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-parse.h"
46 #include "ofp-print.h"
47 #include "ofproto-dpif-governor.h"
48 #include "ofproto-dpif-sflow.h"
49 #include "poll-loop.h"
51 #include "unaligned.h"
53 #include "vlan-bitmap.h"
56 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
58 COVERAGE_DEFINE(ofproto_dpif_ctlr_action);
59 COVERAGE_DEFINE(ofproto_dpif_expired);
60 COVERAGE_DEFINE(ofproto_dpif_xlate);
61 COVERAGE_DEFINE(facet_changed_rule);
62 COVERAGE_DEFINE(facet_invalidated);
63 COVERAGE_DEFINE(facet_revalidate);
64 COVERAGE_DEFINE(facet_unexpected);
65 COVERAGE_DEFINE(facet_suppress);
67 /* Maximum depth of flow table recursion (due to resubmit actions) in a
68 * flow translation. */
69 #define MAX_RESUBMIT_RECURSION 32
71 /* Number of implemented OpenFlow tables. */
72 enum { N_TABLES = 255 };
73 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
74 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
84 * - Do include packets and bytes from facets that have been deleted or
85 * whose own statistics have been folded into the rule.
87 * - Do include packets and bytes sent "by hand" that were accounted to
88 * the rule without any facet being involved (this is a rare corner
89 * case in rule_execute()).
91 * - Do not include packet or bytes that can be obtained from any facet's
92 * packet_count or byte_count member or that can be obtained from the
93 * datapath by, e.g., dpif_flow_get() for any subfacet.
95 uint64_t packet_count; /* Number of packets received. */
96 uint64_t byte_count; /* Number of bytes received. */
98 tag_type tag; /* Caches rule_calculate_tag() result. */
100 struct list facets; /* List of "struct facet"s. */
103 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
105 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
108 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
109 const struct flow *);
110 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
114 static void rule_credit_stats(struct rule_dpif *,
115 const struct dpif_flow_stats *);
116 static void flow_push_stats(struct rule_dpif *, const struct flow *,
117 const struct dpif_flow_stats *);
118 static tag_type rule_calculate_tag(const struct flow *,
119 const struct flow_wildcards *,
121 static void rule_invalidate(const struct rule_dpif *);
123 #define MAX_MIRRORS 32
124 typedef uint32_t mirror_mask_t;
125 #define MIRROR_MASK_C(X) UINT32_C(X)
126 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
128 struct ofproto_dpif *ofproto; /* Owning ofproto. */
129 size_t idx; /* In ofproto's "mirrors" array. */
130 void *aux; /* Key supplied by ofproto's client. */
131 char *name; /* Identifier for log messages. */
133 /* Selection criteria. */
134 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
135 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
136 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
138 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
139 struct ofbundle *out; /* Output port or NULL. */
140 int out_vlan; /* Output VLAN or -1. */
141 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
144 int64_t packet_count; /* Number of packets sent. */
145 int64_t byte_count; /* Number of bytes sent. */
148 static void mirror_destroy(struct ofmirror *);
149 static void update_mirror_stats(struct ofproto_dpif *ofproto,
150 mirror_mask_t mirrors,
151 uint64_t packets, uint64_t bytes);
154 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
155 struct ofproto_dpif *ofproto; /* Owning ofproto. */
156 void *aux; /* Key supplied by ofproto's client. */
157 char *name; /* Identifier for log messages. */
160 struct list ports; /* Contains "struct ofport"s. */
161 enum port_vlan_mode vlan_mode; /* VLAN mode */
162 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
163 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
164 * NULL if all VLANs are trunked. */
165 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
166 struct bond *bond; /* Nonnull iff more than one port. */
167 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
170 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
172 /* Port mirroring info. */
173 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
174 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
175 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
178 static void bundle_remove(struct ofport *);
179 static void bundle_update(struct ofbundle *);
180 static void bundle_destroy(struct ofbundle *);
181 static void bundle_del_port(struct ofport_dpif *);
182 static void bundle_run(struct ofbundle *);
183 static void bundle_wait(struct ofbundle *);
184 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
185 uint16_t in_port, bool warn,
186 struct ofport_dpif **in_ofportp);
188 /* A controller may use OFPP_NONE as the ingress port to indicate that
189 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
190 * when an input bundle is needed for validation (e.g., mirroring or
191 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
192 * any 'port' structs, so care must be taken when dealing with it. */
193 static struct ofbundle ofpp_none_bundle = {
195 .vlan_mode = PORT_VLAN_TRUNK
198 static void stp_run(struct ofproto_dpif *ofproto);
199 static void stp_wait(struct ofproto_dpif *ofproto);
200 static int set_stp_port(struct ofport *,
201 const struct ofproto_port_stp_settings *);
203 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
205 struct action_xlate_ctx {
206 /* action_xlate_ctx_init() initializes these members. */
209 struct ofproto_dpif *ofproto;
211 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
212 * this flow when actions change header fields. */
215 /* The packet corresponding to 'flow', or a null pointer if we are
216 * revalidating without a packet to refer to. */
217 const struct ofpbuf *packet;
219 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
220 * actions update the flow table?
222 * We want to update these tables if we are actually processing a packet,
223 * or if we are accounting for packets that the datapath has processed, but
224 * not if we are just revalidating. */
227 /* The rule that we are currently translating, or NULL. */
228 struct rule_dpif *rule;
230 /* Union of the set of TCP flags seen so far in this flow. (Used only by
231 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
235 /* If nonnull, flow translation calls this function just before executing a
236 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
237 * when the recursion depth is exceeded.
239 * 'rule' is the rule being submitted into. It will be null if the
240 * resubmit or OFPP_TABLE action didn't find a matching rule.
242 * This is normally null so the client has to set it manually after
243 * calling action_xlate_ctx_init(). */
244 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule);
246 /* If nonnull, flow translation credits the specified statistics to each
247 * rule reached through a resubmit or OFPP_TABLE action.
249 * This is normally null so the client has to set it manually after
250 * calling action_xlate_ctx_init(). */
251 const struct dpif_flow_stats *resubmit_stats;
253 /* xlate_actions() initializes and uses these members. The client might want
254 * to look at them after it returns. */
256 struct ofpbuf *odp_actions; /* Datapath actions. */
257 tag_type tags; /* Tags associated with actions. */
258 enum slow_path_reason slow; /* 0 if fast path may be used. */
259 bool has_learn; /* Actions include NXAST_LEARN? */
260 bool has_normal; /* Actions output to OFPP_NORMAL? */
261 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
262 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
263 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
265 /* xlate_actions() initializes and uses these members, but the client has no
266 * reason to look at them. */
268 int recurse; /* Recursion level, via xlate_table_action. */
269 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
270 struct flow base_flow; /* Flow at the last commit. */
271 uint32_t orig_skb_priority; /* Priority when packet arrived. */
272 uint8_t table_id; /* OpenFlow table ID where flow was found. */
273 uint32_t sflow_n_outputs; /* Number of output ports. */
274 uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
275 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
276 bool exit; /* No further actions should be processed. */
277 struct flow orig_flow; /* Copy of original flow. */
280 static void action_xlate_ctx_init(struct action_xlate_ctx *,
281 struct ofproto_dpif *, const struct flow *,
282 ovs_be16 initial_tci, struct rule_dpif *,
283 uint8_t tcp_flags, const struct ofpbuf *);
284 static void xlate_actions(struct action_xlate_ctx *,
285 const union ofp_action *in, size_t n_in,
286 struct ofpbuf *odp_actions);
287 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
288 const union ofp_action *in,
291 static size_t put_userspace_action(const struct ofproto_dpif *,
292 struct ofpbuf *odp_actions,
294 const union user_action_cookie *);
296 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
297 enum slow_path_reason,
298 uint64_t *stub, size_t stub_size,
299 const struct nlattr **actionsp,
300 size_t *actions_lenp);
302 /* A subfacet (see "struct subfacet" below) has three possible installation
305 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
306 * case just after the subfacet is created, just before the subfacet is
307 * destroyed, or if the datapath returns an error when we try to install a
310 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
312 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
313 * ofproto_dpif is installed in the datapath.
316 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
317 SF_FAST_PATH, /* Full actions are installed. */
318 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
321 static const char *subfacet_path_to_string(enum subfacet_path);
323 /* A dpif flow and actions associated with a facet.
325 * See also the large comment on struct facet. */
328 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
329 struct list list_node; /* In struct facet's 'facets' list. */
330 struct facet *facet; /* Owning facet. */
334 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
335 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
336 * regenerate the ODP flow key from ->facet->flow. */
337 enum odp_key_fitness key_fitness;
341 long long int used; /* Time last used; time created if not used. */
343 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
344 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
348 * These should be essentially identical for every subfacet in a facet, but
349 * may differ in trivial ways due to VLAN splinters. */
350 size_t actions_len; /* Number of bytes in actions[]. */
351 struct nlattr *actions; /* Datapath actions. */
353 enum slow_path_reason slow; /* 0 if fast path may be used. */
354 enum subfacet_path path; /* Installed in datapath? */
356 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
357 * splinters can cause it to differ. This value should be removed when
358 * the VLAN splinters feature is no longer needed. */
359 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
362 static struct subfacet *subfacet_create(struct facet *, enum odp_key_fitness,
363 const struct nlattr *key,
364 size_t key_len, ovs_be16 initial_tci);
365 static struct subfacet *subfacet_find(struct ofproto_dpif *,
366 const struct nlattr *key, size_t key_len);
367 static void subfacet_destroy(struct subfacet *);
368 static void subfacet_destroy__(struct subfacet *);
369 static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *,
371 static void subfacet_reset_dp_stats(struct subfacet *,
372 struct dpif_flow_stats *);
373 static void subfacet_update_time(struct subfacet *, long long int used);
374 static void subfacet_update_stats(struct subfacet *,
375 const struct dpif_flow_stats *);
376 static void subfacet_make_actions(struct subfacet *,
377 const struct ofpbuf *packet,
378 struct ofpbuf *odp_actions);
379 static int subfacet_install(struct subfacet *,
380 const struct nlattr *actions, size_t actions_len,
381 struct dpif_flow_stats *, enum slow_path_reason);
382 static void subfacet_uninstall(struct subfacet *);
384 static enum subfacet_path subfacet_want_path(enum slow_path_reason);
386 /* An exact-match instantiation of an OpenFlow flow.
388 * A facet associates a "struct flow", which represents the Open vSwitch
389 * userspace idea of an exact-match flow, with one or more subfacets. Each
390 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
391 * the facet. When the kernel module (or other dpif implementation) and Open
392 * vSwitch userspace agree on the definition of a flow key, there is exactly
393 * one subfacet per facet. If the dpif implementation supports more-specific
394 * flow matching than userspace, however, a facet can have more than one
395 * subfacet, each of which corresponds to some distinction in flow that
396 * userspace simply doesn't understand.
398 * Flow expiration works in terms of subfacets, so a facet must have at least
399 * one subfacet or it will never expire, leaking memory. */
402 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
403 struct list list_node; /* In owning rule's 'facets' list. */
404 struct rule_dpif *rule; /* Owning rule. */
407 struct list subfacets;
408 long long int used; /* Time last used; time created if not used. */
415 * - Do include packets and bytes sent "by hand", e.g. with
418 * - Do include packets and bytes that were obtained from the datapath
419 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
420 * DPIF_FP_ZERO_STATS).
422 * - Do not include packets or bytes that can be obtained from the
423 * datapath for any existing subfacet.
425 uint64_t packet_count; /* Number of packets received. */
426 uint64_t byte_count; /* Number of bytes received. */
428 /* Resubmit statistics. */
429 uint64_t prev_packet_count; /* Number of packets from last stats push. */
430 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
431 long long int prev_used; /* Used time from last stats push. */
434 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
435 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
436 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
438 /* Properties of datapath actions.
440 * Every subfacet has its own actions because actions can differ slightly
441 * between splintered and non-splintered subfacets due to the VLAN tag
442 * being initially different (present vs. absent). All of them have these
443 * properties in common so we just store one copy of them here. */
444 bool has_learn; /* Actions include NXAST_LEARN? */
445 bool has_normal; /* Actions output to OFPP_NORMAL? */
446 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
447 tag_type tags; /* Tags that would require revalidation. */
448 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
450 /* Storage for a single subfacet, to reduce malloc() time and space
451 * overhead. (A facet always has at least one subfacet and in the common
452 * case has exactly one subfacet.) */
453 struct subfacet one_subfacet;
456 static struct facet *facet_create(struct rule_dpif *,
457 const struct flow *, uint32_t hash);
458 static void facet_remove(struct facet *);
459 static void facet_free(struct facet *);
461 static struct facet *facet_find(struct ofproto_dpif *,
462 const struct flow *, uint32_t hash);
463 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
464 const struct flow *, uint32_t hash);
465 static void facet_revalidate(struct facet *);
466 static bool facet_check_consistency(struct facet *);
468 static void facet_flush_stats(struct facet *);
470 static void facet_update_time(struct facet *, long long int used);
471 static void facet_reset_counters(struct facet *);
472 static void facet_push_stats(struct facet *);
473 static void facet_learn(struct facet *);
474 static void facet_account(struct facet *);
476 static bool facet_is_controller_flow(struct facet *);
482 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
483 struct list bundle_node; /* In struct ofbundle's "ports" list. */
484 struct cfm *cfm; /* Connectivity Fault Management, if any. */
485 tag_type tag; /* Tag associated with this port. */
486 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
487 bool may_enable; /* May be enabled in bonds. */
488 long long int carrier_seq; /* Carrier status changes. */
491 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
492 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
493 long long int stp_state_entered;
495 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
497 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
499 * This is deprecated. It is only for compatibility with broken device
500 * drivers in old versions of Linux that do not properly support VLANs when
501 * VLAN devices are not used. When broken device drivers are no longer in
502 * widespread use, we will delete these interfaces. */
503 uint16_t realdev_ofp_port;
507 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
508 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
509 * traffic egressing the 'ofport' with that priority should be marked with. */
510 struct priority_to_dscp {
511 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
512 uint32_t priority; /* Priority of this queue (see struct flow). */
514 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
517 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
519 * This is deprecated. It is only for compatibility with broken device drivers
520 * in old versions of Linux that do not properly support VLANs when VLAN
521 * devices are not used. When broken device drivers are no longer in
522 * widespread use, we will delete these interfaces. */
523 struct vlan_splinter {
524 struct hmap_node realdev_vid_node;
525 struct hmap_node vlandev_node;
526 uint16_t realdev_ofp_port;
527 uint16_t vlandev_ofp_port;
531 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
532 uint32_t realdev, ovs_be16 vlan_tci);
533 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
534 static void vsp_remove(struct ofport_dpif *);
535 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
537 static struct ofport_dpif *
538 ofport_dpif_cast(const struct ofport *ofport)
540 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
541 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
544 static void port_run(struct ofport_dpif *);
545 static void port_wait(struct ofport_dpif *);
546 static int set_cfm(struct ofport *, const struct cfm_settings *);
547 static void ofport_clear_priorities(struct ofport_dpif *);
549 struct dpif_completion {
550 struct list list_node;
551 struct ofoperation *op;
554 /* Extra information about a classifier table.
555 * Currently used just for optimized flow revalidation. */
557 /* If either of these is nonnull, then this table has a form that allows
558 * flows to be tagged to avoid revalidating most flows for the most common
559 * kinds of flow table changes. */
560 struct cls_table *catchall_table; /* Table that wildcards all fields. */
561 struct cls_table *other_table; /* Table with any other wildcard set. */
562 uint32_t basis; /* Keeps each table's tags separate. */
565 struct ofproto_dpif {
566 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
571 /* Special OpenFlow rules. */
572 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
573 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
579 struct netflow *netflow;
580 struct dpif_sflow *sflow;
581 struct hmap bundles; /* Contains "struct ofbundle"s. */
582 struct mac_learning *ml;
583 struct ofmirror *mirrors[MAX_MIRRORS];
585 bool has_bonded_bundles;
588 struct timer next_expiration;
592 struct hmap subfacets;
593 struct governor *governor;
596 struct table_dpif tables[N_TABLES];
597 bool need_revalidate;
598 struct tag_set revalidate_set;
600 /* Support for debugging async flow mods. */
601 struct list completions;
603 bool has_bundle_action; /* True when the first bundle action appears. */
604 struct netdev_stats stats; /* To account packets generated and consumed in
609 long long int stp_last_tick;
611 /* VLAN splinters. */
612 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
613 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
616 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
617 * for debugging the asynchronous flow_mod implementation.) */
620 /* All existing ofproto_dpif instances, indexed by ->up.name. */
621 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
623 static void ofproto_dpif_unixctl_init(void);
625 static struct ofproto_dpif *
626 ofproto_dpif_cast(const struct ofproto *ofproto)
628 assert(ofproto->ofproto_class == &ofproto_dpif_class);
629 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
632 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
634 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
636 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
637 const struct ofpbuf *, ovs_be16 initial_tci,
640 /* Packet processing. */
641 static void update_learning_table(struct ofproto_dpif *,
642 const struct flow *, int vlan,
645 #define FLOW_MISS_MAX_BATCH 50
646 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
648 /* Flow expiration. */
649 static int expire(struct ofproto_dpif *);
652 static void send_netflow_active_timeouts(struct ofproto_dpif *);
655 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
656 static size_t compose_sflow_action(const struct ofproto_dpif *,
657 struct ofpbuf *odp_actions,
658 const struct flow *, uint32_t odp_port);
659 static void add_mirror_actions(struct action_xlate_ctx *ctx,
660 const struct flow *flow);
661 /* Global variables. */
662 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
664 /* Factory functions. */
667 enumerate_types(struct sset *types)
669 dp_enumerate_types(types);
673 enumerate_names(const char *type, struct sset *names)
675 return dp_enumerate_names(type, names);
679 del(const char *type, const char *name)
684 error = dpif_open(name, type, &dpif);
686 error = dpif_delete(dpif);
692 /* Basic life-cycle. */
694 static int add_internal_flows(struct ofproto_dpif *);
696 static struct ofproto *
699 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
704 dealloc(struct ofproto *ofproto_)
706 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
711 construct(struct ofproto *ofproto_)
713 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
714 const char *name = ofproto->up.name;
718 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
720 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
724 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
725 ofproto->n_matches = 0;
727 dpif_flow_flush(ofproto->dpif);
728 dpif_recv_purge(ofproto->dpif);
730 error = dpif_recv_set(ofproto->dpif, true);
732 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
733 dpif_close(ofproto->dpif);
737 ofproto->netflow = NULL;
738 ofproto->sflow = NULL;
740 hmap_init(&ofproto->bundles);
741 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
742 for (i = 0; i < MAX_MIRRORS; i++) {
743 ofproto->mirrors[i] = NULL;
745 ofproto->has_bonded_bundles = false;
747 timer_set_duration(&ofproto->next_expiration, 1000);
749 hmap_init(&ofproto->facets);
750 hmap_init(&ofproto->subfacets);
751 ofproto->governor = NULL;
753 for (i = 0; i < N_TABLES; i++) {
754 struct table_dpif *table = &ofproto->tables[i];
756 table->catchall_table = NULL;
757 table->other_table = NULL;
758 table->basis = random_uint32();
760 ofproto->need_revalidate = false;
761 tag_set_init(&ofproto->revalidate_set);
763 list_init(&ofproto->completions);
765 ofproto_dpif_unixctl_init();
767 ofproto->has_mirrors = false;
768 ofproto->has_bundle_action = false;
770 hmap_init(&ofproto->vlandev_map);
771 hmap_init(&ofproto->realdev_vid_map);
773 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
774 hash_string(ofproto->up.name, 0));
775 memset(&ofproto->stats, 0, sizeof ofproto->stats);
777 ofproto_init_tables(ofproto_, N_TABLES);
778 error = add_internal_flows(ofproto);
779 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
785 add_internal_flow(struct ofproto_dpif *ofproto, int id,
786 const struct ofpbuf *actions, struct rule_dpif **rulep)
788 struct ofputil_flow_mod fm;
791 cls_rule_init_catchall(&fm.cr, 0);
792 cls_rule_set_reg(&fm.cr, 0, id);
793 fm.cookie = htonll(0);
794 fm.cookie_mask = htonll(0);
795 fm.table_id = TBL_INTERNAL;
796 fm.command = OFPFC_ADD;
802 fm.actions = actions->data;
803 fm.n_actions = actions->size / sizeof(union ofp_action);
805 error = ofproto_flow_mod(&ofproto->up, &fm);
807 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
808 id, ofperr_to_string(error));
812 *rulep = rule_dpif_lookup__(ofproto, &fm.cr.flow, TBL_INTERNAL);
813 assert(*rulep != NULL);
819 add_internal_flows(struct ofproto_dpif *ofproto)
821 struct nx_action_controller *nac;
822 uint64_t actions_stub[128 / 8];
823 struct ofpbuf actions;
827 ofpbuf_use_stack(&actions, actions_stub, sizeof actions_stub);
830 nac = ofputil_put_NXAST_CONTROLLER(&actions);
831 nac->max_len = htons(UINT16_MAX);
832 nac->controller_id = htons(0);
833 nac->reason = OFPR_NO_MATCH;
834 error = add_internal_flow(ofproto, id++, &actions, &ofproto->miss_rule);
839 ofpbuf_clear(&actions);
840 error = add_internal_flow(ofproto, id++, &actions,
841 &ofproto->no_packet_in_rule);
846 complete_operations(struct ofproto_dpif *ofproto)
848 struct dpif_completion *c, *next;
850 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
851 ofoperation_complete(c->op, 0);
852 list_remove(&c->list_node);
858 destruct(struct ofproto *ofproto_)
860 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
861 struct rule_dpif *rule, *next_rule;
862 struct oftable *table;
865 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
866 complete_operations(ofproto);
868 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
869 struct cls_cursor cursor;
871 cls_cursor_init(&cursor, &table->cls, NULL);
872 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
873 ofproto_rule_destroy(&rule->up);
877 for (i = 0; i < MAX_MIRRORS; i++) {
878 mirror_destroy(ofproto->mirrors[i]);
881 netflow_destroy(ofproto->netflow);
882 dpif_sflow_destroy(ofproto->sflow);
883 hmap_destroy(&ofproto->bundles);
884 mac_learning_destroy(ofproto->ml);
886 hmap_destroy(&ofproto->facets);
887 hmap_destroy(&ofproto->subfacets);
888 governor_destroy(ofproto->governor);
890 hmap_destroy(&ofproto->vlandev_map);
891 hmap_destroy(&ofproto->realdev_vid_map);
893 dpif_close(ofproto->dpif);
897 run_fast(struct ofproto *ofproto_)
899 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
902 /* Handle one or more batches of upcalls, until there's nothing left to do
903 * or until we do a fixed total amount of work.
