2 * Copyright (c) 2009, 2010, 2011, 2012 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-provider.h"
26 #include "byte-order.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
36 #include "mac-learning.h"
37 #include "meta-flow.h"
38 #include "multipath.h"
45 #include "ofp-actions.h"
46 #include "ofp-parse.h"
47 #include "ofp-print.h"
48 #include "ofproto-dpif-governor.h"
49 #include "ofproto-dpif-sflow.h"
50 #include "poll-loop.h"
53 #include "unaligned.h"
55 #include "vlan-bitmap.h"
58 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
60 COVERAGE_DEFINE(ofproto_dpif_expired);
61 COVERAGE_DEFINE(ofproto_dpif_xlate);
62 COVERAGE_DEFINE(facet_changed_rule);
63 COVERAGE_DEFINE(facet_revalidate);
64 COVERAGE_DEFINE(facet_unexpected);
65 COVERAGE_DEFINE(facet_suppress);
67 /* Maximum depth of flow table recursion (due to resubmit actions) in a
68 * flow translation. */
69 #define MAX_RESUBMIT_RECURSION 64
71 /* Number of implemented OpenFlow tables. */
72 enum { N_TABLES = 255 };
73 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
74 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
84 * - Do include packets and bytes from facets that have been deleted or
85 * whose own statistics have been folded into the rule.
87 * - Do include packets and bytes sent "by hand" that were accounted to
88 * the rule without any facet being involved (this is a rare corner
89 * case in rule_execute()).
91 * - Do not include packet or bytes that can be obtained from any facet's
92 * packet_count or byte_count member or that can be obtained from the
93 * datapath by, e.g., dpif_flow_get() for any subfacet.
95 uint64_t packet_count; /* Number of packets received. */
96 uint64_t byte_count; /* Number of bytes received. */
98 tag_type tag; /* Caches rule_calculate_tag() result. */
100 struct list facets; /* List of "struct facet"s. */
103 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
105 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
108 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
109 const struct flow *);
110 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
113 static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto,
114 const struct flow *flow);
116 static void rule_credit_stats(struct rule_dpif *,
117 const struct dpif_flow_stats *);
118 static void flow_push_stats(struct rule_dpif *, const struct flow *,
119 const struct dpif_flow_stats *);
120 static tag_type rule_calculate_tag(const struct flow *,
121 const struct minimask *, uint32_t basis);
122 static void rule_invalidate(const struct rule_dpif *);
124 #define MAX_MIRRORS 32
125 typedef uint32_t mirror_mask_t;
126 #define MIRROR_MASK_C(X) UINT32_C(X)
127 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
129 struct ofproto_dpif *ofproto; /* Owning ofproto. */
130 size_t idx; /* In ofproto's "mirrors" array. */
131 void *aux; /* Key supplied by ofproto's client. */
132 char *name; /* Identifier for log messages. */
134 /* Selection criteria. */
135 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
136 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
137 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
139 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
140 struct ofbundle *out; /* Output port or NULL. */
141 int out_vlan; /* Output VLAN or -1. */
142 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
145 int64_t packet_count; /* Number of packets sent. */
146 int64_t byte_count; /* Number of bytes sent. */
149 static void mirror_destroy(struct ofmirror *);
150 static void update_mirror_stats(struct ofproto_dpif *ofproto,
151 mirror_mask_t mirrors,
152 uint64_t packets, uint64_t bytes);
155 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
156 struct ofproto_dpif *ofproto; /* Owning ofproto. */
157 void *aux; /* Key supplied by ofproto's client. */
158 char *name; /* Identifier for log messages. */
161 struct list ports; /* Contains "struct ofport"s. */
162 enum port_vlan_mode vlan_mode; /* VLAN mode */
163 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
164 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
165 * NULL if all VLANs are trunked. */
166 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
167 struct bond *bond; /* Nonnull iff more than one port. */
168 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
171 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
173 /* Port mirroring info. */
174 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
175 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
176 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
179 static void bundle_remove(struct ofport *);
180 static void bundle_update(struct ofbundle *);
181 static void bundle_destroy(struct ofbundle *);
182 static void bundle_del_port(struct ofport_dpif *);
183 static void bundle_run(struct ofbundle *);
184 static void bundle_wait(struct ofbundle *);
185 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
186 uint16_t in_port, bool warn,
187 struct ofport_dpif **in_ofportp);
189 /* A controller may use OFPP_NONE as the ingress port to indicate that
190 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
191 * when an input bundle is needed for validation (e.g., mirroring or
192 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
193 * any 'port' structs, so care must be taken when dealing with it. */
194 static struct ofbundle ofpp_none_bundle = {
196 .vlan_mode = PORT_VLAN_TRUNK
199 static void stp_run(struct ofproto_dpif *ofproto);
200 static void stp_wait(struct ofproto_dpif *ofproto);
201 static int set_stp_port(struct ofport *,
202 const struct ofproto_port_stp_settings *);
204 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
206 struct action_xlate_ctx {
207 /* action_xlate_ctx_init() initializes these members. */
210 struct ofproto_dpif *ofproto;
212 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
213 * this flow when actions change header fields. */
216 /* The packet corresponding to 'flow', or a null pointer if we are
217 * revalidating without a packet to refer to. */
218 const struct ofpbuf *packet;
220 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
221 * actions update the flow table?
223 * We want to update these tables if we are actually processing a packet,
224 * or if we are accounting for packets that the datapath has processed, but
225 * not if we are just revalidating. */
228 /* The rule that we are currently translating, or NULL. */
229 struct rule_dpif *rule;
231 /* Union of the set of TCP flags seen so far in this flow. (Used only by
232 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
236 /* If nonnull, flow translation calls this function just before executing a
237 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
238 * when the recursion depth is exceeded.
240 * 'rule' is the rule being submitted into. It will be null if the
241 * resubmit or OFPP_TABLE action didn't find a matching rule.
243 * This is normally null so the client has to set it manually after
244 * calling action_xlate_ctx_init(). */
245 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule);
247 /* If nonnull, flow translation calls this function to report some
248 * significant decision, e.g. to explain why OFPP_NORMAL translation
249 * dropped a packet. */
250 void (*report_hook)(struct action_xlate_ctx *, const char *s);
252 /* If nonnull, flow translation credits the specified statistics to each
253 * rule reached through a resubmit or OFPP_TABLE action.
255 * This is normally null so the client has to set it manually after
256 * calling action_xlate_ctx_init(). */
257 const struct dpif_flow_stats *resubmit_stats;
259 /* xlate_actions() initializes and uses these members. The client might want
260 * to look at them after it returns. */
262 struct ofpbuf *odp_actions; /* Datapath actions. */
263 tag_type tags; /* Tags associated with actions. */
264 enum slow_path_reason slow; /* 0 if fast path may be used. */
265 bool has_learn; /* Actions include NXAST_LEARN? */
266 bool has_normal; /* Actions output to OFPP_NORMAL? */
267 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
268 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
269 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
271 /* xlate_actions() initializes and uses these members, but the client has no
272 * reason to look at them. */
274 int recurse; /* Recursion level, via xlate_table_action. */
275 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
276 struct flow base_flow; /* Flow at the last commit. */
277 uint32_t orig_skb_priority; /* Priority when packet arrived. */
278 uint8_t table_id; /* OpenFlow table ID where flow was found. */
279 uint32_t sflow_n_outputs; /* Number of output ports. */
280 uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
281 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
282 bool exit; /* No further actions should be processed. */
283 struct flow orig_flow; /* Copy of original flow. */
286 static void action_xlate_ctx_init(struct action_xlate_ctx *,
287 struct ofproto_dpif *, const struct flow *,
288 ovs_be16 initial_tci, struct rule_dpif *,
289 uint8_t tcp_flags, const struct ofpbuf *);
290 static void xlate_actions(struct action_xlate_ctx *,
291 const struct ofpact *ofpacts, size_t ofpacts_len,
292 struct ofpbuf *odp_actions);
293 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
294 const struct ofpact *ofpacts,
297 static size_t put_userspace_action(const struct ofproto_dpif *,
298 struct ofpbuf *odp_actions,
300 const union user_action_cookie *);
302 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
303 enum slow_path_reason,
304 uint64_t *stub, size_t stub_size,
305 const struct nlattr **actionsp,
306 size_t *actions_lenp);
308 static void xlate_report(struct action_xlate_ctx *ctx, const char *s);
310 /* A subfacet (see "struct subfacet" below) has three possible installation
313 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
314 * case just after the subfacet is created, just before the subfacet is
315 * destroyed, or if the datapath returns an error when we try to install a
318 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
320 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
321 * ofproto_dpif is installed in the datapath.
324 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
325 SF_FAST_PATH, /* Full actions are installed. */
326 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
329 static const char *subfacet_path_to_string(enum subfacet_path);
331 /* A dpif flow and actions associated with a facet.
333 * See also the large comment on struct facet. */
336 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
337 struct list list_node; /* In struct facet's 'facets' list. */
338 struct facet *facet; /* Owning facet. */
342 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
343 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
344 * regenerate the ODP flow key from ->facet->flow. */
345 enum odp_key_fitness key_fitness;
349 long long int used; /* Time last used; time created if not used. */
351 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
352 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
356 * These should be essentially identical for every subfacet in a facet, but
357 * may differ in trivial ways due to VLAN splinters. */
358 size_t actions_len; /* Number of bytes in actions[]. */
359 struct nlattr *actions; /* Datapath actions. */
361 enum slow_path_reason slow; /* 0 if fast path may be used. */
362 enum subfacet_path path; /* Installed in datapath? */
364 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
365 * splinters can cause it to differ. This value should be removed when
366 * the VLAN splinters feature is no longer needed. */
367 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
370 static struct subfacet *subfacet_create(struct facet *, enum odp_key_fitness,
371 const struct nlattr *key,
372 size_t key_len, ovs_be16 initial_tci,
374 static struct subfacet *subfacet_find(struct ofproto_dpif *,
375 const struct nlattr *key, size_t key_len);
376 static void subfacet_destroy(struct subfacet *);
377 static void subfacet_destroy__(struct subfacet *);
378 static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *,
380 static void subfacet_reset_dp_stats(struct subfacet *,
381 struct dpif_flow_stats *);
382 static void subfacet_update_time(struct subfacet *, long long int used);
383 static void subfacet_update_stats(struct subfacet *,
384 const struct dpif_flow_stats *);
385 static void subfacet_make_actions(struct subfacet *,
386 const struct ofpbuf *packet,
387 struct ofpbuf *odp_actions);
388 static int subfacet_install(struct subfacet *,
389 const struct nlattr *actions, size_t actions_len,
390 struct dpif_flow_stats *, enum slow_path_reason);
391 static void subfacet_uninstall(struct subfacet *);
393 static enum subfacet_path subfacet_want_path(enum slow_path_reason);
395 /* An exact-match instantiation of an OpenFlow flow.
397 * A facet associates a "struct flow", which represents the Open vSwitch
398 * userspace idea of an exact-match flow, with one or more subfacets. Each
399 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
400 * the facet. When the kernel module (or other dpif implementation) and Open
401 * vSwitch userspace agree on the definition of a flow key, there is exactly
402 * one subfacet per facet. If the dpif implementation supports more-specific
403 * flow matching than userspace, however, a facet can have more than one
404 * subfacet, each of which corresponds to some distinction in flow that
405 * userspace simply doesn't understand.
407 * Flow expiration works in terms of subfacets, so a facet must have at least
408 * one subfacet or it will never expire, leaking memory. */
411 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
412 struct list list_node; /* In owning rule's 'facets' list. */
413 struct rule_dpif *rule; /* Owning rule. */
416 struct list subfacets;
417 long long int used; /* Time last used; time created if not used. */
424 * - Do include packets and bytes sent "by hand", e.g. with
427 * - Do include packets and bytes that were obtained from the datapath
428 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
429 * DPIF_FP_ZERO_STATS).
431 * - Do not include packets or bytes that can be obtained from the
432 * datapath for any existing subfacet.
434 uint64_t packet_count; /* Number of packets received. */
435 uint64_t byte_count; /* Number of bytes received. */
437 /* Resubmit statistics. */
438 uint64_t prev_packet_count; /* Number of packets from last stats push. */
439 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
440 long long int prev_used; /* Used time from last stats push. */
443 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
444 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
445 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
447 /* Properties of datapath actions.
449 * Every subfacet has its own actions because actions can differ slightly
450 * between splintered and non-splintered subfacets due to the VLAN tag
451 * being initially different (present vs. absent). All of them have these
452 * properties in common so we just store one copy of them here. */
453 bool has_learn; /* Actions include NXAST_LEARN? */
454 bool has_normal; /* Actions output to OFPP_NORMAL? */
455 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
456 tag_type tags; /* Tags that would require revalidation. */
457 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
459 /* Storage for a single subfacet, to reduce malloc() time and space
460 * overhead. (A facet always has at least one subfacet and in the common
461 * case has exactly one subfacet.) */
462 struct subfacet one_subfacet;
465 static struct facet *facet_create(struct rule_dpif *,
466 const struct flow *, uint32_t hash);
467 static void facet_remove(struct facet *);
468 static void facet_free(struct facet *);
470 static struct facet *facet_find(struct ofproto_dpif *,
471 const struct flow *, uint32_t hash);
472 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
473 const struct flow *, uint32_t hash);
474 static void facet_revalidate(struct facet *);
475 static bool facet_check_consistency(struct facet *);
477 static void facet_flush_stats(struct facet *);
479 static void facet_update_time(struct facet *, long long int used);
480 static void facet_reset_counters(struct facet *);
481 static void facet_push_stats(struct facet *);
482 static void facet_learn(struct facet *);
483 static void facet_account(struct facet *);
485 static bool facet_is_controller_flow(struct facet *);
491 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
492 struct list bundle_node; /* In struct ofbundle's "ports" list. */
493 struct cfm *cfm; /* Connectivity Fault Management, if any. */
494 tag_type tag; /* Tag associated with this port. */
495 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
496 bool may_enable; /* May be enabled in bonds. */
497 long long int carrier_seq; /* Carrier status changes. */
500 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
501 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
502 long long int stp_state_entered;
504 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
506 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
508 * This is deprecated. It is only for compatibility with broken device
509 * drivers in old versions of Linux that do not properly support VLANs when
510 * VLAN devices are not used. When broken device drivers are no longer in
511 * widespread use, we will delete these interfaces. */
512 uint16_t realdev_ofp_port;
516 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
517 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
518 * traffic egressing the 'ofport' with that priority should be marked with. */
519 struct priority_to_dscp {
520 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
521 uint32_t priority; /* Priority of this queue (see struct flow). */
523 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
526 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
528 * This is deprecated. It is only for compatibility with broken device drivers
529 * in old versions of Linux that do not properly support VLANs when VLAN
530 * devices are not used. When broken device drivers are no longer in
531 * widespread use, we will delete these interfaces. */
532 struct vlan_splinter {
533 struct hmap_node realdev_vid_node;
534 struct hmap_node vlandev_node;
535 uint16_t realdev_ofp_port;
536 uint16_t vlandev_ofp_port;
540 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
541 uint32_t realdev, ovs_be16 vlan_tci);
542 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
543 static void vsp_remove(struct ofport_dpif *);
544 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
546 static struct ofport_dpif *
547 ofport_dpif_cast(const struct ofport *ofport)
549 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
550 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
553 static void port_run(struct ofport_dpif *);
554 static void port_run_fast(struct ofport_dpif *);
555 static void port_wait(struct ofport_dpif *);
556 static int set_cfm(struct ofport *, const struct cfm_settings *);
557 static void ofport_clear_priorities(struct ofport_dpif *);
559 struct dpif_completion {
560 struct list list_node;
561 struct ofoperation *op;
564 /* Extra information about a classifier table.
565 * Currently used just for optimized flow revalidation. */
567 /* If either of these is nonnull, then this table has a form that allows
568 * flows to be tagged to avoid revalidating most flows for the most common
569 * kinds of flow table changes. */
570 struct cls_table *catchall_table; /* Table that wildcards all fields. */
571 struct cls_table *other_table; /* Table with any other wildcard set. */
572 uint32_t basis; /* Keeps each table's tags separate. */
575 /* Reasons that we might need to revalidate every facet, and corresponding
578 * A value of 0 means that there is no need to revalidate.
580 * It would be nice to have some cleaner way to integrate with coverage
581 * counters, but with only a few reasons I guess this is good enough for
583 enum revalidate_reason {
584 REV_RECONFIGURE = 1, /* Switch configuration changed. */
585 REV_STP, /* Spanning tree protocol port status change. */
586 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
587 REV_FLOW_TABLE, /* Flow table changed. */
588 REV_INCONSISTENCY /* Facet self-check failed. */
590 COVERAGE_DEFINE(rev_reconfigure);
591 COVERAGE_DEFINE(rev_stp);
592 COVERAGE_DEFINE(rev_port_toggled);
593 COVERAGE_DEFINE(rev_flow_table);
594 COVERAGE_DEFINE(rev_inconsistency);
596 struct ofproto_dpif {
597 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
601 /* Special OpenFlow rules. */
602 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
603 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
609 struct netflow *netflow;
610 struct dpif_sflow *sflow;
611 struct hmap bundles; /* Contains "struct ofbundle"s. */
612 struct mac_learning *ml;
613 struct ofmirror *mirrors[MAX_MIRRORS];
615 bool has_bonded_bundles;
618 struct timer next_expiration;
622 struct hmap subfacets;
623 struct governor *governor;
626 struct table_dpif tables[N_TABLES];
627 enum revalidate_reason need_revalidate;
628 struct tag_set revalidate_set;
630 /* Support for debugging async flow mods. */
631 struct list completions;
633 bool has_bundle_action; /* True when the first bundle action appears. */
634 struct netdev_stats stats; /* To account packets generated and consumed in
639 long long int stp_last_tick;
641 /* VLAN splinters. */
642 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
643 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
646 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
647 * for debugging the asynchronous flow_mod implementation.) */
650 /* All existing ofproto_dpif instances, indexed by ->up.name. */
651 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
653 static void ofproto_dpif_unixctl_init(void);
655 static struct ofproto_dpif *
656 ofproto_dpif_cast(const struct ofproto *ofproto)
658 assert(ofproto->ofproto_class == &ofproto_dpif_class);
659 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
662 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
664 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
666 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
667 const struct ofpbuf *, ovs_be16 initial_tci,
670 /* Packet processing. */
671 static void update_learning_table(struct ofproto_dpif *,
672 const struct flow *, int vlan,
675 #define FLOW_MISS_MAX_BATCH 50
676 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
678 /* Flow expiration. */
679 static int expire(struct ofproto_dpif *);
682 static void send_netflow_active_timeouts(struct ofproto_dpif *);
685 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
686 static size_t compose_sflow_action(const struct ofproto_dpif *,
687 struct ofpbuf *odp_actions,
688 const struct flow *, uint32_t odp_port);
689 static void add_mirror_actions(struct action_xlate_ctx *ctx,
690 const struct flow *flow);
691 /* Global variables. */
692 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
694 /* Factory functions. */
697 enumerate_types(struct sset *types)
699 dp_enumerate_types(types);
703 enumerate_names(const char *type, struct sset *names)
705 return dp_enumerate_names(type, names);
709 del(const char *type, const char *name)
714 error = dpif_open(name, type, &dpif);
716 error = dpif_delete(dpif);
722 /* Basic life-cycle. */
724 static int add_internal_flows(struct ofproto_dpif *);
726 static struct ofproto *
729 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
734 dealloc(struct ofproto *ofproto_)
736 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
741 construct(struct ofproto *ofproto_)
743 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
744 const char *name = ofproto->up.name;
749 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
751 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
755 max_ports = dpif_get_max_ports(ofproto->dpif);
756 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
758 ofproto->n_matches = 0;
760 dpif_flow_flush(ofproto->dpif);
761 dpif_recv_purge(ofproto->dpif);
763 error = dpif_recv_set(ofproto->dpif, true);
765 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
766 dpif_close(ofproto->dpif);
770 ofproto->netflow = NULL;
771 ofproto->sflow = NULL;
773 hmap_init(&ofproto->bundles);
774 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
775 for (i = 0; i < MAX_MIRRORS; i++) {
776 ofproto->mirrors[i] = NULL;
778 ofproto->has_bonded_bundles = false;
780 timer_set_duration(&ofproto->next_expiration, 1000);
782 hmap_init(&ofproto->facets);
783 hmap_init(&ofproto->subfacets);
784 ofproto->governor = NULL;
786 for (i = 0; i < N_TABLES; i++) {
787 struct table_dpif *table = &ofproto->tables[i];
789 table->catchall_table = NULL;
790 table->other_table = NULL;
791 table->basis = random_uint32();
793 ofproto->need_revalidate = 0;
794 tag_set_init(&ofproto->revalidate_set);
796 list_init(&ofproto->completions);
798 ofproto_dpif_unixctl_init();
800 ofproto->has_mirrors = false;
801 ofproto->has_bundle_action = false;
803 hmap_init(&ofproto->vlandev_map);
804 hmap_init(&ofproto->realdev_vid_map);
806 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
807 hash_string(ofproto->up.name, 0));
808 memset(&ofproto->stats, 0, sizeof ofproto->stats);
810 ofproto_init_tables(ofproto_, N_TABLES);
811 error = add_internal_flows(ofproto);
812 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
818 add_internal_flow(struct ofproto_dpif *ofproto, int id,
819 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
821 struct ofputil_flow_mod fm;
824 match_init_catchall(&fm.match);
826 match_set_reg(&fm.match, 0, id);
827 fm.new_cookie = htonll(0);
828 fm.cookie = htonll(0);
829 fm.cookie_mask = htonll(0);
830 fm.table_id = TBL_INTERNAL;
831 fm.command = OFPFC_ADD;
837 fm.ofpacts = ofpacts->data;
838 fm.ofpacts_len = ofpacts->size;
840 error = ofproto_flow_mod(&ofproto->up, &fm);
842 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
843 id, ofperr_to_string(error));
847 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
848 assert(*rulep != NULL);
854 add_internal_flows(struct ofproto_dpif *ofproto)
856 struct ofpact_controller *controller;
857 uint64_t ofpacts_stub[128 / 8];
858 struct ofpbuf ofpacts;
862 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
865 controller = ofpact_put_CONTROLLER(&ofpacts);
866 controller->max_len = UINT16_MAX;
867 controller->controller_id = 0;
868 controller->reason = OFPR_NO_MATCH;
869 ofpact_pad(&ofpacts);
871 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
876 ofpbuf_clear(&ofpacts);
877 error = add_internal_flow(ofproto, id++, &ofpacts,
878 &ofproto->no_packet_in_rule);
883 complete_operations(struct ofproto_dpif *ofproto)
885 struct dpif_completion *c, *next;
887 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
888 ofoperation_complete(c->op, 0);
889 list_remove(&c->list_node);
895 destruct(struct ofproto *ofproto_)
897 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
898 struct rule_dpif *rule, *next_rule;
899 struct oftable *table;
902 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
903 complete_operations(ofproto);
905 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
906 struct cls_cursor cursor;
908 cls_cursor_init(&cursor, &table->cls, NULL);
909 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
910 ofproto_rule_destroy(&rule->up);
914 for (i = 0; i < MAX_MIRRORS; i++) {
915 mirror_destroy(ofproto->mirrors[i]);
918 netflow_destroy(ofproto->netflow);
919 dpif_sflow_destroy(ofproto->sflow);
920 hmap_destroy(&ofproto->bundles);
921 mac_learning_destroy(ofproto->ml);
923 hmap_destroy(&ofproto->facets);
924 hmap_destroy(&ofproto->subfacets);
925 governor_destroy(ofproto->governor);
927 hmap_destroy(&ofproto->vlandev_map);
928 hmap_destroy(&ofproto->realdev_vid_map);
930 dpif_close(ofproto->dpif);
934 run_fast(struct ofproto *ofproto_)
936 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
937 struct ofport_dpif *ofport;
940 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
941 port_run_fast(ofport);
944 /* Handle one or more batches of upcalls, until there's nothing left to do
945 * or until we do a fixed total amount of work.
