5439 ctx->flow.vlan_tci);
5440 if (out_port != odp_port) {
5441 ctx->flow.vlan_tci = htons(0);
5443 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
5444 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
5446 ctx->sflow_odp_port = odp_port;
5447 ctx->sflow_n_outputs++;
5448 ctx->nf_output_iface = ofp_port;
5449 ctx->flow.vlan_tci = flow_vlan_tci;
5450 ctx->flow.nw_tos = flow_nw_tos;
5454 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
5456 compose_output_action__(ctx, ofp_port, true);
5460 xlate_table_action(struct action_xlate_ctx *ctx,
5461 uint16_t in_port, uint8_t table_id, bool may_packet_in)
5463 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
5464 struct ofproto_dpif *ofproto = ctx->ofproto;
5465 struct rule_dpif *rule;
5466 uint16_t old_in_port;
5467 uint8_t old_table_id;
5469 old_table_id = ctx->table_id;
5470 ctx->table_id = table_id;
5472 /* Look up a flow with 'in_port' as the input port. */
5473 old_in_port = ctx->flow.in_port;
5474 ctx->flow.in_port = in_port;
5475 rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id);
5478 if (table_id > 0 && table_id < N_TABLES) {
5479 struct table_dpif *table = &ofproto->tables[table_id];
5480 if (table->other_table) {
5481 ctx->tags |= (rule && rule->tag
5483 : rule_calculate_tag(&ctx->flow,
5484 &table->other_table->mask,
5489 /* Restore the original input port. Otherwise OFPP_NORMAL and
5490 * OFPP_IN_PORT will have surprising behavior. */
5491 ctx->flow.in_port = old_in_port;
5493 if (ctx->resubmit_hook) {
5494 ctx->resubmit_hook(ctx, rule);
5497 if (rule == NULL && may_packet_in) {
5499 * check if table configuration flags
5500 * OFPTC_TABLE_MISS_CONTROLLER, default.
5501 * OFPTC_TABLE_MISS_CONTINUE,
5502 * OFPTC_TABLE_MISS_DROP
5503 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
5505 rule = rule_dpif_miss_rule(ofproto, &ctx->flow);
5509 struct rule_dpif *old_rule = ctx->rule;
5511 if (ctx->resubmit_stats) {
5512 rule_credit_stats(rule, ctx->resubmit_stats);
5517 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
5518 ctx->rule = old_rule;
5522 ctx->table_id = old_table_id;
5524 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5526 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
5527 MAX_RESUBMIT_RECURSION);
5528 ctx->max_resubmit_trigger = true;
5533 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
5534 const struct ofpact_resubmit *resubmit)
5539 in_port = resubmit->in_port;
5540 if (in_port == OFPP_IN_PORT) {
5541 in_port = ctx->flow.in_port;
5544 table_id = resubmit->table_id;
5545 if (table_id == 255) {
5546 table_id = ctx->table_id;
5549 xlate_table_action(ctx, in_port, table_id, false);
5553 flood_packets(struct action_xlate_ctx *ctx, bool all)
5555 struct ofport_dpif *ofport;
5557 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
5558 uint16_t ofp_port = ofport->up.ofp_port;
5560 if (ofp_port == ctx->flow.in_port) {
5565 compose_output_action__(ctx, ofp_port, false);
5566 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
5567 compose_output_action(ctx, ofp_port);
5571 ctx->nf_output_iface = NF_OUT_FLOOD;
5575 execute_controller_action(struct action_xlate_ctx *ctx, int len,
5576 enum ofp_packet_in_reason reason,
5577 uint16_t controller_id)
5579 struct ofputil_packet_in pin;
5580 struct ofpbuf *packet;
5582 ctx->slow |= SLOW_CONTROLLER;
5587 packet = ofpbuf_clone(ctx->packet);
5589 if (packet->l2 && packet->l3) {
5590 struct eth_header *eh;
5592 eth_pop_vlan(packet);
5595 /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
5596 * LLC frame. Calculating the Ethernet type of these frames is more
5597 * trouble than seems appropriate for a simple assertion. */
5598 ovs_assert(ntohs(eh->eth_type) < ETH_TYPE_MIN
5599 || eh->eth_type == ctx->flow.dl_type);
5601 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
5602 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
5604 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
5605 eth_push_vlan(packet, ctx->flow.vlan_tci);
5609 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5610 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
5611 ctx->flow.nw_tos, ctx->flow.nw_ttl);
5615 if (ctx->flow.nw_proto == IPPROTO_TCP) {
5616 packet_set_tcp_port(packet, ctx->flow.tp_src,
5618 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
5619 packet_set_udp_port(packet, ctx->flow.tp_src,
5626 pin.packet = packet->data;
5627 pin.packet_len = packet->size;
5628 pin.reason = reason;
5629 pin.controller_id = controller_id;
5630 pin.table_id = ctx->table_id;
5631 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
5634 flow_get_metadata(&ctx->flow, &pin.fmd);
5636 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
5637 ofpbuf_delete(packet);
5641 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
5643 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
5644 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
5648 if (ctx->flow.nw_ttl > 1) {
5654 for (i = 0; i < ids->n_controllers; i++) {
5655 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
5659 /* Stop processing for current table. */
5665 xlate_output_action(struct action_xlate_ctx *ctx,
5666 uint16_t port, uint16_t max_len, bool may_packet_in)
5668 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
5670 ctx->nf_output_iface = NF_OUT_DROP;
5674 compose_output_action(ctx, ctx->flow.in_port);
5677 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
5683 flood_packets(ctx, false);
5686 flood_packets(ctx, true);
5688 case OFPP_CONTROLLER:
5689 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
5695 if (port != ctx->flow.in_port) {
5696 compose_output_action(ctx, port);
5698 xlate_report(ctx, "skipping output to input port");
5703 if (prev_nf_output_iface == NF_OUT_FLOOD) {
5704 ctx->nf_output_iface = NF_OUT_FLOOD;
5705 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
5706 ctx->nf_output_iface = prev_nf_output_iface;
5707 } else if (prev_nf_output_iface != NF_OUT_DROP &&
5708 ctx->nf_output_iface != NF_OUT_FLOOD) {
5709 ctx->nf_output_iface = NF_OUT_MULTI;
5714 xlate_output_reg_action(struct action_xlate_ctx *ctx,
5715 const struct ofpact_output_reg *or)
5717 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
5718 if (port <= UINT16_MAX) {
5719 xlate_output_action(ctx, port, or->max_len, false);
5724 xlate_enqueue_action(struct action_xlate_ctx *ctx,
5725 const struct ofpact_enqueue *enqueue)
5727 uint16_t ofp_port = enqueue->port;
5728 uint32_t queue_id = enqueue->queue;
5729 uint32_t flow_priority, priority;
5732 /* Translate queue to priority. */
5733 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
5734 queue_id, &priority);
5736 /* Fall back to ordinary output action. */
5737 xlate_output_action(ctx, enqueue->port, 0, false);
5741 /* Check output port. */
5742 if (ofp_port == OFPP_IN_PORT) {
5743 ofp_port = ctx->flow.in_port;
5744 } else if (ofp_port == ctx->flow.in_port) {
5748 /* Add datapath actions. */
5749 flow_priority = ctx->flow.skb_priority;
5750 ctx->flow.skb_priority = priority;
5751 compose_output_action(ctx, ofp_port);
5752 ctx->flow.skb_priority = flow_priority;
5754 /* Update NetFlow output port. */
5755 if (ctx->nf_output_iface == NF_OUT_DROP) {
5756 ctx->nf_output_iface = ofp_port;
5757 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
5758 ctx->nf_output_iface = NF_OUT_MULTI;
5763 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
5765 uint32_t skb_priority;
5767 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
5768 queue_id, &skb_priority)) {
5769 ctx->flow.skb_priority = skb_priority;
5771 /* Couldn't translate queue to a priority. Nothing to do. A warning
5772 * has already been logged. */
5776 struct xlate_reg_state {