905 * We do work in batches because it can be much cheaper to set up a number
906 * of flows and fire off their patches all at once. We do multiple batches
907 * because in some cases handling a packet can cause another packet to be
908 * queued almost immediately as part of the return flow. Both
909 * optimizations can make major improvements on some benchmarks and
910 * presumably for real traffic as well. */
912 while (work < FLOW_MISS_MAX_BATCH) {
913 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
923 run(struct ofproto *ofproto_)
925 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
926 struct ofport_dpif *ofport;
927 struct ofbundle *bundle;
931 complete_operations(ofproto);
933 dpif_run(ofproto->dpif);
935 error = run_fast(ofproto_);
940 if (timer_expired(&ofproto->next_expiration)) {
941 int delay = expire(ofproto);
942 timer_set_duration(&ofproto->next_expiration, delay);
945 if (ofproto->netflow) {
946 if (netflow_run(ofproto->netflow)) {
947 send_netflow_active_timeouts(ofproto);
950 if (ofproto->sflow) {
951 dpif_sflow_run(ofproto->sflow);
954 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
957 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
962 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
964 /* Now revalidate if there's anything to do. */
965 if (ofproto->need_revalidate
966 || !tag_set_is_empty(&ofproto->revalidate_set)) {
967 struct tag_set revalidate_set = ofproto->revalidate_set;
968 bool revalidate_all = ofproto->need_revalidate;
971 /* Clear the revalidation flags. */
972 tag_set_init(&ofproto->revalidate_set);
973 ofproto->need_revalidate = false;
975 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
977 || tag_set_intersects(&revalidate_set, facet->tags)) {
978 facet_revalidate(facet);
983 /* Check the consistency of a random facet, to aid debugging. */
984 if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) {
987 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
988 struct facet, hmap_node);
989 if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) {
990 if (!facet_check_consistency(facet)) {
991 ofproto->need_revalidate = true;
996 if (ofproto->governor) {
999 governor_run(ofproto->governor);
1001 /* If the governor has shrunk to its minimum size and the number of
1002 * subfacets has dwindled, then drop the governor entirely.
1004 * For hysteresis, the number of subfacets to drop the governor is
1005 * smaller than the number needed to trigger its creation. */
1006 n_subfacets = hmap_count(&ofproto->subfacets);
1007 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1008 && governor_is_idle(ofproto->governor)) {
1009 governor_destroy(ofproto->governor);
1010 ofproto->governor = NULL;
1018 wait(struct ofproto *ofproto_)
1020 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1021 struct ofport_dpif *ofport;
1022 struct ofbundle *bundle;
1024 if (!clogged && !list_is_empty(&ofproto->completions)) {
1025 poll_immediate_wake();
1028 dpif_wait(ofproto->dpif);
1029 dpif_recv_wait(ofproto->dpif);
1030 if (ofproto->sflow) {
1031 dpif_sflow_wait(ofproto->sflow);
1033 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
1034 poll_immediate_wake();
1036 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1039 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1040 bundle_wait(bundle);
1042 if (ofproto->netflow) {
1043 netflow_wait(ofproto->netflow);
1045 mac_learning_wait(ofproto->ml);
1047 if (ofproto->need_revalidate) {
1048 /* Shouldn't happen, but if it does just go around again. */
1049 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1050 poll_immediate_wake();
1052 timer_wait(&ofproto->next_expiration);
1054 if (ofproto->governor) {
1055 governor_wait(ofproto->governor);
1060 flush(struct ofproto *ofproto_)
1062 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1063 struct facet *facet, *next_facet;
1065 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1066 /* Mark the facet as not installed so that facet_remove() doesn't
1067 * bother trying to uninstall it. There is no point in uninstalling it
1068 * individually since we are about to blow away all the facets with
1069 * dpif_flow_flush(). */
1070 struct subfacet *subfacet;
1072 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
1073 subfacet->path = SF_NOT_INSTALLED;
1074 subfacet->dp_packet_count = 0;
1075 subfacet->dp_byte_count = 0;
1077 facet_remove(facet);
1079 dpif_flow_flush(ofproto->dpif);
1083 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1084 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1086 *arp_match_ip = true;
1087 *actions = (OFPUTIL_A_OUTPUT |
1088 OFPUTIL_A_SET_VLAN_VID |
1089 OFPUTIL_A_SET_VLAN_PCP |
1090 OFPUTIL_A_STRIP_VLAN |
1091 OFPUTIL_A_SET_DL_SRC |
1092 OFPUTIL_A_SET_DL_DST |
1093 OFPUTIL_A_SET_NW_SRC |
1094 OFPUTIL_A_SET_NW_DST |
1095 OFPUTIL_A_SET_NW_TOS |
1096 OFPUTIL_A_SET_TP_SRC |
1097 OFPUTIL_A_SET_TP_DST |
1102 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
1104 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1105 struct dpif_dp_stats s;
1107 strcpy(ots->name, "classifier");
1109 dpif_get_dp_stats(ofproto->dpif, &s);
1110 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
1111 put_32aligned_be64(&ots->matched_count,
1112 htonll(s.n_hit + ofproto->n_matches));
1115 static struct ofport *
1118 struct ofport_dpif *port = xmalloc(sizeof *port);
1123 port_dealloc(struct ofport *port_)
1125 struct ofport_dpif *port = ofport_dpif_cast(port_);
1130 port_construct(struct ofport *port_)
1132 struct ofport_dpif *port = ofport_dpif_cast(port_);
1133 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1135 ofproto->need_revalidate = true;
1136 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
1137 port->bundle = NULL;
1139 port->tag = tag_create_random();
1140 port->may_enable = true;
1141 port->stp_port = NULL;
1142 port->stp_state = STP_DISABLED;
1143 hmap_init(&port->priorities);
1144 port->realdev_ofp_port = 0;
1145 port->vlandev_vid = 0;
1146 port->carrier_seq = netdev_get_carrier_resets(port->up.netdev);
1148 if (ofproto->sflow) {
1149 dpif_sflow_add_port(ofproto->sflow, port_);
1156 port_destruct(struct ofport *port_)
1158 struct ofport_dpif *port = ofport_dpif_cast(port_);
1159 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1161 ofproto->need_revalidate = true;
1162 bundle_remove(port_);
1163 set_cfm(port_, NULL);
1164 if (ofproto->sflow) {
1165 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1168 ofport_clear_priorities(port);
1169 hmap_destroy(&port->priorities);
1173 port_modified(struct ofport *port_)
1175 struct ofport_dpif *port = ofport_dpif_cast(port_);
1177 if (port->bundle && port->bundle->bond) {
1178 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1183 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1185 struct ofport_dpif *port = ofport_dpif_cast(port_);
1186 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1187 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1189 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1190 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1191 OFPUTIL_PC_NO_PACKET_IN)) {
1192 ofproto->need_revalidate = true;
1194 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1195 bundle_update(port->bundle);
1201 set_sflow(struct ofproto *ofproto_,
1202 const struct ofproto_sflow_options *sflow_options)
1204 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1205 struct dpif_sflow *ds = ofproto->sflow;
1207 if (sflow_options) {
1209 struct ofport_dpif *ofport;
1211 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1212 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1213 dpif_sflow_add_port(ds, &ofport->up);
1215 ofproto->need_revalidate = true;
1217 dpif_sflow_set_options(ds, sflow_options);
1220 dpif_sflow_destroy(ds);
1221 ofproto->need_revalidate = true;
1222 ofproto->sflow = NULL;
1229 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1231 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1238 struct ofproto_dpif *ofproto;
1240 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1241 ofproto->need_revalidate = true;
1242 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1245 if (cfm_configure(ofport->cfm, s)) {
1251 cfm_destroy(ofport->cfm);
1257 get_cfm_fault(const struct ofport *ofport_)
1259 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1261 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1265 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1268 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1271 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1279 get_cfm_health(const struct ofport *ofport_)
1281 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1283 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1286 /* Spanning Tree. */
1289 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1291 struct ofproto_dpif *ofproto = ofproto_;
1292 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1293 struct ofport_dpif *ofport;
1295 ofport = stp_port_get_aux(sp);
1297 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1298 ofproto->up.name, port_num);
1300 struct eth_header *eth = pkt->l2;
1302 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1303 if (eth_addr_is_zero(eth->eth_src)) {
1304 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1305 "with unknown MAC", ofproto->up.name, port_num);
1307 send_packet(ofport, pkt);
1313 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1315 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1317 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1319 /* Only revalidate flows if the configuration changed. */
1320 if (!s != !ofproto->stp) {
1321 ofproto->need_revalidate = true;
1325 if (!ofproto->stp) {
1326 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1327 send_bpdu_cb, ofproto);
1328 ofproto->stp_last_tick = time_msec();
1331 stp_set_bridge_id(ofproto->stp, s->system_id);
1332 stp_set_bridge_priority(ofproto->stp, s->priority);
1333 stp_set_hello_time(ofproto->stp, s->hello_time);
1334 stp_set_max_age(ofproto->stp, s->max_age);
1335 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1337 struct ofport *ofport;
1339 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1340 set_stp_port(ofport, NULL);
1343 stp_destroy(ofproto->stp);
1344 ofproto->stp = NULL;
1351 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1353 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1357 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1358 s->designated_root = stp_get_designated_root(ofproto->stp);
1359 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1368 update_stp_port_state(struct ofport_dpif *ofport)
1370 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1371 enum stp_state state;
1373 /* Figure out new state. */
1374 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1378 if (ofport->stp_state != state) {
1379 enum ofputil_port_state of_state;
1382 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1383 netdev_get_name(ofport->up.netdev),
1384 stp_state_name(ofport->stp_state),
1385 stp_state_name(state));
1386 if (stp_learn_in_state(ofport->stp_state)
1387 != stp_learn_in_state(state)) {
1388 /* xxx Learning action flows should also be flushed. */
1389 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1391 fwd_change = stp_forward_in_state(ofport->stp_state)
1392 != stp_forward_in_state(state);
1394 ofproto->need_revalidate = true;
1395 ofport->stp_state = state;
1396 ofport->stp_state_entered = time_msec();
1398 if (fwd_change && ofport->bundle) {
1399 bundle_update(ofport->bundle);
1402 /* Update the STP state bits in the OpenFlow port description. */
1403 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1404 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1405 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1406 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1407 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1409 ofproto_port_set_state(&ofport->up, of_state);
1413 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1414 * caller is responsible for assigning STP port numbers and ensuring
1415 * there are no duplicates. */
1417 set_stp_port(struct ofport *ofport_,
1418 const struct ofproto_port_stp_settings *s)
1420 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1421 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1422 struct stp_port *sp = ofport->stp_port;
1424 if (!s || !s->enable) {
1426 ofport->stp_port = NULL;
1427 stp_port_disable(sp);
1428 update_stp_port_state(ofport);
1431 } else if (sp && stp_port_no(sp) != s->port_num
1432 && ofport == stp_port_get_aux(sp)) {
1433 /* The port-id changed, so disable the old one if it's not
1434 * already in use by another port. */
1435 stp_port_disable(sp);
1438 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1439 stp_port_enable(sp);
1441 stp_port_set_aux(sp, ofport);
1442 stp_port_set_priority(sp, s->priority);
1443 stp_port_set_path_cost(sp, s->path_cost);
1445 update_stp_port_state(ofport);
1451 get_stp_port_status(struct ofport *ofport_,
1452 struct ofproto_port_stp_status *s)
1454 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1455 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1456 struct stp_port *sp = ofport->stp_port;
1458 if (!ofproto->stp || !sp) {
1464 s->port_id = stp_port_get_id(sp);
1465 s->state = stp_port_get_state(sp);
1466 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1467 s->role = stp_port_get_role(sp);
1468 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1474 stp_run(struct ofproto_dpif *ofproto)
1477 long long int now = time_msec();
1478 long long int elapsed = now - ofproto->stp_last_tick;
1479 struct stp_port *sp;
1482 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1483 ofproto->stp_last_tick = now;
1485 while (stp_get_changed_port(ofproto->stp, &sp)) {
1486 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1489 update_stp_port_state(ofport);
1493 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1494 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1500 stp_wait(struct ofproto_dpif *ofproto)
1503 poll_timer_wait(1000);
1507 /* Returns true if STP should process 'flow'. */
1509 stp_should_process_flow(const struct flow *flow)
1511 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1515 stp_process_packet(const struct ofport_dpif *ofport,
1516 const struct ofpbuf *packet)
1518 struct ofpbuf payload = *packet;
1519 struct eth_header *eth = payload.data;
1520 struct stp_port *sp = ofport->stp_port;
1522 /* Sink packets on ports that have STP disabled when the bridge has
1524 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1528 /* Trim off padding on payload. */
1529 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1530 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1533 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1534 stp_received_bpdu(sp, payload.data, payload.size);
1538 static struct priority_to_dscp *
1539 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1541 struct priority_to_dscp *pdscp;
1544 hash = hash_int(priority, 0);
1545 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1546 if (pdscp->priority == priority) {
1554 ofport_clear_priorities(struct ofport_dpif *ofport)
1556 struct priority_to_dscp *pdscp, *next;
1558 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1559 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1565 set_queues(struct ofport *ofport_,
1566 const struct ofproto_port_queue *qdscp_list,
1569 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1570 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1571 struct hmap new = HMAP_INITIALIZER(&new);
1574 for (i = 0; i < n_qdscp; i++) {
1575 struct priority_to_dscp *pdscp;
1579 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1580 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1585 pdscp = get_priority(ofport, priority);
1587 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1589 pdscp = xmalloc(sizeof *pdscp);
1590 pdscp->priority = priority;
1592 ofproto->need_revalidate = true;
1595 if (pdscp->dscp != dscp) {
1597 ofproto->need_revalidate = true;
1600 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1603 if (!hmap_is_empty(&ofport->priorities)) {
1604 ofport_clear_priorities(ofport);
1605 ofproto->need_revalidate = true;
1608 hmap_swap(&new, &ofport->priorities);
1616 /* Expires all MAC learning entries associated with 'bundle' and forces its
1617 * ofproto to revalidate every flow.