947 * We do work in batches because it can be much cheaper to set up a number
948 * of flows and fire off their patches all at once. We do multiple batches
949 * because in some cases handling a packet can cause another packet to be
950 * queued almost immediately as part of the return flow. Both
951 * optimizations can make major improvements on some benchmarks and
952 * presumably for real traffic as well. */
954 while (work < FLOW_MISS_MAX_BATCH) {
955 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
965 run(struct ofproto *ofproto_)
967 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
968 struct ofport_dpif *ofport;
969 struct ofbundle *bundle;
973 complete_operations(ofproto);
975 dpif_run(ofproto->dpif);
977 error = run_fast(ofproto_);
982 if (timer_expired(&ofproto->next_expiration)) {
983 int delay = expire(ofproto);
984 timer_set_duration(&ofproto->next_expiration, delay);
987 if (ofproto->netflow) {
988 if (netflow_run(ofproto->netflow)) {
989 send_netflow_active_timeouts(ofproto);
992 if (ofproto->sflow) {
993 dpif_sflow_run(ofproto->sflow);
996 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
999 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1004 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
1006 /* Now revalidate if there's anything to do. */
1007 if (ofproto->need_revalidate
1008 || !tag_set_is_empty(&ofproto->revalidate_set)) {
1009 struct tag_set revalidate_set = ofproto->revalidate_set;
1010 bool revalidate_all = ofproto->need_revalidate;
1011 struct facet *facet;
1013 switch (ofproto->need_revalidate) {
1014 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
1015 case REV_STP: COVERAGE_INC(rev_stp); break;
1016 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
1017 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
1018 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
1021 /* Clear the revalidation flags. */
1022 tag_set_init(&ofproto->revalidate_set);
1023 ofproto->need_revalidate = 0;
1025 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
1027 || tag_set_intersects(&revalidate_set, facet->tags)) {
1028 facet_revalidate(facet);
1033 /* Check the consistency of a random facet, to aid debugging. */
1034 if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) {
1035 struct facet *facet;
1037 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1038 struct facet, hmap_node);
1039 if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) {
1040 if (!facet_check_consistency(facet)) {
1041 ofproto->need_revalidate = REV_INCONSISTENCY;
1046 if (ofproto->governor) {
1049 governor_run(ofproto->governor);
1051 /* If the governor has shrunk to its minimum size and the number of
1052 * subfacets has dwindled, then drop the governor entirely.
1054 * For hysteresis, the number of subfacets to drop the governor is
1055 * smaller than the number needed to trigger its creation. */
1056 n_subfacets = hmap_count(&ofproto->subfacets);
1057 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1058 && governor_is_idle(ofproto->governor)) {
1059 governor_destroy(ofproto->governor);
1060 ofproto->governor = NULL;
1068 wait(struct ofproto *ofproto_)
1070 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1071 struct ofport_dpif *ofport;
1072 struct ofbundle *bundle;
1074 if (!clogged && !list_is_empty(&ofproto->completions)) {
1075 poll_immediate_wake();
1078 dpif_wait(ofproto->dpif);
1079 dpif_recv_wait(ofproto->dpif);
1080 if (ofproto->sflow) {
1081 dpif_sflow_wait(ofproto->sflow);
1083 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
1084 poll_immediate_wake();
1086 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1089 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1090 bundle_wait(bundle);
1092 if (ofproto->netflow) {
1093 netflow_wait(ofproto->netflow);
1095 mac_learning_wait(ofproto->ml);
1097 if (ofproto->need_revalidate) {
1098 /* Shouldn't happen, but if it does just go around again. */
1099 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1100 poll_immediate_wake();
1102 timer_wait(&ofproto->next_expiration);
1104 if (ofproto->governor) {
1105 governor_wait(ofproto->governor);
1110 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1112 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1114 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1115 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1119 flush(struct ofproto *ofproto_)
1121 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1122 struct facet *facet, *next_facet;
1124 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1125 /* Mark the facet as not installed so that facet_remove() doesn't
1126 * bother trying to uninstall it. There is no point in uninstalling it
1127 * individually since we are about to blow away all the facets with
1128 * dpif_flow_flush(). */
1129 struct subfacet *subfacet;
1131 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
1132 subfacet->path = SF_NOT_INSTALLED;
1133 subfacet->dp_packet_count = 0;
1134 subfacet->dp_byte_count = 0;
1136 facet_remove(facet);
1138 dpif_flow_flush(ofproto->dpif);
1142 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1143 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1145 *arp_match_ip = true;
1146 *actions = (OFPUTIL_A_OUTPUT |
1147 OFPUTIL_A_SET_VLAN_VID |
1148 OFPUTIL_A_SET_VLAN_PCP |
1149 OFPUTIL_A_STRIP_VLAN |
1150 OFPUTIL_A_SET_DL_SRC |
1151 OFPUTIL_A_SET_DL_DST |
1152 OFPUTIL_A_SET_NW_SRC |
1153 OFPUTIL_A_SET_NW_DST |
1154 OFPUTIL_A_SET_NW_TOS |
1155 OFPUTIL_A_SET_TP_SRC |
1156 OFPUTIL_A_SET_TP_DST |
1161 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1163 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1164 struct dpif_dp_stats s;
1166 strcpy(ots->name, "classifier");
1168 dpif_get_dp_stats(ofproto->dpif, &s);
1169 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1170 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1173 static struct ofport *
1176 struct ofport_dpif *port = xmalloc(sizeof *port);
1181 port_dealloc(struct ofport *port_)
1183 struct ofport_dpif *port = ofport_dpif_cast(port_);
1188 port_construct(struct ofport *port_)
1190 struct ofport_dpif *port = ofport_dpif_cast(port_);
1191 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1193 ofproto->need_revalidate = REV_RECONFIGURE;
1194 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
1195 port->bundle = NULL;
1197 port->tag = tag_create_random();
1198 port->may_enable = true;
1199 port->stp_port = NULL;
1200 port->stp_state = STP_DISABLED;
1201 hmap_init(&port->priorities);
1202 port->realdev_ofp_port = 0;
1203 port->vlandev_vid = 0;
1204 port->carrier_seq = netdev_get_carrier_resets(port->up.netdev);
1206 if (ofproto->sflow) {
1207 dpif_sflow_add_port(ofproto->sflow, port_);
1214 port_destruct(struct ofport *port_)
1216 struct ofport_dpif *port = ofport_dpif_cast(port_);
1217 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1219 ofproto->need_revalidate = REV_RECONFIGURE;
1220 bundle_remove(port_);
1221 set_cfm(port_, NULL);
1222 if (ofproto->sflow) {
1223 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1226 ofport_clear_priorities(port);
1227 hmap_destroy(&port->priorities);
1231 port_modified(struct ofport *port_)
1233 struct ofport_dpif *port = ofport_dpif_cast(port_);
1235 if (port->bundle && port->bundle->bond) {
1236 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1241 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1243 struct ofport_dpif *port = ofport_dpif_cast(port_);
1244 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1245 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1247 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1248 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1249 OFPUTIL_PC_NO_PACKET_IN)) {
1250 ofproto->need_revalidate = REV_RECONFIGURE;
1252 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1253 bundle_update(port->bundle);
1259 set_sflow(struct ofproto *ofproto_,
1260 const struct ofproto_sflow_options *sflow_options)
1262 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1263 struct dpif_sflow *ds = ofproto->sflow;
1265 if (sflow_options) {
1267 struct ofport_dpif *ofport;
1269 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1270 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1271 dpif_sflow_add_port(ds, &ofport->up);
1273 ofproto->need_revalidate = REV_RECONFIGURE;
1275 dpif_sflow_set_options(ds, sflow_options);
1278 dpif_sflow_destroy(ds);
1279 ofproto->need_revalidate = REV_RECONFIGURE;
1280 ofproto->sflow = NULL;
1287 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1289 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1296 struct ofproto_dpif *ofproto;
1298 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1299 ofproto->need_revalidate = REV_RECONFIGURE;
1300 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1303 if (cfm_configure(ofport->cfm, s)) {
1309 cfm_destroy(ofport->cfm);
1315 get_cfm_fault(const struct ofport *ofport_)
1317 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1319 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1323 get_cfm_opup(const struct ofport *ofport_)
1325 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1327 return ofport->cfm ? cfm_get_opup(ofport->cfm) : -1;
1331 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1334 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1337 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1345 get_cfm_health(const struct ofport *ofport_)
1347 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1349 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1352 /* Spanning Tree. */
1355 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1357 struct ofproto_dpif *ofproto = ofproto_;
1358 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1359 struct ofport_dpif *ofport;
1361 ofport = stp_port_get_aux(sp);
1363 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1364 ofproto->up.name, port_num);
1366 struct eth_header *eth = pkt->l2;
1368 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1369 if (eth_addr_is_zero(eth->eth_src)) {
1370 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1371 "with unknown MAC", ofproto->up.name, port_num);
1373 send_packet(ofport, pkt);
1379 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1381 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1383 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1385 /* Only revalidate flows if the configuration changed. */
1386 if (!s != !ofproto->stp) {
1387 ofproto->need_revalidate = REV_RECONFIGURE;
1391 if (!ofproto->stp) {
1392 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1393 send_bpdu_cb, ofproto);
1394 ofproto->stp_last_tick = time_msec();
1397 stp_set_bridge_id(ofproto->stp, s->system_id);
1398 stp_set_bridge_priority(ofproto->stp, s->priority);
1399 stp_set_hello_time(ofproto->stp, s->hello_time);
1400 stp_set_max_age(ofproto->stp, s->max_age);
1401 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1403 struct ofport *ofport;
1405 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1406 set_stp_port(ofport, NULL);
1409 stp_destroy(ofproto->stp);
1410 ofproto->stp = NULL;
1417 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1419 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1423 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1424 s->designated_root = stp_get_designated_root(ofproto->stp);
1425 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1434 update_stp_port_state(struct ofport_dpif *ofport)
1436 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1437 enum stp_state state;
1439 /* Figure out new state. */
1440 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1444 if (ofport->stp_state != state) {
1445 enum ofputil_port_state of_state;
1448 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1449 netdev_get_name(ofport->up.netdev),
1450 stp_state_name(ofport->stp_state),
1451 stp_state_name(state));
1452 if (stp_learn_in_state(ofport->stp_state)
1453 != stp_learn_in_state(state)) {
1454 /* xxx Learning action flows should also be flushed. */
1455 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1457 fwd_change = stp_forward_in_state(ofport->stp_state)
1458 != stp_forward_in_state(state);
1460 ofproto->need_revalidate = REV_STP;
1461 ofport->stp_state = state;
1462 ofport->stp_state_entered = time_msec();
1464 if (fwd_change && ofport->bundle) {
1465 bundle_update(ofport->bundle);
1468 /* Update the STP state bits in the OpenFlow port description. */
1469 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1470 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1471 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1472 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1473 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1475 ofproto_port_set_state(&ofport->up, of_state);
1479 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1480 * caller is responsible for assigning STP port numbers and ensuring
1481 * there are no duplicates. */
1483 set_stp_port(struct ofport *ofport_,
1484 const struct ofproto_port_stp_settings *s)
1486 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1487 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1488 struct stp_port *sp = ofport->stp_port;
1490 if (!s || !s->enable) {
1492 ofport->stp_port = NULL;
1493 stp_port_disable(sp);
1494 update_stp_port_state(ofport);
1497 } else if (sp && stp_port_no(sp) != s->port_num
1498 && ofport == stp_port_get_aux(sp)) {
1499 /* The port-id changed, so disable the old one if it's not
1500 * already in use by another port. */
1501 stp_port_disable(sp);
1504 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1505 stp_port_enable(sp);
1507 stp_port_set_aux(sp, ofport);
1508 stp_port_set_priority(sp, s->priority);
1509 stp_port_set_path_cost(sp, s->path_cost);
1511 update_stp_port_state(ofport);
1517 get_stp_port_status(struct ofport *ofport_,
1518 struct ofproto_port_stp_status *s)
1520 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1521 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1522 struct stp_port *sp = ofport->stp_port;
1524 if (!ofproto->stp || !sp) {
1530 s->port_id = stp_port_get_id(sp);
1531 s->state = stp_port_get_state(sp);
1532 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1533 s->role = stp_port_get_role(sp);
1534 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1540 stp_run(struct ofproto_dpif *ofproto)
1543 long long int now = time_msec();
1544 long long int elapsed = now - ofproto->stp_last_tick;
1545 struct stp_port *sp;
1548 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1549 ofproto->stp_last_tick = now;
1551 while (stp_get_changed_port(ofproto->stp, &sp)) {
1552 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1555 update_stp_port_state(ofport);
1559 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1560 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1566 stp_wait(struct ofproto_dpif *ofproto)
1569 poll_timer_wait(1000);
1573 /* Returns true if STP should process 'flow'. */
1575 stp_should_process_flow(const struct flow *flow)
1577 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1581 stp_process_packet(const struct ofport_dpif *ofport,
1582 const struct ofpbuf *packet)
1584 struct ofpbuf payload = *packet;
1585 struct eth_header *eth = payload.data;
1586 struct stp_port *sp = ofport->stp_port;
1588 /* Sink packets on ports that have STP disabled when the bridge has
1590 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1594 /* Trim off padding on payload. */
1595 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1596 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1599 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1600 stp_received_bpdu(sp, payload.data, payload.size);
1604 static struct priority_to_dscp *
1605 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1607 struct priority_to_dscp *pdscp;
1610 hash = hash_int(priority, 0);
1611 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1612 if (pdscp->priority == priority) {
1620 ofport_clear_priorities(struct ofport_dpif *ofport)
1622 struct priority_to_dscp *pdscp, *next;
1624 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1625 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1631 set_queues(struct ofport *ofport_,
1632 const struct ofproto_port_queue *qdscp_list,
1635 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1636 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1637 struct hmap new = HMAP_INITIALIZER(&new);
1640 for (i = 0; i < n_qdscp; i++) {
1641 struct priority_to_dscp *pdscp;
1645 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1646 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1651 pdscp = get_priority(ofport, priority);
1653 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1655 pdscp = xmalloc(sizeof *pdscp);
1656 pdscp->priority = priority;
1658 ofproto->need_revalidate = REV_RECONFIGURE;
1661 if (pdscp->dscp != dscp) {
1663 ofproto->need_revalidate = REV_RECONFIGURE;
1666 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1669 if (!hmap_is_empty(&ofport->priorities)) {
1670 ofport_clear_priorities(ofport);
1671 ofproto->need_revalidate = REV_RECONFIGURE;
1674 hmap_swap(&new, &ofport->priorities);
1682 /* Expires all MAC learning entries associated with 'bundle' and forces its
1683 * ofproto to revalidate every flow.