5782 xlate_autopath(struct action_xlate_ctx *ctx,
5783 const struct ofpact_autopath *ap)
5785 uint16_t ofp_port = ap->port;
5786 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
5788 if (!port || !port->bundle) {
5789 ofp_port = OFPP_NONE;
5790 } else if (port->bundle->bond) {
5791 /* Autopath does not support VLAN hashing. */
5792 struct ofport_dpif *slave = bond_choose_output_slave(
5793 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
5795 ofp_port = slave->up.ofp_port;
5798 nxm_reg_load(&ap->dst, ofp_port, &ctx->flow);
5802 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
5804 struct ofproto_dpif *ofproto = ofproto_;
5805 struct ofport_dpif *port;
5815 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
5818 port = get_ofp_port(ofproto, ofp_port);
5819 return port ? port->may_enable : false;
5824 xlate_bundle_action(struct action_xlate_ctx *ctx,
5825 const struct ofpact_bundle *bundle)
5829 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
5830 if (bundle->dst.field) {
5831 nxm_reg_load(&bundle->dst, port, &ctx->flow);
5833 xlate_output_action(ctx, port, 0, false);
5838 xlate_learn_action(struct action_xlate_ctx *ctx,
5839 const struct ofpact_learn *learn)
5841 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
5842 struct ofputil_flow_mod fm;
5843 uint64_t ofpacts_stub[1024 / 8];
5844 struct ofpbuf ofpacts;
5847 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
5848 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
5850 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
5851 if (error && !VLOG_DROP_WARN(&rl)) {
5852 VLOG_WARN("learning action failed to modify flow table (%s)",
5853 ofperr_get_name(error));
5856 ofpbuf_uninit(&ofpacts);
5859 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
5860 * means "infinite". */
5862 reduce_timeout(uint16_t max, uint16_t *timeout)
5864 if (max && (!*timeout || *timeout > max)) {
5870 xlate_fin_timeout(struct action_xlate_ctx *ctx,
5871 const struct ofpact_fin_timeout *oft)
5873 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
5874 struct rule_dpif *rule = ctx->rule;
5876 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
5877 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
5882 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
5884 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
5885 ? OFPUTIL_PC_NO_RECV_STP
5886 : OFPUTIL_PC_NO_RECV)) {
5890 /* Only drop packets here if both forwarding and learning are
5891 * disabled. If just learning is enabled, we need to have
5892 * OFPP_NORMAL and the learning action have a look at the packet
5893 * before we can drop it. */
5894 if (!stp_forward_in_state(port->stp_state)
5895 && !stp_learn_in_state(port->stp_state)) {
5903 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
5904 struct action_xlate_ctx *ctx)
5906 const struct ofport_dpif *port;
5907 bool was_evictable = true;
5908 const struct ofpact *a;
5910 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5911 if (port && !may_receive(port, ctx)) {
5912 /* Drop this flow. */
5917 /* Don't let the rule we're working on get evicted underneath us. */
5918 was_evictable = ctx->rule->up.evictable;
5919 ctx->rule->up.evictable = false;
5921 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
5922 struct ofpact_controller *controller;
5923 const struct ofpact_metadata *metadata;
5931 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
5932 ofpact_get_OUTPUT(a)->max_len, true);
5935 case OFPACT_CONTROLLER:
5936 controller = ofpact_get_CONTROLLER(a);
5937 execute_controller_action(ctx, controller->max_len,
5939 controller->controller_id);
5942 case OFPACT_ENQUEUE:
5943 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
5946 case OFPACT_SET_VLAN_VID:
5947 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
5948 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
5952 case OFPACT_SET_VLAN_PCP:
5953 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
5954 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
5959 case OFPACT_STRIP_VLAN:
5960 ctx->flow.vlan_tci = htons(0);
5963 case OFPACT_PUSH_VLAN:
5964 /* XXX 802.1AD(QinQ) */
5965 ctx->flow.vlan_tci = htons(VLAN_CFI);
5968 case OFPACT_SET_ETH_SRC:
5969 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
5973 case OFPACT_SET_ETH_DST:
5974 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
5978 case OFPACT_SET_IPV4_SRC:
5979 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
5982 case OFPACT_SET_IPV4_DST:
5983 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
5986 case OFPACT_SET_IPV4_DSCP:
5987 /* OpenFlow 1.0 only supports IPv4. */
5988 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5989 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5990 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
5994 case OFPACT_SET_L4_SRC_PORT:
5995 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
5998 case OFPACT_SET_L4_DST_PORT:
5999 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6002 case OFPACT_RESUBMIT:
6003 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6006 case OFPACT_SET_TUNNEL:
6007 ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6010 case OFPACT_SET_QUEUE:
6011 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6014 case OFPACT_POP_QUEUE:
6015 ctx->flow.skb_priority = ctx->orig_skb_priority;
6018 case OFPACT_REG_MOVE:
6019 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
6022 case OFPACT_REG_LOAD:
6023 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
6026 case OFPACT_DEC_TTL:
6027 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6033 /* Nothing to do. */
6036 case OFPACT_MULTIPATH:
6037 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
6040 case OFPACT_AUTOPATH:
6041 xlate_autopath(ctx, ofpact_get_AUTOPATH(a));
6045 ctx->ofproto->has_bundle_action = true;
6046 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6049 case OFPACT_OUTPUT_REG:
6050 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6054 ctx->has_learn = true;
6055 if (ctx->may_learn) {
6056 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6064 case OFPACT_FIN_TIMEOUT:
6065 ctx->has_fin_timeout = true;
6066 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6069 case OFPACT_CLEAR_ACTIONS:
6071 * Nothing to do because writa-actions is not supported for now.
6072 * When writa-actions is supported, clear-actions also must
6073 * be supported at the same time.
6077 case OFPACT_WRITE_METADATA:
6078 metadata = ofpact_get_WRITE_METADATA(a);
6079 ctx->flow.metadata &= ~metadata->mask;
6080 ctx->flow.metadata |= metadata->metadata & metadata->mask;
6083 case OFPACT_GOTO_TABLE: {
6084 /* XXX remove recursion */
6085 /* It is assumed that goto-table is last action */
6086 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6087 ovs_assert(ctx->table_id < ogt->table_id);
6088 xlate_table_action(ctx, ctx->flow.in_port, ogt->table_id, true);
6095 /* We've let OFPP_NORMAL and the learning action look at the packet,
6096 * so drop it now if forwarding is disabled. */
6097 if (port && !stp_forward_in_state(port->stp_state)) {
6098 ofpbuf_clear(ctx->odp_actions);
6099 add_sflow_action(ctx);
6102 ctx->rule->up.evictable = was_evictable;
6107 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
6108 struct ofproto_dpif *ofproto, const struct flow *flow,
6109 ovs_be16 initial_tci, struct rule_dpif *rule,
6110 uint8_t tcp_flags, const struct ofpbuf *packet)
6112 ovs_be64 initial_tun_id = flow->tunnel.tun_id;
6114 /* Flow initialization rules:
6115 * - 'base_flow' must match the kernel's view of the packet at the
6116 * time that action processing starts. 'flow' represents any
6117 * transformations we wish to make through actions.