1619 * Normally MAC learning entries are removed only from the ofproto associated
1620 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1621 * are removed from every ofproto. When patch ports and SLB bonds are in use
1622 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1623 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1624 * with the host from which it migrated. */
1626 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1628 struct ofproto_dpif *ofproto = bundle->ofproto;
1629 struct mac_learning *ml = ofproto->ml;
1630 struct mac_entry *mac, *next_mac;
1632 ofproto->need_revalidate = true;
1633 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1634 if (mac->port.p == bundle) {
1636 struct ofproto_dpif *o;
1638 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1640 struct mac_entry *e;
1642 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1645 tag_set_add(&o->revalidate_set, e->tag);
1646 mac_learning_expire(o->ml, e);
1652 mac_learning_expire(ml, mac);
1657 static struct ofbundle *
1658 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1660 struct ofbundle *bundle;
1662 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1663 &ofproto->bundles) {
1664 if (bundle->aux == aux) {
1671 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1672 * ones that are found to 'bundles'. */
1674 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1675 void **auxes, size_t n_auxes,
1676 struct hmapx *bundles)
1680 hmapx_init(bundles);
1681 for (i = 0; i < n_auxes; i++) {
1682 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1684 hmapx_add(bundles, bundle);
1690 bundle_update(struct ofbundle *bundle)
1692 struct ofport_dpif *port;
1694 bundle->floodable = true;
1695 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1696 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1697 || !stp_forward_in_state(port->stp_state)) {
1698 bundle->floodable = false;
1705 bundle_del_port(struct ofport_dpif *port)
1707 struct ofbundle *bundle = port->bundle;
1709 bundle->ofproto->need_revalidate = true;
1711 list_remove(&port->bundle_node);
1712 port->bundle = NULL;
1715 lacp_slave_unregister(bundle->lacp, port);
1718 bond_slave_unregister(bundle->bond, port);
1721 bundle_update(bundle);
1725 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1726 struct lacp_slave_settings *lacp,
1727 uint32_t bond_stable_id)
1729 struct ofport_dpif *port;
1731 port = get_ofp_port(bundle->ofproto, ofp_port);
1736 if (port->bundle != bundle) {
1737 bundle->ofproto->need_revalidate = true;
1739 bundle_del_port(port);
1742 port->bundle = bundle;
1743 list_push_back(&bundle->ports, &port->bundle_node);
1744 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1745 || !stp_forward_in_state(port->stp_state)) {
1746 bundle->floodable = false;
1750 port->bundle->ofproto->need_revalidate = true;
1751 lacp_slave_register(bundle->lacp, port, lacp);
1754 port->bond_stable_id = bond_stable_id;
1760 bundle_destroy(struct ofbundle *bundle)
1762 struct ofproto_dpif *ofproto;
1763 struct ofport_dpif *port, *next_port;
1770 ofproto = bundle->ofproto;
1771 for (i = 0; i < MAX_MIRRORS; i++) {
1772 struct ofmirror *m = ofproto->mirrors[i];
1774 if (m->out == bundle) {
1776 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1777 || hmapx_find_and_delete(&m->dsts, bundle)) {
1778 ofproto->need_revalidate = true;
1783 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1784 bundle_del_port(port);
1787 bundle_flush_macs(bundle, true);
1788 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1790 free(bundle->trunks);
1791 lacp_destroy(bundle->lacp);
1792 bond_destroy(bundle->bond);
1797 bundle_set(struct ofproto *ofproto_, void *aux,
1798 const struct ofproto_bundle_settings *s)
1800 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1801 bool need_flush = false;
1802 struct ofport_dpif *port;
1803 struct ofbundle *bundle;
1804 unsigned long *trunks;
1810 bundle_destroy(bundle_lookup(ofproto, aux));
1814 assert(s->n_slaves == 1 || s->bond != NULL);
1815 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1817 bundle = bundle_lookup(ofproto, aux);
1819 bundle = xmalloc(sizeof *bundle);
1821 bundle->ofproto = ofproto;
1822 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1823 hash_pointer(aux, 0));
1825 bundle->name = NULL;
1827 list_init(&bundle->ports);
1828 bundle->vlan_mode = PORT_VLAN_TRUNK;
1830 bundle->trunks = NULL;
1831 bundle->use_priority_tags = s->use_priority_tags;
1832 bundle->lacp = NULL;
1833 bundle->bond = NULL;
1835 bundle->floodable = true;
1837 bundle->src_mirrors = 0;
1838 bundle->dst_mirrors = 0;
1839 bundle->mirror_out = 0;
1842 if (!bundle->name || strcmp(s->name, bundle->name)) {
1844 bundle->name = xstrdup(s->name);
1849 if (!bundle->lacp) {
1850 ofproto->need_revalidate = true;
1851 bundle->lacp = lacp_create();
1853 lacp_configure(bundle->lacp, s->lacp);
1855 lacp_destroy(bundle->lacp);
1856 bundle->lacp = NULL;
1859 /* Update set of ports. */
1861 for (i = 0; i < s->n_slaves; i++) {
1862 if (!bundle_add_port(bundle, s->slaves[i],
1863 s->lacp ? &s->lacp_slaves[i] : NULL,
1864 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1868 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1869 struct ofport_dpif *next_port;
1871 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1872 for (i = 0; i < s->n_slaves; i++) {
1873 if (s->slaves[i] == port->up.ofp_port) {
1878 bundle_del_port(port);
1882 assert(list_size(&bundle->ports) <= s->n_slaves);
1884 if (list_is_empty(&bundle->ports)) {
1885 bundle_destroy(bundle);
1889 /* Set VLAN tagging mode */
1890 if (s->vlan_mode != bundle->vlan_mode
1891 || s->use_priority_tags != bundle->use_priority_tags) {
1892 bundle->vlan_mode = s->vlan_mode;
1893 bundle->use_priority_tags = s->use_priority_tags;
1898 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1899 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1901 if (vlan != bundle->vlan) {
1902 bundle->vlan = vlan;
1906 /* Get trunked VLANs. */
1907 switch (s->vlan_mode) {
1908 case PORT_VLAN_ACCESS:
1912 case PORT_VLAN_TRUNK:
1913 trunks = (unsigned long *) s->trunks;
1916 case PORT_VLAN_NATIVE_UNTAGGED:
1917 case PORT_VLAN_NATIVE_TAGGED:
1918 if (vlan != 0 && (!s->trunks
1919 || !bitmap_is_set(s->trunks, vlan)
1920 || bitmap_is_set(s->trunks, 0))) {
1921 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1923 trunks = bitmap_clone(s->trunks, 4096);
1925 trunks = bitmap_allocate1(4096);
1927 bitmap_set1(trunks, vlan);
1928 bitmap_set0(trunks, 0);
1930 trunks = (unsigned long *) s->trunks;
1937 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1938 free(bundle->trunks);
1939 if (trunks == s->trunks) {
1940 bundle->trunks = vlan_bitmap_clone(trunks);
1942 bundle->trunks = trunks;
1947 if (trunks != s->trunks) {
1952 if (!list_is_short(&bundle->ports)) {
1953 bundle->ofproto->has_bonded_bundles = true;
1955 if (bond_reconfigure(bundle->bond, s->bond)) {
1956 ofproto->need_revalidate = true;
1959 bundle->bond = bond_create(s->bond);
1960 ofproto->need_revalidate = true;
1963 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1964 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1968 bond_destroy(bundle->bond);
1969 bundle->bond = NULL;
1972 /* If we changed something that would affect MAC learning, un-learn
1973 * everything on this port and force flow revalidation. */
1975 bundle_flush_macs(bundle, false);
1982 bundle_remove(struct ofport *port_)
1984 struct ofport_dpif *port = ofport_dpif_cast(port_);
1985 struct ofbundle *bundle = port->bundle;
1988 bundle_del_port(port);
1989 if (list_is_empty(&bundle->ports)) {
1990 bundle_destroy(bundle);
1991 } else if (list_is_short(&bundle->ports)) {
1992 bond_destroy(bundle->bond);
1993 bundle->bond = NULL;
1999 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2001 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2002 struct ofport_dpif *port = port_;
2003 uint8_t ea[ETH_ADDR_LEN];
2006 error = netdev_get_etheraddr(port->up.netdev, ea);
2008 struct ofpbuf packet;
2011 ofpbuf_init(&packet, 0);
2012 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2014 memcpy(packet_pdu, pdu, pdu_size);
2016 send_packet(port, &packet);
2017 ofpbuf_uninit(&packet);
2019 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2020 "%s (%s)", port->bundle->name,
2021 netdev_get_name(port->up.netdev), strerror(error));
2026 bundle_send_learning_packets(struct ofbundle *bundle)
2028 struct ofproto_dpif *ofproto = bundle->ofproto;
2029 int error, n_packets, n_errors;
2030 struct mac_entry *e;
2032 error = n_packets = n_errors = 0;
2033 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2034 if (e->port.p != bundle) {
2035 struct ofpbuf *learning_packet;
2036 struct ofport_dpif *port;
2040 /* The assignment to "port" is unnecessary but makes "grep"ing for
2041 * struct ofport_dpif more effective. */
2042 learning_packet = bond_compose_learning_packet(bundle->bond,
2046 ret = send_packet(port, learning_packet);
2047 ofpbuf_delete(learning_packet);
2057 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2058 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2059 "packets, last error was: %s",
2060 bundle->name, n_errors, n_packets, strerror(error));
2062 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2063 bundle->name, n_packets);
2068 bundle_run(struct ofbundle *bundle)
2071 lacp_run(bundle->lacp, send_pdu_cb);
2074 struct ofport_dpif *port;
2076 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2077 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2080 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
2081 lacp_status(bundle->lacp));
2082 if (bond_should_send_learning_packets(bundle->bond)) {
2083 bundle_send_learning_packets(bundle);
2089 bundle_wait(struct ofbundle *bundle)
2092 lacp_wait(bundle->lacp);
2095 bond_wait(bundle->bond);
2102 mirror_scan(struct ofproto_dpif *ofproto)
2106 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2107 if (!ofproto->mirrors[idx]) {
2114 static struct ofmirror *
2115 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2119 for (i = 0; i < MAX_MIRRORS; i++) {
2120 struct ofmirror *mirror = ofproto->mirrors[i];
2121 if (mirror && mirror->aux == aux) {
2129 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2131 mirror_update_dups(struct ofproto_dpif *ofproto)
2135 for (i = 0; i < MAX_MIRRORS; i++) {
2136 struct ofmirror *m = ofproto->mirrors[i];
2139 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2143 for (i = 0; i < MAX_MIRRORS; i++) {
2144 struct ofmirror *m1 = ofproto->mirrors[i];
2151 for (j = i + 1; j < MAX_MIRRORS; j++) {
2152 struct ofmirror *m2 = ofproto->mirrors[j];
2154 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2155 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2156 m2->dup_mirrors |= m1->dup_mirrors;
2163 mirror_set(struct ofproto *ofproto_, void *aux,
2164 const struct ofproto_mirror_settings *s)
2166 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2167 mirror_mask_t mirror_bit;
2168 struct ofbundle *bundle;
2169 struct ofmirror *mirror;
2170 struct ofbundle *out;
2171 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2172 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2175 mirror = mirror_lookup(ofproto, aux);
2177 mirror_destroy(mirror);
2183 idx = mirror_scan(ofproto);
2185 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2187 ofproto->up.name, MAX_MIRRORS, s->name);
2191 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2192 mirror->ofproto = ofproto;
2195 mirror->out_vlan = -1;
2196 mirror->name = NULL;
2199 if (!mirror->name || strcmp(s->name, mirror->name)) {
2201 mirror->name = xstrdup(s->name);
2204 /* Get the new configuration. */
2205 if (s->out_bundle) {
2206 out = bundle_lookup(ofproto, s->out_bundle);
2208 mirror_destroy(mirror);
2214 out_vlan = s->out_vlan;
2216 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2217 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2219 /* If the configuration has not changed, do nothing. */
2220 if (hmapx_equals(&srcs, &mirror->srcs)
2221 && hmapx_equals(&dsts, &mirror->dsts)
2222 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2223 && mirror->out == out
2224 && mirror->out_vlan == out_vlan)
2226 hmapx_destroy(&srcs);
2227 hmapx_destroy(&dsts);
2231 hmapx_swap(&srcs, &mirror->srcs);
2232 hmapx_destroy(&srcs);
2234 hmapx_swap(&dsts, &mirror->dsts);
2235 hmapx_destroy(&dsts);
2237 free(mirror->vlans);
2238 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2241 mirror->out_vlan = out_vlan;
2243 /* Update bundles. */
2244 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2245 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2246 if (hmapx_contains(&mirror->srcs, bundle)) {
2247 bundle->src_mirrors |= mirror_bit;
2249 bundle->src_mirrors &= ~mirror_bit;
2252 if (hmapx_contains(&mirror->dsts, bundle)) {
2253 bundle->dst_mirrors |= mirror_bit;
2255 bundle->dst_mirrors &= ~mirror_bit;
2258 if (mirror->out == bundle) {
2259 bundle->mirror_out |= mirror_bit;
2261 bundle->mirror_out &= ~mirror_bit;
2265 ofproto->need_revalidate = true;
2266 ofproto->has_mirrors = true;
2267 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2268 mirror_update_dups(ofproto);
2274 mirror_destroy(struct ofmirror *mirror)
2276 struct ofproto_dpif *ofproto;
2277 mirror_mask_t mirror_bit;
2278 struct ofbundle *bundle;
2285 ofproto = mirror->ofproto;
2286 ofproto->need_revalidate = true;
2287 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2289 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2290 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2291 bundle->src_mirrors &= ~mirror_bit;
2292 bundle->dst_mirrors &= ~mirror_bit;
2293 bundle->mirror_out &= ~mirror_bit;
2296 hmapx_destroy(&mirror->srcs);
2297 hmapx_destroy(&mirror->dsts);
2298 free(mirror->vlans);
2300 ofproto->mirrors[mirror->idx] = NULL;
2304 mirror_update_dups(ofproto);
2306 ofproto->has_mirrors = false;
2307 for (i = 0; i < MAX_MIRRORS; i++) {
2308 if (ofproto->mirrors[i]) {
2309 ofproto->has_mirrors = true;
2316 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2317 uint64_t *packets, uint64_t *bytes)
2319 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2320 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2323 *packets = *bytes = UINT64_MAX;
2327 *packets = mirror->packet_count;
2328 *bytes = mirror->byte_count;
2334 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2336 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2337 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2338 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2344 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2346 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2347 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2348 return bundle && bundle->mirror_out != 0;
2352 forward_bpdu_changed(struct ofproto *ofproto_)
2354 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2355 /* Revalidate cached flows whenever forward_bpdu option changes. */
2356 ofproto->need_revalidate = true;
2360 set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time)
2362 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2363 mac_learning_set_idle_time(ofproto->ml, idle_time);
2368 static struct ofport_dpif *
2369 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2371 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2372 return ofport ? ofport_dpif_cast(ofport) : NULL;
2375 static struct ofport_dpif *
2376 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2378 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2382 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2383 struct dpif_port *dpif_port)
2385 ofproto_port->name = dpif_port->name;
2386 ofproto_port->type = dpif_port->type;
2387 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2391 port_run(struct ofport_dpif *ofport)
2393 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2394 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2395 bool enable = netdev_get_carrier(ofport->up.netdev);
2397 ofport->carrier_seq = carrier_seq;
2400 cfm_run(ofport->cfm);
2402 if (cfm_should_send_ccm(ofport->cfm)) {
2403 struct ofpbuf packet;
2405 ofpbuf_init(&packet, 0);
2406 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2407 send_packet(ofport, &packet);
2408 ofpbuf_uninit(&packet);
2411 enable = enable && !cfm_get_fault(ofport->cfm)
2412 && cfm_get_opup(ofport->cfm);
2415 if (ofport->bundle) {
2416 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2417 if (carrier_changed) {
2418 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2422 if (ofport->may_enable != enable) {
2423 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2425 if (ofproto->has_bundle_action) {
2426 ofproto->need_revalidate = true;
2430 ofport->may_enable = enable;
2434 port_wait(struct ofport_dpif *ofport)
2437 cfm_wait(ofport->cfm);
2442 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2443 struct ofproto_port *ofproto_port)
2445 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2446 struct dpif_port dpif_port;
2449 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2451 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2457 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2459 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2463 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2465 *ofp_portp = odp_port_to_ofp_port(odp_port);
2471 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2473 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2476 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2478 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2480 /* The caller is going to close ofport->up.netdev. If this is a
2481 * bonded port, then the bond is using that netdev, so remove it
2482 * from the bond. The client will need to reconfigure everything
2483 * after deleting ports, so then the slave will get re-added. */
2484 bundle_remove(&ofport->up);
2491 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2493 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2496 error = netdev_get_stats(ofport->up.netdev, stats);
2498 if (!error && ofport->odp_port == OVSP_LOCAL) {
2499 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2501 /* ofproto->stats.tx_packets represents packets that we created
2502 * internally and sent to some port (e.g. packets sent with
2503 * send_packet()). Account for them as if they had come from
2504 * OFPP_LOCAL and got forwarded. */
2506 if (stats->rx_packets != UINT64_MAX) {
2507 stats->rx_packets += ofproto->stats.tx_packets;
2510 if (stats->rx_bytes != UINT64_MAX) {
2511 stats->rx_bytes += ofproto->stats.tx_bytes;
2514 /* ofproto->stats.rx_packets represents packets that were received on
2515 * some port and we processed internally and dropped (e.g. STP).
2516 * Account fro them as if they had been forwarded to OFPP_LOCAL. */
2518 if (stats->tx_packets != UINT64_MAX) {
2519 stats->tx_packets += ofproto->stats.rx_packets;
2522 if (stats->tx_bytes != UINT64_MAX) {
2523 stats->tx_bytes += ofproto->stats.rx_bytes;
2530 /* Account packets for LOCAL port. */
2532 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2533 size_t tx_size, size_t rx_size)
2535 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2538 ofproto->stats.rx_packets++;
2539 ofproto->stats.rx_bytes += rx_size;
2542 ofproto->stats.tx_packets++;
2543 ofproto->stats.tx_bytes += tx_size;
2547 struct port_dump_state {
2548 struct dpif_port_dump dump;
2553 port_dump_start(const struct ofproto *ofproto_, void **statep)
2555 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2556 struct port_dump_state *state;
2558 *statep = state = xmalloc(sizeof *state);
2559 dpif_port_dump_start(&state->dump, ofproto->dpif);
2560 state->done = false;
2565 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2566 struct ofproto_port *port)
2568 struct port_dump_state *state = state_;
2569 struct dpif_port dpif_port;
2571 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2572 ofproto_port_from_dpif_port(port, &dpif_port);
2575 int error = dpif_port_dump_done(&state->dump);
2577 return error ? error : EOF;
2582 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2584 struct port_dump_state *state = state_;
2587 dpif_port_dump_done(&state->dump);
2594 port_poll(const struct ofproto *ofproto_, char **devnamep)
2596 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2597 return dpif_port_poll(ofproto->dpif, devnamep);
2601 port_poll_wait(const struct ofproto *ofproto_)
2603 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2604 dpif_port_poll_wait(ofproto->dpif);
2608 port_is_lacp_current(const struct ofport *ofport_)
2610 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2611 return (ofport->bundle && ofport->bundle->lacp
2612 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2616 /* Upcall handling. */
2618 /* Flow miss batching.
2620 * Some dpifs implement operations faster when you hand them off in a batch.
2621 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2622 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2623 * more packets, plus possibly installing the flow in the dpif.
2625 * So far we only batch the operations that affect flow setup time the most.