1685 * Normally MAC learning entries are removed only from the ofproto associated
1686 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1687 * are removed from every ofproto. When patch ports and SLB bonds are in use
1688 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1689 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1690 * with the host from which it migrated. */
1692 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1694 struct ofproto_dpif *ofproto = bundle->ofproto;
1695 struct mac_learning *ml = ofproto->ml;
1696 struct mac_entry *mac, *next_mac;
1698 ofproto->need_revalidate = REV_RECONFIGURE;
1699 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1700 if (mac->port.p == bundle) {
1702 struct ofproto_dpif *o;
1704 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1706 struct mac_entry *e;
1708 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1711 tag_set_add(&o->revalidate_set, e->tag);
1712 mac_learning_expire(o->ml, e);
1718 mac_learning_expire(ml, mac);
1723 static struct ofbundle *
1724 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1726 struct ofbundle *bundle;
1728 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1729 &ofproto->bundles) {
1730 if (bundle->aux == aux) {
1737 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1738 * ones that are found to 'bundles'. */
1740 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1741 void **auxes, size_t n_auxes,
1742 struct hmapx *bundles)
1746 hmapx_init(bundles);
1747 for (i = 0; i < n_auxes; i++) {
1748 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1750 hmapx_add(bundles, bundle);
1756 bundle_update(struct ofbundle *bundle)
1758 struct ofport_dpif *port;
1760 bundle->floodable = true;
1761 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1762 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1763 || !stp_forward_in_state(port->stp_state)) {
1764 bundle->floodable = false;
1771 bundle_del_port(struct ofport_dpif *port)
1773 struct ofbundle *bundle = port->bundle;
1775 bundle->ofproto->need_revalidate = REV_RECONFIGURE;
1777 list_remove(&port->bundle_node);
1778 port->bundle = NULL;
1781 lacp_slave_unregister(bundle->lacp, port);
1784 bond_slave_unregister(bundle->bond, port);
1787 bundle_update(bundle);
1791 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1792 struct lacp_slave_settings *lacp,
1793 uint32_t bond_stable_id)
1795 struct ofport_dpif *port;
1797 port = get_ofp_port(bundle->ofproto, ofp_port);
1802 if (port->bundle != bundle) {
1803 bundle->ofproto->need_revalidate = REV_RECONFIGURE;
1805 bundle_del_port(port);
1808 port->bundle = bundle;
1809 list_push_back(&bundle->ports, &port->bundle_node);
1810 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1811 || !stp_forward_in_state(port->stp_state)) {
1812 bundle->floodable = false;
1816 port->bundle->ofproto->need_revalidate = REV_RECONFIGURE;
1817 lacp_slave_register(bundle->lacp, port, lacp);
1820 port->bond_stable_id = bond_stable_id;
1826 bundle_destroy(struct ofbundle *bundle)
1828 struct ofproto_dpif *ofproto;
1829 struct ofport_dpif *port, *next_port;
1836 ofproto = bundle->ofproto;
1837 for (i = 0; i < MAX_MIRRORS; i++) {
1838 struct ofmirror *m = ofproto->mirrors[i];
1840 if (m->out == bundle) {
1842 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1843 || hmapx_find_and_delete(&m->dsts, bundle)) {
1844 ofproto->need_revalidate = REV_RECONFIGURE;
1849 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1850 bundle_del_port(port);
1853 bundle_flush_macs(bundle, true);
1854 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1856 free(bundle->trunks);
1857 lacp_destroy(bundle->lacp);
1858 bond_destroy(bundle->bond);
1863 bundle_set(struct ofproto *ofproto_, void *aux,
1864 const struct ofproto_bundle_settings *s)
1866 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1867 bool need_flush = false;
1868 struct ofport_dpif *port;
1869 struct ofbundle *bundle;
1870 unsigned long *trunks;
1876 bundle_destroy(bundle_lookup(ofproto, aux));
1880 assert(s->n_slaves == 1 || s->bond != NULL);
1881 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1883 bundle = bundle_lookup(ofproto, aux);
1885 bundle = xmalloc(sizeof *bundle);
1887 bundle->ofproto = ofproto;
1888 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1889 hash_pointer(aux, 0));
1891 bundle->name = NULL;
1893 list_init(&bundle->ports);
1894 bundle->vlan_mode = PORT_VLAN_TRUNK;
1896 bundle->trunks = NULL;
1897 bundle->use_priority_tags = s->use_priority_tags;
1898 bundle->lacp = NULL;
1899 bundle->bond = NULL;
1901 bundle->floodable = true;
1903 bundle->src_mirrors = 0;
1904 bundle->dst_mirrors = 0;
1905 bundle->mirror_out = 0;
1908 if (!bundle->name || strcmp(s->name, bundle->name)) {
1910 bundle->name = xstrdup(s->name);
1915 if (!bundle->lacp) {
1916 ofproto->need_revalidate = REV_RECONFIGURE;
1917 bundle->lacp = lacp_create();
1919 lacp_configure(bundle->lacp, s->lacp);
1921 lacp_destroy(bundle->lacp);
1922 bundle->lacp = NULL;
1925 /* Update set of ports. */
1927 for (i = 0; i < s->n_slaves; i++) {
1928 if (!bundle_add_port(bundle, s->slaves[i],
1929 s->lacp ? &s->lacp_slaves[i] : NULL,
1930 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1934 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1935 struct ofport_dpif *next_port;
1937 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1938 for (i = 0; i < s->n_slaves; i++) {
1939 if (s->slaves[i] == port->up.ofp_port) {
1944 bundle_del_port(port);
1948 assert(list_size(&bundle->ports) <= s->n_slaves);
1950 if (list_is_empty(&bundle->ports)) {
1951 bundle_destroy(bundle);
1955 /* Set VLAN tagging mode */
1956 if (s->vlan_mode != bundle->vlan_mode
1957 || s->use_priority_tags != bundle->use_priority_tags) {
1958 bundle->vlan_mode = s->vlan_mode;
1959 bundle->use_priority_tags = s->use_priority_tags;
1964 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1965 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1967 if (vlan != bundle->vlan) {
1968 bundle->vlan = vlan;
1972 /* Get trunked VLANs. */
1973 switch (s->vlan_mode) {
1974 case PORT_VLAN_ACCESS:
1978 case PORT_VLAN_TRUNK:
1979 trunks = CONST_CAST(unsigned long *, s->trunks);
1982 case PORT_VLAN_NATIVE_UNTAGGED:
1983 case PORT_VLAN_NATIVE_TAGGED:
1984 if (vlan != 0 && (!s->trunks
1985 || !bitmap_is_set(s->trunks, vlan)
1986 || bitmap_is_set(s->trunks, 0))) {
1987 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1989 trunks = bitmap_clone(s->trunks, 4096);
1991 trunks = bitmap_allocate1(4096);
1993 bitmap_set1(trunks, vlan);
1994 bitmap_set0(trunks, 0);
1996 trunks = CONST_CAST(unsigned long *, s->trunks);
2003 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2004 free(bundle->trunks);
2005 if (trunks == s->trunks) {
2006 bundle->trunks = vlan_bitmap_clone(trunks);
2008 bundle->trunks = trunks;
2013 if (trunks != s->trunks) {
2018 if (!list_is_short(&bundle->ports)) {
2019 bundle->ofproto->has_bonded_bundles = true;
2021 if (bond_reconfigure(bundle->bond, s->bond)) {
2022 ofproto->need_revalidate = REV_RECONFIGURE;
2025 bundle->bond = bond_create(s->bond);
2026 ofproto->need_revalidate = REV_RECONFIGURE;
2029 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2030 bond_slave_register(bundle->bond, port, port->bond_stable_id,
2034 bond_destroy(bundle->bond);
2035 bundle->bond = NULL;
2038 /* If we changed something that would affect MAC learning, un-learn
2039 * everything on this port and force flow revalidation. */
2041 bundle_flush_macs(bundle, false);
2048 bundle_remove(struct ofport *port_)
2050 struct ofport_dpif *port = ofport_dpif_cast(port_);
2051 struct ofbundle *bundle = port->bundle;
2054 bundle_del_port(port);
2055 if (list_is_empty(&bundle->ports)) {
2056 bundle_destroy(bundle);
2057 } else if (list_is_short(&bundle->ports)) {
2058 bond_destroy(bundle->bond);
2059 bundle->bond = NULL;
2065 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2067 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2068 struct ofport_dpif *port = port_;
2069 uint8_t ea[ETH_ADDR_LEN];
2072 error = netdev_get_etheraddr(port->up.netdev, ea);
2074 struct ofpbuf packet;
2077 ofpbuf_init(&packet, 0);
2078 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2080 memcpy(packet_pdu, pdu, pdu_size);
2082 send_packet(port, &packet);
2083 ofpbuf_uninit(&packet);
2085 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2086 "%s (%s)", port->bundle->name,
2087 netdev_get_name(port->up.netdev), strerror(error));
2092 bundle_send_learning_packets(struct ofbundle *bundle)
2094 struct ofproto_dpif *ofproto = bundle->ofproto;
2095 int error, n_packets, n_errors;
2096 struct mac_entry *e;
2098 error = n_packets = n_errors = 0;
2099 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2100 if (e->port.p != bundle) {
2101 struct ofpbuf *learning_packet;
2102 struct ofport_dpif *port;
2106 /* The assignment to "port" is unnecessary but makes "grep"ing for
2107 * struct ofport_dpif more effective. */
2108 learning_packet = bond_compose_learning_packet(bundle->bond,
2112 ret = send_packet(port, learning_packet);
2113 ofpbuf_delete(learning_packet);
2123 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2124 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2125 "packets, last error was: %s",
2126 bundle->name, n_errors, n_packets, strerror(error));
2128 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2129 bundle->name, n_packets);
2134 bundle_run(struct ofbundle *bundle)
2137 lacp_run(bundle->lacp, send_pdu_cb);
2140 struct ofport_dpif *port;
2142 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2143 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2146 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
2147 lacp_status(bundle->lacp));
2148 if (bond_should_send_learning_packets(bundle->bond)) {
2149 bundle_send_learning_packets(bundle);
2155 bundle_wait(struct ofbundle *bundle)
2158 lacp_wait(bundle->lacp);
2161 bond_wait(bundle->bond);
2168 mirror_scan(struct ofproto_dpif *ofproto)
2172 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2173 if (!ofproto->mirrors[idx]) {
2180 static struct ofmirror *
2181 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2185 for (i = 0; i < MAX_MIRRORS; i++) {
2186 struct ofmirror *mirror = ofproto->mirrors[i];
2187 if (mirror && mirror->aux == aux) {
2195 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2197 mirror_update_dups(struct ofproto_dpif *ofproto)
2201 for (i = 0; i < MAX_MIRRORS; i++) {
2202 struct ofmirror *m = ofproto->mirrors[i];
2205 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2209 for (i = 0; i < MAX_MIRRORS; i++) {
2210 struct ofmirror *m1 = ofproto->mirrors[i];
2217 for (j = i + 1; j < MAX_MIRRORS; j++) {
2218 struct ofmirror *m2 = ofproto->mirrors[j];
2220 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2221 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2222 m2->dup_mirrors |= m1->dup_mirrors;
2229 mirror_set(struct ofproto *ofproto_, void *aux,
2230 const struct ofproto_mirror_settings *s)
2232 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2233 mirror_mask_t mirror_bit;
2234 struct ofbundle *bundle;
2235 struct ofmirror *mirror;
2236 struct ofbundle *out;
2237 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2238 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2241 mirror = mirror_lookup(ofproto, aux);
2243 mirror_destroy(mirror);
2249 idx = mirror_scan(ofproto);
2251 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2253 ofproto->up.name, MAX_MIRRORS, s->name);
2257 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2258 mirror->ofproto = ofproto;
2261 mirror->out_vlan = -1;
2262 mirror->name = NULL;
2265 if (!mirror->name || strcmp(s->name, mirror->name)) {
2267 mirror->name = xstrdup(s->name);
2270 /* Get the new configuration. */
2271 if (s->out_bundle) {
2272 out = bundle_lookup(ofproto, s->out_bundle);
2274 mirror_destroy(mirror);
2280 out_vlan = s->out_vlan;
2282 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2283 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2285 /* If the configuration has not changed, do nothing. */
2286 if (hmapx_equals(&srcs, &mirror->srcs)
2287 && hmapx_equals(&dsts, &mirror->dsts)
2288 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2289 && mirror->out == out
2290 && mirror->out_vlan == out_vlan)
2292 hmapx_destroy(&srcs);
2293 hmapx_destroy(&dsts);
2297 hmapx_swap(&srcs, &mirror->srcs);
2298 hmapx_destroy(&srcs);
2300 hmapx_swap(&dsts, &mirror->dsts);
2301 hmapx_destroy(&dsts);
2303 free(mirror->vlans);
2304 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2307 mirror->out_vlan = out_vlan;
2309 /* Update bundles. */
2310 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2311 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2312 if (hmapx_contains(&mirror->srcs, bundle)) {
2313 bundle->src_mirrors |= mirror_bit;
2315 bundle->src_mirrors &= ~mirror_bit;
2318 if (hmapx_contains(&mirror->dsts, bundle)) {
2319 bundle->dst_mirrors |= mirror_bit;
2321 bundle->dst_mirrors &= ~mirror_bit;
2324 if (mirror->out == bundle) {
2325 bundle->mirror_out |= mirror_bit;
2327 bundle->mirror_out &= ~mirror_bit;
2331 ofproto->need_revalidate = REV_RECONFIGURE;
2332 ofproto->has_mirrors = true;
2333 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2334 mirror_update_dups(ofproto);
2340 mirror_destroy(struct ofmirror *mirror)
2342 struct ofproto_dpif *ofproto;
2343 mirror_mask_t mirror_bit;
2344 struct ofbundle *bundle;
2351 ofproto = mirror->ofproto;
2352 ofproto->need_revalidate = REV_RECONFIGURE;
2353 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2355 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2356 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2357 bundle->src_mirrors &= ~mirror_bit;
2358 bundle->dst_mirrors &= ~mirror_bit;
2359 bundle->mirror_out &= ~mirror_bit;
2362 hmapx_destroy(&mirror->srcs);
2363 hmapx_destroy(&mirror->dsts);
2364 free(mirror->vlans);
2366 ofproto->mirrors[mirror->idx] = NULL;
2370 mirror_update_dups(ofproto);
2372 ofproto->has_mirrors = false;
2373 for (i = 0; i < MAX_MIRRORS; i++) {
2374 if (ofproto->mirrors[i]) {
2375 ofproto->has_mirrors = true;
2382 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2383 uint64_t *packets, uint64_t *bytes)
2385 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2386 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2389 *packets = *bytes = UINT64_MAX;
2393 *packets = mirror->packet_count;
2394 *bytes = mirror->byte_count;
2400 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2402 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2403 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2404 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2410 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2412 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2413 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2414 return bundle && bundle->mirror_out != 0;
2418 forward_bpdu_changed(struct ofproto *ofproto_)
2420 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2421 ofproto->need_revalidate = REV_RECONFIGURE;
2425 set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time)
2427 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2428 mac_learning_set_idle_time(ofproto->ml, idle_time);
2433 static struct ofport_dpif *
2434 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2436 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2437 return ofport ? ofport_dpif_cast(ofport) : NULL;
2440 static struct ofport_dpif *
2441 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2443 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2447 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2448 struct dpif_port *dpif_port)
2450 ofproto_port->name = dpif_port->name;
2451 ofproto_port->type = dpif_port->type;
2452 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2456 port_run_fast(struct ofport_dpif *ofport)
2458 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2459 struct ofpbuf packet;
2461 ofpbuf_init(&packet, 0);
2462 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2463 send_packet(ofport, &packet);
2464 ofpbuf_uninit(&packet);
2469 port_run(struct ofport_dpif *ofport)
2471 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2472 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2473 bool enable = netdev_get_carrier(ofport->up.netdev);
2475 ofport->carrier_seq = carrier_seq;
2477 port_run_fast(ofport);
2479 int cfm_opup = cfm_get_opup(ofport->cfm);
2481 cfm_run(ofport->cfm);
2482 enable = enable && !cfm_get_fault(ofport->cfm);
2484 if (cfm_opup >= 0) {
2485 enable = enable && cfm_opup;
2489 if (ofport->bundle) {
2490 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2491 if (carrier_changed) {
2492 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2496 if (ofport->may_enable != enable) {
2497 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2499 if (ofproto->has_bundle_action) {
2500 ofproto->need_revalidate = REV_PORT_TOGGLED;
2504 ofport->may_enable = enable;
2508 port_wait(struct ofport_dpif *ofport)
2511 cfm_wait(ofport->cfm);
2516 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2517 struct ofproto_port *ofproto_port)
2519 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2520 struct dpif_port dpif_port;
2523 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2525 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2531 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2533 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2534 uint16_t odp_port = UINT16_MAX;
2537 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2539 *ofp_portp = odp_port_to_ofp_port(odp_port);
2545 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2547 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2550 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2552 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2554 /* The caller is going to close ofport->up.netdev. If this is a
2555 * bonded port, then the bond is using that netdev, so remove it
2556 * from the bond. The client will need to reconfigure everything
2557 * after deleting ports, so then the slave will get re-added. */
2558 bundle_remove(&ofport->up);
2565 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2567 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2570 error = netdev_get_stats(ofport->up.netdev, stats);
2572 if (!error && ofport->odp_port == OVSP_LOCAL) {
2573 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2575 /* ofproto->stats.tx_packets represents packets that we created
2576 * internally and sent to some port (e.g. packets sent with
2577 * send_packet()). Account for them as if they had come from
2578 * OFPP_LOCAL and got forwarded. */
2580 if (stats->rx_packets != UINT64_MAX) {
2581 stats->rx_packets += ofproto->stats.tx_packets;
2584 if (stats->rx_bytes != UINT64_MAX) {
2585 stats->rx_bytes += ofproto->stats.tx_bytes;
2588 /* ofproto->stats.rx_packets represents packets that were received on
2589 * some port and we processed internally and dropped (e.g. STP).
2590 * Account for them as if they had been forwarded to OFPP_LOCAL. */
2592 if (stats->tx_packets != UINT64_MAX) {
2593 stats->tx_packets += ofproto->stats.rx_packets;
2596 if (stats->tx_bytes != UINT64_MAX) {
2597 stats->tx_bytes += ofproto->stats.rx_bytes;
2604 /* Account packets for LOCAL port. */
2606 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2607 size_t tx_size, size_t rx_size)
2609 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2612 ofproto->stats.rx_packets++;
2613 ofproto->stats.rx_bytes += rx_size;
2616 ofproto->stats.tx_packets++;
2617 ofproto->stats.tx_bytes += tx_size;
2621 struct port_dump_state {
2622 struct dpif_port_dump dump;
2627 port_dump_start(const struct ofproto *ofproto_, void **statep)
2629 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2630 struct port_dump_state *state;
2632 *statep = state = xmalloc(sizeof *state);
2633 dpif_port_dump_start(&state->dump, ofproto->dpif);
2634 state->done = false;
2639 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2640 struct ofproto_port *port)
2642 struct port_dump_state *state = state_;
2643 struct dpif_port dpif_port;
2645 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2646 ofproto_port_from_dpif_port(port, &dpif_port);
2649 int error = dpif_port_dump_done(&state->dump);
2651 return error ? error : EOF;
2656 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2658 struct port_dump_state *state = state_;
2661 dpif_port_dump_done(&state->dump);
2668 port_poll(const struct ofproto *ofproto_, char **devnamep)
2670 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2671 return dpif_port_poll(ofproto->dpif, devnamep);
2675 port_poll_wait(const struct ofproto *ofproto_)
2677 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2678 dpif_port_poll_wait(ofproto->dpif);
2682 port_is_lacp_current(const struct ofport *ofport_)
2684 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2685 return (ofport->bundle && ofport->bundle->lacp
2686 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2690 /* Upcall handling. */
2692 /* Flow miss batching.
2694 * Some dpifs implement operations faster when you hand them off in a batch.
2695 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2696 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2697 * more packets, plus possibly installing the flow in the dpif.
2699 * So far we only batch the operations that affect flow setup time the most.
2700 * It's possible to batch more than that, but the benefit might be minimal. */
2702 struct hmap_node hmap_node;
2704 enum odp_key_fitness key_fitness;
2705 const struct nlattr *key;
2707 ovs_be16 initial_tci;
2708 struct list packets;
2709 enum dpif_upcall_type upcall_type;
2712 struct flow_miss_op {
2713 struct dpif_op dpif_op;
2714 struct subfacet *subfacet; /* Subfacet */
2715 void *garbage; /* Pointer to pass to free(), NULL if none. */
2716 uint64_t stub[1024 / 8]; /* Temporary buffer. */
2719 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2720 * OpenFlow controller as necessary according to their individual
2721 * configurations. */
2723 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
2724 const struct flow *flow)
2726 struct ofputil_packet_in pin;
2728 pin.packet = packet->data;
2729 pin.packet_len = packet->size;
2730 pin.reason = OFPR_NO_MATCH;
2731 pin.controller_id = 0;
2736 pin.send_len = 0; /* not used for flow table misses */
2738 flow_get_metadata(flow, &pin.fmd);
2740 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
2743 static enum slow_path_reason
2744 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2745 const struct ofpbuf *packet)
2747 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2753 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2755 cfm_process_heartbeat(ofport->cfm, packet);
2758 } else if (ofport->bundle && ofport->bundle->lacp
2759 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2761 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2764 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2766 stp_process_packet(ofport, packet);
2773 static struct flow_miss *
2774 flow_miss_find(struct hmap *todo, const struct flow *flow, uint32_t hash)
2776 struct flow_miss *miss;
2778 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2779 if (flow_equal(&miss->flow, flow)) {
2787 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
2788 * 'packet'. The caller must initialize op->actions and op->actions_len. If
2789 * 'miss' is associated with a subfacet the caller must also initialize the
2790 * returned op->subfacet, and if anything needs to be freed after processing
2791 * the op, the caller must initialize op->garbage also. */
2793 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
2794 struct flow_miss_op *op)
2796 if (miss->flow.vlan_tci != miss->initial_tci) {
2797 /* This packet was received on a VLAN splinter port. We
2798 * added a VLAN to the packet to make the packet resemble
2799 * the flow, but the actions were composed assuming that
2800 * the packet contained no VLAN. So, we must remove the
2801 * VLAN header from the packet before trying to execute the
2803 eth_pop_vlan(packet);
2806 op->subfacet = NULL;
2808 op->dpif_op.type = DPIF_OP_EXECUTE;
2809 op->dpif_op.u.execute.key = miss->key;
2810 op->dpif_op.u.execute.key_len = miss->key_len;
2811 op->dpif_op.u.execute.packet = packet;
2814 /* Helper for handle_flow_miss_without_facet() and
2815 * handle_flow_miss_with_facet(). */
2817 handle_flow_miss_common(struct rule_dpif *rule,
2818 struct ofpbuf *packet, const struct flow *flow)
2820 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2822 ofproto->n_matches++;
2824 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2826 * Extra-special case for fail-open mode.
2828 * We are in fail-open mode and the packet matched the fail-open
2829 * rule, but we are connected to a controller too. We should send
2830 * the packet up to the controller in the hope that it will try to
2831 * set up a flow and thereby allow us to exit fail-open.
2833 * See the top-level comment in fail-open.c for more information.