6118 * - By default 'base_flow' and 'flow' are the same since the input
6119 * packet matches the output before any actions are applied.
6120 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
6121 * of the received packet as seen by the kernel. If we later output
6122 * to another device without any modifications this will cause us to
6123 * insert a new tag since the original one was stripped off by the
6125 * - Tunnel 'flow' is largely cleared when transitioning between
6126 * the input and output stages since it does not make sense to output
6127 * a packet with the exact headers that it was received with (i.e.
6128 * the destination IP is us). The one exception is the tun_id, which
6129 * is preserved to allow use in later resubmit lookups and loads into
6131 * - Tunnel 'base_flow' is completely cleared since that is what the
6132 * kernel does. If we wish to maintain the original values an action
6133 * needs to be generated. */
6135 ctx->ofproto = ofproto;
6137 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
6138 ctx->base_flow = ctx->flow;
6139 ctx->base_flow.vlan_tci = initial_tci;
6140 ctx->flow.tunnel.tun_id = initial_tun_id;
6142 ctx->packet = packet;
6143 ctx->may_learn = packet != NULL;
6144 ctx->tcp_flags = tcp_flags;
6145 ctx->resubmit_hook = NULL;
6146 ctx->report_hook = NULL;
6147 ctx->resubmit_stats = NULL;
6150 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6151 * into datapath actions in 'odp_actions', using 'ctx'. */
6153 xlate_actions(struct action_xlate_ctx *ctx,
6154 const struct ofpact *ofpacts, size_t ofpacts_len,
6155 struct ofpbuf *odp_actions)
6157 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6158 * that in the future we always keep a copy of the original flow for
6159 * tracing purposes. */
6160 static bool hit_resubmit_limit;
6162 enum slow_path_reason special;
6164 COVERAGE_INC(ofproto_dpif_xlate);
6166 ofpbuf_clear(odp_actions);
6167 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
6169 ctx->odp_actions = odp_actions;
6172 ctx->has_learn = false;
6173 ctx->has_normal = false;
6174 ctx->has_fin_timeout = false;
6175 ctx->nf_output_iface = NF_OUT_DROP;
6178 ctx->max_resubmit_trigger = false;
6179 ctx->orig_skb_priority = ctx->flow.skb_priority;
6183 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
6184 /* Do this conditionally because the copy is expensive enough that it
6185 * shows up in profiles.
6187 * We keep orig_flow in 'ctx' only because I couldn't make GCC 4.4
6188 * believe that I wasn't using it without initializing it if I kept it
6189 * in a local variable. */
6190 ctx->orig_flow = ctx->flow;
6193 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6194 switch (ctx->ofproto->up.frag_handling) {
6195 case OFPC_FRAG_NORMAL:
6196 /* We must pretend that transport ports are unavailable. */
6197 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
6198 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
6201 case OFPC_FRAG_DROP:
6204 case OFPC_FRAG_REASM:
6207 case OFPC_FRAG_NX_MATCH:
6208 /* Nothing to do. */
6211 case OFPC_INVALID_TTL_TO_CONTROLLER:
6216 special = process_special(ctx->ofproto, &ctx->flow, ctx->packet);
6218 ctx->slow |= special;
6220 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6221 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
6222 uint32_t local_odp_port;
6224 add_sflow_action(ctx);
6225 do_xlate_actions(ofpacts, ofpacts_len, ctx);
6227 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
6228 if (!hit_resubmit_limit) {
6229 /* We didn't record the original flow. Make sure we do from
6231 hit_resubmit_limit = true;
6232 } else if (!VLOG_DROP_ERR(&trace_rl)) {
6233 struct ds ds = DS_EMPTY_INITIALIZER;
6235 ofproto_trace(ctx->ofproto, &ctx->orig_flow, ctx->packet,
6237 VLOG_ERR("Trace triggered by excessive resubmit "
6238 "recursion:\n%s", ds_cstr(&ds));
6243 local_odp_port = ofp_port_to_odp_port(ctx->ofproto, OFPP_LOCAL);
6244 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
6246 ctx->odp_actions->data,
6247 ctx->odp_actions->size)) {
6248 ctx->slow |= SLOW_IN_BAND;
6250 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
6252 compose_output_action(ctx, OFPP_LOCAL);
6255 if (ctx->ofproto->has_mirrors) {
6256 add_mirror_actions(ctx, &ctx->orig_flow);
6258 fix_sflow_action(ctx);
6262 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
6263 * into datapath actions, using 'ctx', and discards the datapath actions. */
6265 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
6266 const struct ofpact *ofpacts,
6269 uint64_t odp_actions_stub[1024 / 8];
6270 struct ofpbuf odp_actions;
6272 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
6273 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
6274 ofpbuf_uninit(&odp_actions);
6278 xlate_report(struct action_xlate_ctx *ctx, const char *s)
6280 if (ctx->report_hook) {
6281 ctx->report_hook(ctx, s);
6285 /* OFPP_NORMAL implementation. */
6287 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
6289 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
6290 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
6291 * the bundle on which the packet was received, returns the VLAN to which the
6294 * Both 'vid' and the return value are in the range 0...4095. */
6296 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
6298 switch (in_bundle->vlan_mode) {
6299 case PORT_VLAN_ACCESS:
6300 return in_bundle->vlan;
6303 case PORT_VLAN_TRUNK:
6306 case PORT_VLAN_NATIVE_UNTAGGED:
6307 case PORT_VLAN_NATIVE_TAGGED:
6308 return vid ? vid : in_bundle->vlan;
6315 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
6316 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
6319 * 'vid' should be the VID obtained from the 802.1Q header that was received as
6320 * part of a packet (specify 0 if there was no 802.1Q header), in the range
6323 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
6325 /* Allow any VID on the OFPP_NONE port. */
6326 if (in_bundle == &ofpp_none_bundle) {
6330 switch (in_bundle->vlan_mode) {
6331 case PORT_VLAN_ACCESS:
6334 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6335 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
6336 "packet received on port %s configured as VLAN "
6337 "%"PRIu16" access port",
6338 in_bundle->ofproto->up.name, vid,
6339 in_bundle->name, in_bundle->vlan);
6345 case PORT_VLAN_NATIVE_UNTAGGED:
6346 case PORT_VLAN_NATIVE_TAGGED:
6348 /* Port must always carry its native VLAN. */
6352 case PORT_VLAN_TRUNK:
6353 if (!ofbundle_includes_vlan(in_bundle, vid)) {
6355 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6356 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
6357 "received on port %s not configured for trunking "
6359 in_bundle->ofproto->up.name, vid,
6360 in_bundle->name, vid);
6372 /* Given 'vlan', the VLAN that a packet belongs to, and
6373 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
6374 * that should be included in the 802.1Q header. (If the return value is 0,
6375 * then the 802.1Q header should only be included in the packet if there is a
6378 * Both 'vlan' and the return value are in the range 0...4095. */
6380 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
6382 switch (out_bundle->vlan_mode) {
6383 case PORT_VLAN_ACCESS:
6386 case PORT_VLAN_TRUNK:
6387 case PORT_VLAN_NATIVE_TAGGED:
6390 case PORT_VLAN_NATIVE_UNTAGGED:
6391 return vlan == out_bundle->vlan ? 0 : vlan;