2626 * It's possible to batch more than that, but the benefit might be minimal. */
2628 struct hmap_node hmap_node;
2630 enum odp_key_fitness key_fitness;
2631 const struct nlattr *key;
2633 ovs_be16 initial_tci;
2634 struct list packets;
2635 enum dpif_upcall_type upcall_type;
2638 struct flow_miss_op {
2639 struct dpif_op dpif_op;
2640 struct subfacet *subfacet; /* Subfacet */
2641 void *garbage; /* Pointer to pass to free(), NULL if none. */
2642 uint64_t stub[1024 / 8]; /* Temporary buffer. */
2645 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2646 * OpenFlow controller as necessary according to their individual
2647 * configurations. */
2649 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
2650 const struct flow *flow)
2652 struct ofputil_packet_in pin;
2654 pin.packet = packet->data;
2655 pin.packet_len = packet->size;
2656 pin.reason = OFPR_NO_MATCH;
2657 pin.controller_id = 0;
2662 pin.send_len = 0; /* not used for flow table misses */
2664 flow_get_metadata(flow, &pin.fmd);
2666 /* Registers aren't meaningful on a miss. */
2667 memset(pin.fmd.reg_masks, 0, sizeof pin.fmd.reg_masks);
2669 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
2672 static enum slow_path_reason
2673 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2674 const struct ofpbuf *packet)
2676 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2682 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2684 cfm_process_heartbeat(ofport->cfm, packet);
2687 } else if (ofport->bundle && ofport->bundle->lacp
2688 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2690 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2693 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2695 stp_process_packet(ofport, packet);
2702 static struct flow_miss *
2703 flow_miss_find(struct hmap *todo, const struct flow *flow, uint32_t hash)
2705 struct flow_miss *miss;
2707 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2708 if (flow_equal(&miss->flow, flow)) {
2716 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
2717 * 'packet'. The caller must initialize op->actions and op->actions_len. If
2718 * 'miss' is associated with a subfacet the caller must also initialize the
2719 * returned op->subfacet, and if anything needs to be freed after processing
2720 * the op, the caller must initialize op->garbage also. */
2722 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
2723 struct flow_miss_op *op)
2725 if (miss->flow.vlan_tci != miss->initial_tci) {
2726 /* This packet was received on a VLAN splinter port. We
2727 * added a VLAN to the packet to make the packet resemble
2728 * the flow, but the actions were composed assuming that
2729 * the packet contained no VLAN. So, we must remove the
2730 * VLAN header from the packet before trying to execute the
2732 eth_pop_vlan(packet);
2735 op->subfacet = NULL;
2737 op->dpif_op.type = DPIF_OP_EXECUTE;
2738 op->dpif_op.u.execute.key = miss->key;
2739 op->dpif_op.u.execute.key_len = miss->key_len;
2740 op->dpif_op.u.execute.packet = packet;
2743 /* Helper for handle_flow_miss_without_facet() and
2744 * handle_flow_miss_with_facet(). */
2746 handle_flow_miss_common(struct rule_dpif *rule,
2747 struct ofpbuf *packet, const struct flow *flow)
2749 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2751 ofproto->n_matches++;
2753 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2755 * Extra-special case for fail-open mode.
2757 * We are in fail-open mode and the packet matched the fail-open
2758 * rule, but we are connected to a controller too. We should send
2759 * the packet up to the controller in the hope that it will try to
2760 * set up a flow and thereby allow us to exit fail-open.
2762 * See the top-level comment in fail-open.c for more information.
2764 send_packet_in_miss(ofproto, packet, flow);
2768 /* Figures out whether a flow that missed in 'ofproto', whose details are in
2769 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
2770 * installing a datapath flow. The answer is usually "yes" (a return value of
2771 * true). However, for short flows the cost of bookkeeping is much higher than
2772 * the benefits, so when the datapath holds a large number of flows we impose
2773 * some heuristics to decide which flows are likely to be worth tracking. */
2775 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
2776 struct flow_miss *miss, uint32_t hash)
2778 if (!ofproto->governor) {
2781 n_subfacets = hmap_count(&ofproto->subfacets);
2782 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
2786 ofproto->governor = governor_create(ofproto->up.name);
2789 return governor_should_install_flow(ofproto->governor, hash,
2790 list_size(&miss->packets));
2793 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
2794 * or creating any datapath flow. May add an "execute" operation to 'ops' and
2795 * increment '*n_ops'. */
2797 handle_flow_miss_without_facet(struct flow_miss *miss,
2798 struct rule_dpif *rule,
2799 struct flow_miss_op *ops, size_t *n_ops)
2801 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2802 struct action_xlate_ctx ctx;
2803 struct ofpbuf *packet;
2805 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2806 struct flow_miss_op *op = &ops[*n_ops];
2807 struct dpif_flow_stats stats;
2808 struct ofpbuf odp_actions;
2810 COVERAGE_INC(facet_suppress);
2812 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
2814 dpif_flow_stats_extract(&miss->flow, packet, &stats);
2815 rule_credit_stats(rule, &stats);
2817 action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci,
2819 ctx.resubmit_stats = &stats;
2820 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions,
2823 if (odp_actions.size) {
2824 struct dpif_execute *execute = &op->dpif_op.u.execute;
2826 init_flow_miss_execute_op(miss, packet, op);
2827 execute->actions = odp_actions.data;
2828 execute->actions_len = odp_actions.size;
2829 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
2833 ofpbuf_uninit(&odp_actions);
2838 /* Handles 'miss', which matches 'facet'. May add any required datapath
2839 * operations to 'ops', incrementing '*n_ops' for each new op. */
2841 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
2842 struct flow_miss_op *ops, size_t *n_ops)
2844 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
2845 enum subfacet_path want_path;
2846 struct subfacet *subfacet;
2847 struct ofpbuf *packet;
2849 subfacet = subfacet_create(facet,
2850 miss->key_fitness, miss->key, miss->key_len,
2853 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2854 struct flow_miss_op *op = &ops[*n_ops];
2855 struct dpif_flow_stats stats;
2856 struct ofpbuf odp_actions;
2858 handle_flow_miss_common(facet->rule, packet, &miss->flow);
2860 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
2861 if (!subfacet->actions || subfacet->slow) {
2862 subfacet_make_actions(subfacet, packet, &odp_actions);
2865 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2866 subfacet_update_stats(subfacet, &stats);
2868 if (subfacet->actions_len) {
2869 struct dpif_execute *execute = &op->dpif_op.u.execute;
2871 init_flow_miss_execute_op(miss, packet, op);
2872 op->subfacet = subfacet;
2873 if (!subfacet->slow) {
2874 execute->actions = subfacet->actions;
2875 execute->actions_len = subfacet->actions_len;
2876 ofpbuf_uninit(&odp_actions);
2878 execute->actions = odp_actions.data;
2879 execute->actions_len = odp_actions.size;
2880 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
2885 ofpbuf_uninit(&odp_actions);
2889 want_path = subfacet_want_path(subfacet->slow);
2890 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
2891 struct flow_miss_op *op = &ops[(*n_ops)++];
2892 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
2894 op->subfacet = subfacet;
2896 op->dpif_op.type = DPIF_OP_FLOW_PUT;
2897 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2898 put->key = miss->key;
2899 put->key_len = miss->key_len;
2900 if (want_path == SF_FAST_PATH) {
2901 put->actions = subfacet->actions;
2902 put->actions_len = subfacet->actions_len;
2904 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
2905 op->stub, sizeof op->stub,
2906 &put->actions, &put->actions_len);
2912 /* Handles flow miss 'miss' on 'ofproto'. May add any required datapath
2913 * operations to 'ops', incrementing '*n_ops' for each new op. */
2915 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2916 struct flow_miss_op *ops, size_t *n_ops)
2918 struct facet *facet;
2921 /* The caller must ensure that miss->hmap_node.hash contains
2922 * flow_hash(miss->flow, 0). */
2923 hash = miss->hmap_node.hash;
2925 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
2927 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
2929 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
2930 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
2934 facet = facet_create(rule, &miss->flow, hash);
2936 handle_flow_miss_with_facet(miss, facet, ops, n_ops);
2939 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
2940 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
2941 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
2942 * what a flow key should contain.
2944 * This function also includes some logic to help make VLAN splinters
2945 * transparent to the rest of the upcall processing logic. In particular, if
2946 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
2947 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
2948 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
2950 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
2951 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
2952 * (This differs from the value returned in flow->vlan_tci only for packets
2953 * received on VLAN splinters.)
2955 static enum odp_key_fitness
2956 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
2957 const struct nlattr *key, size_t key_len,
2958 struct flow *flow, ovs_be16 *initial_tci,
2959 struct ofpbuf *packet)
2961 enum odp_key_fitness fitness;
2963 fitness = odp_flow_key_to_flow(key, key_len, flow);
2964 if (fitness == ODP_FIT_ERROR) {
2967 *initial_tci = flow->vlan_tci;
2969 if (vsp_adjust_flow(ofproto, flow)) {
2971 /* Make the packet resemble the flow, so that it gets sent to an
2972 * OpenFlow controller properly, so that it looks correct for
2973 * sFlow, and so that flow_extract() will get the correct vlan_tci
2974 * if it is called on 'packet'.
2976 * The allocated space inside 'packet' probably also contains
2977 * 'key', that is, both 'packet' and 'key' are probably part of a
2978 * struct dpif_upcall (see the large comment on that structure
2979 * definition), so pushing data on 'packet' is in general not a
2980 * good idea since it could overwrite 'key' or free it as a side
2981 * effect. However, it's OK in this special case because we know
2982 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
2983 * will just overwrite the 4-byte "struct nlattr", which is fine
2984 * since we don't need that header anymore. */
2985 eth_push_vlan(packet, flow->vlan_tci);
2988 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
2989 if (fitness == ODP_FIT_PERFECT) {
2990 fitness = ODP_FIT_TOO_MUCH;
2998 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
3001 struct dpif_upcall *upcall;
3002 struct flow_miss *miss;
3003 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3004 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3005 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3015 /* Construct the to-do list.
3017 * This just amounts to extracting the flow from each packet and sticking
3018 * the packets that have the same flow in the same "flow_miss" structure so
3019 * that we can process them together. */
3022 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3023 struct flow_miss *miss = &misses[n_misses];
3024 struct flow_miss *existing_miss;
3027 /* Obtain metadata and check userspace/kernel agreement on flow match,
3028 * then set 'flow''s header pointers. */
3029 miss->key_fitness = ofproto_dpif_extract_flow_key(
3030 ofproto, upcall->key, upcall->key_len,
3031 &miss->flow, &miss->initial_tci, upcall->packet);
3032 if (miss->key_fitness == ODP_FIT_ERROR) {
3035 flow_extract(upcall->packet, miss->flow.skb_priority,
3036 miss->flow.tun_id, miss->flow.in_port, &miss->flow);
3038 /* Add other packets to a to-do list. */
3039 hash = flow_hash(&miss->flow, 0);
3040 existing_miss = flow_miss_find(&todo, &miss->flow, hash);
3041 if (!existing_miss) {
3042 hmap_insert(&todo, &miss->hmap_node, hash);
3043 miss->key = upcall->key;
3044 miss->key_len = upcall->key_len;
3045 miss->upcall_type = upcall->type;
3046 list_init(&miss->packets);
3050 miss = existing_miss;
3052 list_push_back(&miss->packets, &upcall->packet->list_node);
3055 /* Process each element in the to-do list, constructing the set of
3056 * operations to batch. */
3058 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3059 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
3061 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3063 /* Execute batch. */
3064 for (i = 0; i < n_ops; i++) {
3065 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3067 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
3069 /* Free memory and update facets. */
3070 for (i = 0; i < n_ops; i++) {
3071 struct flow_miss_op *op = &flow_miss_ops[i];
3073 switch (op->dpif_op.type) {
3074 case DPIF_OP_EXECUTE:
3077 case DPIF_OP_FLOW_PUT:
3078 if (!op->dpif_op.error) {
3079 op->subfacet->path = subfacet_want_path(op->subfacet->slow);
3083 case DPIF_OP_FLOW_DEL:
3089 hmap_destroy(&todo);
3092 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
3093 classify_upcall(const struct dpif_upcall *upcall)
3095 union user_action_cookie cookie;
3097 /* First look at the upcall type. */
3098 switch (upcall->type) {
3099 case DPIF_UC_ACTION:
3105 case DPIF_N_UC_TYPES:
3107 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3111 /* "action" upcalls need a closer look. */
3112 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3113 switch (cookie.type) {
3114 case USER_ACTION_COOKIE_SFLOW:
3115 return SFLOW_UPCALL;
3117 case USER_ACTION_COOKIE_SLOW_PATH:
3120 case USER_ACTION_COOKIE_UNSPEC:
3122 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
3128 handle_sflow_upcall(struct ofproto_dpif *ofproto,
3129 const struct dpif_upcall *upcall)
3131 union user_action_cookie cookie;
3132 enum odp_key_fitness fitness;
3133 ovs_be16 initial_tci;
3136 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
3137 upcall->key_len, &flow,
3138 &initial_tci, upcall->packet);
3139 if (fitness == ODP_FIT_ERROR) {
3143 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3144 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, &cookie);
3148 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
3150 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3151 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3152 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3157 assert(max_batch <= FLOW_MISS_MAX_BATCH);
3160 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3161 struct dpif_upcall *upcall = &misses[n_misses];
3162 struct ofpbuf *buf = &miss_bufs[n_misses];
3165 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3166 sizeof miss_buf_stubs[n_misses]);
3167 error = dpif_recv(ofproto->dpif, upcall, buf);
3173 switch (classify_upcall(upcall)) {
3175 /* Handle it later. */
3180 if (ofproto->sflow) {
3181 handle_sflow_upcall(ofproto, upcall);
3192 /* Handle deferred MISS_UPCALL processing. */
3193 handle_miss_upcalls(ofproto, misses, n_misses);
3194 for (i = 0; i < n_misses; i++) {
3195 ofpbuf_uninit(&miss_bufs[i]);
3201 /* Flow expiration. */
3203 static int subfacet_max_idle(const struct ofproto_dpif *);
3204 static void update_stats(struct ofproto_dpif *);
3205 static void rule_expire(struct rule_dpif *);
3206 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
3208 /* This function is called periodically by run(). Its job is to collect
3209 * updates for the flows that have been installed into the datapath, most
3210 * importantly when they last were used, and then use that information to
3211 * expire flows that have not been used recently.
3213 * Returns the number of milliseconds after which it should be called again. */
3215 expire(struct ofproto_dpif *ofproto)
3217 struct rule_dpif *rule, *next_rule;
3218 struct oftable *table;
3221 /* Update stats for each flow in the datapath. */
3222 update_stats(ofproto);
3224 /* Expire subfacets that have been idle too long. */
3225 dp_max_idle = subfacet_max_idle(ofproto);
3226 expire_subfacets(ofproto, dp_max_idle);
3228 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
3229 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
3230 struct cls_cursor cursor;
3232 cls_cursor_init(&cursor, &table->cls, NULL);
3233 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
3238 /* All outstanding data in existing flows has been accounted, so it's a
3239 * good time to do bond rebalancing. */
3240 if (ofproto->has_bonded_bundles) {
3241 struct ofbundle *bundle;
3243 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3245 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
3250 return MIN(dp_max_idle, 1000);
3253 /* Updates flow table statistics given that the datapath just reported 'stats'
3254 * as 'subfacet''s statistics. */
3256 update_subfacet_stats(struct subfacet *subfacet,
3257 const struct dpif_flow_stats *stats)
3259 struct facet *facet = subfacet->facet;
3261 if (stats->n_packets >= subfacet->dp_packet_count) {
3262 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
3263 facet->packet_count += extra;
3265 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3268 if (stats->n_bytes >= subfacet->dp_byte_count) {
3269 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
3271 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3274 subfacet->dp_packet_count = stats->n_packets;
3275 subfacet->dp_byte_count = stats->n_bytes;
3277 facet->tcp_flags |= stats->tcp_flags;
3279 subfacet_update_time(subfacet, stats->used);
3280 if (facet->accounted_bytes < facet->byte_count) {
3282 facet_account(facet);
3283 facet->accounted_bytes = facet->byte_count;
3285 facet_push_stats(facet);
3288 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3289 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3291 delete_unexpected_flow(struct dpif *dpif,
3292 const struct nlattr *key, size_t key_len)
3294 if (!VLOG_DROP_WARN(&rl)) {
3298 odp_flow_key_format(key, key_len, &s);
3299 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
3303 COVERAGE_INC(facet_unexpected);
3304 dpif_flow_del(dpif, key, key_len, NULL);
3307 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3309 * This function also pushes statistics updates to rules which each facet
3310 * resubmits into. Generally these statistics will be accurate. However, if a
3311 * facet changes the rule it resubmits into at some time in between
3312 * update_stats() runs, it is possible that statistics accrued to the
3313 * old rule will be incorrectly attributed to the new rule. This could be
3314 * avoided by calling update_stats() whenever rules are created or
3315 * deleted. However, the performance impact of making so many calls to the
3316 * datapath do not justify the benefit of having perfectly accurate statistics.
3319 update_stats(struct ofproto_dpif *p)
3321 const struct dpif_flow_stats *stats;
3322 struct dpif_flow_dump dump;
3323 const struct nlattr *key;
3326 dpif_flow_dump_start(&dump, p->dpif);
3327 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
3328 struct subfacet *subfacet;
3330 subfacet = subfacet_find(p, key, key_len);
3331 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3333 update_subfacet_stats(subfacet, stats);
3337 /* Stats are updated per-packet. */
3340 case SF_NOT_INSTALLED:
3342 delete_unexpected_flow(p->dpif, key, key_len);
3346 dpif_flow_dump_done(&dump);
3349 /* Calculates and returns the number of milliseconds of idle time after which
3350 * subfacets should expire from the datapath. When a subfacet expires, we fold
3351 * its statistics into its facet, and when a facet's last subfacet expires, we
3352 * fold its statistic into its rule. */
3354 subfacet_max_idle(const struct ofproto_dpif *ofproto)
3357 * Idle time histogram.
3359 * Most of the time a switch has a relatively small number of subfacets.
3360 * When this is the case we might as well keep statistics for all of them
3361 * in userspace and to cache them in the kernel datapath for performance as
3364 * As the number of subfacets increases, the memory required to maintain
3365 * statistics about them in userspace and in the kernel becomes
3366 * significant. However, with a large number of subfacets it is likely
3367 * that only a few of them are "heavy hitters" that consume a large amount
3368 * of bandwidth. At this point, only heavy hitters are worth caching in
3369 * the kernel and maintaining in userspaces; other subfacets we can
3372 * The technique used to compute the idle time is to build a histogram with
3373 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3374 * that is installed in the kernel gets dropped in the appropriate bucket.
3375 * After the histogram has been built, we compute the cutoff so that only
3376 * the most-recently-used 1% of subfacets (but at least
3377 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3378 * the most-recently-used bucket of subfacets is kept, so actually an
3379 * arbitrary number of subfacets can be kept in any given expiration run
3380 * (though the next run will delete most of those unless they receive
3383 * This requires a second pass through the subfacets, in addition to the
3384 * pass made by update_stats(), because the former function never looks at
3385 * uninstallable subfacets.