2835 send_packet_in_miss(ofproto, packet, flow);
2839 /* Figures out whether a flow that missed in 'ofproto', whose details are in
2840 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
2841 * installing a datapath flow. The answer is usually "yes" (a return value of
2842 * true). However, for short flows the cost of bookkeeping is much higher than
2843 * the benefits, so when the datapath holds a large number of flows we impose
2844 * some heuristics to decide which flows are likely to be worth tracking. */
2846 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
2847 struct flow_miss *miss, uint32_t hash)
2849 if (!ofproto->governor) {
2852 n_subfacets = hmap_count(&ofproto->subfacets);
2853 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
2857 ofproto->governor = governor_create(ofproto->up.name);
2860 return governor_should_install_flow(ofproto->governor, hash,
2861 list_size(&miss->packets));
2864 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
2865 * or creating any datapath flow. May add an "execute" operation to 'ops' and
2866 * increment '*n_ops'. */
2868 handle_flow_miss_without_facet(struct flow_miss *miss,
2869 struct rule_dpif *rule,
2870 struct flow_miss_op *ops, size_t *n_ops)
2872 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2873 long long int now = time_msec();
2874 struct action_xlate_ctx ctx;
2875 struct ofpbuf *packet;
2877 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2878 struct flow_miss_op *op = &ops[*n_ops];
2879 struct dpif_flow_stats stats;
2880 struct ofpbuf odp_actions;
2882 COVERAGE_INC(facet_suppress);
2884 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
2886 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
2887 rule_credit_stats(rule, &stats);
2889 action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci,
2891 ctx.resubmit_stats = &stats;
2892 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
2895 if (odp_actions.size) {
2896 struct dpif_execute *execute = &op->dpif_op.u.execute;
2898 init_flow_miss_execute_op(miss, packet, op);
2899 execute->actions = odp_actions.data;
2900 execute->actions_len = odp_actions.size;
2901 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
2905 ofpbuf_uninit(&odp_actions);
2910 /* Handles 'miss', which matches 'facet'. May add any required datapath
2911 * operations to 'ops', incrementing '*n_ops' for each new op.
2913 * All of the packets in 'miss' are considered to have arrived at time 'now'.
2914 * This is really important only for new facets: if we just called time_msec()
2915 * here, then the new subfacet or its packets could look (occasionally) as
2916 * though it was used some time after the facet was used. That can make a
2917 * one-packet flow look like it has a nonzero duration, which looks odd in
2918 * e.g. NetFlow statistics. */
2920 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
2922 struct flow_miss_op *ops, size_t *n_ops)
2924 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
2925 enum subfacet_path want_path;
2926 struct subfacet *subfacet;
2927 struct ofpbuf *packet;
2929 subfacet = subfacet_create(facet,
2930 miss->key_fitness, miss->key, miss->key_len,
2931 miss->initial_tci, now);
2933 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2934 struct flow_miss_op *op = &ops[*n_ops];
2935 struct dpif_flow_stats stats;
2936 struct ofpbuf odp_actions;
2938 handle_flow_miss_common(facet->rule, packet, &miss->flow);
2940 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
2941 if (!subfacet->actions || subfacet->slow) {
2942 subfacet_make_actions(subfacet, packet, &odp_actions);
2945 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
2946 subfacet_update_stats(subfacet, &stats);
2948 if (subfacet->actions_len) {
2949 struct dpif_execute *execute = &op->dpif_op.u.execute;
2951 init_flow_miss_execute_op(miss, packet, op);
2952 op->subfacet = subfacet;
2953 if (!subfacet->slow) {
2954 execute->actions = subfacet->actions;
2955 execute->actions_len = subfacet->actions_len;
2956 ofpbuf_uninit(&odp_actions);
2958 execute->actions = odp_actions.data;
2959 execute->actions_len = odp_actions.size;
2960 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
2965 ofpbuf_uninit(&odp_actions);
2969 want_path = subfacet_want_path(subfacet->slow);
2970 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
2971 struct flow_miss_op *op = &ops[(*n_ops)++];
2972 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
2974 op->subfacet = subfacet;
2976 op->dpif_op.type = DPIF_OP_FLOW_PUT;
2977 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2978 put->key = miss->key;
2979 put->key_len = miss->key_len;
2980 if (want_path == SF_FAST_PATH) {
2981 put->actions = subfacet->actions;
2982 put->actions_len = subfacet->actions_len;
2984 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
2985 op->stub, sizeof op->stub,
2986 &put->actions, &put->actions_len);
2992 /* Handles flow miss 'miss' on 'ofproto'. May add any required datapath
2993 * operations to 'ops', incrementing '*n_ops' for each new op. */
2995 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2996 struct flow_miss_op *ops, size_t *n_ops)
2998 struct facet *facet;
3002 /* The caller must ensure that miss->hmap_node.hash contains
3003 * flow_hash(miss->flow, 0). */
3004 hash = miss->hmap_node.hash;
3006 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3008 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
3010 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
3011 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
3015 facet = facet_create(rule, &miss->flow, hash);
3020 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3023 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
3024 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
3025 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
3026 * what a flow key should contain.
3028 * This function also includes some logic to help make VLAN splinters
3029 * transparent to the rest of the upcall processing logic. In particular, if
3030 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
3031 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
3032 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
3034 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
3035 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
3036 * (This differs from the value returned in flow->vlan_tci only for packets
3037 * received on VLAN splinters.)
3039 static enum odp_key_fitness
3040 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
3041 const struct nlattr *key, size_t key_len,
3042 struct flow *flow, ovs_be16 *initial_tci,
3043 struct ofpbuf *packet)
3045 enum odp_key_fitness fitness;
3047 fitness = odp_flow_key_to_flow(key, key_len, flow);
3048 if (fitness == ODP_FIT_ERROR) {
3051 *initial_tci = flow->vlan_tci;
3053 if (vsp_adjust_flow(ofproto, flow)) {
3055 /* Make the packet resemble the flow, so that it gets sent to an
3056 * OpenFlow controller properly, so that it looks correct for
3057 * sFlow, and so that flow_extract() will get the correct vlan_tci
3058 * if it is called on 'packet'.
3060 * The allocated space inside 'packet' probably also contains
3061 * 'key', that is, both 'packet' and 'key' are probably part of a
3062 * struct dpif_upcall (see the large comment on that structure
3063 * definition), so pushing data on 'packet' is in general not a
3064 * good idea since it could overwrite 'key' or free it as a side
3065 * effect. However, it's OK in this special case because we know
3066 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
3067 * will just overwrite the 4-byte "struct nlattr", which is fine
3068 * since we don't need that header anymore. */
3069 eth_push_vlan(packet, flow->vlan_tci);
3072 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
3073 if (fitness == ODP_FIT_PERFECT) {
3074 fitness = ODP_FIT_TOO_MUCH;
3082 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
3085 struct dpif_upcall *upcall;
3086 struct flow_miss *miss;
3087 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3088 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3089 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3099 /* Construct the to-do list.
3101 * This just amounts to extracting the flow from each packet and sticking
3102 * the packets that have the same flow in the same "flow_miss" structure so
3103 * that we can process them together. */
3106 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3107 struct flow_miss *miss = &misses[n_misses];
3108 struct flow_miss *existing_miss;
3112 /* Obtain metadata and check userspace/kernel agreement on flow match,
3113 * then set 'flow''s header pointers. */
3114 miss->key_fitness = ofproto_dpif_extract_flow_key(
3115 ofproto, upcall->key, upcall->key_len,
3116 &flow, &miss->initial_tci, upcall->packet);
3117 if (miss->key_fitness == ODP_FIT_ERROR) {
3120 flow_extract(upcall->packet, flow.skb_priority,
3121 &flow.tunnel, flow.in_port, &miss->flow);
3123 /* Add other packets to a to-do list. */
3124 hash = flow_hash(&miss->flow, 0);
3125 existing_miss = flow_miss_find(&todo, &miss->flow, hash);
3126 if (!existing_miss) {
3127 hmap_insert(&todo, &miss->hmap_node, hash);
3128 miss->key = upcall->key;
3129 miss->key_len = upcall->key_len;
3130 miss->upcall_type = upcall->type;
3131 list_init(&miss->packets);
3135 miss = existing_miss;
3137 list_push_back(&miss->packets, &upcall->packet->list_node);
3140 /* Process each element in the to-do list, constructing the set of
3141 * operations to batch. */
3143 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3144 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
3146 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3148 /* Execute batch. */
3149 for (i = 0; i < n_ops; i++) {
3150 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3152 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
3154 /* Free memory and update facets. */
3155 for (i = 0; i < n_ops; i++) {
3156 struct flow_miss_op *op = &flow_miss_ops[i];
3158 switch (op->dpif_op.type) {
3159 case DPIF_OP_EXECUTE:
3162 case DPIF_OP_FLOW_PUT:
3163 if (!op->dpif_op.error) {
3164 op->subfacet->path = subfacet_want_path(op->subfacet->slow);
3168 case DPIF_OP_FLOW_DEL:
3174 hmap_destroy(&todo);
3177 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
3178 classify_upcall(const struct dpif_upcall *upcall)
3180 union user_action_cookie cookie;
3182 /* First look at the upcall type. */
3183 switch (upcall->type) {
3184 case DPIF_UC_ACTION:
3190 case DPIF_N_UC_TYPES:
3192 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3196 /* "action" upcalls need a closer look. */
3197 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3198 switch (cookie.type) {
3199 case USER_ACTION_COOKIE_SFLOW:
3200 return SFLOW_UPCALL;
3202 case USER_ACTION_COOKIE_SLOW_PATH:
3205 case USER_ACTION_COOKIE_UNSPEC:
3207 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
3213 handle_sflow_upcall(struct ofproto_dpif *ofproto,
3214 const struct dpif_upcall *upcall)
3216 union user_action_cookie cookie;
3217 enum odp_key_fitness fitness;
3218 ovs_be16 initial_tci;
3221 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
3222 upcall->key_len, &flow,
3223 &initial_tci, upcall->packet);
3224 if (fitness == ODP_FIT_ERROR) {
3228 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3229 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, &cookie);
3233 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
3235 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3236 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3237 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3242 assert(max_batch <= FLOW_MISS_MAX_BATCH);
3245 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3246 struct dpif_upcall *upcall = &misses[n_misses];
3247 struct ofpbuf *buf = &miss_bufs[n_misses];
3250 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3251 sizeof miss_buf_stubs[n_misses]);
3252 error = dpif_recv(ofproto->dpif, upcall, buf);
3258 switch (classify_upcall(upcall)) {
3260 /* Handle it later. */
3265 if (ofproto->sflow) {
3266 handle_sflow_upcall(ofproto, upcall);
3277 /* Handle deferred MISS_UPCALL processing. */
3278 handle_miss_upcalls(ofproto, misses, n_misses);
3279 for (i = 0; i < n_misses; i++) {
3280 ofpbuf_uninit(&miss_bufs[i]);
3286 /* Flow expiration. */
3288 static int subfacet_max_idle(const struct ofproto_dpif *);
3289 static void update_stats(struct ofproto_dpif *);
3290 static void rule_expire(struct rule_dpif *);
3291 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
3293 /* This function is called periodically by run(). Its job is to collect
3294 * updates for the flows that have been installed into the datapath, most
3295 * importantly when they last were used, and then use that information to
3296 * expire flows that have not been used recently.
3298 * Returns the number of milliseconds after which it should be called again. */
3300 expire(struct ofproto_dpif *ofproto)
3302 struct rule_dpif *rule, *next_rule;
3303 struct oftable *table;
3306 /* Update stats for each flow in the datapath. */
3307 update_stats(ofproto);
3309 /* Expire subfacets that have been idle too long. */
3310 dp_max_idle = subfacet_max_idle(ofproto);
3311 expire_subfacets(ofproto, dp_max_idle);
3313 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
3314 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
3315 struct cls_cursor cursor;
3317 cls_cursor_init(&cursor, &table->cls, NULL);
3318 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
3323 /* All outstanding data in existing flows has been accounted, so it's a
3324 * good time to do bond rebalancing. */
3325 if (ofproto->has_bonded_bundles) {
3326 struct ofbundle *bundle;
3328 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3330 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
3335 return MIN(dp_max_idle, 1000);
3338 /* Updates flow table statistics given that the datapath just reported 'stats'
3339 * as 'subfacet''s statistics. */
3341 update_subfacet_stats(struct subfacet *subfacet,
3342 const struct dpif_flow_stats *stats)
3344 struct facet *facet = subfacet->facet;
3346 if (stats->n_packets >= subfacet->dp_packet_count) {
3347 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
3348 facet->packet_count += extra;
3350 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3353 if (stats->n_bytes >= subfacet->dp_byte_count) {
3354 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
3356 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3359 subfacet->dp_packet_count = stats->n_packets;
3360 subfacet->dp_byte_count = stats->n_bytes;
3362 facet->tcp_flags |= stats->tcp_flags;
3364 subfacet_update_time(subfacet, stats->used);
3365 if (facet->accounted_bytes < facet->byte_count) {
3367 facet_account(facet);
3368 facet->accounted_bytes = facet->byte_count;
3370 facet_push_stats(facet);
3373 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3374 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3376 delete_unexpected_flow(struct dpif *dpif,
3377 const struct nlattr *key, size_t key_len)
3379 if (!VLOG_DROP_WARN(&rl)) {
3383 odp_flow_key_format(key, key_len, &s);
3384 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
3388 COVERAGE_INC(facet_unexpected);
3389 dpif_flow_del(dpif, key, key_len, NULL);
3392 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3394 * This function also pushes statistics updates to rules which each facet
3395 * resubmits into. Generally these statistics will be accurate. However, if a
3396 * facet changes the rule it resubmits into at some time in between
3397 * update_stats() runs, it is possible that statistics accrued to the
3398 * old rule will be incorrectly attributed to the new rule. This could be
3399 * avoided by calling update_stats() whenever rules are created or
3400 * deleted. However, the performance impact of making so many calls to the
3401 * datapath do not justify the benefit of having perfectly accurate statistics.
3404 update_stats(struct ofproto_dpif *p)
3406 const struct dpif_flow_stats *stats;
3407 struct dpif_flow_dump dump;
3408 const struct nlattr *key;
3411 dpif_flow_dump_start(&dump, p->dpif);
3412 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
3413 struct subfacet *subfacet;
3415 subfacet = subfacet_find(p, key, key_len);
3416 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3418 update_subfacet_stats(subfacet, stats);
3422 /* Stats are updated per-packet. */
3425 case SF_NOT_INSTALLED:
3427 delete_unexpected_flow(p->dpif, key, key_len);
3431 dpif_flow_dump_done(&dump);
3434 /* Calculates and returns the number of milliseconds of idle time after which
3435 * subfacets should expire from the datapath. When a subfacet expires, we fold
3436 * its statistics into its facet, and when a facet's last subfacet expires, we
3437 * fold its statistic into its rule. */
3439 subfacet_max_idle(const struct ofproto_dpif *ofproto)
3442 * Idle time histogram.
3444 * Most of the time a switch has a relatively small number of subfacets.
3445 * When this is the case we might as well keep statistics for all of them
3446 * in userspace and to cache them in the kernel datapath for performance as
3449 * As the number of subfacets increases, the memory required to maintain
3450 * statistics about them in userspace and in the kernel becomes
3451 * significant. However, with a large number of subfacets it is likely
3452 * that only a few of them are "heavy hitters" that consume a large amount
3453 * of bandwidth. At this point, only heavy hitters are worth caching in
3454 * the kernel and maintaining in userspaces; other subfacets we can
3457 * The technique used to compute the idle time is to build a histogram with
3458 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3459 * that is installed in the kernel gets dropped in the appropriate bucket.
3460 * After the histogram has been built, we compute the cutoff so that only
3461 * the most-recently-used 1% of subfacets (but at least
3462 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3463 * the most-recently-used bucket of subfacets is kept, so actually an
3464 * arbitrary number of subfacets can be kept in any given expiration run
3465 * (though the next run will delete most of those unless they receive
3468 * This requires a second pass through the subfacets, in addition to the
3469 * pass made by update_stats(), because the former function never looks at
3470 * uninstallable subfacets.
3472 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3473 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3474 int buckets[N_BUCKETS] = { 0 };
3475 int total, subtotal, bucket;
3476 struct subfacet *subfacet;
3480 total = hmap_count(&ofproto->subfacets);
3481 if (total <= ofproto->up.flow_eviction_threshold) {
3482 return N_BUCKETS * BUCKET_WIDTH;
3485 /* Build histogram. */
3487 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3488 long long int idle = now - subfacet->used;
3489 int bucket = (idle <= 0 ? 0
3490 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3491 : (unsigned int) idle / BUCKET_WIDTH);
3495 /* Find the first bucket whose flows should be expired. */
3496 subtotal = bucket = 0;
3498 subtotal += buckets[bucket++];
3499 } while (bucket < N_BUCKETS &&
3500 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3502 if (VLOG_IS_DBG_ENABLED()) {
3506 ds_put_cstr(&s, "keep");
3507 for (i = 0; i < N_BUCKETS; i++) {
3509 ds_put_cstr(&s, ", drop");
3512 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3515 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3519 return bucket * BUCKET_WIDTH;
3522 enum { EXPIRE_MAX_BATCH = 50 };
3525 expire_batch(struct ofproto_dpif *ofproto, struct subfacet **subfacets, int n)
3527 struct odputil_keybuf keybufs[EXPIRE_MAX_BATCH];
3528 struct dpif_op ops[EXPIRE_MAX_BATCH];
3529 struct dpif_op *opsp[EXPIRE_MAX_BATCH];
3530 struct ofpbuf keys[EXPIRE_MAX_BATCH];
3531 struct dpif_flow_stats stats[EXPIRE_MAX_BATCH];
3534 for (i = 0; i < n; i++) {
3535 ops[i].type = DPIF_OP_FLOW_DEL;
3536 subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]);
3537 ops[i].u.flow_del.key = keys[i].data;
3538 ops[i].u.flow_del.key_len = keys[i].size;
3539 ops[i].u.flow_del.stats = &stats[i];
3543 dpif_operate(ofproto->dpif, opsp, n);
3544 for (i = 0; i < n; i++) {
3545 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
3546 subfacets[i]->path = SF_NOT_INSTALLED;
3547 subfacet_destroy(subfacets[i]);
3552 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3554 /* Cutoff time for most flows. */
3555 long long int normal_cutoff = time_msec() - dp_max_idle;
3557 /* We really want to keep flows for special protocols around, so use a more
3558 * conservative cutoff. */
3559 long long int special_cutoff = time_msec() - 10000;
3561 struct subfacet *subfacet, *next_subfacet;
3562 struct subfacet *batch[EXPIRE_MAX_BATCH];
3566 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3567 &ofproto->subfacets) {
3568 long long int cutoff;
3570 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
3573 if (subfacet->used < cutoff) {
3574 if (subfacet->path != SF_NOT_INSTALLED) {
3575 batch[n_batch++] = subfacet;
3576 if (n_batch >= EXPIRE_MAX_BATCH) {
3577 expire_batch(ofproto, batch, n_batch);
3581 subfacet_destroy(subfacet);
3587 expire_batch(ofproto, batch, n_batch);
3591 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3592 * then delete it entirely. */
3594 rule_expire(struct rule_dpif *rule)
3596 struct facet *facet, *next_facet;
3600 if (rule->up.pending) {
3601 /* We'll have to expire it later. */
3605 /* Has 'rule' expired? */
3607 if (rule->up.hard_timeout
3608 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3609 reason = OFPRR_HARD_TIMEOUT;
3610 } else if (rule->up.idle_timeout
3611 && now > rule->up.used + rule->up.idle_timeout * 1000) {
3612 reason = OFPRR_IDLE_TIMEOUT;
3617 COVERAGE_INC(ofproto_dpif_expired);
3619 /* Update stats. (This is a no-op if the rule expired due to an idle
3620 * timeout, because that only happens when the rule has no facets left.) */
3621 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3622 facet_remove(facet);
3625 /* Get rid of the rule. */
3626 ofproto_rule_expire(&rule->up, reason);
3631 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3633 * The caller must already have determined that no facet with an identical
3634 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3635 * the ofproto's classifier table.
3637 * 'hash' must be the return value of flow_hash(flow, 0).
3639 * The facet will initially have no subfacets. The caller should create (at
3640 * least) one subfacet with subfacet_create(). */
3641 static struct facet *
3642 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
3644 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3645 struct facet *facet;
3647 facet = xzalloc(sizeof *facet);
3648 facet->used = time_msec();
3649 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
3650 list_push_back(&rule->facets, &facet->list_node);
3652 facet->flow = *flow;
3653 list_init(&facet->subfacets);
3654 netflow_flow_init(&facet->nf_flow);
3655 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3661 facet_free(struct facet *facet)
3666 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3667 * 'packet', which arrived on 'in_port'.