6399 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
6402 struct ofport_dpif *port;
6404 ovs_be16 tci, old_tci;
6406 vid = output_vlan_to_vid(out_bundle, vlan);
6407 if (!out_bundle->bond) {
6408 port = ofbundle_get_a_port(out_bundle);
6410 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
6413 /* No slaves enabled, so drop packet. */
6418 old_tci = ctx->flow.vlan_tci;
6420 if (tci || out_bundle->use_priority_tags) {
6421 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
6423 tci |= htons(VLAN_CFI);
6426 ctx->flow.vlan_tci = tci;
6428 compose_output_action(ctx, port->up.ofp_port);
6429 ctx->flow.vlan_tci = old_tci;
6433 mirror_mask_ffs(mirror_mask_t mask)
6435 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
6440 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
6442 return (bundle->vlan_mode != PORT_VLAN_ACCESS
6443 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
6447 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
6449 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
6452 /* Returns an arbitrary interface within 'bundle'. */
6453 static struct ofport_dpif *
6454 ofbundle_get_a_port(const struct ofbundle *bundle)
6456 return CONTAINER_OF(list_front(&bundle->ports),
6457 struct ofport_dpif, bundle_node);
6461 vlan_is_mirrored(const struct ofmirror *m, int vlan)
6463 return !m->vlans || bitmap_is_set(m->vlans, vlan);
6467 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
6469 struct ofproto_dpif *ofproto = ctx->ofproto;
6470 mirror_mask_t mirrors;
6471 struct ofbundle *in_bundle;
6474 const struct nlattr *a;
6477 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
6478 ctx->packet != NULL, NULL);
6482 mirrors = in_bundle->src_mirrors;
6484 /* Drop frames on bundles reserved for mirroring. */
6485 if (in_bundle->mirror_out) {
6486 if (ctx->packet != NULL) {
6487 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6488 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6489 "%s, which is reserved exclusively for mirroring",
6490 ctx->ofproto->up.name, in_bundle->name);
6496 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
6497 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6500 vlan = input_vid_to_vlan(in_bundle, vid);
6502 /* Look at the output ports to check for destination selections. */
6504 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
6505 ctx->odp_actions->size) {
6506 enum ovs_action_attr type = nl_attr_type(a);
6507 struct ofport_dpif *ofport;
6509 if (type != OVS_ACTION_ATTR_OUTPUT) {
6513 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
6514 if (ofport && ofport->bundle) {
6515 mirrors |= ofport->bundle->dst_mirrors;
6523 /* Restore the original packet before adding the mirror actions. */
6524 ctx->flow = *orig_flow;
6529 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6531 if (!vlan_is_mirrored(m, vlan)) {
6532 mirrors = zero_rightmost_1bit(mirrors);
6536 mirrors &= ~m->dup_mirrors;
6537 ctx->mirrors |= m->dup_mirrors;
6539 output_normal(ctx, m->out, vlan);
6540 } else if (vlan != m->out_vlan
6541 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
6542 struct ofbundle *bundle;
6544 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
6545 if (ofbundle_includes_vlan(bundle, m->out_vlan)
6546 && !bundle->mirror_out) {
6547 output_normal(ctx, bundle, m->out_vlan);
6555 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
6556 uint64_t packets, uint64_t bytes)
6562 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
6565 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6568 /* In normal circumstances 'm' will not be NULL. However,
6569 * if mirrors are reconfigured, we can temporarily get out
6570 * of sync in facet_revalidate(). We could "correct" the
6571 * mirror list before reaching here, but doing that would
6572 * not properly account the traffic stats we've currently
6573 * accumulated for previous mirror configuration. */
6577 m->packet_count += packets;
6578 m->byte_count += bytes;
6582 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
6583 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
6584 * indicate this; newer upstream kernels use gratuitous ARP requests. */
6586 is_gratuitous_arp(const struct flow *flow)
6588 return (flow->dl_type == htons(ETH_TYPE_ARP)
6589 && eth_addr_is_broadcast(flow->dl_dst)
6590 && (flow->nw_proto == ARP_OP_REPLY
6591 || (flow->nw_proto == ARP_OP_REQUEST
6592 && flow->nw_src == flow->nw_dst)));
6596 update_learning_table(struct ofproto_dpif *ofproto,
6597 const struct flow *flow, int vlan,
6598 struct ofbundle *in_bundle)
6600 struct mac_entry *mac;
6602 /* Don't learn the OFPP_NONE port. */
6603 if (in_bundle == &ofpp_none_bundle) {
6607 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
6611 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
6612 if (is_gratuitous_arp(flow)) {
6613 /* We don't want to learn from gratuitous ARP packets that are
6614 * reflected back over bond slaves so we lock the learning table. */
6615 if (!in_bundle->bond) {
6616 mac_entry_set_grat_arp_lock(mac);
6617 } else if (mac_entry_is_grat_arp_locked(mac)) {
6622 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
6623 /* The log messages here could actually be useful in debugging,
6624 * so keep the rate limit relatively high. */
6625 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
6626 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
6627 "on port %s in VLAN %d",
6628 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
6629 in_bundle->name, vlan);
6631 mac->port.p = in_bundle;
6632 tag_set_add(&ofproto->backer->revalidate_set,
6633 mac_learning_changed(ofproto->ml, mac));
6637 static struct ofbundle *
6638 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
6639 bool warn, struct ofport_dpif **in_ofportp)
6641 struct ofport_dpif *ofport;
6643 /* Find the port and bundle for the received packet. */
6644 ofport = get_ofp_port(ofproto, in_port);
6646 *in_ofportp = ofport;
6648 if (ofport && ofport->bundle) {
6649 return ofport->bundle;
6652 /* Special-case OFPP_NONE, which a controller may use as the ingress
6653 * port for traffic that it is sourcing. */
6654 if (in_port == OFPP_NONE) {
6655 return &ofpp_none_bundle;
6658 /* Odd. A few possible reasons here:
6660 * - We deleted a port but there are still a few packets queued up
6663 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
6664 * we don't know about.
6666 * - The ofproto client didn't configure the port as part of a bundle.
6667 * This is particularly likely to happen if a packet was received on the
6668 * port after it was created, but before the client had a chance to
6669 * configure its bundle.
6672 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6674 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
6675 "port %"PRIu16, ofproto->up.name, in_port);
6680 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
6681 * dropped. Returns true if they may be forwarded, false if they should be
6684 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
6685 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
6687 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
6688 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
6689 * checked by input_vid_is_valid().
6691 * May also add tags to '*tags', although the current implementation only does
6692 * so in one special case.