3387 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3388 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3389 int buckets[N_BUCKETS] = { 0 };
3390 int total, subtotal, bucket;
3391 struct subfacet *subfacet;
3395 total = hmap_count(&ofproto->subfacets);
3396 if (total <= ofproto->up.flow_eviction_threshold) {
3397 return N_BUCKETS * BUCKET_WIDTH;
3400 /* Build histogram. */
3402 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3403 long long int idle = now - subfacet->used;
3404 int bucket = (idle <= 0 ? 0
3405 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3406 : (unsigned int) idle / BUCKET_WIDTH);
3410 /* Find the first bucket whose flows should be expired. */
3411 subtotal = bucket = 0;
3413 subtotal += buckets[bucket++];
3414 } while (bucket < N_BUCKETS &&
3415 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3417 if (VLOG_IS_DBG_ENABLED()) {
3421 ds_put_cstr(&s, "keep");
3422 for (i = 0; i < N_BUCKETS; i++) {
3424 ds_put_cstr(&s, ", drop");
3427 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3430 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3434 return bucket * BUCKET_WIDTH;
3437 enum { EXPIRE_MAX_BATCH = 50 };
3440 expire_batch(struct ofproto_dpif *ofproto, struct subfacet **subfacets, int n)
3442 struct odputil_keybuf keybufs[EXPIRE_MAX_BATCH];
3443 struct dpif_op ops[EXPIRE_MAX_BATCH];
3444 struct dpif_op *opsp[EXPIRE_MAX_BATCH];
3445 struct ofpbuf keys[EXPIRE_MAX_BATCH];
3446 struct dpif_flow_stats stats[EXPIRE_MAX_BATCH];
3449 for (i = 0; i < n; i++) {
3450 ops[i].type = DPIF_OP_FLOW_DEL;
3451 subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]);
3452 ops[i].u.flow_del.key = keys[i].data;
3453 ops[i].u.flow_del.key_len = keys[i].size;
3454 ops[i].u.flow_del.stats = &stats[i];
3458 dpif_operate(ofproto->dpif, opsp, n);
3459 for (i = 0; i < n; i++) {
3460 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
3461 subfacets[i]->path = SF_NOT_INSTALLED;
3462 subfacet_destroy(subfacets[i]);
3467 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3469 /* Cutoff time for most flows. */
3470 long long int normal_cutoff = time_msec() - dp_max_idle;
3472 /* We really want to keep flows for special protocols around, so use a more
3473 * conservative cutoff. */
3474 long long int special_cutoff = time_msec() - 10000;
3476 struct subfacet *subfacet, *next_subfacet;
3477 struct subfacet *batch[EXPIRE_MAX_BATCH];
3481 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3482 &ofproto->subfacets) {
3483 long long int cutoff;
3485 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
3488 if (subfacet->used < cutoff) {
3489 if (subfacet->path != SF_NOT_INSTALLED) {
3490 batch[n_batch++] = subfacet;
3491 if (n_batch >= EXPIRE_MAX_BATCH) {
3492 expire_batch(ofproto, batch, n_batch);
3496 subfacet_destroy(subfacet);
3502 expire_batch(ofproto, batch, n_batch);
3506 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3507 * then delete it entirely. */
3509 rule_expire(struct rule_dpif *rule)
3511 struct facet *facet, *next_facet;
3515 /* Has 'rule' expired? */
3517 if (rule->up.hard_timeout
3518 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3519 reason = OFPRR_HARD_TIMEOUT;
3520 } else if (rule->up.idle_timeout
3521 && now > rule->up.used + rule->up.idle_timeout * 1000) {
3522 reason = OFPRR_IDLE_TIMEOUT;
3527 COVERAGE_INC(ofproto_dpif_expired);
3529 /* Update stats. (This is a no-op if the rule expired due to an idle
3530 * timeout, because that only happens when the rule has no facets left.) */
3531 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3532 facet_remove(facet);
3535 /* Get rid of the rule. */
3536 ofproto_rule_expire(&rule->up, reason);
3541 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3543 * The caller must already have determined that no facet with an identical
3544 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3545 * the ofproto's classifier table.
3547 * 'hash' must be the return value of flow_hash(flow, 0).
3549 * The facet will initially have no subfacets. The caller should create (at
3550 * least) one subfacet with subfacet_create(). */
3551 static struct facet *
3552 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
3554 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3555 struct facet *facet;
3557 facet = xzalloc(sizeof *facet);
3558 facet->used = time_msec();
3559 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
3560 list_push_back(&rule->facets, &facet->list_node);
3562 facet->flow = *flow;
3563 list_init(&facet->subfacets);
3564 netflow_flow_init(&facet->nf_flow);
3565 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3571 facet_free(struct facet *facet)
3576 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3577 * 'packet', which arrived on 'in_port'.
3579 * Takes ownership of 'packet'. */
3581 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3582 const struct nlattr *odp_actions, size_t actions_len,
3583 struct ofpbuf *packet)
3585 struct odputil_keybuf keybuf;
3589 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3590 odp_flow_key_from_flow(&key, flow);
3592 error = dpif_execute(ofproto->dpif, key.data, key.size,
3593 odp_actions, actions_len, packet);
3595 ofpbuf_delete(packet);
3599 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3601 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3602 * rule's statistics, via subfacet_uninstall().
3604 * - Removes 'facet' from its rule and from ofproto->facets.
3607 facet_remove(struct facet *facet)
3609 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3610 struct subfacet *subfacet, *next_subfacet;
3612 assert(!list_is_empty(&facet->subfacets));
3614 /* First uninstall all of the subfacets to get final statistics. */
3615 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3616 subfacet_uninstall(subfacet);
3619 /* Flush the final stats to the rule.
3621 * This might require us to have at least one subfacet around so that we
3622 * can use its actions for accounting in facet_account(), which is why we
3623 * have uninstalled but not yet destroyed the subfacets. */
3624 facet_flush_stats(facet);
3626 /* Now we're really all done so destroy everything. */
3627 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3628 &facet->subfacets) {
3629 subfacet_destroy__(subfacet);
3631 hmap_remove(&ofproto->facets, &facet->hmap_node);
3632 list_remove(&facet->list_node);
3636 /* Feed information from 'facet' back into the learning table to keep it in
3637 * sync with what is actually flowing through the datapath. */
3639 facet_learn(struct facet *facet)
3641 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3642 struct action_xlate_ctx ctx;
3644 if (!facet->has_learn
3645 && !facet->has_normal
3646 && (!facet->has_fin_timeout
3647 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
3651 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3652 facet->flow.vlan_tci,
3653 facet->rule, facet->tcp_flags, NULL);
3654 ctx.may_learn = true;
3655 xlate_actions_for_side_effects(&ctx, facet->rule->up.actions,
3656 facet->rule->up.n_actions);
3660 facet_account(struct facet *facet)
3662 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3663 struct subfacet *subfacet;
3664 const struct nlattr *a;
3669 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3672 n_bytes = facet->byte_count - facet->accounted_bytes;
3674 /* This loop feeds byte counters to bond_account() for rebalancing to use
3675 * as a basis. We also need to track the actual VLAN on which the packet
3676 * is going to be sent to ensure that it matches the one passed to
3677 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3680 * We use the actions from an arbitrary subfacet because they should all
3681 * be equally valid for our purpose. */
3682 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3683 struct subfacet, list_node);
3684 vlan_tci = facet->flow.vlan_tci;
3685 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3686 subfacet->actions, subfacet->actions_len) {
3687 const struct ovs_action_push_vlan *vlan;
3688 struct ofport_dpif *port;
3690 switch (nl_attr_type(a)) {
3691 case OVS_ACTION_ATTR_OUTPUT:
3692 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3693 if (port && port->bundle && port->bundle->bond) {
3694 bond_account(port->bundle->bond, &facet->flow,
3695 vlan_tci_to_vid(vlan_tci), n_bytes);
3699 case OVS_ACTION_ATTR_POP_VLAN:
3700 vlan_tci = htons(0);
3703 case OVS_ACTION_ATTR_PUSH_VLAN:
3704 vlan = nl_attr_get(a);
3705 vlan_tci = vlan->vlan_tci;
3711 /* Returns true if the only action for 'facet' is to send to the controller.
3712 * (We don't report NetFlow expiration messages for such facets because they
3713 * are just part of the control logic for the network, not real traffic). */
3715 facet_is_controller_flow(struct facet *facet)
3718 && facet->rule->up.n_actions == 1
3719 && action_outputs_to_port(&facet->rule->up.actions[0],
3720 htons(OFPP_CONTROLLER)));
3723 /* Folds all of 'facet''s statistics into its rule. Also updates the
3724 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3725 * 'facet''s statistics in the datapath should have been zeroed and folded into
3726 * its packet and byte counts before this function is called. */
3728 facet_flush_stats(struct facet *facet)
3730 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3731 struct subfacet *subfacet;
3733 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3734 assert(!subfacet->dp_byte_count);
3735 assert(!subfacet->dp_packet_count);
3738 facet_push_stats(facet);
3739 if (facet->accounted_bytes < facet->byte_count) {
3740 facet_account(facet);
3741 facet->accounted_bytes = facet->byte_count;
3744 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3745 struct ofexpired expired;
3746 expired.flow = facet->flow;
3747 expired.packet_count = facet->packet_count;
3748 expired.byte_count = facet->byte_count;
3749 expired.used = facet->used;
3750 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3753 facet->rule->packet_count += facet->packet_count;
3754 facet->rule->byte_count += facet->byte_count;
3756 /* Reset counters to prevent double counting if 'facet' ever gets
3758 facet_reset_counters(facet);
3760 netflow_flow_clear(&facet->nf_flow);
3761 facet->tcp_flags = 0;
3764 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3765 * Returns it if found, otherwise a null pointer.
3767 * 'hash' must be the return value of flow_hash(flow, 0).
3769 * The returned facet might need revalidation; use facet_lookup_valid()
3770 * instead if that is important. */
3771 static struct facet *
3772 facet_find(struct ofproto_dpif *ofproto,
3773 const struct flow *flow, uint32_t hash)
3775 struct facet *facet;
3777 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
3778 if (flow_equal(flow, &facet->flow)) {
3786 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3787 * Returns it if found, otherwise a null pointer.
3789 * 'hash' must be the return value of flow_hash(flow, 0).
3791 * The returned facet is guaranteed to be valid. */
3792 static struct facet *
3793 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
3796 struct facet *facet;
3798 facet = facet_find(ofproto, flow, hash);
3800 && (ofproto->need_revalidate
3801 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))) {
3802 facet_revalidate(facet);
3809 subfacet_path_to_string(enum subfacet_path path)
3812 case SF_NOT_INSTALLED:
3813 return "not installed";
3815 return "in fast path";
3817 return "in slow path";
3823 /* Returns the path in which a subfacet should be installed if its 'slow'
3824 * member has the specified value. */
3825 static enum subfacet_path
3826 subfacet_want_path(enum slow_path_reason slow)
3828 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
3831 /* Returns true if 'subfacet' needs to have its datapath flow updated,
3832 * supposing that its actions have been recalculated as 'want_actions' and that
3833 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
3835 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
3836 const struct ofpbuf *want_actions)
3838 enum subfacet_path want_path = subfacet_want_path(slow);
3839 return (want_path != subfacet->path
3840 || (want_path == SF_FAST_PATH
3841 && (subfacet->actions_len != want_actions->size
3842 || memcmp(subfacet->actions, want_actions->data,
3843 subfacet->actions_len))));
3847 facet_check_consistency(struct facet *facet)
3849 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3851 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3853 uint64_t odp_actions_stub[1024 / 8];
3854 struct ofpbuf odp_actions;
3856 struct rule_dpif *rule;
3857 struct subfacet *subfacet;
3858 bool may_log = false;
3861 /* Check the rule for consistency. */
3862 rule = rule_dpif_lookup(ofproto, &facet->flow);
3863 ok = rule == facet->rule;
3865 may_log = !VLOG_DROP_WARN(&rl);
3870 flow_format(&s, &facet->flow);
3871 ds_put_format(&s, ": facet associated with wrong rule (was "
3872 "table=%"PRIu8",", facet->rule->up.table_id);
3873 cls_rule_format(&facet->rule->up.cr, &s);
3874 ds_put_format(&s, ") (should have been table=%"PRIu8",",
3876 cls_rule_format(&rule->up.cr, &s);
3877 ds_put_char(&s, ')');
3879 VLOG_WARN("%s", ds_cstr(&s));
3884 /* Check the datapath actions for consistency. */
3885 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
3886 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3887 enum subfacet_path want_path;
3888 struct odputil_keybuf keybuf;
3889 struct action_xlate_ctx ctx;
3893 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3894 subfacet->initial_tci, rule, 0, NULL);
3895 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions,
3898 if (subfacet->path == SF_NOT_INSTALLED) {
3899 /* This only happens if the datapath reported an error when we
3900 * tried to install the flow. Don't flag another error here. */
3904 want_path = subfacet_want_path(subfacet->slow);
3905 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
3906 /* The actions for slow-path flows may legitimately vary from one
3907 * packet to the next. We're done. */
3911 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
3915 /* Inconsistency! */
3917 may_log = !VLOG_DROP_WARN(&rl);
3921 /* Rate-limited, skip reporting. */
3926 subfacet_get_key(subfacet, &keybuf, &key);
3927 odp_flow_key_format(key.data, key.size, &s);
3929 ds_put_cstr(&s, ": inconsistency in subfacet");
3930 if (want_path != subfacet->path) {
3931 enum odp_key_fitness fitness = subfacet->key_fitness;
3933 ds_put_format(&s, " (%s, fitness=%s)",
3934 subfacet_path_to_string(subfacet->path),
3935 odp_key_fitness_to_string(fitness));
3936 ds_put_format(&s, " (should have been %s)",
3937 subfacet_path_to_string(want_path));
3938 } else if (want_path == SF_FAST_PATH) {
3939 ds_put_cstr(&s, " (actions were: ");
3940 format_odp_actions(&s, subfacet->actions,
3941 subfacet->actions_len);
3942 ds_put_cstr(&s, ") (correct actions: ");
3943 format_odp_actions(&s, odp_actions.data, odp_actions.size);
3944 ds_put_char(&s, ')');
3946 ds_put_cstr(&s, " (actions: ");
3947 format_odp_actions(&s, subfacet->actions,
3948 subfacet->actions_len);
3949 ds_put_char(&s, ')');
3951 VLOG_WARN("%s", ds_cstr(&s));
3954 ofpbuf_uninit(&odp_actions);
3959 /* Re-searches the classifier for 'facet':
3961 * - If the rule found is different from 'facet''s current rule, moves
3962 * 'facet' to the new rule and recompiles its actions.
3964 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3965 * where it is and recompiles its actions anyway. */
3967 facet_revalidate(struct facet *facet)
3969 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3971 struct nlattr *odp_actions;
3974 struct actions *new_actions;
3976 struct action_xlate_ctx ctx;
3977 uint64_t odp_actions_stub[1024 / 8];
3978 struct ofpbuf odp_actions;
3980 struct rule_dpif *new_rule;
3981 struct subfacet *subfacet;
3984 COVERAGE_INC(facet_revalidate);
3986 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
3988 /* Calculate new datapath actions.
3990 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3991 * emit a NetFlow expiration and, if so, we need to have the old state
3992 * around to properly compose it. */
3994 /* If the datapath actions changed or the installability changed,
3995 * then we need to talk to the datapath. */
3998 memset(&ctx, 0, sizeof ctx);
3999 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4000 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4001 enum slow_path_reason slow;
4003 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4004 subfacet->initial_tci, new_rule, 0, NULL);
4005 xlate_actions(&ctx, new_rule->up.actions, new_rule->up.n_actions,
4008 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4009 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
4010 struct dpif_flow_stats stats;
4012 subfacet_install(subfacet,
4013 odp_actions.data, odp_actions.size, &stats, slow);
4014 subfacet_update_stats(subfacet, &stats);
4017 new_actions = xcalloc(list_size(&facet->subfacets),
4018 sizeof *new_actions);
4020 new_actions[i].odp_actions = xmemdup(odp_actions.data,
4022 new_actions[i].actions_len = odp_actions.size;
4027 ofpbuf_uninit(&odp_actions);
4030 facet_flush_stats(facet);
4033 /* Update 'facet' now that we've taken care of all the old state. */
4034 facet->tags = ctx.tags;
4035 facet->nf_flow.output_iface = ctx.nf_output_iface;
4036 facet->has_learn = ctx.has_learn;
4037 facet->has_normal = ctx.has_normal;
4038 facet->has_fin_timeout = ctx.has_fin_timeout;
4039 facet->mirrors = ctx.mirrors;
4042 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4043 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4045 if (new_actions && new_actions[i].odp_actions) {
4046 free(subfacet->actions);
4047 subfacet->actions = new_actions[i].odp_actions;
4048 subfacet->actions_len = new_actions[i].actions_len;
4054 if (facet->rule != new_rule) {
4055 COVERAGE_INC(facet_changed_rule);
4056 list_remove(&facet->list_node);
4057 list_push_back(&new_rule->facets, &facet->list_node);
4058 facet->rule = new_rule;
4059 facet->used = new_rule->up.created;
4060 facet->prev_used = facet->used;
4064 /* Updates 'facet''s used time. Caller is responsible for calling
4065 * facet_push_stats() to update the flows which 'facet' resubmits into. */
4067 facet_update_time(struct facet *facet, long long int used)
4069 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4070 if (used > facet->used) {
4072 ofproto_rule_update_used(&facet->rule->up, used);
4073 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
4078 facet_reset_counters(struct facet *facet)
4080 facet->packet_count = 0;
4081 facet->byte_count = 0;
4082 facet->prev_packet_count = 0;
4083 facet->prev_byte_count = 0;
4084 facet->accounted_bytes = 0;
4088 facet_push_stats(struct facet *facet)
4090 struct dpif_flow_stats stats;
4092 assert(facet->packet_count >= facet->prev_packet_count);
4093 assert(facet->byte_count >= facet->prev_byte_count);
4094 assert(facet->used >= facet->prev_used);
4096 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4097 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4098 stats.used = facet->used;
4099 stats.tcp_flags = 0;
4101 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
4102 facet->prev_packet_count = facet->packet_count;
4103 facet->prev_byte_count = facet->byte_count;
4104 facet->prev_used = facet->used;
4106 flow_push_stats(facet->rule, &facet->flow, &stats);
4108 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
4109 facet->mirrors, stats.n_packets, stats.n_bytes);
4114 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4116 rule->packet_count += stats->n_packets;
4117 rule->byte_count += stats->n_bytes;
4118 ofproto_rule_update_used(&rule->up, stats->used);
4121 /* Pushes flow statistics to the rules which 'flow' resubmits into given
4122 * 'rule''s actions and mirrors. */
4124 flow_push_stats(struct rule_dpif *rule,
4125 const struct flow *flow, const struct dpif_flow_stats *stats)
4127 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4128 struct action_xlate_ctx ctx;
4130 ofproto_rule_update_used(&rule->up, stats->used);
4132 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule,
4134 ctx.resubmit_stats = stats;
4135 xlate_actions_for_side_effects(&ctx, rule->up.actions, rule->up.n_actions);
4140 static struct subfacet *
4141 subfacet_find__(struct ofproto_dpif *ofproto,
4142 const struct nlattr *key, size_t key_len, uint32_t key_hash,
4143 const struct flow *flow)
4145 struct subfacet *subfacet;
4147 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4148 &ofproto->subfacets) {
4150 ? (subfacet->key_len == key_len
4151 && !memcmp(key, subfacet->key, key_len))
4152 : flow_equal(flow, &subfacet->facet->flow)) {
4160 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4161 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
4162 * there is one, otherwise creates and returns a new subfacet.