3669 * Takes ownership of 'packet'. */
3671 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3672 const struct nlattr *odp_actions, size_t actions_len,
3673 struct ofpbuf *packet)
3675 struct odputil_keybuf keybuf;
3679 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3680 odp_flow_key_from_flow(&key, flow);
3682 error = dpif_execute(ofproto->dpif, key.data, key.size,
3683 odp_actions, actions_len, packet);
3685 ofpbuf_delete(packet);
3689 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3691 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3692 * rule's statistics, via subfacet_uninstall().
3694 * - Removes 'facet' from its rule and from ofproto->facets.
3697 facet_remove(struct facet *facet)
3699 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3700 struct subfacet *subfacet, *next_subfacet;
3702 assert(!list_is_empty(&facet->subfacets));
3704 /* First uninstall all of the subfacets to get final statistics. */
3705 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3706 subfacet_uninstall(subfacet);
3709 /* Flush the final stats to the rule.
3711 * This might require us to have at least one subfacet around so that we
3712 * can use its actions for accounting in facet_account(), which is why we
3713 * have uninstalled but not yet destroyed the subfacets. */
3714 facet_flush_stats(facet);
3716 /* Now we're really all done so destroy everything. */
3717 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3718 &facet->subfacets) {
3719 subfacet_destroy__(subfacet);
3721 hmap_remove(&ofproto->facets, &facet->hmap_node);
3722 list_remove(&facet->list_node);
3726 /* Feed information from 'facet' back into the learning table to keep it in
3727 * sync with what is actually flowing through the datapath. */
3729 facet_learn(struct facet *facet)
3731 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3732 struct action_xlate_ctx ctx;
3734 if (!facet->has_learn
3735 && !facet->has_normal
3736 && (!facet->has_fin_timeout
3737 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
3741 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3742 facet->flow.vlan_tci,
3743 facet->rule, facet->tcp_flags, NULL);
3744 ctx.may_learn = true;
3745 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
3746 facet->rule->up.ofpacts_len);
3750 facet_account(struct facet *facet)
3752 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3753 struct subfacet *subfacet;
3754 const struct nlattr *a;
3759 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3762 n_bytes = facet->byte_count - facet->accounted_bytes;
3764 /* This loop feeds byte counters to bond_account() for rebalancing to use
3765 * as a basis. We also need to track the actual VLAN on which the packet
3766 * is going to be sent to ensure that it matches the one passed to
3767 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3770 * We use the actions from an arbitrary subfacet because they should all
3771 * be equally valid for our purpose. */
3772 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3773 struct subfacet, list_node);
3774 vlan_tci = facet->flow.vlan_tci;
3775 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3776 subfacet->actions, subfacet->actions_len) {
3777 const struct ovs_action_push_vlan *vlan;
3778 struct ofport_dpif *port;
3780 switch (nl_attr_type(a)) {
3781 case OVS_ACTION_ATTR_OUTPUT:
3782 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3783 if (port && port->bundle && port->bundle->bond) {
3784 bond_account(port->bundle->bond, &facet->flow,
3785 vlan_tci_to_vid(vlan_tci), n_bytes);
3789 case OVS_ACTION_ATTR_POP_VLAN:
3790 vlan_tci = htons(0);
3793 case OVS_ACTION_ATTR_PUSH_VLAN:
3794 vlan = nl_attr_get(a);
3795 vlan_tci = vlan->vlan_tci;
3801 /* Returns true if the only action for 'facet' is to send to the controller.
3802 * (We don't report NetFlow expiration messages for such facets because they
3803 * are just part of the control logic for the network, not real traffic). */
3805 facet_is_controller_flow(struct facet *facet)
3808 const struct rule *rule = &facet->rule->up;
3809 const struct ofpact *ofpacts = rule->ofpacts;
3810 size_t ofpacts_len = rule->ofpacts_len;
3812 if (ofpacts_len > 0 &&
3813 ofpacts->type == OFPACT_CONTROLLER &&
3814 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
3821 /* Folds all of 'facet''s statistics into its rule. Also updates the
3822 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3823 * 'facet''s statistics in the datapath should have been zeroed and folded into
3824 * its packet and byte counts before this function is called. */
3826 facet_flush_stats(struct facet *facet)
3828 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3829 struct subfacet *subfacet;
3831 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3832 assert(!subfacet->dp_byte_count);
3833 assert(!subfacet->dp_packet_count);
3836 facet_push_stats(facet);
3837 if (facet->accounted_bytes < facet->byte_count) {
3838 facet_account(facet);
3839 facet->accounted_bytes = facet->byte_count;
3842 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3843 struct ofexpired expired;
3844 expired.flow = facet->flow;
3845 expired.packet_count = facet->packet_count;
3846 expired.byte_count = facet->byte_count;
3847 expired.used = facet->used;
3848 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3851 facet->rule->packet_count += facet->packet_count;
3852 facet->rule->byte_count += facet->byte_count;
3854 /* Reset counters to prevent double counting if 'facet' ever gets
3856 facet_reset_counters(facet);
3858 netflow_flow_clear(&facet->nf_flow);
3859 facet->tcp_flags = 0;
3862 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3863 * Returns it if found, otherwise a null pointer.
3865 * 'hash' must be the return value of flow_hash(flow, 0).
3867 * The returned facet might need revalidation; use facet_lookup_valid()
3868 * instead if that is important. */
3869 static struct facet *
3870 facet_find(struct ofproto_dpif *ofproto,
3871 const struct flow *flow, uint32_t hash)
3873 struct facet *facet;
3875 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
3876 if (flow_equal(flow, &facet->flow)) {
3884 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3885 * Returns it if found, otherwise a null pointer.
3887 * 'hash' must be the return value of flow_hash(flow, 0).
3889 * The returned facet is guaranteed to be valid. */
3890 static struct facet *
3891 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
3894 struct facet *facet;
3896 facet = facet_find(ofproto, flow, hash);
3898 && (ofproto->need_revalidate
3899 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))) {
3900 facet_revalidate(facet);
3907 subfacet_path_to_string(enum subfacet_path path)
3910 case SF_NOT_INSTALLED:
3911 return "not installed";
3913 return "in fast path";
3915 return "in slow path";
3921 /* Returns the path in which a subfacet should be installed if its 'slow'
3922 * member has the specified value. */
3923 static enum subfacet_path
3924 subfacet_want_path(enum slow_path_reason slow)
3926 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
3929 /* Returns true if 'subfacet' needs to have its datapath flow updated,
3930 * supposing that its actions have been recalculated as 'want_actions' and that
3931 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
3933 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
3934 const struct ofpbuf *want_actions)
3936 enum subfacet_path want_path = subfacet_want_path(slow);
3937 return (want_path != subfacet->path
3938 || (want_path == SF_FAST_PATH
3939 && (subfacet->actions_len != want_actions->size
3940 || memcmp(subfacet->actions, want_actions->data,
3941 subfacet->actions_len))));
3945 facet_check_consistency(struct facet *facet)
3947 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3949 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3951 uint64_t odp_actions_stub[1024 / 8];
3952 struct ofpbuf odp_actions;
3954 struct rule_dpif *rule;
3955 struct subfacet *subfacet;
3956 bool may_log = false;
3959 /* Check the rule for consistency. */
3960 rule = rule_dpif_lookup(ofproto, &facet->flow);
3961 ok = rule == facet->rule;
3963 may_log = !VLOG_DROP_WARN(&rl);
3968 flow_format(&s, &facet->flow);
3969 ds_put_format(&s, ": facet associated with wrong rule (was "
3970 "table=%"PRIu8",", facet->rule->up.table_id);
3971 cls_rule_format(&facet->rule->up.cr, &s);
3972 ds_put_format(&s, ") (should have been table=%"PRIu8",",
3974 cls_rule_format(&rule->up.cr, &s);
3975 ds_put_char(&s, ')');
3977 VLOG_WARN("%s", ds_cstr(&s));
3982 /* Check the datapath actions for consistency. */
3983 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
3984 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3985 enum subfacet_path want_path;
3986 struct odputil_keybuf keybuf;
3987 struct action_xlate_ctx ctx;
3991 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3992 subfacet->initial_tci, rule, 0, NULL);
3993 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3996 if (subfacet->path == SF_NOT_INSTALLED) {
3997 /* This only happens if the datapath reported an error when we
3998 * tried to install the flow. Don't flag another error here. */
4002 want_path = subfacet_want_path(subfacet->slow);
4003 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
4004 /* The actions for slow-path flows may legitimately vary from one
4005 * packet to the next. We're done. */
4009 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
4013 /* Inconsistency! */
4015 may_log = !VLOG_DROP_WARN(&rl);
4019 /* Rate-limited, skip reporting. */
4024 subfacet_get_key(subfacet, &keybuf, &key);
4025 odp_flow_key_format(key.data, key.size, &s);
4027 ds_put_cstr(&s, ": inconsistency in subfacet");
4028 if (want_path != subfacet->path) {
4029 enum odp_key_fitness fitness = subfacet->key_fitness;
4031 ds_put_format(&s, " (%s, fitness=%s)",
4032 subfacet_path_to_string(subfacet->path),
4033 odp_key_fitness_to_string(fitness));
4034 ds_put_format(&s, " (should have been %s)",
4035 subfacet_path_to_string(want_path));
4036 } else if (want_path == SF_FAST_PATH) {
4037 ds_put_cstr(&s, " (actions were: ");
4038 format_odp_actions(&s, subfacet->actions,
4039 subfacet->actions_len);
4040 ds_put_cstr(&s, ") (correct actions: ");
4041 format_odp_actions(&s, odp_actions.data, odp_actions.size);
4042 ds_put_char(&s, ')');
4044 ds_put_cstr(&s, " (actions: ");
4045 format_odp_actions(&s, subfacet->actions,
4046 subfacet->actions_len);
4047 ds_put_char(&s, ')');
4049 VLOG_WARN("%s", ds_cstr(&s));
4052 ofpbuf_uninit(&odp_actions);
4057 /* Re-searches the classifier for 'facet':
4059 * - If the rule found is different from 'facet''s current rule, moves
4060 * 'facet' to the new rule and recompiles its actions.
4062 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4063 * where it is and recompiles its actions anyway. */
4065 facet_revalidate(struct facet *facet)
4067 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4069 struct nlattr *odp_actions;
4072 struct actions *new_actions;
4074 struct action_xlate_ctx ctx;
4075 uint64_t odp_actions_stub[1024 / 8];
4076 struct ofpbuf odp_actions;
4078 struct rule_dpif *new_rule;
4079 struct subfacet *subfacet;
4082 COVERAGE_INC(facet_revalidate);
4084 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
4086 /* Calculate new datapath actions.
4088 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4089 * emit a NetFlow expiration and, if so, we need to have the old state
4090 * around to properly compose it. */
4092 /* If the datapath actions changed or the installability changed,
4093 * then we need to talk to the datapath. */
4096 memset(&ctx, 0, sizeof ctx);
4097 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4098 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4099 enum slow_path_reason slow;
4101 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4102 subfacet->initial_tci, new_rule, 0, NULL);
4103 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
4106 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4107 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
4108 struct dpif_flow_stats stats;
4110 subfacet_install(subfacet,
4111 odp_actions.data, odp_actions.size, &stats, slow);
4112 subfacet_update_stats(subfacet, &stats);
4115 new_actions = xcalloc(list_size(&facet->subfacets),
4116 sizeof *new_actions);
4118 new_actions[i].odp_actions = xmemdup(odp_actions.data,
4120 new_actions[i].actions_len = odp_actions.size;
4125 ofpbuf_uninit(&odp_actions);
4128 facet_flush_stats(facet);
4131 /* Update 'facet' now that we've taken care of all the old state. */
4132 facet->tags = ctx.tags;
4133 facet->nf_flow.output_iface = ctx.nf_output_iface;
4134 facet->has_learn = ctx.has_learn;
4135 facet->has_normal = ctx.has_normal;
4136 facet->has_fin_timeout = ctx.has_fin_timeout;
4137 facet->mirrors = ctx.mirrors;
4140 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4141 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4143 if (new_actions && new_actions[i].odp_actions) {
4144 free(subfacet->actions);
4145 subfacet->actions = new_actions[i].odp_actions;
4146 subfacet->actions_len = new_actions[i].actions_len;
4152 if (facet->rule != new_rule) {
4153 COVERAGE_INC(facet_changed_rule);
4154 list_remove(&facet->list_node);
4155 list_push_back(&new_rule->facets, &facet->list_node);
4156 facet->rule = new_rule;
4157 facet->used = new_rule->up.created;
4158 facet->prev_used = facet->used;
4162 /* Updates 'facet''s used time. Caller is responsible for calling
4163 * facet_push_stats() to update the flows which 'facet' resubmits into. */
4165 facet_update_time(struct facet *facet, long long int used)
4167 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4168 if (used > facet->used) {
4170 ofproto_rule_update_used(&facet->rule->up, used);
4171 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
4176 facet_reset_counters(struct facet *facet)
4178 facet->packet_count = 0;
4179 facet->byte_count = 0;
4180 facet->prev_packet_count = 0;
4181 facet->prev_byte_count = 0;
4182 facet->accounted_bytes = 0;
4186 facet_push_stats(struct facet *facet)
4188 struct dpif_flow_stats stats;
4190 assert(facet->packet_count >= facet->prev_packet_count);
4191 assert(facet->byte_count >= facet->prev_byte_count);
4192 assert(facet->used >= facet->prev_used);
4194 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4195 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4196 stats.used = facet->used;
4197 stats.tcp_flags = 0;
4199 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
4200 facet->prev_packet_count = facet->packet_count;
4201 facet->prev_byte_count = facet->byte_count;
4202 facet->prev_used = facet->used;
4204 flow_push_stats(facet->rule, &facet->flow, &stats);
4206 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
4207 facet->mirrors, stats.n_packets, stats.n_bytes);
4212 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4214 rule->packet_count += stats->n_packets;
4215 rule->byte_count += stats->n_bytes;
4216 ofproto_rule_update_used(&rule->up, stats->used);
4219 /* Pushes flow statistics to the rules which 'flow' resubmits into given
4220 * 'rule''s actions and mirrors. */
4222 flow_push_stats(struct rule_dpif *rule,
4223 const struct flow *flow, const struct dpif_flow_stats *stats)
4225 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4226 struct action_xlate_ctx ctx;
4228 ofproto_rule_update_used(&rule->up, stats->used);
4230 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule,
4232 ctx.resubmit_stats = stats;
4233 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
4234 rule->up.ofpacts_len);
4239 static struct subfacet *
4240 subfacet_find__(struct ofproto_dpif *ofproto,
4241 const struct nlattr *key, size_t key_len, uint32_t key_hash,
4242 const struct flow *flow)
4244 struct subfacet *subfacet;
4246 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4247 &ofproto->subfacets) {
4249 ? (subfacet->key_len == key_len
4250 && !memcmp(key, subfacet->key, key_len))
4251 : flow_equal(flow, &subfacet->facet->flow)) {
4259 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4260 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
4261 * there is one, otherwise creates and returns a new subfacet.
4263 * If the returned subfacet is new, then subfacet->actions will be NULL, in
4264 * which case the caller must populate the actions with
4265 * subfacet_make_actions(). */
4266 static struct subfacet *
4267 subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness,
4268 const struct nlattr *key, size_t key_len,
4269 ovs_be16 initial_tci, long long int now)
4271 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4272 uint32_t key_hash = odp_flow_key_hash(key, key_len);
4273 struct subfacet *subfacet;
4275 if (list_is_empty(&facet->subfacets)) {
4276 subfacet = &facet->one_subfacet;
4278 subfacet = subfacet_find__(ofproto, key, key_len, key_hash,
4281 if (subfacet->facet == facet) {
4285 /* This shouldn't happen. */
4286 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4287 subfacet_destroy(subfacet);
4290 subfacet = xmalloc(sizeof *subfacet);
4293 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
4294 list_push_back(&facet->subfacets, &subfacet->list_node);
4295 subfacet->facet = facet;
4296 subfacet->key_fitness = key_fitness;
4297 if (key_fitness != ODP_FIT_PERFECT) {
4298 subfacet->key = xmemdup(key, key_len);
4299 subfacet->key_len = key_len;
4301 subfacet->key = NULL;
4302 subfacet->key_len = 0;
4304 subfacet->used = now;
4305 subfacet->dp_packet_count = 0;
4306 subfacet->dp_byte_count = 0;
4307 subfacet->actions_len = 0;
4308 subfacet->actions = NULL;
4309 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
4312 subfacet->path = SF_NOT_INSTALLED;
4313 subfacet->initial_tci = initial_tci;
4318 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
4319 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
4320 static struct subfacet *
4321 subfacet_find(struct ofproto_dpif *ofproto,
4322 const struct nlattr *key, size_t key_len)
4324 uint32_t key_hash = odp_flow_key_hash(key, key_len);
4325 enum odp_key_fitness fitness;
4328 fitness = odp_flow_key_to_flow(key, key_len, &flow);
4329 if (fitness == ODP_FIT_ERROR) {
4333 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
4336 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4337 * its facet within 'ofproto', and frees it. */
4339 subfacet_destroy__(struct subfacet *subfacet)
4341 struct facet *facet = subfacet->facet;
4342 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4344 subfacet_uninstall(subfacet);
4345 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
4346 list_remove(&subfacet->list_node);
4347 free(subfacet->key);
4348 free(subfacet->actions);
4349 if (subfacet != &facet->one_subfacet) {
4354 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4355 * last remaining subfacet in its facet destroys the facet too. */
4357 subfacet_destroy(struct subfacet *subfacet)
4359 struct facet *facet = subfacet->facet;
4361 if (list_is_singleton(&facet->subfacets)) {
4362 /* facet_remove() needs at least one subfacet (it will remove it). */
4363 facet_remove(facet);
4365 subfacet_destroy__(subfacet);
4369 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
4370 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
4371 * for use as temporary storage. */
4373 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
4376 if (!subfacet->key) {
4377 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
4378 odp_flow_key_from_flow(key, &subfacet->facet->flow);
4380 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
4384 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
4385 * Translates the actions into 'odp_actions', which the caller must have
4386 * initialized and is responsible for uninitializing. */
4388 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
4389 struct ofpbuf *odp_actions)
4391 struct facet *facet = subfacet->facet;
4392 struct rule_dpif *rule = facet->rule;
4393 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4395 struct action_xlate_ctx ctx;
4397 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
4399 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
4400 facet->tags = ctx.tags;
4401 facet->has_learn = ctx.has_learn;
4402 facet->has_normal = ctx.has_normal;
4403 facet->has_fin_timeout = ctx.has_fin_timeout;
4404 facet->nf_flow.output_iface = ctx.nf_output_iface;
4405 facet->mirrors = ctx.mirrors;
4407 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4408 if (subfacet->actions_len != odp_actions->size
4409 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
4410 free(subfacet->actions);
4411 subfacet->actions_len = odp_actions->size;
4412 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
4416 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4417 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4418 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4419 * since 'subfacet' was last updated.