6695 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
6698 struct ofproto_dpif *ofproto = ctx->ofproto;
6699 struct flow *flow = &ctx->flow;
6700 struct ofbundle *in_bundle = in_port->bundle;
6702 /* Drop frames for reserved multicast addresses
6703 * only if forward_bpdu option is absent. */
6704 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
6705 xlate_report(ctx, "packet has reserved destination MAC, dropping");
6709 if (in_bundle->bond) {
6710 struct mac_entry *mac;
6712 switch (bond_check_admissibility(in_bundle->bond, in_port,
6713 flow->dl_dst, &ctx->tags)) {
6718 xlate_report(ctx, "bonding refused admissibility, dropping");
6721 case BV_DROP_IF_MOVED:
6722 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
6723 if (mac && mac->port.p != in_bundle &&
6724 (!is_gratuitous_arp(flow)
6725 || mac_entry_is_grat_arp_locked(mac))) {
6726 xlate_report(ctx, "SLB bond thinks this packet looped back, "
6738 xlate_normal(struct action_xlate_ctx *ctx)
6740 struct ofport_dpif *in_port;
6741 struct ofbundle *in_bundle;
6742 struct mac_entry *mac;
6746 ctx->has_normal = true;
6748 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
6749 ctx->packet != NULL, &in_port);
6751 xlate_report(ctx, "no input bundle, dropping");
6755 /* Drop malformed frames. */
6756 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
6757 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
6758 if (ctx->packet != NULL) {
6759 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6760 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
6761 "VLAN tag received on port %s",
6762 ctx->ofproto->up.name, in_bundle->name);
6764 xlate_report(ctx, "partial VLAN tag, dropping");
6768 /* Drop frames on bundles reserved for mirroring. */
6769 if (in_bundle->mirror_out) {
6770 if (ctx->packet != NULL) {
6771 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6772 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6773 "%s, which is reserved exclusively for mirroring",
6774 ctx->ofproto->up.name, in_bundle->name);
6776 xlate_report(ctx, "input port is mirror output port, dropping");
6781 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
6782 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6783 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
6786 vlan = input_vid_to_vlan(in_bundle, vid);
6788 /* Check other admissibility requirements. */
6789 if (in_port && !is_admissible(ctx, in_port, vlan)) {
6793 /* Learn source MAC. */
6794 if (ctx->may_learn) {
6795 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
6798 /* Determine output bundle. */
6799 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
6802 if (mac->port.p != in_bundle) {
6803 xlate_report(ctx, "forwarding to learned port");
6804 output_normal(ctx, mac->port.p, vlan);
6806 xlate_report(ctx, "learned port is input port, dropping");
6809 struct ofbundle *bundle;
6811 xlate_report(ctx, "no learned MAC for destination, flooding");
6812 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
6813 if (bundle != in_bundle
6814 && ofbundle_includes_vlan(bundle, vlan)
6815 && bundle->floodable
6816 && !bundle->mirror_out) {
6817 output_normal(ctx, bundle, vlan);
6820 ctx->nf_output_iface = NF_OUT_FLOOD;
6824 /* Optimized flow revalidation.
6826 * It's a difficult problem, in general, to tell which facets need to have
6827 * their actions recalculated whenever the OpenFlow flow table changes. We
6828 * don't try to solve that general problem: for most kinds of OpenFlow flow
6829 * table changes, we recalculate the actions for every facet. This is
6830 * relatively expensive, but it's good enough if the OpenFlow flow table
6831 * doesn't change very often.
6833 * However, we can expect one particular kind of OpenFlow flow table change to
6834 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
6835 * of CPU on revalidating every facet whenever MAC learning modifies the flow
6836 * table, we add a special case that applies to flow tables in which every rule
6837 * has the same form (that is, the same wildcards), except that the table is
6838 * also allowed to have a single "catch-all" flow that matches all packets. We
6839 * optimize this case by tagging all of the facets that resubmit into the table
6840 * and invalidating the same tag whenever a flow changes in that table. The
6841 * end result is that we revalidate just the facets that need it (and sometimes
6842 * a few more, but not all of the facets or even all of the facets that
6843 * resubmit to the table modified by MAC learning). */
6845 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
6846 * into an OpenFlow table with the given 'basis'. */
6848 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
6851 if (minimask_is_catchall(mask)) {
6854 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
6855 return tag_create_deterministic(hash);
6859 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
6860 * taggability of that table.
6862 * This function must be called after *each* change to a flow table. If you
6863 * skip calling it on some changes then the pointer comparisons at the end can
6864 * be invalid if you get unlucky. For example, if a flow removal causes a
6865 * cls_table to be destroyed and then a flow insertion causes a cls_table with
6866 * different wildcards to be created with the same address, then this function
6867 * will incorrectly skip revalidation. */
6869 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
6871 struct table_dpif *table = &ofproto->tables[table_id];
6872 const struct oftable *oftable = &ofproto->up.tables[table_id];
6873 struct cls_table *catchall, *other;
6874 struct cls_table *t;
6876 catchall = other = NULL;
6878 switch (hmap_count(&oftable->cls.tables)) {
6880 /* We could tag this OpenFlow table but it would make the logic a
6881 * little harder and it's a corner case that doesn't seem worth it
6887 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
6888 if (cls_table_is_catchall(t)) {
6890 } else if (!other) {
6893 /* Indicate that we can't tag this by setting both tables to
6894 * NULL. (We know that 'catchall' is already NULL.) */
6901 /* Can't tag this table. */
6905 if (table->catchall_table != catchall || table->other_table != other) {
6906 table->catchall_table = catchall;
6907 table->other_table = other;
6908 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
6912 /* Given 'rule' that has changed in some way (either it is a rule being
6913 * inserted, a rule being deleted, or a rule whose actions are being
6914 * modified), marks facets for revalidation to ensure that packets will be
6915 * forwarded correctly according to the new state of the flow table.