4164 * If the returned subfacet is new, then subfacet->actions will be NULL, in
4165 * which case the caller must populate the actions with
4166 * subfacet_make_actions(). */
4167 static struct subfacet *
4168 subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness,
4169 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
4171 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4172 uint32_t key_hash = odp_flow_key_hash(key, key_len);
4173 struct subfacet *subfacet;
4175 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
4177 if (subfacet->facet == facet) {
4181 /* This shouldn't happen. */
4182 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4183 subfacet_destroy(subfacet);
4186 subfacet = (list_is_empty(&facet->subfacets)
4187 ? &facet->one_subfacet
4188 : xmalloc(sizeof *subfacet));
4189 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
4190 list_push_back(&facet->subfacets, &subfacet->list_node);
4191 subfacet->facet = facet;
4192 subfacet->key_fitness = key_fitness;
4193 if (key_fitness != ODP_FIT_PERFECT) {
4194 subfacet->key = xmemdup(key, key_len);
4195 subfacet->key_len = key_len;
4197 subfacet->key = NULL;
4198 subfacet->key_len = 0;
4200 subfacet->used = time_msec();
4201 subfacet->dp_packet_count = 0;
4202 subfacet->dp_byte_count = 0;
4203 subfacet->actions_len = 0;
4204 subfacet->actions = NULL;
4205 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
4208 subfacet->path = SF_NOT_INSTALLED;
4209 subfacet->initial_tci = initial_tci;
4214 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
4215 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
4216 static struct subfacet *
4217 subfacet_find(struct ofproto_dpif *ofproto,
4218 const struct nlattr *key, size_t key_len)
4220 uint32_t key_hash = odp_flow_key_hash(key, key_len);
4221 enum odp_key_fitness fitness;
4224 fitness = odp_flow_key_to_flow(key, key_len, &flow);
4225 if (fitness == ODP_FIT_ERROR) {
4229 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
4232 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4233 * its facet within 'ofproto', and frees it. */
4235 subfacet_destroy__(struct subfacet *subfacet)
4237 struct facet *facet = subfacet->facet;
4238 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4240 subfacet_uninstall(subfacet);
4241 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
4242 list_remove(&subfacet->list_node);
4243 free(subfacet->key);
4244 free(subfacet->actions);
4245 if (subfacet != &facet->one_subfacet) {
4250 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4251 * last remaining subfacet in its facet destroys the facet too. */
4253 subfacet_destroy(struct subfacet *subfacet)
4255 struct facet *facet = subfacet->facet;
4257 if (list_is_singleton(&facet->subfacets)) {
4258 /* facet_remove() needs at least one subfacet (it will remove it). */
4259 facet_remove(facet);
4261 subfacet_destroy__(subfacet);
4265 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
4266 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
4267 * for use as temporary storage. */
4269 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
4272 if (!subfacet->key) {
4273 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
4274 odp_flow_key_from_flow(key, &subfacet->facet->flow);
4276 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
4280 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
4281 * Translates the actions into 'odp_actions', which the caller must have
4282 * initialized and is responsible for uninitializing. */
4284 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
4285 struct ofpbuf *odp_actions)
4287 struct facet *facet = subfacet->facet;
4288 struct rule_dpif *rule = facet->rule;
4289 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4291 struct action_xlate_ctx ctx;
4293 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
4295 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions, odp_actions);
4296 facet->tags = ctx.tags;
4297 facet->has_learn = ctx.has_learn;
4298 facet->has_normal = ctx.has_normal;
4299 facet->has_fin_timeout = ctx.has_fin_timeout;
4300 facet->nf_flow.output_iface = ctx.nf_output_iface;
4301 facet->mirrors = ctx.mirrors;
4303 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4304 if (subfacet->actions_len != odp_actions->size
4305 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
4306 free(subfacet->actions);
4307 subfacet->actions_len = odp_actions->size;
4308 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
4312 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4313 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4314 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4315 * since 'subfacet' was last updated.
4317 * Returns 0 if successful, otherwise a positive errno value. */
4319 subfacet_install(struct subfacet *subfacet,
4320 const struct nlattr *actions, size_t actions_len,
4321 struct dpif_flow_stats *stats,
4322 enum slow_path_reason slow)
4324 struct facet *facet = subfacet->facet;
4325 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4326 enum subfacet_path path = subfacet_want_path(slow);
4327 uint64_t slow_path_stub[128 / 8];
4328 struct odputil_keybuf keybuf;
4329 enum dpif_flow_put_flags flags;
4333 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
4335 flags |= DPIF_FP_ZERO_STATS;
4338 if (path == SF_SLOW_PATH) {
4339 compose_slow_path(ofproto, &facet->flow, slow,
4340 slow_path_stub, sizeof slow_path_stub,
4341 &actions, &actions_len);
4344 subfacet_get_key(subfacet, &keybuf, &key);
4345 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
4346 actions, actions_len, stats);
4349 subfacet_reset_dp_stats(subfacet, stats);
4353 subfacet->path = path;
4359 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
4361 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
4362 stats, subfacet->slow);
4365 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4367 subfacet_uninstall(struct subfacet *subfacet)
4369 if (subfacet->path != SF_NOT_INSTALLED) {
4370 struct rule_dpif *rule = subfacet->facet->rule;
4371 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4372 struct odputil_keybuf keybuf;
4373 struct dpif_flow_stats stats;
4377 subfacet_get_key(subfacet, &keybuf, &key);
4378 error = dpif_flow_del(ofproto->dpif, key.data, key.size, &stats);
4379 subfacet_reset_dp_stats(subfacet, &stats);
4381 subfacet_update_stats(subfacet, &stats);
4383 subfacet->path = SF_NOT_INSTALLED;
4385 assert(subfacet->dp_packet_count == 0);
4386 assert(subfacet->dp_byte_count == 0);
4390 /* Resets 'subfacet''s datapath statistics counters. This should be called
4391 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4392 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4393 * was reset in the datapath. 'stats' will be modified to include only
4394 * statistics new since 'subfacet' was last updated. */
4396 subfacet_reset_dp_stats(struct subfacet *subfacet,
4397 struct dpif_flow_stats *stats)
4400 && subfacet->dp_packet_count <= stats->n_packets
4401 && subfacet->dp_byte_count <= stats->n_bytes) {
4402 stats->n_packets -= subfacet->dp_packet_count;
4403 stats->n_bytes -= subfacet->dp_byte_count;
4406 subfacet->dp_packet_count = 0;
4407 subfacet->dp_byte_count = 0;
4410 /* Updates 'subfacet''s used time. The caller is responsible for calling
4411 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
4413 subfacet_update_time(struct subfacet *subfacet, long long int used)
4415 if (used > subfacet->used) {
4416 subfacet->used = used;
4417 facet_update_time(subfacet->facet, used);
4421 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4423 * Because of the meaning of a subfacet's counters, it only makes sense to do
4424 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4425 * represents a packet that was sent by hand or if it represents statistics
4426 * that have been cleared out of the datapath. */
4428 subfacet_update_stats(struct subfacet *subfacet,
4429 const struct dpif_flow_stats *stats)
4431 if (stats->n_packets || stats->used > subfacet->used) {
4432 struct facet *facet = subfacet->facet;
4434 subfacet_update_time(subfacet, stats->used);
4435 facet->packet_count += stats->n_packets;
4436 facet->byte_count += stats->n_bytes;
4437 facet->tcp_flags |= stats->tcp_flags;
4438 facet_push_stats(facet);
4439 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
4445 static struct rule_dpif *
4446 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
4448 struct ofport_dpif *port;
4449 struct rule_dpif *rule;
4451 rule = rule_dpif_lookup__(ofproto, flow, 0);
4456 port = get_ofp_port(ofproto, flow->in_port);
4458 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
4459 return ofproto->miss_rule;
4462 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
4463 return ofproto->no_packet_in_rule;
4465 return ofproto->miss_rule;
4468 static struct rule_dpif *
4469 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
4472 struct cls_rule *cls_rule;
4473 struct classifier *cls;
4475 if (table_id >= N_TABLES) {
4479 cls = &ofproto->up.tables[table_id].cls;
4480 if (flow->nw_frag & FLOW_NW_FRAG_ANY
4481 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4482 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
4483 * are unavailable. */
4484 struct flow ofpc_normal_flow = *flow;
4485 ofpc_normal_flow.tp_src = htons(0);
4486 ofpc_normal_flow.tp_dst = htons(0);
4487 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
4489 cls_rule = classifier_lookup(cls, flow);
4491 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
4495 complete_operation(struct rule_dpif *rule)
4497 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4499 rule_invalidate(rule);
4501 struct dpif_completion *c = xmalloc(sizeof *c);
4502 c->op = rule->up.pending;
4503 list_push_back(&ofproto->completions, &c->list_node);
4505 ofoperation_complete(rule->up.pending, 0);
4509 static struct rule *
4512 struct rule_dpif *rule = xmalloc(sizeof *rule);
4517 rule_dealloc(struct rule *rule_)
4519 struct rule_dpif *rule = rule_dpif_cast(rule_);
4524 rule_construct(struct rule *rule_)
4526 struct rule_dpif *rule = rule_dpif_cast(rule_);
4527 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4528 struct rule_dpif *victim;
4532 error = validate_actions(rule->up.actions, rule->up.n_actions,
4533 &rule->up.cr.flow, ofproto->max_ports);
4538 rule->packet_count = 0;
4539 rule->byte_count = 0;
4541 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
4542 if (victim && !list_is_empty(&victim->facets)) {
4543 struct facet *facet;
4545 rule->facets = victim->facets;
4546 list_moved(&rule->facets);
4547 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4548 /* XXX: We're only clearing our local counters here. It's possible
4549 * that quite a few packets are unaccounted for in the datapath
4550 * statistics. These will be accounted to the new rule instead of
4551 * cleared as required. This could be fixed by clearing out the
4552 * datapath statistics for this facet, but currently it doesn't
4554 facet_reset_counters(facet);
4558 /* Must avoid list_moved() in this case. */
4559 list_init(&rule->facets);
4562 table_id = rule->up.table_id;
4563 rule->tag = (victim ? victim->tag
4565 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
4566 ofproto->tables[table_id].basis));
4568 complete_operation(rule);
4573 rule_destruct(struct rule *rule_)
4575 struct rule_dpif *rule = rule_dpif_cast(rule_);
4576 struct facet *facet, *next_facet;
4578 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4579 facet_revalidate(facet);
4582 complete_operation(rule);
4586 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4588 struct rule_dpif *rule = rule_dpif_cast(rule_);
4589 struct facet *facet;
4591 /* Start from historical data for 'rule' itself that are no longer tracked
4592 * in facets. This counts, for example, facets that have expired. */
4593 *packets = rule->packet_count;
4594 *bytes = rule->byte_count;
4596 /* Add any statistics that are tracked by facets. This includes
4597 * statistical data recently updated by ofproto_update_stats() as well as
4598 * stats for packets that were executed "by hand" via dpif_execute(). */
4599 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4600 *packets += facet->packet_count;
4601 *bytes += facet->byte_count;
4606 rule_execute(struct rule *rule_, const struct flow *flow,
4607 struct ofpbuf *packet)
4609 struct rule_dpif *rule = rule_dpif_cast(rule_);
4610 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4612 struct dpif_flow_stats stats;
4614 struct action_xlate_ctx ctx;
4615 uint64_t odp_actions_stub[1024 / 8];
4616 struct ofpbuf odp_actions;
4618 dpif_flow_stats_extract(flow, packet, &stats);
4619 rule_credit_stats(rule, &stats);
4621 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4622 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
4623 rule, stats.tcp_flags, packet);
4624 ctx.resubmit_stats = &stats;
4625 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions, &odp_actions);
4627 execute_odp_actions(ofproto, flow, odp_actions.data,
4628 odp_actions.size, packet);
4630 ofpbuf_uninit(&odp_actions);
4636 rule_modify_actions(struct rule *rule_)
4638 struct rule_dpif *rule = rule_dpif_cast(rule_);
4639 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4642 error = validate_actions(rule->up.actions, rule->up.n_actions,
4643 &rule->up.cr.flow, ofproto->max_ports);
4645 ofoperation_complete(rule->up.pending, error);
4649 complete_operation(rule);
4652 /* Sends 'packet' out 'ofport'.
4653 * May modify 'packet'.
4654 * Returns 0 if successful, otherwise a positive errno value. */
4656 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4658 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4659 struct ofpbuf key, odp_actions;
4660 struct odputil_keybuf keybuf;
4665 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
4666 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4668 if (odp_port != ofport->odp_port) {
4669 eth_pop_vlan(packet);
4670 flow.vlan_tci = htons(0);
4673 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4674 odp_flow_key_from_flow(&key, &flow);
4676 ofpbuf_init(&odp_actions, 32);
4677 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4679 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4680 error = dpif_execute(ofproto->dpif,
4682 odp_actions.data, odp_actions.size,
4684 ofpbuf_uninit(&odp_actions);
4687 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4688 ofproto->up.name, odp_port, strerror(error));
4690 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
4694 /* OpenFlow to datapath action translation. */
4696 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
4697 struct action_xlate_ctx *ctx);
4698 static void xlate_normal(struct action_xlate_ctx *);
4700 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
4701 * The action will state 'slow' as the reason that the action is in the slow
4702 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
4703 * dump-flows" output to see why a flow is in the slow path.)
4705 * The 'stub_size' bytes in 'stub' will be used to store the action.
4706 * 'stub_size' must be large enough for the action.
4708 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
4711 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
4712 enum slow_path_reason slow,
4713 uint64_t *stub, size_t stub_size,
4714 const struct nlattr **actionsp, size_t *actions_lenp)
4716 union user_action_cookie cookie;
4719 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
4720 cookie.slow_path.unused = 0;
4721 cookie.slow_path.reason = slow;
4723 ofpbuf_use_stack(&buf, stub, stub_size);
4724 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
4725 uint32_t pid = dpif_port_get_pid(ofproto->dpif, UINT16_MAX);
4726 odp_put_userspace_action(pid, &cookie, &buf);
4728 put_userspace_action(ofproto, &buf, flow, &cookie);
4730 *actionsp = buf.data;
4731 *actions_lenp = buf.size;
4735 put_userspace_action(const struct ofproto_dpif *ofproto,
4736 struct ofpbuf *odp_actions,
4737 const struct flow *flow,
4738 const union user_action_cookie *cookie)
4742 pid = dpif_port_get_pid(ofproto->dpif,
4743 ofp_port_to_odp_port(flow->in_port));
4745 return odp_put_userspace_action(pid, cookie, odp_actions);
4749 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
4750 ovs_be16 vlan_tci, uint32_t odp_port,
4751 unsigned int n_outputs, union user_action_cookie *cookie)
4755 cookie->type = USER_ACTION_COOKIE_SFLOW;
4756 cookie->sflow.vlan_tci = vlan_tci;
4758 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
4759 * port information") for the interpretation of cookie->output. */
4760 switch (n_outputs) {
4762 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
4763 cookie->sflow.output = 0x40000000 | 256;
4767 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4769 cookie->sflow.output = ifindex;
4774 /* 0x80000000 means "multiple output ports. */
4775 cookie->sflow.output = 0x80000000 | n_outputs;
4780 /* Compose SAMPLE action for sFlow. */
4782 compose_sflow_action(const struct ofproto_dpif *ofproto,
4783 struct ofpbuf *odp_actions,
4784 const struct flow *flow,
4787 uint32_t probability;
4788 union user_action_cookie cookie;
4789 size_t sample_offset, actions_offset;
4792 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4796 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4798 /* Number of packets out of UINT_MAX to sample. */
4799 probability = dpif_sflow_get_probability(ofproto->sflow);
4800 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4802 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4803 compose_sflow_cookie(ofproto, htons(0), odp_port,
4804 odp_port == OVSP_NONE ? 0 : 1, &cookie);
4805 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4807 nl_msg_end_nested(odp_actions, actions_offset);
4808 nl_msg_end_nested(odp_actions, sample_offset);
4809 return cookie_offset;
4812 /* SAMPLE action must be first action in any given list of actions.
4813 * At this point we do not have all information required to build it. So try to
4814 * build sample action as complete as possible. */
4816 add_sflow_action(struct action_xlate_ctx *ctx)
4818 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4820 &ctx->flow, OVSP_NONE);
4821 ctx->sflow_odp_port = 0;
4822 ctx->sflow_n_outputs = 0;
4825 /* Fix SAMPLE action according to data collected while composing ODP actions.