4421 * Returns 0 if successful, otherwise a positive errno value. */
4423 subfacet_install(struct subfacet *subfacet,
4424 const struct nlattr *actions, size_t actions_len,
4425 struct dpif_flow_stats *stats,
4426 enum slow_path_reason slow)
4428 struct facet *facet = subfacet->facet;
4429 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4430 enum subfacet_path path = subfacet_want_path(slow);
4431 uint64_t slow_path_stub[128 / 8];
4432 struct odputil_keybuf keybuf;
4433 enum dpif_flow_put_flags flags;
4437 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
4439 flags |= DPIF_FP_ZERO_STATS;
4442 if (path == SF_SLOW_PATH) {
4443 compose_slow_path(ofproto, &facet->flow, slow,
4444 slow_path_stub, sizeof slow_path_stub,
4445 &actions, &actions_len);
4448 subfacet_get_key(subfacet, &keybuf, &key);
4449 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
4450 actions, actions_len, stats);
4453 subfacet_reset_dp_stats(subfacet, stats);
4457 subfacet->path = path;
4463 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
4465 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
4466 stats, subfacet->slow);
4469 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4471 subfacet_uninstall(struct subfacet *subfacet)
4473 if (subfacet->path != SF_NOT_INSTALLED) {
4474 struct rule_dpif *rule = subfacet->facet->rule;
4475 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4476 struct odputil_keybuf keybuf;
4477 struct dpif_flow_stats stats;
4481 subfacet_get_key(subfacet, &keybuf, &key);
4482 error = dpif_flow_del(ofproto->dpif, key.data, key.size, &stats);
4483 subfacet_reset_dp_stats(subfacet, &stats);
4485 subfacet_update_stats(subfacet, &stats);
4487 subfacet->path = SF_NOT_INSTALLED;
4489 assert(subfacet->dp_packet_count == 0);
4490 assert(subfacet->dp_byte_count == 0);
4494 /* Resets 'subfacet''s datapath statistics counters. This should be called
4495 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4496 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4497 * was reset in the datapath. 'stats' will be modified to include only
4498 * statistics new since 'subfacet' was last updated. */
4500 subfacet_reset_dp_stats(struct subfacet *subfacet,
4501 struct dpif_flow_stats *stats)
4504 && subfacet->dp_packet_count <= stats->n_packets
4505 && subfacet->dp_byte_count <= stats->n_bytes) {
4506 stats->n_packets -= subfacet->dp_packet_count;
4507 stats->n_bytes -= subfacet->dp_byte_count;
4510 subfacet->dp_packet_count = 0;
4511 subfacet->dp_byte_count = 0;
4514 /* Updates 'subfacet''s used time. The caller is responsible for calling
4515 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
4517 subfacet_update_time(struct subfacet *subfacet, long long int used)
4519 if (used > subfacet->used) {
4520 subfacet->used = used;
4521 facet_update_time(subfacet->facet, used);
4525 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4527 * Because of the meaning of a subfacet's counters, it only makes sense to do
4528 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4529 * represents a packet that was sent by hand or if it represents statistics
4530 * that have been cleared out of the datapath. */
4532 subfacet_update_stats(struct subfacet *subfacet,
4533 const struct dpif_flow_stats *stats)
4535 if (stats->n_packets || stats->used > subfacet->used) {
4536 struct facet *facet = subfacet->facet;
4538 subfacet_update_time(subfacet, stats->used);
4539 facet->packet_count += stats->n_packets;
4540 facet->byte_count += stats->n_bytes;
4541 facet->tcp_flags |= stats->tcp_flags;
4542 facet_push_stats(facet);
4543 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
4549 static struct rule_dpif *
4550 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
4552 struct rule_dpif *rule;
4554 rule = rule_dpif_lookup__(ofproto, flow, 0);
4559 return rule_dpif_miss_rule(ofproto, flow);
4562 static struct rule_dpif *
4563 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
4566 struct cls_rule *cls_rule;
4567 struct classifier *cls;
4569 if (table_id >= N_TABLES) {
4573 cls = &ofproto->up.tables[table_id].cls;
4574 if (flow->nw_frag & FLOW_NW_FRAG_ANY
4575 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4576 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
4577 * are unavailable. */
4578 struct flow ofpc_normal_flow = *flow;
4579 ofpc_normal_flow.tp_src = htons(0);
4580 ofpc_normal_flow.tp_dst = htons(0);
4581 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
4583 cls_rule = classifier_lookup(cls, flow);
4585 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
4588 static struct rule_dpif *
4589 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
4591 struct ofport_dpif *port;
4593 port = get_ofp_port(ofproto, flow->in_port);
4595 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
4596 return ofproto->miss_rule;
4599 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
4600 return ofproto->no_packet_in_rule;
4602 return ofproto->miss_rule;
4606 complete_operation(struct rule_dpif *rule)
4608 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4610 rule_invalidate(rule);
4612 struct dpif_completion *c = xmalloc(sizeof *c);
4613 c->op = rule->up.pending;
4614 list_push_back(&ofproto->completions, &c->list_node);
4616 ofoperation_complete(rule->up.pending, 0);
4620 static struct rule *
4623 struct rule_dpif *rule = xmalloc(sizeof *rule);
4628 rule_dealloc(struct rule *rule_)
4630 struct rule_dpif *rule = rule_dpif_cast(rule_);
4635 rule_construct(struct rule *rule_)
4637 struct rule_dpif *rule = rule_dpif_cast(rule_);
4638 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4639 struct rule_dpif *victim;
4642 rule->packet_count = 0;
4643 rule->byte_count = 0;
4645 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
4646 if (victim && !list_is_empty(&victim->facets)) {
4647 struct facet *facet;
4649 rule->facets = victim->facets;
4650 list_moved(&rule->facets);
4651 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4652 /* XXX: We're only clearing our local counters here. It's possible
4653 * that quite a few packets are unaccounted for in the datapath
4654 * statistics. These will be accounted to the new rule instead of
4655 * cleared as required. This could be fixed by clearing out the
4656 * datapath statistics for this facet, but currently it doesn't
4658 facet_reset_counters(facet);
4662 /* Must avoid list_moved() in this case. */
4663 list_init(&rule->facets);
4666 table_id = rule->up.table_id;
4668 rule->tag = victim->tag;
4669 } else if (table_id == 0) {
4674 miniflow_expand(&rule->up.cr.match.flow, &flow);
4675 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
4676 ofproto->tables[table_id].basis);
4679 complete_operation(rule);
4684 rule_destruct(struct rule *rule_)
4686 struct rule_dpif *rule = rule_dpif_cast(rule_);
4687 struct facet *facet, *next_facet;
4689 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4690 facet_revalidate(facet);
4693 complete_operation(rule);
4697 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4699 struct rule_dpif *rule = rule_dpif_cast(rule_);
4700 struct facet *facet;
4702 /* Start from historical data for 'rule' itself that are no longer tracked
4703 * in facets. This counts, for example, facets that have expired. */
4704 *packets = rule->packet_count;
4705 *bytes = rule->byte_count;
4707 /* Add any statistics that are tracked by facets. This includes
4708 * statistical data recently updated by ofproto_update_stats() as well as
4709 * stats for packets that were executed "by hand" via dpif_execute(). */
4710 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4711 *packets += facet->packet_count;
4712 *bytes += facet->byte_count;
4717 rule_execute(struct rule *rule_, const struct flow *flow,
4718 struct ofpbuf *packet)
4720 struct rule_dpif *rule = rule_dpif_cast(rule_);
4721 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4723 struct dpif_flow_stats stats;
4725 struct action_xlate_ctx ctx;
4726 uint64_t odp_actions_stub[1024 / 8];
4727 struct ofpbuf odp_actions;
4729 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
4730 rule_credit_stats(rule, &stats);
4732 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4733 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
4734 rule, stats.tcp_flags, packet);
4735 ctx.resubmit_stats = &stats;
4736 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
4738 execute_odp_actions(ofproto, flow, odp_actions.data,
4739 odp_actions.size, packet);
4741 ofpbuf_uninit(&odp_actions);
4747 rule_modify_actions(struct rule *rule_)
4749 struct rule_dpif *rule = rule_dpif_cast(rule_);
4751 complete_operation(rule);
4754 /* Sends 'packet' out 'ofport'.
4755 * May modify 'packet'.
4756 * Returns 0 if successful, otherwise a positive errno value. */
4758 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4760 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4761 struct ofpbuf key, odp_actions;
4762 struct odputil_keybuf keybuf;
4767 flow_extract(packet, 0, NULL, 0, &flow);
4768 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4770 if (odp_port != ofport->odp_port) {
4771 eth_pop_vlan(packet);
4772 flow.vlan_tci = htons(0);
4775 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4776 odp_flow_key_from_flow(&key, &flow);
4778 ofpbuf_init(&odp_actions, 32);
4779 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4781 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4782 error = dpif_execute(ofproto->dpif,
4784 odp_actions.data, odp_actions.size,
4786 ofpbuf_uninit(&odp_actions);
4789 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4790 ofproto->up.name, odp_port, strerror(error));
4792 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
4796 /* OpenFlow to datapath action translation. */
4798 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
4799 struct action_xlate_ctx *);
4800 static void xlate_normal(struct action_xlate_ctx *);
4802 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
4803 * The action will state 'slow' as the reason that the action is in the slow
4804 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
4805 * dump-flows" output to see why a flow is in the slow path.)
4807 * The 'stub_size' bytes in 'stub' will be used to store the action.
4808 * 'stub_size' must be large enough for the action.
4810 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
4813 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
4814 enum slow_path_reason slow,
4815 uint64_t *stub, size_t stub_size,
4816 const struct nlattr **actionsp, size_t *actions_lenp)
4818 union user_action_cookie cookie;
4821 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
4822 cookie.slow_path.unused = 0;
4823 cookie.slow_path.reason = slow;
4825 ofpbuf_use_stack(&buf, stub, stub_size);
4826 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
4827 uint32_t pid = dpif_port_get_pid(ofproto->dpif, UINT16_MAX);
4828 odp_put_userspace_action(pid, &cookie, &buf);
4830 put_userspace_action(ofproto, &buf, flow, &cookie);
4832 *actionsp = buf.data;
4833 *actions_lenp = buf.size;
4837 put_userspace_action(const struct ofproto_dpif *ofproto,
4838 struct ofpbuf *odp_actions,
4839 const struct flow *flow,
4840 const union user_action_cookie *cookie)
4844 pid = dpif_port_get_pid(ofproto->dpif,
4845 ofp_port_to_odp_port(flow->in_port));
4847 return odp_put_userspace_action(pid, cookie, odp_actions);
4851 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
4852 ovs_be16 vlan_tci, uint32_t odp_port,
4853 unsigned int n_outputs, union user_action_cookie *cookie)
4857 cookie->type = USER_ACTION_COOKIE_SFLOW;
4858 cookie->sflow.vlan_tci = vlan_tci;
4860 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
4861 * port information") for the interpretation of cookie->output. */
4862 switch (n_outputs) {
4864 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
4865 cookie->sflow.output = 0x40000000 | 256;
4869 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4871 cookie->sflow.output = ifindex;
4876 /* 0x80000000 means "multiple output ports. */
4877 cookie->sflow.output = 0x80000000 | n_outputs;
4882 /* Compose SAMPLE action for sFlow. */
4884 compose_sflow_action(const struct ofproto_dpif *ofproto,
4885 struct ofpbuf *odp_actions,
4886 const struct flow *flow,
4889 uint32_t probability;
4890 union user_action_cookie cookie;
4891 size_t sample_offset, actions_offset;
4894 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4898 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4900 /* Number of packets out of UINT_MAX to sample. */
4901 probability = dpif_sflow_get_probability(ofproto->sflow);
4902 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4904 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4905 compose_sflow_cookie(ofproto, htons(0), odp_port,
4906 odp_port == OVSP_NONE ? 0 : 1, &cookie);
4907 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4909 nl_msg_end_nested(odp_actions, actions_offset);
4910 nl_msg_end_nested(odp_actions, sample_offset);
4911 return cookie_offset;
4914 /* SAMPLE action must be first action in any given list of actions.
4915 * At this point we do not have all information required to build it. So try to
4916 * build sample action as complete as possible. */
4918 add_sflow_action(struct action_xlate_ctx *ctx)
4920 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4922 &ctx->flow, OVSP_NONE);
4923 ctx->sflow_odp_port = 0;
4924 ctx->sflow_n_outputs = 0;
4927 /* Fix SAMPLE action according to data collected while composing ODP actions.
4928 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4929 * USERSPACE action's user-cookie which is required for sflow. */
4931 fix_sflow_action(struct action_xlate_ctx *ctx)
4933 const struct flow *base = &ctx->base_flow;
4934 union user_action_cookie *cookie;
4936 if (!ctx->user_cookie_offset) {
4940 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4942 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4944 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
4945 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
4949 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4952 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4953 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4954 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4955 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4959 struct priority_to_dscp *pdscp;
4961 if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
4962 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
4964 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
4965 xlate_report(ctx, "STP not in forwarding state, skipping output");
4969 pdscp = get_priority(ofport, ctx->flow.skb_priority);
4971 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4972 ctx->flow.nw_tos |= pdscp->dscp;
4975 /* We may not have an ofport record for this port, but it doesn't hurt
4976 * to allow forwarding to it anyhow. Maybe such a port will appear
4977 * later and we're pre-populating the flow table. */
4980 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4981 ctx->flow.vlan_tci);
4982 if (out_port != odp_port) {
4983 ctx->flow.vlan_tci = htons(0);
4985 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4986 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4988 ctx->sflow_odp_port = odp_port;
4989 ctx->sflow_n_outputs++;
4990 ctx->nf_output_iface = ofp_port;
4991 ctx->flow.vlan_tci = flow_vlan_tci;
4992 ctx->flow.nw_tos = flow_nw_tos;
4996 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4998 compose_output_action__(ctx, ofp_port, true);
5002 xlate_table_action(struct action_xlate_ctx *ctx,
5003 uint16_t in_port, uint8_t table_id, bool may_packet_in)
5005 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
5006 struct ofproto_dpif *ofproto = ctx->ofproto;
5007 struct rule_dpif *rule;
5008 uint16_t old_in_port;
5009 uint8_t old_table_id;
5011 old_table_id = ctx->table_id;
5012 ctx->table_id = table_id;
5014 /* Look up a flow with 'in_port' as the input port. */
5015 old_in_port = ctx->flow.in_port;
5016 ctx->flow.in_port = in_port;
5017 rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id);
5020 if (table_id > 0 && table_id < N_TABLES) {
5021 struct table_dpif *table = &ofproto->tables[table_id];
5022 if (table->other_table) {
5023 ctx->tags |= (rule && rule->tag
5025 : rule_calculate_tag(&ctx->flow,
5026 &table->other_table->mask,
5031 /* Restore the original input port. Otherwise OFPP_NORMAL and
5032 * OFPP_IN_PORT will have surprising behavior. */
5033 ctx->flow.in_port = old_in_port;
5035 if (ctx->resubmit_hook) {
5036 ctx->resubmit_hook(ctx, rule);
5039 if (rule == NULL && may_packet_in) {
5041 * check if table configuration flags
5042 * OFPTC_TABLE_MISS_CONTROLLER, default.
5043 * OFPTC_TABLE_MISS_CONTINUE,
5044 * OFPTC_TABLE_MISS_DROP
5045 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
5047 rule = rule_dpif_miss_rule(ofproto, &ctx->flow);
5051 struct rule_dpif *old_rule = ctx->rule;
5053 if (ctx->resubmit_stats) {
5054 rule_credit_stats(rule, ctx->resubmit_stats);
5059 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
5060 ctx->rule = old_rule;
5064 ctx->table_id = old_table_id;
5066 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5068 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
5069 MAX_RESUBMIT_RECURSION);
5070 ctx->max_resubmit_trigger = true;
5075 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
5076 const struct ofpact_resubmit *resubmit)
5081 in_port = resubmit->in_port;
5082 if (in_port == OFPP_IN_PORT) {
5083 in_port = ctx->flow.in_port;
5086 table_id = resubmit->table_id;
5087 if (table_id == 255) {
5088 table_id = ctx->table_id;
5091 xlate_table_action(ctx, in_port, table_id, false);
5095 flood_packets(struct action_xlate_ctx *ctx, bool all)
5097 struct ofport_dpif *ofport;
5099 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
5100 uint16_t ofp_port = ofport->up.ofp_port;
5102 if (ofp_port == ctx->flow.in_port) {
5107 compose_output_action__(ctx, ofp_port, false);
5108 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
5109 compose_output_action(ctx, ofp_port);
5113 ctx->nf_output_iface = NF_OUT_FLOOD;
5117 execute_controller_action(struct action_xlate_ctx *ctx, int len,
5118 enum ofp_packet_in_reason reason,
5119 uint16_t controller_id)
5121 struct ofputil_packet_in pin;
5122 struct ofpbuf *packet;
5124 ctx->slow |= SLOW_CONTROLLER;
5129 packet = ofpbuf_clone(ctx->packet);
5131 if (packet->l2 && packet->l3) {
5132 struct eth_header *eh;
5134 eth_pop_vlan(packet);
5137 /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
5138 * LLC frame. Calculating the Ethernet type of these frames is more
5139 * trouble than seems appropriate for a simple assertion. */
5140 assert(ntohs(eh->eth_type) < ETH_TYPE_MIN
5141 || eh->eth_type == ctx->flow.dl_type);
5143 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
5144 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
5146 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
5147 eth_push_vlan(packet, ctx->flow.vlan_tci);
5151 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5152 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
5153 ctx->flow.nw_tos, ctx->flow.nw_ttl);
5157 if (ctx->flow.nw_proto == IPPROTO_TCP) {
5158 packet_set_tcp_port(packet, ctx->flow.tp_src,
5160 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
5161 packet_set_udp_port(packet, ctx->flow.tp_src,
5168 pin.packet = packet->data;
5169 pin.packet_len = packet->size;
5170 pin.reason = reason;
5171 pin.controller_id = controller_id;
5172 pin.table_id = ctx->table_id;
5173 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
5176 flow_get_metadata(&ctx->flow, &pin.fmd);
5178 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
5179 ofpbuf_delete(packet);
5183 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
5185 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
5186 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
5190 if (ctx->flow.nw_ttl > 1) {
5196 for (i = 0; i < ids->n_controllers; i++) {
5197 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
5201 /* Stop processing for current table. */
5207 xlate_output_action(struct action_xlate_ctx *ctx,
5208 uint16_t port, uint16_t max_len, bool may_packet_in)
5210 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
5212 ctx->nf_output_iface = NF_OUT_DROP;
5216 compose_output_action(ctx, ctx->flow.in_port);
5219 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
5225 flood_packets(ctx, false);
5228 flood_packets(ctx, true);
5230 case OFPP_CONTROLLER:
5231 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
5237 if (port != ctx->flow.in_port) {
5238 compose_output_action(ctx, port);
5240 xlate_report(ctx, "skipping output to input port");
5245 if (prev_nf_output_iface == NF_OUT_FLOOD) {
5246 ctx->nf_output_iface = NF_OUT_FLOOD;
5247 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
5248 ctx->nf_output_iface = prev_nf_output_iface;
5249 } else if (prev_nf_output_iface != NF_OUT_DROP &&
5250 ctx->nf_output_iface != NF_OUT_FLOOD) {
5251 ctx->nf_output_iface = NF_OUT_MULTI;
5256 xlate_output_reg_action(struct action_xlate_ctx *ctx,
5257 const struct ofpact_output_reg *or)
5259 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
5260 if (port <= UINT16_MAX) {
5261 xlate_output_action(ctx, port, or->max_len, false);
5266 xlate_enqueue_action(struct action_xlate_ctx *ctx,
5267 const struct ofpact_enqueue *enqueue)
5269 uint16_t ofp_port = enqueue->port;
5270 uint32_t queue_id = enqueue->queue;
5271 uint32_t flow_priority, priority;
5274 /* Translate queue to priority. */
5275 error = dpif_queue_to_priority(ctx->ofproto->dpif, queue_id, &priority);
5277 /* Fall back to ordinary output action. */
5278 xlate_output_action(ctx, enqueue->port, 0, false);
5282 /* Check output port. */
5283 if (ofp_port == OFPP_IN_PORT) {
5284 ofp_port = ctx->flow.in_port;
5285 } else if (ofp_port == ctx->flow.in_port) {
5289 /* Add datapath actions. */
5290 flow_priority = ctx->flow.skb_priority;
5291 ctx->flow.skb_priority = priority;
5292 compose_output_action(ctx, ofp_port);
5293 ctx->flow.skb_priority = flow_priority;
5295 /* Update NetFlow output port. */
5296 if (ctx->nf_output_iface == NF_OUT_DROP) {
5297 ctx->nf_output_iface = ofp_port;
5298 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
5299 ctx->nf_output_iface = NF_OUT_MULTI;
5304 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
5306 uint32_t skb_priority;
5308 if (!dpif_queue_to_priority(ctx->ofproto->dpif, queue_id, &skb_priority)) {
5309 ctx->flow.skb_priority = skb_priority;
5311 /* Couldn't translate queue to a priority. Nothing to do. A warning
5312 * has already been logged. */
5316 struct xlate_reg_state {
5322 xlate_autopath(struct action_xlate_ctx *ctx,
5323 const struct ofpact_autopath *ap)
5325 uint16_t ofp_port = ap->port;
5326 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
5328 if (!port || !port->bundle) {
5329 ofp_port = OFPP_NONE;
5330 } else if (port->bundle->bond) {
5331 /* Autopath does not support VLAN hashing. */
5332 struct ofport_dpif *slave = bond_choose_output_slave(
5333 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
5335 ofp_port = slave->up.ofp_port;
5338 nxm_reg_load(&ap->dst, ofp_port, &ctx->flow);
5342 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
5344 struct ofproto_dpif *ofproto = ofproto_;
5345 struct ofport_dpif *port;
5355 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
5358 port = get_ofp_port(ofproto, ofp_port);
5359 return port ? port->may_enable : false;
5364 xlate_bundle_action(struct action_xlate_ctx *ctx,
5365 const struct ofpact_bundle *bundle)
5369 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
5370 if (bundle->dst.field) {
5371 nxm_reg_load(&bundle->dst, port, &ctx->flow);
5373 xlate_output_action(ctx, port, 0, false);
5378 xlate_learn_action(struct action_xlate_ctx *ctx,
5379 const struct ofpact_learn *learn)
5381 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
5382 struct ofputil_flow_mod fm;
5383 uint64_t ofpacts_stub[1024 / 8];
5384 struct ofpbuf ofpacts;
5387 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
5388 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
5390 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
5391 if (error && !VLOG_DROP_WARN(&rl)) {
5392 VLOG_WARN("learning action failed to modify flow table (%s)",
5393 ofperr_get_name(error));
5396 ofpbuf_uninit(&ofpacts);
5399 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
5400 * means "infinite". */
5402 reduce_timeout(uint16_t max, uint16_t *timeout)
5404 if (max && (!*timeout || *timeout > max)) {
5410 xlate_fin_timeout(struct action_xlate_ctx *ctx,
5411 const struct ofpact_fin_timeout *oft)
5413 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
5414 struct rule_dpif *rule = ctx->rule;
5416 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
5417 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
5422 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
5424 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
5425 ? OFPUTIL_PC_NO_RECV_STP
5426 : OFPUTIL_PC_NO_RECV)) {
5430 /* Only drop packets here if both forwarding and learning are
5431 * disabled. If just learning is enabled, we need to have
5432 * OFPP_NORMAL and the learning action have a look at the packet
5433 * before we can drop it. */
5434 if (!stp_forward_in_state(port->stp_state)
5435 && !stp_learn_in_state(port->stp_state)) {
5443 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
5444 struct action_xlate_ctx *ctx)
5446 const struct ofport_dpif *port;
5447 bool was_evictable = true;
5448 const struct ofpact *a;
5450 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5451 if (port && !may_receive(port, ctx)) {
5452 /* Drop this flow. */
5457 /* Don't let the rule we're working on get evicted underneath us. */
5458 was_evictable = ctx->rule->up.evictable;
5459 ctx->rule->up.evictable = false;
5461 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
5462 struct ofpact_controller *controller;
5463 const struct ofpact_metadata *metadata;
5471 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
5472 ofpact_get_OUTPUT(a)->max_len, true);
5475 case OFPACT_CONTROLLER:
5476 controller = ofpact_get_CONTROLLER(a);
5477 execute_controller_action(ctx, controller->max_len,
5479 controller->controller_id);
5482 case OFPACT_ENQUEUE:
5483 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
5486 case OFPACT_SET_VLAN_VID:
5487 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
5488 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
5492 case OFPACT_SET_VLAN_PCP:
5493 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
5494 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
5499 case OFPACT_STRIP_VLAN:
5500 ctx->flow.vlan_tci = htons(0);
5503 case OFPACT_PUSH_VLAN:
5504 /* TODO:XXX 802.1AD(QinQ) */
5505 ctx->flow.vlan_tci = htons(VLAN_CFI);
5508 case OFPACT_SET_ETH_SRC:
5509 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
5513 case OFPACT_SET_ETH_DST:
5514 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
5518 case OFPACT_SET_IPV4_SRC:
5519 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
5522 case OFPACT_SET_IPV4_DST:
5523 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
5526 case OFPACT_SET_IPV4_DSCP:
5527 /* OpenFlow 1.0 only supports IPv4. */
5528 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5529 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5530 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
5534 case OFPACT_SET_L4_SRC_PORT:
5535 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
5538 case OFPACT_SET_L4_DST_PORT:
5539 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
5542 case OFPACT_RESUBMIT:
5543 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
5546 case OFPACT_SET_TUNNEL:
5547 ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
5550 case OFPACT_SET_QUEUE:
5551 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
5554 case OFPACT_POP_QUEUE:
5555 ctx->flow.skb_priority = ctx->orig_skb_priority;
5558 case OFPACT_REG_MOVE:
5559 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
5562 case OFPACT_REG_LOAD:
5563 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
5566 case OFPACT_DEC_TTL:
5567 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
5573 /* Nothing to do. */
5576 case OFPACT_MULTIPATH:
5577 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
5580 case OFPACT_AUTOPATH:
5581 xlate_autopath(ctx, ofpact_get_AUTOPATH(a));
5585 ctx->ofproto->has_bundle_action = true;
5586 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
5589 case OFPACT_OUTPUT_REG:
5590 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
5594 ctx->has_learn = true;
5595 if (ctx->may_learn) {
5596 xlate_learn_action(ctx, ofpact_get_LEARN(a));
5604 case OFPACT_FIN_TIMEOUT:
5605 ctx->has_fin_timeout = true;
5606 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
5609 case OFPACT_CLEAR_ACTIONS:
5611 * Nothing to do because writa-actions is not supported for now.