6917 * This function must be called after *each* change to a flow table. See
6918 * the comment on table_update_taggable() for more information. */
6920 rule_invalidate(const struct rule_dpif *rule)
6922 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
6924 table_update_taggable(ofproto, rule->up.table_id);
6926 if (!ofproto->backer->need_revalidate) {
6927 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
6929 if (table->other_table && rule->tag) {
6930 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
6932 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
6938 set_frag_handling(struct ofproto *ofproto_,
6939 enum ofp_config_flags frag_handling)
6941 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6942 if (frag_handling != OFPC_FRAG_REASM) {
6943 ofproto->backer->need_revalidate = REV_RECONFIGURE;
6951 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
6952 const struct flow *flow,
6953 const struct ofpact *ofpacts, size_t ofpacts_len)
6955 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6956 struct odputil_keybuf keybuf;
6957 struct dpif_flow_stats stats;
6961 struct action_xlate_ctx ctx;
6962 uint64_t odp_actions_stub[1024 / 8];
6963 struct ofpbuf odp_actions;
6965 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
6966 odp_flow_key_from_flow(&key, flow,
6967 ofp_port_to_odp_port(ofproto, flow->in_port));
6969 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
6971 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL,
6972 packet_get_tcp_flags(packet, flow), packet);
6973 ctx.resubmit_stats = &stats;
6975 ofpbuf_use_stub(&odp_actions,
6976 odp_actions_stub, sizeof odp_actions_stub);
6977 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
6978 dpif_execute(ofproto->backer->dpif, key.data, key.size,
6979 odp_actions.data, odp_actions.size, packet);
6980 ofpbuf_uninit(&odp_actions);
6988 set_netflow(struct ofproto *ofproto_,
6989 const struct netflow_options *netflow_options)
6991 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6993 if (netflow_options) {
6994 if (!ofproto->netflow) {
6995 ofproto->netflow = netflow_create();
6997 return netflow_set_options(ofproto->netflow, netflow_options);
6999 netflow_destroy(ofproto->netflow);
7000 ofproto->netflow = NULL;
7006 get_netflow_ids(const struct ofproto *ofproto_,
7007 uint8_t *engine_type, uint8_t *engine_id)
7009 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7011 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7015 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7017 if (!facet_is_controller_flow(facet) &&
7018 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7019 struct subfacet *subfacet;
7020 struct ofexpired expired;
7022 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7023 if (subfacet->path == SF_FAST_PATH) {
7024 struct dpif_flow_stats stats;
7026 subfacet_reinstall(subfacet, &stats);
7027 subfacet_update_stats(subfacet, &stats);
7031 expired.flow = facet->flow;
7032 expired.packet_count = facet->packet_count;
7033 expired.byte_count = facet->byte_count;
7034 expired.used = facet->used;
7035 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
7040 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7042 struct facet *facet;
7044 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7045 send_active_timeout(ofproto, facet);
7049 static struct ofproto_dpif *
7050 ofproto_dpif_lookup(const char *name)
7052 struct ofproto_dpif *ofproto;
7054 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7055 hash_string(name, 0), &all_ofproto_dpifs) {
7056 if (!strcmp(ofproto->up.name, name)) {
7064 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7065 const char *argv[], void *aux OVS_UNUSED)
7067 struct ofproto_dpif *ofproto;
7070 ofproto = ofproto_dpif_lookup(argv[1]);
7072 unixctl_command_reply_error(conn, "no such bridge");
7075 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7077 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7078 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7082 unixctl_command_reply(conn, "table successfully flushed");
7086 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7087 const char *argv[], void *aux OVS_UNUSED)
7089 struct ds ds = DS_EMPTY_INITIALIZER;
7090 const struct ofproto_dpif *ofproto;
7091 const struct mac_entry *e;
7093 ofproto = ofproto_dpif_lookup(argv[1]);
7095 unixctl_command_reply_error(conn, "no such bridge");
7099 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7100 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7101 struct ofbundle *bundle = e->port.p;
7102 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7103 ofbundle_get_a_port(bundle)->odp_port,
7104 e->vlan, ETH_ADDR_ARGS(e->mac),
7105 mac_entry_age(ofproto->ml, e));
7107 unixctl_command_reply(conn, ds_cstr(&ds));
7112 struct action_xlate_ctx ctx;
7118 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7119 const struct rule_dpif *rule)
7121 ds_put_char_multiple(result, '\t', level);
7123 ds_put_cstr(result, "No match\n");
7127 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7128 table_id, ntohll(rule->up.flow_cookie));
7129 cls_rule_format(&rule->up.cr, result);
7130 ds_put_char(result, '\n');
7132 ds_put_char_multiple(result, '\t', level);
7133 ds_put_cstr(result, "OpenFlow ");
7134 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7135 ds_put_char(result, '\n');
7139 trace_format_flow(struct ds *result, int level, const char *title,
7140 struct trace_ctx *trace)
7142 ds_put_char_multiple(result, '\t', level);
7143 ds_put_format(result, "%s: ", title);
7144 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
7145 ds_put_cstr(result, "unchanged");
7147 flow_format(result, &trace->ctx.flow);
7148 trace->flow = trace->ctx.flow;
7150 ds_put_char(result, '\n');
7154 trace_format_regs(struct ds *result, int level, const char *title,
7155 struct trace_ctx *trace)
7159 ds_put_char_multiple(result, '\t', level);
7160 ds_put_format(result, "%s:", title);
7161 for (i = 0; i < FLOW_N_REGS; i++) {
7162 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7164 ds_put_char(result, '\n');
7168 trace_format_odp(struct ds *result, int level, const char *title,
7169 struct trace_ctx *trace)
7171 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
7173 ds_put_char_multiple(result, '\t', level);
7174 ds_put_format(result, "%s: ", title);
7175 format_odp_actions(result, odp_actions->data, odp_actions->size);
7176 ds_put_char(result, '\n');
7180 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
7182 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7183 struct ds *result = trace->result;
7185 ds_put_char(result, '\n');
7186 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7187 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
7188 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
7189 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
7193 trace_report(struct action_xlate_ctx *ctx, const char *s)
7195 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7196 struct ds *result = trace->result;
7198 ds_put_char_multiple(result, '\t', ctx->recurse);
7199 ds_put_cstr(result, s);
7200 ds_put_char(result, '\n');
7204 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
7205 void *aux OVS_UNUSED)
7207 const char *dpname = argv[1];
7208 struct ofproto_dpif *ofproto;
7209 struct ofpbuf odp_key;
7210 struct ofpbuf *packet;
7211 ovs_be16 initial_tci;
7217 ofpbuf_init(&odp_key, 0);
7220 ofproto = ofproto_dpif_lookup(dpname);
7222 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
7226 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
7227 /* ofproto/trace dpname flow [-generate] */
7228 const char *flow_s = argv[2];
7229 const char *generate_s = argv[3];
7231 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
7232 * flow. We guess which type it is based on whether 'flow_s' contains
7233 * an '(', since a datapath flow always contains '(') but an
7234 * OpenFlow-like flow should not (in fact it's allowed but I believe
7235 * that's not documented anywhere).
7237 * An alternative would be to try to parse 'flow_s' both ways, but then
7238 * it would be tricky giving a sensible error message. After all, do
7239 * you just say "syntax error" or do you present both error messages?
7240 * Both choices seem lousy. */
7241 if (strchr(flow_s, '(')) {
7244 /* Convert string to datapath key. */
7245 ofpbuf_init(&odp_key, 0);
7246 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
7248 unixctl_command_reply_error(conn, "Bad flow syntax");
7252 /* XXX: Since we allow the user to specify an ofproto, it's
7253 * possible they will specify a different ofproto than the one the