4826 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4827 * USERSPACE action's user-cookie which is required for sflow. */
4829 fix_sflow_action(struct action_xlate_ctx *ctx)
4831 const struct flow *base = &ctx->base_flow;
4832 union user_action_cookie *cookie;
4834 if (!ctx->user_cookie_offset) {
4838 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4840 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4842 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
4843 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
4847 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4850 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4851 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4852 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4853 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4857 struct priority_to_dscp *pdscp;
4859 if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD
4860 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4864 pdscp = get_priority(ofport, ctx->flow.skb_priority);
4866 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4867 ctx->flow.nw_tos |= pdscp->dscp;
4870 /* We may not have an ofport record for this port, but it doesn't hurt
4871 * to allow forwarding to it anyhow. Maybe such a port will appear
4872 * later and we're pre-populating the flow table. */
4875 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4876 ctx->flow.vlan_tci);
4877 if (out_port != odp_port) {
4878 ctx->flow.vlan_tci = htons(0);
4880 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4881 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4883 ctx->sflow_odp_port = odp_port;
4884 ctx->sflow_n_outputs++;
4885 ctx->nf_output_iface = ofp_port;
4886 ctx->flow.vlan_tci = flow_vlan_tci;
4887 ctx->flow.nw_tos = flow_nw_tos;
4891 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4893 compose_output_action__(ctx, ofp_port, true);
4897 xlate_table_action(struct action_xlate_ctx *ctx,
4898 uint16_t in_port, uint8_t table_id)
4900 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4901 struct ofproto_dpif *ofproto = ctx->ofproto;
4902 struct rule_dpif *rule;
4903 uint16_t old_in_port;
4904 uint8_t old_table_id;
4906 old_table_id = ctx->table_id;
4907 ctx->table_id = table_id;
4909 /* Look up a flow with 'in_port' as the input port. */
4910 old_in_port = ctx->flow.in_port;
4911 ctx->flow.in_port = in_port;
4912 rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id);
4915 if (table_id > 0 && table_id < N_TABLES) {
4916 struct table_dpif *table = &ofproto->tables[table_id];
4917 if (table->other_table) {
4918 ctx->tags |= (rule && rule->tag
4920 : rule_calculate_tag(&ctx->flow,
4921 &table->other_table->wc,
4926 /* Restore the original input port. Otherwise OFPP_NORMAL and
4927 * OFPP_IN_PORT will have surprising behavior. */
4928 ctx->flow.in_port = old_in_port;
4930 if (ctx->resubmit_hook) {
4931 ctx->resubmit_hook(ctx, rule);
4935 struct rule_dpif *old_rule = ctx->rule;
4937 if (ctx->resubmit_stats) {
4938 rule_credit_stats(rule, ctx->resubmit_stats);
4943 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
4944 ctx->rule = old_rule;
4948 ctx->table_id = old_table_id;
4950 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4952 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
4953 MAX_RESUBMIT_RECURSION);
4954 ctx->max_resubmit_trigger = true;
4959 xlate_resubmit_table(struct action_xlate_ctx *ctx,
4960 const struct nx_action_resubmit *nar)
4965 in_port = (nar->in_port == htons(OFPP_IN_PORT)
4967 : ntohs(nar->in_port));
4968 table_id = nar->table == 255 ? ctx->table_id : nar->table;
4970 xlate_table_action(ctx, in_port, table_id);
4974 flood_packets(struct action_xlate_ctx *ctx, bool all)
4976 struct ofport_dpif *ofport;
4978 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
4979 uint16_t ofp_port = ofport->up.ofp_port;
4981 if (ofp_port == ctx->flow.in_port) {
4986 compose_output_action__(ctx, ofp_port, false);
4987 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
4988 compose_output_action(ctx, ofp_port);
4992 ctx->nf_output_iface = NF_OUT_FLOOD;
4996 execute_controller_action(struct action_xlate_ctx *ctx, int len,
4997 enum ofp_packet_in_reason reason,
4998 uint16_t controller_id)
5000 struct ofputil_packet_in pin;
5001 struct ofpbuf *packet;
5003 ctx->slow |= SLOW_CONTROLLER;
5008 packet = ofpbuf_clone(ctx->packet);
5010 if (packet->l2 && packet->l3) {
5011 struct eth_header *eh;
5013 eth_pop_vlan(packet);
5016 /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
5017 * LLC frame. Calculating the Ethernet type of these frames is more
5018 * trouble than seems appropriate for a simple assertion. */
5019 assert(ntohs(eh->eth_type) < ETH_TYPE_MIN
5020 || eh->eth_type == ctx->flow.dl_type);
5022 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
5023 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
5025 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
5026 eth_push_vlan(packet, ctx->flow.vlan_tci);
5030 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5031 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
5032 ctx->flow.nw_tos, ctx->flow.nw_ttl);
5036 if (ctx->flow.nw_proto == IPPROTO_TCP) {
5037 packet_set_tcp_port(packet, ctx->flow.tp_src,
5039 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
5040 packet_set_udp_port(packet, ctx->flow.tp_src,
5047 pin.packet = packet->data;
5048 pin.packet_len = packet->size;
5049 pin.reason = reason;
5050 pin.controller_id = controller_id;
5051 pin.table_id = ctx->table_id;
5052 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
5055 flow_get_metadata(&ctx->flow, &pin.fmd);
5057 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
5058 ofpbuf_delete(packet);
5062 compose_dec_ttl(struct action_xlate_ctx *ctx)
5064 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
5065 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
5069 if (ctx->flow.nw_ttl > 1) {
5073 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
5075 /* Stop processing for current table. */
5081 xlate_output_action__(struct action_xlate_ctx *ctx,
5082 uint16_t port, uint16_t max_len)
5084 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
5086 ctx->nf_output_iface = NF_OUT_DROP;
5090 compose_output_action(ctx, ctx->flow.in_port);
5093 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
5099 flood_packets(ctx, false);
5102 flood_packets(ctx, true);
5104 case OFPP_CONTROLLER:
5105 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
5111 if (port != ctx->flow.in_port) {
5112 compose_output_action(ctx, port);
5117 if (prev_nf_output_iface == NF_OUT_FLOOD) {
5118 ctx->nf_output_iface = NF_OUT_FLOOD;
5119 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
5120 ctx->nf_output_iface = prev_nf_output_iface;
5121 } else if (prev_nf_output_iface != NF_OUT_DROP &&
5122 ctx->nf_output_iface != NF_OUT_FLOOD) {
5123 ctx->nf_output_iface = NF_OUT_MULTI;
5128 xlate_output_reg_action(struct action_xlate_ctx *ctx,
5129 const struct nx_action_output_reg *naor)
5131 struct mf_subfield src;
5134 nxm_decode(&src, naor->src, naor->ofs_nbits);
5135 ofp_port = mf_get_subfield(&src, &ctx->flow);
5137 if (ofp_port <= UINT16_MAX) {
5138 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
5143 xlate_output_action(struct action_xlate_ctx *ctx,
5144 const struct ofp_action_output *oao)
5146 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
5150 xlate_enqueue_action(struct action_xlate_ctx *ctx,
5151 const struct ofp_action_enqueue *oae)
5154 uint32_t flow_priority, priority;
5157 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
5160 /* Fall back to ordinary output action. */
5161 xlate_output_action__(ctx, ntohs(oae->port), 0);
5165 /* Figure out datapath output port. */
5166 ofp_port = ntohs(oae->port);
5167 if (ofp_port == OFPP_IN_PORT) {
5168 ofp_port = ctx->flow.in_port;
5169 } else if (ofp_port == ctx->flow.in_port) {
5173 /* Add datapath actions. */
5174 flow_priority = ctx->flow.skb_priority;
5175 ctx->flow.skb_priority = priority;
5176 compose_output_action(ctx, ofp_port);
5177 ctx->flow.skb_priority = flow_priority;
5179 /* Update NetFlow output port. */
5180 if (ctx->nf_output_iface == NF_OUT_DROP) {
5181 ctx->nf_output_iface = ofp_port;
5182 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
5183 ctx->nf_output_iface = NF_OUT_MULTI;
5188 xlate_set_queue_action(struct action_xlate_ctx *ctx,
5189 const struct nx_action_set_queue *nasq)
5194 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
5197 /* Couldn't translate queue to a priority, so ignore. A warning
5198 * has already been logged. */
5202 ctx->flow.skb_priority = priority;
5205 struct xlate_reg_state {
5211 xlate_autopath(struct action_xlate_ctx *ctx,
5212 const struct nx_action_autopath *naa)
5214 uint16_t ofp_port = ntohl(naa->id);
5215 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
5217 if (!port || !port->bundle) {
5218 ofp_port = OFPP_NONE;
5219 } else if (port->bundle->bond) {
5220 /* Autopath does not support VLAN hashing. */
5221 struct ofport_dpif *slave = bond_choose_output_slave(
5222 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
5224 ofp_port = slave->up.ofp_port;
5227 autopath_execute(naa, &ctx->flow, ofp_port);
5231 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
5233 struct ofproto_dpif *ofproto = ofproto_;
5234 struct ofport_dpif *port;
5244 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
5247 port = get_ofp_port(ofproto, ofp_port);
5248 return port ? port->may_enable : false;
5253 xlate_learn_action(struct action_xlate_ctx *ctx,
5254 const struct nx_action_learn *learn)
5256 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
5257 struct ofputil_flow_mod fm;
5260 learn_execute(learn, &ctx->flow, &fm);
5262 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
5263 if (error && !VLOG_DROP_WARN(&rl)) {
5264 VLOG_WARN("learning action failed to modify flow table (%s)",
5265 ofperr_get_name(error));
5271 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
5272 * means "infinite". */
5274 reduce_timeout(uint16_t max, uint16_t *timeout)
5276 if (max && (!*timeout || *timeout > max)) {
5282 xlate_fin_timeout(struct action_xlate_ctx *ctx,
5283 const struct nx_action_fin_timeout *naft)
5285 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
5286 struct rule_dpif *rule = ctx->rule;
5288 reduce_timeout(ntohs(naft->fin_idle_timeout), &rule->up.idle_timeout);
5289 reduce_timeout(ntohs(naft->fin_hard_timeout), &rule->up.hard_timeout);
5294 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
5296 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
5297 ? OFPUTIL_PC_NO_RECV_STP
5298 : OFPUTIL_PC_NO_RECV)) {
5302 /* Only drop packets here if both forwarding and learning are
5303 * disabled. If just learning is enabled, we need to have
5304 * OFPP_NORMAL and the learning action have a look at the packet
5305 * before we can drop it. */
5306 if (!stp_forward_in_state(port->stp_state)
5307 && !stp_learn_in_state(port->stp_state)) {
5315 do_xlate_actions(const union ofp_action *in, size_t n_in,
5316 struct action_xlate_ctx *ctx)
5318 const struct ofport_dpif *port;
5319 const union ofp_action *ia;
5320 bool was_evictable = true;
5323 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5324 if (port && !may_receive(port, ctx)) {
5325 /* Drop this flow. */
5330 /* Don't let the rule we're working on get evicted underneath us. */
5331 was_evictable = ctx->rule->up.evictable;
5332 ctx->rule->up.evictable = false;
5334 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
5335 const struct ofp_action_dl_addr *oada;
5336 const struct nx_action_resubmit *nar;
5337 const struct nx_action_set_tunnel *nast;
5338 const struct nx_action_set_queue *nasq;
5339 const struct nx_action_multipath *nam;
5340 const struct nx_action_autopath *naa;
5341 const struct nx_action_bundle *nab;
5342 const struct nx_action_output_reg *naor;
5343 const struct nx_action_controller *nac;
5344 enum ofputil_action_code code;
5351 code = ofputil_decode_action_unsafe(ia);
5353 case OFPUTIL_OFPAT10_OUTPUT:
5354 xlate_output_action(ctx, &ia->output);
5357 case OFPUTIL_OFPAT10_SET_VLAN_VID:
5358 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
5359 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
5362 case OFPUTIL_OFPAT10_SET_VLAN_PCP:
5363 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
5364 ctx->flow.vlan_tci |= htons(
5365 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
5368 case OFPUTIL_OFPAT10_STRIP_VLAN:
5369 ctx->flow.vlan_tci = htons(0);
5372 case OFPUTIL_OFPAT10_SET_DL_SRC:
5373 oada = ((struct ofp_action_dl_addr *) ia);
5374 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
5377 case OFPUTIL_OFPAT10_SET_DL_DST:
5378 oada = ((struct ofp_action_dl_addr *) ia);
5379 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
5382 case OFPUTIL_OFPAT10_SET_NW_SRC:
5383 ctx->flow.nw_src = ia->nw_addr.nw_addr;
5386 case OFPUTIL_OFPAT10_SET_NW_DST:
5387 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
5390 case OFPUTIL_OFPAT10_SET_NW_TOS:
5391 /* OpenFlow 1.0 only supports IPv4. */
5392 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5393 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5394 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
5398 case OFPUTIL_OFPAT10_SET_TP_SRC:
5399 ctx->flow.tp_src = ia->tp_port.tp_port;
5402 case OFPUTIL_OFPAT10_SET_TP_DST:
5403 ctx->flow.tp_dst = ia->tp_port.tp_port;
5406 case OFPUTIL_OFPAT10_ENQUEUE:
5407 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
5410 case OFPUTIL_NXAST_RESUBMIT:
5411 nar = (const struct nx_action_resubmit *) ia;
5412 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
5415 case OFPUTIL_NXAST_RESUBMIT_TABLE:
5416 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
5419 case OFPUTIL_NXAST_SET_TUNNEL:
5420 nast = (const struct nx_action_set_tunnel *) ia;
5421 tun_id = htonll(ntohl(nast->tun_id));
5422 ctx->flow.tun_id = tun_id;
5425 case OFPUTIL_NXAST_SET_QUEUE:
5426 nasq = (const struct nx_action_set_queue *) ia;
5427 xlate_set_queue_action(ctx, nasq);
5430 case OFPUTIL_NXAST_POP_QUEUE:
5431 ctx->flow.skb_priority = ctx->orig_skb_priority;
5434 case OFPUTIL_NXAST_REG_MOVE:
5435 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
5439 case OFPUTIL_NXAST_REG_LOAD:
5440 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
5444 case OFPUTIL_NXAST_NOTE:
5445 /* Nothing to do. */
5448 case OFPUTIL_NXAST_SET_TUNNEL64:
5449 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
5450 ctx->flow.tun_id = tun_id;
5453 case OFPUTIL_NXAST_MULTIPATH:
5454 nam = (const struct nx_action_multipath *) ia;
5455 multipath_execute(nam, &ctx->flow);
5458 case OFPUTIL_NXAST_AUTOPATH:
5459 naa = (const struct nx_action_autopath *) ia;
5460 xlate_autopath(ctx, naa);
5463 case OFPUTIL_NXAST_BUNDLE:
5464 ctx->ofproto->has_bundle_action = true;
5465 nab = (const struct nx_action_bundle *) ia;
5466 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
5471 case OFPUTIL_NXAST_BUNDLE_LOAD:
5472 ctx->ofproto->has_bundle_action = true;
5473 nab = (const struct nx_action_bundle *) ia;
5474 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
5478 case OFPUTIL_NXAST_OUTPUT_REG:
5479 naor = (const struct nx_action_output_reg *) ia;
5480 xlate_output_reg_action(ctx, naor);
5483 case OFPUTIL_NXAST_LEARN:
5484 ctx->has_learn = true;
5485 if (ctx->may_learn) {
5486 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
5490 case OFPUTIL_NXAST_DEC_TTL:
5491 if (compose_dec_ttl(ctx)) {
5496 case OFPUTIL_NXAST_EXIT:
5500 case OFPUTIL_NXAST_FIN_TIMEOUT:
5501 ctx->has_fin_timeout = true;
5502 xlate_fin_timeout(ctx, (const struct nx_action_fin_timeout *) ia);
5505 case OFPUTIL_NXAST_CONTROLLER:
5506 nac = (const struct nx_action_controller *) ia;
5507 execute_controller_action(ctx, ntohs(nac->max_len), nac->reason,
5508 ntohs(nac->controller_id));
5514 /* We've let OFPP_NORMAL and the learning action look at the packet,
5515 * so drop it now if forwarding is disabled. */
5516 if (port && !stp_forward_in_state(port->stp_state)) {
5517 ofpbuf_clear(ctx->odp_actions);
5518 add_sflow_action(ctx);
5521 ctx->rule->up.evictable = was_evictable;
5526 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
5527 struct ofproto_dpif *ofproto, const struct flow *flow,
5528 ovs_be16 initial_tci, struct rule_dpif *rule,
5529 uint8_t tcp_flags, const struct ofpbuf *packet)
5531 ctx->ofproto = ofproto;
5533 ctx->base_flow = ctx->flow;
5534 ctx->base_flow.tun_id = 0;
5535 ctx->base_flow.vlan_tci = initial_tci;
5537 ctx->packet = packet;
5538 ctx->may_learn = packet != NULL;
5539 ctx->tcp_flags = tcp_flags;
5540 ctx->resubmit_hook = NULL;
5541 ctx->resubmit_stats = NULL;
5544 /* Translates the 'n_in' "union ofp_action"s in 'in' into datapath actions in
5545 * 'odp_actions', using 'ctx'. */
5547 xlate_actions(struct action_xlate_ctx *ctx,
5548 const union ofp_action *in, size_t n_in,
5549 struct ofpbuf *odp_actions)
5551 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
5552 * that in the future we always keep a copy of the original flow for
5553 * tracing purposes. */
5554 static bool hit_resubmit_limit;
5556 enum slow_path_reason special;
5558 COVERAGE_INC(ofproto_dpif_xlate);
5560 ofpbuf_clear(odp_actions);
5561 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
5563 ctx->odp_actions = odp_actions;
5566 ctx->has_learn = false;
5567 ctx->has_normal = false;
5568 ctx->has_fin_timeout = false;
5569 ctx->nf_output_iface = NF_OUT_DROP;
5572 ctx->max_resubmit_trigger = false;
5573 ctx->orig_skb_priority = ctx->flow.skb_priority;
5577 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
5578 /* Do this conditionally because the copy is expensive enough that it
5579 * shows up in profiles.