5612 * When writa-actions is supported, clear-actions also must
5613 * be supported at the same time.
5617 case OFPACT_WRITE_METADATA:
5618 metadata = ofpact_get_WRITE_METADATA(a);
5619 ctx->flow.metadata &= ~metadata->mask;
5620 ctx->flow.metadata |= metadata->metadata & metadata->mask;
5623 case OFPACT_GOTO_TABLE: {
5624 /* TODO:XXX remove recursion */
5625 /* It is assumed that goto-table is last action */
5626 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
5627 assert(ctx->table_id < ogt->table_id);
5628 xlate_table_action(ctx, ctx->flow.in_port, ogt->table_id, true);
5635 /* We've let OFPP_NORMAL and the learning action look at the packet,
5636 * so drop it now if forwarding is disabled. */
5637 if (port && !stp_forward_in_state(port->stp_state)) {
5638 ofpbuf_clear(ctx->odp_actions);
5639 add_sflow_action(ctx);
5642 ctx->rule->up.evictable = was_evictable;
5647 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
5648 struct ofproto_dpif *ofproto, const struct flow *flow,
5649 ovs_be16 initial_tci, struct rule_dpif *rule,
5650 uint8_t tcp_flags, const struct ofpbuf *packet)
5652 ctx->ofproto = ofproto;
5654 ctx->base_flow = ctx->flow;
5655 memset(&ctx->base_flow.tunnel, 0, sizeof ctx->base_flow.tunnel);
5656 ctx->base_flow.vlan_tci = initial_tci;
5658 ctx->packet = packet;
5659 ctx->may_learn = packet != NULL;
5660 ctx->tcp_flags = tcp_flags;
5661 ctx->resubmit_hook = NULL;
5662 ctx->report_hook = NULL;
5663 ctx->resubmit_stats = NULL;
5666 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
5667 * into datapath actions in 'odp_actions', using 'ctx'. */
5669 xlate_actions(struct action_xlate_ctx *ctx,
5670 const struct ofpact *ofpacts, size_t ofpacts_len,
5671 struct ofpbuf *odp_actions)
5673 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
5674 * that in the future we always keep a copy of the original flow for
5675 * tracing purposes. */
5676 static bool hit_resubmit_limit;
5678 enum slow_path_reason special;
5680 COVERAGE_INC(ofproto_dpif_xlate);
5682 ofpbuf_clear(odp_actions);
5683 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
5685 ctx->odp_actions = odp_actions;
5688 ctx->has_learn = false;
5689 ctx->has_normal = false;
5690 ctx->has_fin_timeout = false;
5691 ctx->nf_output_iface = NF_OUT_DROP;
5694 ctx->max_resubmit_trigger = false;
5695 ctx->orig_skb_priority = ctx->flow.skb_priority;
5699 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
5700 /* Do this conditionally because the copy is expensive enough that it
5701 * shows up in profiles.
5703 * We keep orig_flow in 'ctx' only because I couldn't make GCC 4.4
5704 * believe that I wasn't using it without initializing it if I kept it
5705 * in a local variable. */
5706 ctx->orig_flow = ctx->flow;
5709 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
5710 switch (ctx->ofproto->up.frag_handling) {
5711 case OFPC_FRAG_NORMAL:
5712 /* We must pretend that transport ports are unavailable. */
5713 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
5714 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
5717 case OFPC_FRAG_DROP:
5720 case OFPC_FRAG_REASM:
5723 case OFPC_FRAG_NX_MATCH:
5724 /* Nothing to do. */
5727 case OFPC_INVALID_TTL_TO_CONTROLLER:
5732 special = process_special(ctx->ofproto, &ctx->flow, ctx->packet);
5734 ctx->slow |= special;
5736 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5737 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
5739 add_sflow_action(ctx);
5740 do_xlate_actions(ofpacts, ofpacts_len, ctx);
5742 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
5743 if (!hit_resubmit_limit) {
5744 /* We didn't record the original flow. Make sure we do from
5746 hit_resubmit_limit = true;
5747 } else if (!VLOG_DROP_ERR(&trace_rl)) {
5748 struct ds ds = DS_EMPTY_INITIALIZER;
5750 ofproto_trace(ctx->ofproto, &ctx->orig_flow, ctx->packet,
5752 VLOG_ERR("Trace triggered by excessive resubmit "
5753 "recursion:\n%s", ds_cstr(&ds));
5758 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
5759 ctx->odp_actions->data,
5760 ctx->odp_actions->size)) {
5761 ctx->slow |= SLOW_IN_BAND;
5763 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
5765 compose_output_action(ctx, OFPP_LOCAL);
5768 if (ctx->ofproto->has_mirrors) {
5769 add_mirror_actions(ctx, &ctx->orig_flow);
5771 fix_sflow_action(ctx);
5775 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
5776 * into datapath actions, using 'ctx', and discards the datapath actions. */
5778 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
5779 const struct ofpact *ofpacts,
5782 uint64_t odp_actions_stub[1024 / 8];
5783 struct ofpbuf odp_actions;
5785 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5786 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
5787 ofpbuf_uninit(&odp_actions);
5791 xlate_report(struct action_xlate_ctx *ctx, const char *s)
5793 if (ctx->report_hook) {
5794 ctx->report_hook(ctx, s);
5798 /* OFPP_NORMAL implementation. */
5800 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
5802 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
5803 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
5804 * the bundle on which the packet was received, returns the VLAN to which the
5807 * Both 'vid' and the return value are in the range 0...4095. */
5809 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
5811 switch (in_bundle->vlan_mode) {
5812 case PORT_VLAN_ACCESS:
5813 return in_bundle->vlan;
5816 case PORT_VLAN_TRUNK:
5819 case PORT_VLAN_NATIVE_UNTAGGED:
5820 case PORT_VLAN_NATIVE_TAGGED:
5821 return vid ? vid : in_bundle->vlan;
5828 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
5829 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
5832 * 'vid' should be the VID obtained from the 802.1Q header that was received as
5833 * part of a packet (specify 0 if there was no 802.1Q header), in the range
5836 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
5838 /* Allow any VID on the OFPP_NONE port. */
5839 if (in_bundle == &ofpp_none_bundle) {
5843 switch (in_bundle->vlan_mode) {
5844 case PORT_VLAN_ACCESS:
5847 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5848 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
5849 "packet received on port %s configured as VLAN "
5850 "%"PRIu16" access port",
5851 in_bundle->ofproto->up.name, vid,
5852 in_bundle->name, in_bundle->vlan);
5858 case PORT_VLAN_NATIVE_UNTAGGED:
5859 case PORT_VLAN_NATIVE_TAGGED:
5861 /* Port must always carry its native VLAN. */
5865 case PORT_VLAN_TRUNK:
5866 if (!ofbundle_includes_vlan(in_bundle, vid)) {
5868 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5869 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
5870 "received on port %s not configured for trunking "
5872 in_bundle->ofproto->up.name, vid,
5873 in_bundle->name, vid);
5885 /* Given 'vlan', the VLAN that a packet belongs to, and
5886 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
5887 * that should be included in the 802.1Q header. (If the return value is 0,
5888 * then the 802.1Q header should only be included in the packet if there is a
5891 * Both 'vlan' and the return value are in the range 0...4095. */
5893 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
5895 switch (out_bundle->vlan_mode) {
5896 case PORT_VLAN_ACCESS:
5899 case PORT_VLAN_TRUNK:
5900 case PORT_VLAN_NATIVE_TAGGED:
5903 case PORT_VLAN_NATIVE_UNTAGGED:
5904 return vlan == out_bundle->vlan ? 0 : vlan;
5912 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
5915 struct ofport_dpif *port;
5917 ovs_be16 tci, old_tci;
5919 vid = output_vlan_to_vid(out_bundle, vlan);
5920 if (!out_bundle->bond) {
5921 port = ofbundle_get_a_port(out_bundle);
5923 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
5926 /* No slaves enabled, so drop packet. */
5931 old_tci = ctx->flow.vlan_tci;
5933 if (tci || out_bundle->use_priority_tags) {
5934 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
5936 tci |= htons(VLAN_CFI);
5939 ctx->flow.vlan_tci = tci;
5941 compose_output_action(ctx, port->up.ofp_port);
5942 ctx->flow.vlan_tci = old_tci;
5946 mirror_mask_ffs(mirror_mask_t mask)
5948 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
5953 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
5955 return (bundle->vlan_mode != PORT_VLAN_ACCESS
5956 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
5960 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
5962 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
5965 /* Returns an arbitrary interface within 'bundle'. */
5966 static struct ofport_dpif *
5967 ofbundle_get_a_port(const struct ofbundle *bundle)
5969 return CONTAINER_OF(list_front(&bundle->ports),
5970 struct ofport_dpif, bundle_node);
5974 vlan_is_mirrored(const struct ofmirror *m, int vlan)
5976 return !m->vlans || bitmap_is_set(m->vlans, vlan);
5980 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5982 struct ofproto_dpif *ofproto = ctx->ofproto;
5983 mirror_mask_t mirrors;
5984 struct ofbundle *in_bundle;
5987 const struct nlattr *a;
5990 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5991 ctx->packet != NULL, NULL);
5995 mirrors = in_bundle->src_mirrors;
5997 /* Drop frames on bundles reserved for mirroring. */
5998 if (in_bundle->mirror_out) {
5999 if (ctx->packet != NULL) {
6000 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6001 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6002 "%s, which is reserved exclusively for mirroring",
6003 ctx->ofproto->up.name, in_bundle->name);
6009 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
6010 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6013 vlan = input_vid_to_vlan(in_bundle, vid);
6015 /* Look at the output ports to check for destination selections. */
6017 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
6018 ctx->odp_actions->size) {
6019 enum ovs_action_attr type = nl_attr_type(a);
6020 struct ofport_dpif *ofport;
6022 if (type != OVS_ACTION_ATTR_OUTPUT) {
6026 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
6027 if (ofport && ofport->bundle) {
6028 mirrors |= ofport->bundle->dst_mirrors;
6036 /* Restore the original packet before adding the mirror actions. */
6037 ctx->flow = *orig_flow;
6042 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6044 if (!vlan_is_mirrored(m, vlan)) {
6045 mirrors = zero_rightmost_1bit(mirrors);
6049 mirrors &= ~m->dup_mirrors;
6050 ctx->mirrors |= m->dup_mirrors;
6052 output_normal(ctx, m->out, vlan);
6053 } else if (vlan != m->out_vlan
6054 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
6055 struct ofbundle *bundle;
6057 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
6058 if (ofbundle_includes_vlan(bundle, m->out_vlan)
6059 && !bundle->mirror_out) {
6060 output_normal(ctx, bundle, m->out_vlan);
6068 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
6069 uint64_t packets, uint64_t bytes)
6075 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
6078 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6081 /* In normal circumstances 'm' will not be NULL. However,
6082 * if mirrors are reconfigured, we can temporarily get out
6083 * of sync in facet_revalidate(). We could "correct" the
6084 * mirror list before reaching here, but doing that would
6085 * not properly account the traffic stats we've currently
6086 * accumulated for previous mirror configuration. */
6090 m->packet_count += packets;
6091 m->byte_count += bytes;
6095 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
6096 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
6097 * indicate this; newer upstream kernels use gratuitous ARP requests. */
6099 is_gratuitous_arp(const struct flow *flow)
6101 return (flow->dl_type == htons(ETH_TYPE_ARP)
6102 && eth_addr_is_broadcast(flow->dl_dst)
6103 && (flow->nw_proto == ARP_OP_REPLY
6104 || (flow->nw_proto == ARP_OP_REQUEST
6105 && flow->nw_src == flow->nw_dst)));
6109 update_learning_table(struct ofproto_dpif *ofproto,
6110 const struct flow *flow, int vlan,
6111 struct ofbundle *in_bundle)
6113 struct mac_entry *mac;
6115 /* Don't learn the OFPP_NONE port. */
6116 if (in_bundle == &ofpp_none_bundle) {
6120 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
6124 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
6125 if (is_gratuitous_arp(flow)) {
6126 /* We don't want to learn from gratuitous ARP packets that are
6127 * reflected back over bond slaves so we lock the learning table. */
6128 if (!in_bundle->bond) {
6129 mac_entry_set_grat_arp_lock(mac);
6130 } else if (mac_entry_is_grat_arp_locked(mac)) {
6135 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
6136 /* The log messages here could actually be useful in debugging,
6137 * so keep the rate limit relatively high. */
6138 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
6139 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
6140 "on port %s in VLAN %d",
6141 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
6142 in_bundle->name, vlan);
6144 mac->port.p = in_bundle;
6145 tag_set_add(&ofproto->revalidate_set,
6146 mac_learning_changed(ofproto->ml, mac));
6150 static struct ofbundle *
6151 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
6152 bool warn, struct ofport_dpif **in_ofportp)
6154 struct ofport_dpif *ofport;
6156 /* Find the port and bundle for the received packet. */
6157 ofport = get_ofp_port(ofproto, in_port);
6159 *in_ofportp = ofport;
6161 if (ofport && ofport->bundle) {
6162 return ofport->bundle;
6165 /* Special-case OFPP_NONE, which a controller may use as the ingress
6166 * port for traffic that it is sourcing. */
6167 if (in_port == OFPP_NONE) {
6168 return &ofpp_none_bundle;
6171 /* Odd. A few possible reasons here:
6173 * - We deleted a port but there are still a few packets queued up
6176 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
6177 * we don't know about.
6179 * - The ofproto client didn't configure the port as part of a bundle.
6180 * This is particularly likely to happen if a packet was received on the
6181 * port after it was created, but before the client had a chance to
6182 * configure its bundle.
6185 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6187 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
6188 "port %"PRIu16, ofproto->up.name, in_port);
6193 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
6194 * dropped. Returns true if they may be forwarded, false if they should be
6197 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
6198 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
6200 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
6201 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
6202 * checked by input_vid_is_valid().
6204 * May also add tags to '*tags', although the current implementation only does
6205 * so in one special case.
6208 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
6211 struct ofproto_dpif *ofproto = ctx->ofproto;
6212 struct flow *flow = &ctx->flow;
6213 struct ofbundle *in_bundle = in_port->bundle;
6215 /* Drop frames for reserved multicast addresses
6216 * only if forward_bpdu option is absent. */
6217 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
6218 xlate_report(ctx, "packet has reserved destination MAC, dropping");
6222 if (in_bundle->bond) {
6223 struct mac_entry *mac;
6225 switch (bond_check_admissibility(in_bundle->bond, in_port,
6226 flow->dl_dst, &ctx->tags)) {
6231 xlate_report(ctx, "bonding refused admissibility, dropping");
6234 case BV_DROP_IF_MOVED:
6235 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
6236 if (mac && mac->port.p != in_bundle &&
6237 (!is_gratuitous_arp(flow)
6238 || mac_entry_is_grat_arp_locked(mac))) {
6239 xlate_report(ctx, "SLB bond thinks this packet looped back, "
6251 xlate_normal(struct action_xlate_ctx *ctx)
6253 struct ofport_dpif *in_port;
6254 struct ofbundle *in_bundle;
6255 struct mac_entry *mac;
6259 ctx->has_normal = true;
6261 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
6262 ctx->packet != NULL, &in_port);
6264 xlate_report(ctx, "no input bundle, dropping");
6268 /* Drop malformed frames. */
6269 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
6270 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
6271 if (ctx->packet != NULL) {
6272 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6273 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
6274 "VLAN tag received on port %s",
6275 ctx->ofproto->up.name, in_bundle->name);
6277 xlate_report(ctx, "partial VLAN tag, dropping");
6281 /* Drop frames on bundles reserved for mirroring. */
6282 if (in_bundle->mirror_out) {
6283 if (ctx->packet != NULL) {
6284 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6285 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6286 "%s, which is reserved exclusively for mirroring",
6287 ctx->ofproto->up.name, in_bundle->name);
6289 xlate_report(ctx, "input port is mirror output port, dropping");
6294 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
6295 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6296 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
6299 vlan = input_vid_to_vlan(in_bundle, vid);
6301 /* Check other admissibility requirements. */
6302 if (in_port && !is_admissible(ctx, in_port, vlan)) {
6306 /* Learn source MAC. */
6307 if (ctx->may_learn) {
6308 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
6311 /* Determine output bundle. */
6312 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
6315 if (mac->port.p != in_bundle) {
6316 xlate_report(ctx, "forwarding to learned port");
6317 output_normal(ctx, mac->port.p, vlan);
6319 xlate_report(ctx, "learned port is input port, dropping");
6322 struct ofbundle *bundle;
6324 xlate_report(ctx, "no learned MAC for destination, flooding");
6325 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
6326 if (bundle != in_bundle
6327 && ofbundle_includes_vlan(bundle, vlan)
6328 && bundle->floodable
6329 && !bundle->mirror_out) {
6330 output_normal(ctx, bundle, vlan);
6333 ctx->nf_output_iface = NF_OUT_FLOOD;
6337 /* Optimized flow revalidation.