7254 * port actually belongs too. Ideally we should simply remove the
7255 * ability to specify the ofproto. */
7256 if (ofproto_receive(ofproto->backer, NULL, odp_key.data,
7257 odp_key.size, &flow, NULL, NULL, NULL,
7259 unixctl_command_reply_error(conn, "Invalid flow");
7265 error_s = parse_ofp_exact_flow(&flow, argv[2]);
7267 unixctl_command_reply_error(conn, error_s);
7272 initial_tci = flow.vlan_tci;
7275 /* Generate a packet, if requested. */
7277 packet = ofpbuf_new(0);
7278 flow_compose(packet, &flow);
7280 } else if (argc == 7) {
7281 /* ofproto/trace dpname priority tun_id in_port mark packet */
7282 const char *priority_s = argv[2];
7283 const char *tun_id_s = argv[3];
7284 const char *in_port_s = argv[4];
7285 const char *mark_s = argv[5];
7286 const char *packet_s = argv[6];
7287 uint32_t in_port = atoi(in_port_s);
7288 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
7289 uint32_t priority = atoi(priority_s);
7290 uint32_t mark = atoi(mark_s);
7293 msg = eth_from_hex(packet_s, &packet);
7295 unixctl_command_reply_error(conn, msg);
7299 ds_put_cstr(&result, "Packet: ");
7300 s = ofp_packet_to_string(packet->data, packet->size);
7301 ds_put_cstr(&result, s);
7304 flow_extract(packet, priority, mark, NULL, in_port, &flow);
7305 flow.tunnel.tun_id = tun_id;
7306 initial_tci = flow.vlan_tci;
7308 unixctl_command_reply_error(conn, "Bad command syntax");
7312 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
7313 unixctl_command_reply(conn, ds_cstr(&result));
7316 ds_destroy(&result);
7317 ofpbuf_delete(packet);
7318 ofpbuf_uninit(&odp_key);
7322 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
7323 const struct ofpbuf *packet, ovs_be16 initial_tci,
7326 struct rule_dpif *rule;
7328 ds_put_cstr(ds, "Flow: ");
7329 flow_format(ds, flow);
7330 ds_put_char(ds, '\n');
7332 rule = rule_dpif_lookup(ofproto, flow);
7334 trace_format_rule(ds, 0, 0, rule);
7335 if (rule == ofproto->miss_rule) {
7336 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
7337 } else if (rule == ofproto->no_packet_in_rule) {
7338 ds_put_cstr(ds, "\nNo match, packets dropped because "
7339 "OFPPC_NO_PACKET_IN is set on in_port.\n");
7343 uint64_t odp_actions_stub[1024 / 8];
7344 struct ofpbuf odp_actions;
7346 struct trace_ctx trace;
7349 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
7352 ofpbuf_use_stub(&odp_actions,
7353 odp_actions_stub, sizeof odp_actions_stub);
7354 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
7355 rule, tcp_flags, packet);
7356 trace.ctx.resubmit_hook = trace_resubmit;
7357 trace.ctx.report_hook = trace_report;
7358 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
7361 ds_put_char(ds, '\n');
7362 trace_format_flow(ds, 0, "Final flow", &trace);
7363 ds_put_cstr(ds, "Datapath actions: ");
7364 format_odp_actions(ds, odp_actions.data, odp_actions.size);
7365 ofpbuf_uninit(&odp_actions);
7367 if (trace.ctx.slow) {
7368 enum slow_path_reason slow;
7370 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
7371 "slow path because it:");
7372 for (slow = trace.ctx.slow; slow; ) {
7373 enum slow_path_reason bit = rightmost_1bit(slow);
7377 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
7380 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
7383 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
7386 ds_put_cstr(ds, "\n\t- Needs in-band special case "
7389 ds_put_cstr(ds, "\n\t (The datapath actions are "
7390 "incomplete--for complete actions, "
7391 "please supply a packet.)");
7394 case SLOW_CONTROLLER:
7395 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
7396 "to the OpenFlow controller.");
7399 ds_put_cstr(ds, "\n\t- Needs more specific matching "
7400 "than the datapath supports.");
7407 if (slow & ~SLOW_MATCH) {
7408 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
7409 "the special slow-path processing.");
7416 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7417 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7420 unixctl_command_reply(conn, NULL);
7424 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7425 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7428 unixctl_command_reply(conn, NULL);
7431 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
7432 * 'reply' describing the results. */
7434 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
7436 struct facet *facet;
7440 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7441 if (!facet_check_consistency(facet)) {
7446 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
7450 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
7451 ofproto->up.name, errors);
7453 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
7458 ofproto_dpif_self_check(struct unixctl_conn *conn,
7459 int argc, const char *argv[], void *aux OVS_UNUSED)
7461 struct ds reply = DS_EMPTY_INITIALIZER;
7462 struct ofproto_dpif *ofproto;
7465 ofproto = ofproto_dpif_lookup(argv[1]);
7467 unixctl_command_reply_error(conn, "Unknown ofproto (use "
7468 "ofproto/list for help)");
7471 ofproto_dpif_self_check__(ofproto, &reply);
7473 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7474 ofproto_dpif_self_check__(ofproto, &reply);
7478 unixctl_command_reply(conn, ds_cstr(&reply));
7482 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
7483 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
7484 * to destroy 'ofproto_shash' and free the returned value. */
7485 static const struct shash_node **
7486 get_ofprotos(struct shash *ofproto_shash)
7488 const struct ofproto_dpif *ofproto;
7490 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7491 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
7492 shash_add_nocopy(ofproto_shash, name, ofproto);
7495 return shash_sort(ofproto_shash);
7499 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
7500 const char *argv[] OVS_UNUSED,
7501 void *aux OVS_UNUSED)
7503 struct ds ds = DS_EMPTY_INITIALIZER;
7504 struct shash ofproto_shash;
7505 const struct shash_node **sorted_ofprotos;
7508 shash_init(&ofproto_shash);
7509 sorted_ofprotos = get_ofprotos(&ofproto_shash);
7510 for (i = 0; i < shash_count(&ofproto_shash); i++) {
7511 const struct shash_node *node = sorted_ofprotos[i];
7512 ds_put_format(&ds, "%s\n", node->name);
7515 shash_destroy(&ofproto_shash);
7516 free(sorted_ofprotos);
7518 unixctl_command_reply(conn, ds_cstr(&ds));
7523 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
7525 struct dpif_dp_stats s;
7526 const struct shash_node **ports;
7529 dpif_get_dp_stats(ofproto->backer->dpif, &s);
7531 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
7532 dpif_name(ofproto->backer->dpif));
7533 /* xxx It would be better to show bridge-specific stats instead
7534 * xxx of dp ones. */
7536 "\tlookups: hit:%"PRIu64" missed:%"PRIu64" lost:%"PRIu64"\n",
7537 s.n_hit, s.n_missed, s.n_lost);
7538 ds_put_format(ds, "\tflows: %zu\n",
7539 hmap_count(&ofproto->subfacets));
7541 ports = shash_sort(&ofproto->up.port_by_name);
7542 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
7543 const struct shash_node *node = ports[i];
7544 struct ofport *ofport = node->data;
7545 const char *name = netdev_get_name(ofport->netdev);
7546 const char *type = netdev_get_type(ofport->netdev);
7548 ds_put_format(ds, "\t%s %u/%u:", name, ofport->ofp_port,
7549 ofp_port_to_odp_port(ofproto, ofport->ofp_port));
7550 if (strcmp(type, "system")) {
7551 struct netdev *netdev;
7554 ds_put_format(ds, " (%s", type);
7556 error = netdev_open(name, type, &netdev);
7561 error = netdev_get_config(netdev, &config);
7563 const struct smap_node **nodes;
7566 nodes = smap_sort(&config);
7567 for (i = 0; i < smap_count(&config); i++) {
7568 const struct smap_node *node = nodes[i];
7569 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
7570 node->key, node->value);
7574 smap_destroy(&config);
7576 netdev_close(netdev);
7578 ds_put_char(ds, ')');
7580 ds_put_char(ds, '\n');
7586 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
7587 const char *argv[], void *aux OVS_UNUSED)
7589 struct ds ds = DS_EMPTY_INITIALIZER;
7590 const struct ofproto_dpif *ofproto;
7594 for (i = 1; i < argc; i++) {
7595 ofproto = ofproto_dpif_lookup(argv[i]);
7597 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
7598 "for help)", argv[i]);
7599 unixctl_command_reply_error(conn, ds_cstr(&ds));
7602 show_dp_format(ofproto, &ds);
7605 struct shash ofproto_shash;
7606 const struct shash_node **sorted_ofprotos;
7609 shash_init(&ofproto_shash);
7610 sorted_ofprotos = get_ofprotos(&ofproto_shash);
7611 for (i = 0; i < shash_count(&ofproto_shash); i++) {
7612 const struct shash_node *node = sorted_ofprotos[i];
7613 show_dp_format(node->data, &ds);
7616 shash_destroy(&ofproto_shash);