6339 * It's a difficult problem, in general, to tell which facets need to have
6340 * their actions recalculated whenever the OpenFlow flow table changes. We
6341 * don't try to solve that general problem: for most kinds of OpenFlow flow
6342 * table changes, we recalculate the actions for every facet. This is
6343 * relatively expensive, but it's good enough if the OpenFlow flow table
6344 * doesn't change very often.
6346 * However, we can expect one particular kind of OpenFlow flow table change to
6347 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
6348 * of CPU on revalidating every facet whenever MAC learning modifies the flow
6349 * table, we add a special case that applies to flow tables in which every rule
6350 * has the same form (that is, the same wildcards), except that the table is
6351 * also allowed to have a single "catch-all" flow that matches all packets. We
6352 * optimize this case by tagging all of the facets that resubmit into the table
6353 * and invalidating the same tag whenever a flow changes in that table. The
6354 * end result is that we revalidate just the facets that need it (and sometimes
6355 * a few more, but not all of the facets or even all of the facets that
6356 * resubmit to the table modified by MAC learning). */
6358 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
6359 * into an OpenFlow table with the given 'basis'. */
6361 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
6364 if (minimask_is_catchall(mask)) {
6367 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
6368 return tag_create_deterministic(hash);
6372 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
6373 * taggability of that table.
6375 * This function must be called after *each* change to a flow table. If you
6376 * skip calling it on some changes then the pointer comparisons at the end can
6377 * be invalid if you get unlucky. For example, if a flow removal causes a
6378 * cls_table to be destroyed and then a flow insertion causes a cls_table with
6379 * different wildcards to be created with the same address, then this function
6380 * will incorrectly skip revalidation. */
6382 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
6384 struct table_dpif *table = &ofproto->tables[table_id];
6385 const struct oftable *oftable = &ofproto->up.tables[table_id];
6386 struct cls_table *catchall, *other;
6387 struct cls_table *t;
6389 catchall = other = NULL;
6391 switch (hmap_count(&oftable->cls.tables)) {
6393 /* We could tag this OpenFlow table but it would make the logic a
6394 * little harder and it's a corner case that doesn't seem worth it
6400 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
6401 if (cls_table_is_catchall(t)) {
6403 } else if (!other) {
6406 /* Indicate that we can't tag this by setting both tables to
6407 * NULL. (We know that 'catchall' is already NULL.) */
6414 /* Can't tag this table. */
6418 if (table->catchall_table != catchall || table->other_table != other) {
6419 table->catchall_table = catchall;
6420 table->other_table = other;
6421 ofproto->need_revalidate = REV_FLOW_TABLE;
6425 /* Given 'rule' that has changed in some way (either it is a rule being
6426 * inserted, a rule being deleted, or a rule whose actions are being
6427 * modified), marks facets for revalidation to ensure that packets will be
6428 * forwarded correctly according to the new state of the flow table.
6430 * This function must be called after *each* change to a flow table. See
6431 * the comment on table_update_taggable() for more information. */
6433 rule_invalidate(const struct rule_dpif *rule)
6435 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
6437 table_update_taggable(ofproto, rule->up.table_id);
6439 if (!ofproto->need_revalidate) {
6440 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
6442 if (table->other_table && rule->tag) {
6443 tag_set_add(&ofproto->revalidate_set, rule->tag);
6445 ofproto->need_revalidate = REV_FLOW_TABLE;
6451 set_frag_handling(struct ofproto *ofproto_,
6452 enum ofp_config_flags frag_handling)
6454 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6456 if (frag_handling != OFPC_FRAG_REASM) {
6457 ofproto->need_revalidate = REV_RECONFIGURE;
6465 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
6466 const struct flow *flow,
6467 const struct ofpact *ofpacts, size_t ofpacts_len)
6469 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6470 struct odputil_keybuf keybuf;
6471 struct dpif_flow_stats stats;
6475 struct action_xlate_ctx ctx;
6476 uint64_t odp_actions_stub[1024 / 8];
6477 struct ofpbuf odp_actions;
6479 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
6480 odp_flow_key_from_flow(&key, flow);
6482 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
6484 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL,
6485 packet_get_tcp_flags(packet, flow), packet);
6486 ctx.resubmit_stats = &stats;
6488 ofpbuf_use_stub(&odp_actions,
6489 odp_actions_stub, sizeof odp_actions_stub);
6490 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
6491 dpif_execute(ofproto->dpif, key.data, key.size,
6492 odp_actions.data, odp_actions.size, packet);
6493 ofpbuf_uninit(&odp_actions);
6501 set_netflow(struct ofproto *ofproto_,
6502 const struct netflow_options *netflow_options)
6504 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6506 if (netflow_options) {
6507 if (!ofproto->netflow) {
6508 ofproto->netflow = netflow_create();
6510 return netflow_set_options(ofproto->netflow, netflow_options);
6512 netflow_destroy(ofproto->netflow);
6513 ofproto->netflow = NULL;
6519 get_netflow_ids(const struct ofproto *ofproto_,
6520 uint8_t *engine_type, uint8_t *engine_id)
6522 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6524 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
6528 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
6530 if (!facet_is_controller_flow(facet) &&
6531 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
6532 struct subfacet *subfacet;
6533 struct ofexpired expired;
6535 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
6536 if (subfacet->path == SF_FAST_PATH) {
6537 struct dpif_flow_stats stats;
6539 subfacet_reinstall(subfacet, &stats);
6540 subfacet_update_stats(subfacet, &stats);
6544 expired.flow = facet->flow;
6545 expired.packet_count = facet->packet_count;
6546 expired.byte_count = facet->byte_count;
6547 expired.used = facet->used;
6548 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
6553 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
6555 struct facet *facet;
6557 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6558 send_active_timeout(ofproto, facet);
6562 static struct ofproto_dpif *
6563 ofproto_dpif_lookup(const char *name)
6565 struct ofproto_dpif *ofproto;
6567 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
6568 hash_string(name, 0), &all_ofproto_dpifs) {
6569 if (!strcmp(ofproto->up.name, name)) {
6577 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
6578 const char *argv[], void *aux OVS_UNUSED)
6580 struct ofproto_dpif *ofproto;
6583 ofproto = ofproto_dpif_lookup(argv[1]);
6585 unixctl_command_reply_error(conn, "no such bridge");
6588 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6590 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6591 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6595 unixctl_command_reply(conn, "table successfully flushed");
6599 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6600 const char *argv[], void *aux OVS_UNUSED)
6602 struct ds ds = DS_EMPTY_INITIALIZER;
6603 const struct ofproto_dpif *ofproto;
6604 const struct mac_entry *e;
6606 ofproto = ofproto_dpif_lookup(argv[1]);
6608 unixctl_command_reply_error(conn, "no such bridge");
6612 ds_put_cstr(&ds, " port VLAN MAC Age\n");
6613 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
6614 struct ofbundle *bundle = e->port.p;
6615 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
6616 ofbundle_get_a_port(bundle)->odp_port,
6617 e->vlan, ETH_ADDR_ARGS(e->mac),
6618 mac_entry_age(ofproto->ml, e));
6620 unixctl_command_reply(conn, ds_cstr(&ds));
6625 struct action_xlate_ctx ctx;
6631 trace_format_rule(struct ds *result, uint8_t table_id, int level,
6632 const struct rule_dpif *rule)
6634 ds_put_char_multiple(result, '\t', level);
6636 ds_put_cstr(result, "No match\n");
6640 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
6641 table_id, ntohll(rule->up.flow_cookie));
6642 cls_rule_format(&rule->up.cr, result);
6643 ds_put_char(result, '\n');
6645 ds_put_char_multiple(result, '\t', level);
6646 ds_put_cstr(result, "OpenFlow ");
6647 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
6648 ds_put_char(result, '\n');
6652 trace_format_flow(struct ds *result, int level, const char *title,
6653 struct trace_ctx *trace)
6655 ds_put_char_multiple(result, '\t', level);
6656 ds_put_format(result, "%s: ", title);
6657 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
6658 ds_put_cstr(result, "unchanged");
6660 flow_format(result, &trace->ctx.flow);
6661 trace->flow = trace->ctx.flow;
6663 ds_put_char(result, '\n');
6667 trace_format_regs(struct ds *result, int level, const char *title,
6668 struct trace_ctx *trace)
6672 ds_put_char_multiple(result, '\t', level);
6673 ds_put_format(result, "%s:", title);
6674 for (i = 0; i < FLOW_N_REGS; i++) {
6675 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
6677 ds_put_char(result, '\n');
6681 trace_format_odp(struct ds *result, int level, const char *title,
6682 struct trace_ctx *trace)
6684 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
6686 ds_put_char_multiple(result, '\t', level);
6687 ds_put_format(result, "%s: ", title);
6688 format_odp_actions(result, odp_actions->data, odp_actions->size);
6689 ds_put_char(result, '\n');
6693 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
6695 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
6696 struct ds *result = trace->result;
6698 ds_put_char(result, '\n');
6699 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
6700 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
6701 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
6702 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
6706 trace_report(struct action_xlate_ctx *ctx, const char *s)
6708 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
6709 struct ds *result = trace->result;
6711 ds_put_char_multiple(result, '\t', ctx->recurse);
6712 ds_put_cstr(result, s);
6713 ds_put_char(result, '\n');
6717 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
6718 void *aux OVS_UNUSED)
6720 const char *dpname = argv[1];
6721 struct ofproto_dpif *ofproto;
6722 struct ofpbuf odp_key;
6723 struct ofpbuf *packet;
6724 ovs_be16 initial_tci;
6730 ofpbuf_init(&odp_key, 0);
6733 ofproto = ofproto_dpif_lookup(dpname);
6735 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
6739 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
6740 /* ofproto/trace dpname flow [-generate] */
6741 const char *flow_s = argv[2];
6742 const char *generate_s = argv[3];
6744 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
6745 * flow. We guess which type it is based on whether 'flow_s' contains
6746 * an '(', since a datapath flow always contains '(') but an
6747 * OpenFlow-like flow should not (in fact it's allowed but I believe
6748 * that's not documented anywhere).
6750 * An alternative would be to try to parse 'flow_s' both ways, but then
6751 * it would be tricky giving a sensible error message. After all, do
6752 * you just say "syntax error" or do you present both error messages?
6753 * Both choices seem lousy. */
6754 if (strchr(flow_s, '(')) {
6757 /* Convert string to datapath key. */
6758 ofpbuf_init(&odp_key, 0);
6759 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
6761 unixctl_command_reply_error(conn, "Bad flow syntax");
6765 /* Convert odp_key to flow. */
6766 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
6767 odp_key.size, &flow,
6768 &initial_tci, NULL);
6769 if (error == ODP_FIT_ERROR) {
6770 unixctl_command_reply_error(conn, "Invalid flow");
6776 error_s = parse_ofp_exact_flow(&flow, argv[2]);
6778 unixctl_command_reply_error(conn, error_s);
6783 initial_tci = flow.vlan_tci;
6784 vsp_adjust_flow(ofproto, &flow);
6787 /* Generate a packet, if requested. */
6789 packet = ofpbuf_new(0);
6790 flow_compose(packet, &flow);
6792 } else if (argc == 6) {
6793 /* ofproto/trace dpname priority tun_id in_port packet */
6794 const char *priority_s = argv[2];
6795 const char *tun_id_s = argv[3];
6796 const char *in_port_s = argv[4];
6797 const char *packet_s = argv[5];
6798 uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s));
6799 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
6800 uint32_t priority = atoi(priority_s);
6803 msg = eth_from_hex(packet_s, &packet);
6805 unixctl_command_reply_error(conn, msg);
6809 ds_put_cstr(&result, "Packet: ");
6810 s = ofp_packet_to_string(packet->data, packet->size);
6811 ds_put_cstr(&result, s);
6814 flow_extract(packet, priority, NULL, in_port, &flow);
6815 flow.tunnel.tun_id = tun_id;
6816 initial_tci = flow.vlan_tci;
6818 unixctl_command_reply_error(conn, "Bad command syntax");
6822 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
6823 unixctl_command_reply(conn, ds_cstr(&result));
6826 ds_destroy(&result);
6827 ofpbuf_delete(packet);
6828 ofpbuf_uninit(&odp_key);
6832 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
6833 const struct ofpbuf *packet, ovs_be16 initial_tci,
6836 struct rule_dpif *rule;
6838 ds_put_cstr(ds, "Flow: ");
6839 flow_format(ds, flow);
6840 ds_put_char(ds, '\n');
6842 rule = rule_dpif_lookup(ofproto, flow);
6844 trace_format_rule(ds, 0, 0, rule);
6845 if (rule == ofproto->miss_rule) {
6846 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
6847 } else if (rule == ofproto->no_packet_in_rule) {
6848 ds_put_cstr(ds, "\nNo match, packets dropped because "
6849 "OFPPC_NO_PACKET_IN is set on in_port.\n");
6853 uint64_t odp_actions_stub[1024 / 8];
6854 struct ofpbuf odp_actions;
6856 struct trace_ctx trace;
6859 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
6862 ofpbuf_use_stub(&odp_actions,
6863 odp_actions_stub, sizeof odp_actions_stub);
6864 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
6865 rule, tcp_flags, packet);
6866 trace.ctx.resubmit_hook = trace_resubmit;
6867 trace.ctx.report_hook = trace_report;
6868 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
6871 ds_put_char(ds, '\n');
6872 trace_format_flow(ds, 0, "Final flow", &trace);
6873 ds_put_cstr(ds, "Datapath actions: ");
6874 format_odp_actions(ds, odp_actions.data, odp_actions.size);
6875 ofpbuf_uninit(&odp_actions);
6877 if (trace.ctx.slow) {
6878 enum slow_path_reason slow;
6880 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
6881 "slow path because it:");
6882 for (slow = trace.ctx.slow; slow; ) {
6883 enum slow_path_reason bit = rightmost_1bit(slow);
6887 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
6890 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
6893 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
6896 ds_put_cstr(ds, "\n\t- Needs in-band special case "
6899 ds_put_cstr(ds, "\n\t (The datapath actions are "
6900 "incomplete--for complete actions, "
6901 "please supply a packet.)");
6904 case SLOW_CONTROLLER:
6905 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
6906 "to the OpenFlow controller.");
6909 ds_put_cstr(ds, "\n\t- Needs more specific matching "
6910 "than the datapath supports.");
6917 if (slow & ~SLOW_MATCH) {
6918 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
6919 "the special slow-path processing.");
6926 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6927 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6930 unixctl_command_reply(conn, NULL);
6934 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6935 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6938 unixctl_command_reply(conn, NULL);
6941 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
6942 * 'reply' describing the results. */
6944 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
6946 struct facet *facet;
6950 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6951 if (!facet_check_consistency(facet)) {
6956 ofproto->need_revalidate = REV_INCONSISTENCY;
6960 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
6961 ofproto->up.name, errors);
6963 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
6968 ofproto_dpif_self_check(struct unixctl_conn *conn,
6969 int argc, const char *argv[], void *aux OVS_UNUSED)
6971 struct ds reply = DS_EMPTY_INITIALIZER;
6972 struct ofproto_dpif *ofproto;
6975 ofproto = ofproto_dpif_lookup(argv[1]);
6977 unixctl_command_reply_error(conn, "Unknown ofproto (use "
6978 "ofproto/list for help)");
6981 ofproto_dpif_self_check__(ofproto, &reply);
6983 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6984 ofproto_dpif_self_check__(ofproto, &reply);
6988 unixctl_command_reply(conn, ds_cstr(&reply));
6993 ofproto_dpif_unixctl_init(void)
6995 static bool registered;
7001 unixctl_command_register(
7003 "bridge {tun_id in_port packet | odp_flow [-generate]}",
7004 2, 5, ofproto_unixctl_trace, NULL);
7005 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
7006 ofproto_unixctl_fdb_flush, NULL);
7007 unixctl_command_register("fdb/show", "bridge", 1, 1,
7008 ofproto_unixctl_fdb_show, NULL);
7009 unixctl_command_register("ofproto/clog", "", 0, 0,
7010 ofproto_dpif_clog, NULL);
7011 unixctl_command_register("ofproto/unclog", "", 0, 0,
7012 ofproto_dpif_unclog, NULL);
7013 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
7014 ofproto_dpif_self_check, NULL);
7017 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
7019 * This is deprecated. It is only for compatibility with broken device drivers
7020 * in old versions of Linux that do not properly support VLANs when VLAN
7021 * devices are not used. When broken device drivers are no longer in
7022 * widespread use, we will delete these interfaces. */
7025 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
7027 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
7028 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
7030 if (realdev_ofp_port == ofport->realdev_ofp_port
7031 && vid == ofport->vlandev_vid) {
7035 ofproto->need_revalidate = REV_RECONFIGURE;
7037 if (ofport->realdev_ofp_port) {
7040 if (realdev_ofp_port && ofport->bundle) {
7041 /* vlandevs are enslaved to their realdevs, so they are not allowed to
7042 * themselves be part of a bundle. */
7043 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
7046 ofport->realdev_ofp_port = realdev_ofp_port;
7047 ofport->vlandev_vid = vid;
7049 if (realdev_ofp_port) {
7050 vsp_add(ofport, realdev_ofp_port, vid);
7057 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
7059 return hash_2words(realdev_ofp_port, vid);
7062 /* Returns the ODP port number of the Linux VLAN device that corresponds to
7063 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
7064 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
7065 * it would return the port number of eth0.9.
7067 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
7068 * function just returns its 'realdev_odp_port' argument. */
7070 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
7071 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
7073 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
7074 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
7075 int vid = vlan_tci_to_vid(vlan_tci);
7076 const struct vlan_splinter *vsp;
7078 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
7079 hash_realdev_vid(realdev_ofp_port, vid),
7080 &ofproto->realdev_vid_map) {
7081 if (vsp->realdev_ofp_port == realdev_ofp_port
7082 && vsp->vid == vid) {
7083 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
7087 return realdev_odp_port;
7090 static struct vlan_splinter *
7091 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
7093 struct vlan_splinter *vsp;
7095 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
7096 &ofproto->vlandev_map) {
7097 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
7105 /* Returns the OpenFlow port number of the "real" device underlying the Linux
7106 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
7107 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
7108 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
7109 * eth0 and store 9 in '*vid'.
7111 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
7112 * VLAN device. Unless VLAN splinters are enabled, this is what this function
7115 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
7116 uint16_t vlandev_ofp_port, int *vid)
7118 if (!hmap_is_empty(&ofproto->vlandev_map)) {
7119 const struct vlan_splinter *vsp;
7121 vsp = vlandev_find(ofproto, vlandev_ofp_port);
7126 return vsp->realdev_ofp_port;
7132 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
7133 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
7134 * 'flow->in_port' to the "real" device backing the VLAN device, sets
7135 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
7136 * always the case unless VLAN splinters are enabled), returns false without
7137 * making any changes. */
7139 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
7144 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
7149 /* Cause the flow to be processed as if it came in on the real device with
7150 * the VLAN device's VLAN ID. */
7151 flow->in_port = realdev;
7152 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
7157 vsp_remove(struct ofport_dpif *port)
7159 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7160 struct vlan_splinter *vsp;
7162 vsp = vlandev_find(ofproto, port->up.ofp_port);
7164 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
7165 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
7168 port->realdev_ofp_port = 0;
7170 VLOG_ERR("missing vlan device record");
7175 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
7177 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7179 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
7180 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
7181 == realdev_ofp_port)) {
7182 struct vlan_splinter *vsp;
7184 vsp = xmalloc(sizeof *vsp);
7185 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
7186 hash_int(port->up.ofp_port, 0));
7187 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
7188 hash_realdev_vid(realdev_ofp_port, vid));
7189 vsp->realdev_ofp_port = realdev_ofp_port;
7190 vsp->vlandev_ofp_port = port->up.ofp_port;
7193 port->realdev_ofp_port = realdev_ofp_port;
7195 VLOG_ERR("duplicate vlan device record");
7199 const struct ofproto_class ofproto_dpif_class = {
7229 port_is_lacp_current,
7230 NULL, /* rule_choose_table */
7237 rule_modify_actions,
7246 get_cfm_remote_mpids,
7251 get_stp_port_status,
7258 is_mirror_output_bundle,
7259 forward_bpdu_changed,