7617 free(sorted_ofprotos);
7620 unixctl_command_reply(conn, ds_cstr(&ds));
7625 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
7626 int argc OVS_UNUSED, const char *argv[],
7627 void *aux OVS_UNUSED)
7629 struct ds ds = DS_EMPTY_INITIALIZER;
7630 const struct ofproto_dpif *ofproto;
7631 struct subfacet *subfacet;
7633 ofproto = ofproto_dpif_lookup(argv[1]);
7635 unixctl_command_reply_error(conn, "no such bridge");
7639 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
7640 struct odputil_keybuf keybuf;
7643 subfacet_get_key(subfacet, &keybuf, &key);
7644 odp_flow_key_format(key.data, key.size, &ds);
7646 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
7647 subfacet->dp_packet_count, subfacet->dp_byte_count);
7648 if (subfacet->used) {
7649 ds_put_format(&ds, "%.3fs",
7650 (time_msec() - subfacet->used) / 1000.0);
7652 ds_put_format(&ds, "never");
7654 if (subfacet->facet->tcp_flags) {
7655 ds_put_cstr(&ds, ", flags:");
7656 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
7659 ds_put_cstr(&ds, ", actions:");
7660 format_odp_actions(&ds, subfacet->actions, subfacet->actions_len);
7661 ds_put_char(&ds, '\n');
7664 unixctl_command_reply(conn, ds_cstr(&ds));
7669 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
7670 int argc OVS_UNUSED, const char *argv[],
7671 void *aux OVS_UNUSED)
7673 struct ds ds = DS_EMPTY_INITIALIZER;
7674 struct ofproto_dpif *ofproto;
7676 ofproto = ofproto_dpif_lookup(argv[1]);
7678 unixctl_command_reply_error(conn, "no such bridge");
7682 flush(&ofproto->up);
7684 unixctl_command_reply(conn, ds_cstr(&ds));
7689 ofproto_dpif_unixctl_init(void)
7691 static bool registered;
7697 unixctl_command_register(
7699 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
7700 2, 6, ofproto_unixctl_trace, NULL);
7701 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
7702 ofproto_unixctl_fdb_flush, NULL);
7703 unixctl_command_register("fdb/show", "bridge", 1, 1,
7704 ofproto_unixctl_fdb_show, NULL);
7705 unixctl_command_register("ofproto/clog", "", 0, 0,
7706 ofproto_dpif_clog, NULL);
7707 unixctl_command_register("ofproto/unclog", "", 0, 0,
7708 ofproto_dpif_unclog, NULL);
7709 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
7710 ofproto_dpif_self_check, NULL);
7711 unixctl_command_register("dpif/dump-dps", "", 0, 0,
7712 ofproto_unixctl_dpif_dump_dps, NULL);
7713 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
7714 ofproto_unixctl_dpif_show, NULL);
7715 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
7716 ofproto_unixctl_dpif_dump_flows, NULL);
7717 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
7718 ofproto_unixctl_dpif_del_flows, NULL);
7721 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
7723 * This is deprecated. It is only for compatibility with broken device drivers
7724 * in old versions of Linux that do not properly support VLANs when VLAN
7725 * devices are not used. When broken device drivers are no longer in
7726 * widespread use, we will delete these interfaces. */
7729 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
7731 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
7732 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
7734 if (realdev_ofp_port == ofport->realdev_ofp_port
7735 && vid == ofport->vlandev_vid) {
7739 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7741 if (ofport->realdev_ofp_port) {
7744 if (realdev_ofp_port && ofport->bundle) {
7745 /* vlandevs are enslaved to their realdevs, so they are not allowed to
7746 * themselves be part of a bundle. */
7747 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
7750 ofport->realdev_ofp_port = realdev_ofp_port;
7751 ofport->vlandev_vid = vid;
7753 if (realdev_ofp_port) {
7754 vsp_add(ofport, realdev_ofp_port, vid);
7761 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
7763 return hash_2words(realdev_ofp_port, vid);
7766 /* Returns the ODP port number of the Linux VLAN device that corresponds to
7767 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
7768 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
7769 * it would return the port number of eth0.9.
7771 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
7772 * function just returns its 'realdev_odp_port' argument. */
7774 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
7775 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
7777 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
7778 uint16_t realdev_ofp_port;
7779 int vid = vlan_tci_to_vid(vlan_tci);
7780 const struct vlan_splinter *vsp;
7782 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
7783 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
7784 hash_realdev_vid(realdev_ofp_port, vid),
7785 &ofproto->realdev_vid_map) {
7786 if (vsp->realdev_ofp_port == realdev_ofp_port
7787 && vsp->vid == vid) {
7788 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
7792 return realdev_odp_port;
7795 static struct vlan_splinter *
7796 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
7798 struct vlan_splinter *vsp;
7800 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
7801 &ofproto->vlandev_map) {
7802 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
7810 /* Returns the OpenFlow port number of the "real" device underlying the Linux
7811 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
7812 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
7813 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
7814 * eth0 and store 9 in '*vid'.
7816 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
7817 * VLAN device. Unless VLAN splinters are enabled, this is what this function
7820 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
7821 uint16_t vlandev_ofp_port, int *vid)
7823 if (!hmap_is_empty(&ofproto->vlandev_map)) {
7824 const struct vlan_splinter *vsp;
7826 vsp = vlandev_find(ofproto, vlandev_ofp_port);
7831 return vsp->realdev_ofp_port;
7837 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
7838 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
7839 * 'flow->in_port' to the "real" device backing the VLAN device, sets
7840 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
7841 * always the case unless VLAN splinters are enabled), returns false without
7842 * making any changes. */
7844 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
7849 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
7854 /* Cause the flow to be processed as if it came in on the real device with
7855 * the VLAN device's VLAN ID. */
7856 flow->in_port = realdev;
7857 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
7862 vsp_remove(struct ofport_dpif *port)
7864 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7865 struct vlan_splinter *vsp;
7867 vsp = vlandev_find(ofproto, port->up.ofp_port);
7869 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
7870 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
7873 port->realdev_ofp_port = 0;
7875 VLOG_ERR("missing vlan device record");
7880 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
7882 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7884 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
7885 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
7886 == realdev_ofp_port)) {
7887 struct vlan_splinter *vsp;
7889 vsp = xmalloc(sizeof *vsp);
7890 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
7891 hash_int(port->up.ofp_port, 0));
7892 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
7893 hash_realdev_vid(realdev_ofp_port, vid));
7894 vsp->realdev_ofp_port = realdev_ofp_port;
7895 vsp->vlandev_ofp_port = port->up.ofp_port;
7898 port->realdev_ofp_port = realdev_ofp_port;
7900 VLOG_ERR("duplicate vlan device record");
7905 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
7907 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
7908 return ofport ? ofport->odp_port : OVSP_NONE;
7911 static struct ofport_dpif *
7912 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
7914 struct ofport_dpif *port;
7916 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
7917 hash_int(odp_port, 0),
7918 &backer->odp_to_ofport_map) {
7919 if (port->odp_port == odp_port) {
7928 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
7930 struct ofport_dpif *port;
7932 port = odp_port_to_ofport(ofproto->backer, odp_port);
7933 if (port && &ofproto->up == port->up.ofproto) {
7934 return port->up.ofp_port;
7940 const struct ofproto_class ofproto_dpif_class = {
7975 port_is_lacp_current,
7976 NULL, /* rule_choose_table */
7983 rule_modify_actions,
7992 get_cfm_remote_mpids,
7997 get_stp_port_status,
8004 is_mirror_output_bundle,
8005 forward_bpdu_changed,
8006 set_mac_table_config,