5581 * We keep orig_flow in 'ctx' only because I couldn't make GCC 4.4
5582 * believe that I wasn't using it without initializing it if I kept it
5583 * in a local variable. */
5584 ctx->orig_flow = ctx->flow;
5587 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
5588 switch (ctx->ofproto->up.frag_handling) {
5589 case OFPC_FRAG_NORMAL:
5590 /* We must pretend that transport ports are unavailable. */
5591 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
5592 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
5595 case OFPC_FRAG_DROP:
5598 case OFPC_FRAG_REASM:
5601 case OFPC_FRAG_NX_MATCH:
5602 /* Nothing to do. */
5605 case OFPC_INVALID_TTL_TO_CONTROLLER:
5610 special = process_special(ctx->ofproto, &ctx->flow, ctx->packet);
5612 ctx->slow |= special;
5614 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5615 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
5617 add_sflow_action(ctx);
5618 do_xlate_actions(in, n_in, ctx);
5620 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
5621 if (!hit_resubmit_limit) {
5622 /* We didn't record the original flow. Make sure we do from
5624 hit_resubmit_limit = true;
5625 } else if (!VLOG_DROP_ERR(&trace_rl)) {
5626 struct ds ds = DS_EMPTY_INITIALIZER;
5628 ofproto_trace(ctx->ofproto, &ctx->orig_flow, ctx->packet,
5630 VLOG_ERR("Trace triggered by excessive resubmit "
5631 "recursion:\n%s", ds_cstr(&ds));
5636 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
5637 ctx->odp_actions->data,
5638 ctx->odp_actions->size)) {
5639 ctx->slow |= SLOW_IN_BAND;
5641 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
5643 compose_output_action(ctx, OFPP_LOCAL);
5646 if (ctx->ofproto->has_mirrors) {
5647 add_mirror_actions(ctx, &ctx->orig_flow);
5649 fix_sflow_action(ctx);
5653 /* Translates the 'n_in' "union ofp_action"s in 'in' into datapath actions,
5654 * using 'ctx', and discards the datapath actions. */
5656 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
5657 const union ofp_action *in, size_t n_in)
5659 uint64_t odp_actions_stub[1024 / 8];
5660 struct ofpbuf odp_actions;
5662 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5663 xlate_actions(ctx, in, n_in, &odp_actions);
5664 ofpbuf_uninit(&odp_actions);
5667 /* OFPP_NORMAL implementation. */
5669 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
5671 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
5672 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
5673 * the bundle on which the packet was received, returns the VLAN to which the
5676 * Both 'vid' and the return value are in the range 0...4095. */
5678 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
5680 switch (in_bundle->vlan_mode) {
5681 case PORT_VLAN_ACCESS:
5682 return in_bundle->vlan;
5685 case PORT_VLAN_TRUNK:
5688 case PORT_VLAN_NATIVE_UNTAGGED:
5689 case PORT_VLAN_NATIVE_TAGGED:
5690 return vid ? vid : in_bundle->vlan;
5697 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
5698 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
5701 * 'vid' should be the VID obtained from the 802.1Q header that was received as
5702 * part of a packet (specify 0 if there was no 802.1Q header), in the range
5705 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
5707 /* Allow any VID on the OFPP_NONE port. */
5708 if (in_bundle == &ofpp_none_bundle) {
5712 switch (in_bundle->vlan_mode) {
5713 case PORT_VLAN_ACCESS:
5716 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5717 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
5718 "packet received on port %s configured as VLAN "
5719 "%"PRIu16" access port",
5720 in_bundle->ofproto->up.name, vid,
5721 in_bundle->name, in_bundle->vlan);
5727 case PORT_VLAN_NATIVE_UNTAGGED:
5728 case PORT_VLAN_NATIVE_TAGGED:
5730 /* Port must always carry its native VLAN. */
5734 case PORT_VLAN_TRUNK:
5735 if (!ofbundle_includes_vlan(in_bundle, vid)) {
5737 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5738 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
5739 "received on port %s not configured for trunking "
5741 in_bundle->ofproto->up.name, vid,
5742 in_bundle->name, vid);
5754 /* Given 'vlan', the VLAN that a packet belongs to, and
5755 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
5756 * that should be included in the 802.1Q header. (If the return value is 0,
5757 * then the 802.1Q header should only be included in the packet if there is a
5760 * Both 'vlan' and the return value are in the range 0...4095. */
5762 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
5764 switch (out_bundle->vlan_mode) {
5765 case PORT_VLAN_ACCESS:
5768 case PORT_VLAN_TRUNK:
5769 case PORT_VLAN_NATIVE_TAGGED:
5772 case PORT_VLAN_NATIVE_UNTAGGED:
5773 return vlan == out_bundle->vlan ? 0 : vlan;
5781 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
5784 struct ofport_dpif *port;
5786 ovs_be16 tci, old_tci;
5788 vid = output_vlan_to_vid(out_bundle, vlan);
5789 if (!out_bundle->bond) {
5790 port = ofbundle_get_a_port(out_bundle);
5792 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
5795 /* No slaves enabled, so drop packet. */
5800 old_tci = ctx->flow.vlan_tci;
5802 if (tci || out_bundle->use_priority_tags) {
5803 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
5805 tci |= htons(VLAN_CFI);
5808 ctx->flow.vlan_tci = tci;
5810 compose_output_action(ctx, port->up.ofp_port);
5811 ctx->flow.vlan_tci = old_tci;
5815 mirror_mask_ffs(mirror_mask_t mask)
5817 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
5822 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
5824 return (bundle->vlan_mode != PORT_VLAN_ACCESS
5825 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
5829 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
5831 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
5834 /* Returns an arbitrary interface within 'bundle'. */
5835 static struct ofport_dpif *
5836 ofbundle_get_a_port(const struct ofbundle *bundle)
5838 return CONTAINER_OF(list_front(&bundle->ports),
5839 struct ofport_dpif, bundle_node);
5843 vlan_is_mirrored(const struct ofmirror *m, int vlan)
5845 return !m->vlans || bitmap_is_set(m->vlans, vlan);
5848 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
5849 * to a VLAN. In general most packets may be mirrored but we want to drop
5850 * protocols that may confuse switches. */
5852 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
5854 /* If you change this function's behavior, please update corresponding
5855 * documentation in vswitch.xml at the same time. */
5856 if (dst[0] != 0x01) {
5857 /* All the currently banned MACs happen to start with 01 currently, so
5858 * this is a quick way to eliminate most of the good ones. */
5860 if (eth_addr_is_reserved(dst)) {
5861 /* Drop STP, IEEE pause frames, and other reserved protocols
5862 * (01-80-c2-00-00-0x). */
5866 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
5868 if ((dst[3] & 0xfe) == 0xcc &&
5869 (dst[4] & 0xfe) == 0xcc &&
5870 (dst[5] & 0xfe) == 0xcc) {
5871 /* Drop the following protocols plus others following the same
5874 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
5875 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
5876 STP Uplink Fast (01-00-0c-cd-cd-cd) */
5880 if (!(dst[3] | dst[4] | dst[5])) {
5881 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
5890 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5892 struct ofproto_dpif *ofproto = ctx->ofproto;
5893 mirror_mask_t mirrors;
5894 struct ofbundle *in_bundle;
5897 const struct nlattr *a;
5900 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5901 ctx->packet != NULL, NULL);
5905 mirrors = in_bundle->src_mirrors;
5907 /* Drop frames on bundles reserved for mirroring. */
5908 if (in_bundle->mirror_out) {
5909 if (ctx->packet != NULL) {
5910 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5911 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5912 "%s, which is reserved exclusively for mirroring",
5913 ctx->ofproto->up.name, in_bundle->name);
5919 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5920 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5923 vlan = input_vid_to_vlan(in_bundle, vid);
5925 /* Look at the output ports to check for destination selections. */
5927 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5928 ctx->odp_actions->size) {
5929 enum ovs_action_attr type = nl_attr_type(a);
5930 struct ofport_dpif *ofport;
5932 if (type != OVS_ACTION_ATTR_OUTPUT) {
5936 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5937 if (ofport && ofport->bundle) {
5938 mirrors |= ofport->bundle->dst_mirrors;
5946 /* Restore the original packet before adding the mirror actions. */
5947 ctx->flow = *orig_flow;
5952 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5954 if (!vlan_is_mirrored(m, vlan)) {
5955 mirrors &= mirrors - 1;
5959 mirrors &= ~m->dup_mirrors;
5960 ctx->mirrors |= m->dup_mirrors;
5962 output_normal(ctx, m->out, vlan);
5963 } else if (eth_dst_may_rspan(orig_flow->dl_dst)
5964 && vlan != m->out_vlan) {
5965 struct ofbundle *bundle;
5967 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5968 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5969 && !bundle->mirror_out) {
5970 output_normal(ctx, bundle, m->out_vlan);
5978 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5979 uint64_t packets, uint64_t bytes)
5985 for (; mirrors; mirrors &= mirrors - 1) {
5988 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5991 /* In normal circumstances 'm' will not be NULL. However,
5992 * if mirrors are reconfigured, we can temporarily get out
5993 * of sync in facet_revalidate(). We could "correct" the
5994 * mirror list before reaching here, but doing that would
5995 * not properly account the traffic stats we've currently
5996 * accumulated for previous mirror configuration. */
6000 m->packet_count += packets;
6001 m->byte_count += bytes;
6005 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
6006 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
6007 * indicate this; newer upstream kernels use gratuitous ARP requests. */
6009 is_gratuitous_arp(const struct flow *flow)
6011 return (flow->dl_type == htons(ETH_TYPE_ARP)
6012 && eth_addr_is_broadcast(flow->dl_dst)
6013 && (flow->nw_proto == ARP_OP_REPLY
6014 || (flow->nw_proto == ARP_OP_REQUEST
6015 && flow->nw_src == flow->nw_dst)));
6019 update_learning_table(struct ofproto_dpif *ofproto,
6020 const struct flow *flow, int vlan,
6021 struct ofbundle *in_bundle)
6023 struct mac_entry *mac;
6025 /* Don't learn the OFPP_NONE port. */
6026 if (in_bundle == &ofpp_none_bundle) {
6030 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
6034 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
6035 if (is_gratuitous_arp(flow)) {
6036 /* We don't want to learn from gratuitous ARP packets that are
6037 * reflected back over bond slaves so we lock the learning table. */
6038 if (!in_bundle->bond) {
6039 mac_entry_set_grat_arp_lock(mac);
6040 } else if (mac_entry_is_grat_arp_locked(mac)) {
6045 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
6046 /* The log messages here could actually be useful in debugging,
6047 * so keep the rate limit relatively high. */
6048 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
6049 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
6050 "on port %s in VLAN %d",
6051 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
6052 in_bundle->name, vlan);
6054 mac->port.p = in_bundle;
6055 tag_set_add(&ofproto->revalidate_set,
6056 mac_learning_changed(ofproto->ml, mac));
6060 static struct ofbundle *
6061 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
6062 bool warn, struct ofport_dpif **in_ofportp)
6064 struct ofport_dpif *ofport;
6066 /* Find the port and bundle for the received packet. */
6067 ofport = get_ofp_port(ofproto, in_port);
6069 *in_ofportp = ofport;
6071 if (ofport && ofport->bundle) {
6072 return ofport->bundle;
6075 /* Special-case OFPP_NONE, which a controller may use as the ingress
6076 * port for traffic that it is sourcing. */
6077 if (in_port == OFPP_NONE) {
6078 return &ofpp_none_bundle;
6081 /* Odd. A few possible reasons here:
6083 * - We deleted a port but there are still a few packets queued up
6086 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
6087 * we don't know about.
6089 * - The ofproto client didn't configure the port as part of a bundle.
6092 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6094 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
6095 "port %"PRIu16, ofproto->up.name, in_port);
6100 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
6101 * dropped. Returns true if they may be forwarded, false if they should be
6104 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
6105 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
6107 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
6108 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
6109 * checked by input_vid_is_valid().
6111 * May also add tags to '*tags', although the current implementation only does
6112 * so in one special case.
6115 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
6116 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
6118 struct ofbundle *in_bundle = in_port->bundle;
6120 /* Drop frames for reserved multicast addresses
6121 * only if forward_bpdu option is absent. */
6122 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
6126 if (in_bundle->bond) {
6127 struct mac_entry *mac;
6129 switch (bond_check_admissibility(in_bundle->bond, in_port,
6130 flow->dl_dst, tags)) {
6137 case BV_DROP_IF_MOVED:
6138 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
6139 if (mac && mac->port.p != in_bundle &&
6140 (!is_gratuitous_arp(flow)
6141 || mac_entry_is_grat_arp_locked(mac))) {
6152 xlate_normal(struct action_xlate_ctx *ctx)
6154 struct ofport_dpif *in_port;
6155 struct ofbundle *in_bundle;
6156 struct mac_entry *mac;
6160 ctx->has_normal = true;
6162 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
6163 ctx->packet != NULL, &in_port);
6168 /* Drop malformed frames. */
6169 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
6170 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
6171 if (ctx->packet != NULL) {
6172 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6173 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
6174 "VLAN tag received on port %s",
6175 ctx->ofproto->up.name, in_bundle->name);
6180 /* Drop frames on bundles reserved for mirroring. */
6181 if (in_bundle->mirror_out) {
6182 if (ctx->packet != NULL) {
6183 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6184 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6185 "%s, which is reserved exclusively for mirroring",
6186 ctx->ofproto->up.name, in_bundle->name);
6192 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
6193 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6196 vlan = input_vid_to_vlan(in_bundle, vid);
6198 /* Check other admissibility requirements. */
6200 !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
6204 /* Learn source MAC. */
6205 if (ctx->may_learn) {
6206 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
6209 /* Determine output bundle. */
6210 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
6213 if (mac->port.p != in_bundle) {
6214 output_normal(ctx, mac->port.p, vlan);
6217 struct ofbundle *bundle;
6219 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
6220 if (bundle != in_bundle
6221 && ofbundle_includes_vlan(bundle, vlan)
6222 && bundle->floodable
6223 && !bundle->mirror_out) {
6224 output_normal(ctx, bundle, vlan);
6227 ctx->nf_output_iface = NF_OUT_FLOOD;
6231 /* Optimized flow revalidation.
6233 * It's a difficult problem, in general, to tell which facets need to have
6234 * their actions recalculated whenever the OpenFlow flow table changes. We
6235 * don't try to solve that general problem: for most kinds of OpenFlow flow
6236 * table changes, we recalculate the actions for every facet. This is
6237 * relatively expensive, but it's good enough if the OpenFlow flow table
6238 * doesn't change very often.
6240 * However, we can expect one particular kind of OpenFlow flow table change to
6241 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
6242 * of CPU on revalidating every facet whenever MAC learning modifies the flow
6243 * table, we add a special case that applies to flow tables in which every rule
6244 * has the same form (that is, the same wildcards), except that the table is
6245 * also allowed to have a single "catch-all" flow that matches all packets. We
6246 * optimize this case by tagging all of the facets that resubmit into the table
6247 * and invalidating the same tag whenever a flow changes in that table. The
6248 * end result is that we revalidate just the facets that need it (and sometimes
6249 * a few more, but not all of the facets or even all of the facets that
6250 * resubmit to the table modified by MAC learning). */
6252 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
6253 * into an OpenFlow table with the given 'basis'. */
6255 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
6258 if (flow_wildcards_is_catchall(wc)) {
6261 struct flow tag_flow = *flow;
6262 flow_zero_wildcards(&tag_flow, wc);
6263 return tag_create_deterministic(flow_hash(&tag_flow, secret));
6267 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
6268 * taggability of that table.
6270 * This function must be called after *each* change to a flow table. If you
6271 * skip calling it on some changes then the pointer comparisons at the end can
6272 * be invalid if you get unlucky. For example, if a flow removal causes a
6273 * cls_table to be destroyed and then a flow insertion causes a cls_table with
6274 * different wildcards to be created with the same address, then this function
6275 * will incorrectly skip revalidation. */
6277 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
6279 struct table_dpif *table = &ofproto->tables[table_id];
6280 const struct oftable *oftable = &ofproto->up.tables[table_id];
6281 struct cls_table *catchall, *other;
6282 struct cls_table *t;
6284 catchall = other = NULL;
6286 switch (hmap_count(&oftable->cls.tables)) {
6288 /* We could tag this OpenFlow table but it would make the logic a
6289 * little harder and it's a corner case that doesn't seem worth it
6295 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
6296 if (cls_table_is_catchall(t)) {
6298 } else if (!other) {
6301 /* Indicate that we can't tag this by setting both tables to
6302 * NULL. (We know that 'catchall' is already NULL.) */
6309 /* Can't tag this table. */
6313 if (table->catchall_table != catchall || table->other_table != other) {
6314 table->catchall_table = catchall;
6315 table->other_table = other;
6316 ofproto->need_revalidate = true;
6320 /* Given 'rule' that has changed in some way (either it is a rule being
6321 * inserted, a rule being deleted, or a rule whose actions are being
6322 * modified), marks facets for revalidation to ensure that packets will be
6323 * forwarded correctly according to the new state of the flow table.
6325 * This function must be called after *each* change to a flow table. See
6326 * the comment on table_update_taggable() for more information. */
6328 rule_invalidate(const struct rule_dpif *rule)
6330 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
6332 table_update_taggable(ofproto, rule->up.table_id);
6334 if (!ofproto->need_revalidate) {
6335 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
6337 if (table->other_table && rule->tag) {
6338 tag_set_add(&ofproto->revalidate_set, rule->tag);
6340 ofproto->need_revalidate = true;
6346 set_frag_handling(struct ofproto *ofproto_,
6347 enum ofp_config_flags frag_handling)
6349 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6351 if (frag_handling != OFPC_FRAG_REASM) {
6352 ofproto->need_revalidate = true;
6360 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
6361 const struct flow *flow,
6362 const union ofp_action *ofp_actions, size_t n_ofp_actions)
6364 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6367 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
6368 return OFPERR_NXBRC_BAD_IN_PORT;
6371 error = validate_actions(ofp_actions, n_ofp_actions, flow,
6372 ofproto->max_ports);
6374 struct odputil_keybuf keybuf;
6375 struct dpif_flow_stats stats;
6379 struct action_xlate_ctx ctx;
6380 uint64_t odp_actions_stub[1024 / 8];
6381 struct ofpbuf odp_actions;
6383 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
6384 odp_flow_key_from_flow(&key, flow);
6386 dpif_flow_stats_extract(flow, packet, &stats);
6388 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL,
6389 packet_get_tcp_flags(packet, flow), packet);
6390 ctx.resubmit_stats = &stats;
6392 ofpbuf_use_stub(&odp_actions,
6393 odp_actions_stub, sizeof odp_actions_stub);
6394 xlate_actions(&ctx, ofp_actions, n_ofp_actions, &odp_actions);
6395 dpif_execute(ofproto->dpif, key.data, key.size,
6396 odp_actions.data, odp_actions.size, packet);
6397 ofpbuf_uninit(&odp_actions);
6405 set_netflow(struct ofproto *ofproto_,
6406 const struct netflow_options *netflow_options)
6408 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6410 if (netflow_options) {
6411 if (!ofproto->netflow) {
6412 ofproto->netflow = netflow_create();
6414 return netflow_set_options(ofproto->netflow, netflow_options);
6416 netflow_destroy(ofproto->netflow);
6417 ofproto->netflow = NULL;
6423 get_netflow_ids(const struct ofproto *ofproto_,
6424 uint8_t *engine_type, uint8_t *engine_id)
6426 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6428 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
6432 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
6434 if (!facet_is_controller_flow(facet) &&
6435 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
6436 struct subfacet *subfacet;
6437 struct ofexpired expired;
6439 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
6440 if (subfacet->path == SF_FAST_PATH) {
6441 struct dpif_flow_stats stats;
6443 subfacet_reinstall(subfacet, &stats);
6444 subfacet_update_stats(subfacet, &stats);
6448 expired.flow = facet->flow;
6449 expired.packet_count = facet->packet_count;
6450 expired.byte_count = facet->byte_count;
6451 expired.used = facet->used;
6452 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
6457 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
6459 struct facet *facet;
6461 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6462 send_active_timeout(ofproto, facet);
6466 static struct ofproto_dpif *
6467 ofproto_dpif_lookup(const char *name)
6469 struct ofproto_dpif *ofproto;
6471 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
6472 hash_string(name, 0), &all_ofproto_dpifs) {
6473 if (!strcmp(ofproto->up.name, name)) {
6481 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
6482 const char *argv[], void *aux OVS_UNUSED)
6484 struct ofproto_dpif *ofproto;
6487 ofproto = ofproto_dpif_lookup(argv[1]);
6489 unixctl_command_reply_error(conn, "no such bridge");
6492 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6494 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6495 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6499 unixctl_command_reply(conn, "table successfully flushed");
6503 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6504 const char *argv[], void *aux OVS_UNUSED)
6506 struct ds ds = DS_EMPTY_INITIALIZER;
6507 const struct ofproto_dpif *ofproto;
6508 const struct mac_entry *e;
6510 ofproto = ofproto_dpif_lookup(argv[1]);
6512 unixctl_command_reply_error(conn, "no such bridge");
6516 ds_put_cstr(&ds, " port VLAN MAC Age\n");
6517 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
6518 struct ofbundle *bundle = e->port.p;
6519 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
6520 ofbundle_get_a_port(bundle)->odp_port,
6521 e->vlan, ETH_ADDR_ARGS(e->mac),
6522 mac_entry_age(ofproto->ml, e));
6524 unixctl_command_reply(conn, ds_cstr(&ds));
6529 struct action_xlate_ctx ctx;
6535 trace_format_rule(struct ds *result, uint8_t table_id, int level,
6536 const struct rule_dpif *rule)
6538 ds_put_char_multiple(result, '\t', level);
6540 ds_put_cstr(result, "No match\n");
6544 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
6545 table_id, ntohll(rule->up.flow_cookie));
6546 cls_rule_format(&rule->up.cr, result);
6547 ds_put_char(result, '\n');
6549 ds_put_char_multiple(result, '\t', level);
6550 ds_put_cstr(result, "OpenFlow ");
6551 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
6552 ds_put_char(result, '\n');
6556 trace_format_flow(struct ds *result, int level, const char *title,
6557 struct trace_ctx *trace)
6559 ds_put_char_multiple(result, '\t', level);
6560 ds_put_format(result, "%s: ", title);
6561 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
6562 ds_put_cstr(result, "unchanged");
6564 flow_format(result, &trace->ctx.flow);
6565 trace->flow = trace->ctx.flow;
6567 ds_put_char(result, '\n');
6571 trace_format_regs(struct ds *result, int level, const char *title,
6572 struct trace_ctx *trace)
6576 ds_put_char_multiple(result, '\t', level);
6577 ds_put_format(result, "%s:", title);
6578 for (i = 0; i < FLOW_N_REGS; i++) {
6579 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
6581 ds_put_char(result, '\n');
6585 trace_format_odp(struct ds *result, int level, const char *title,
6586 struct trace_ctx *trace)
6588 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
6590 ds_put_char_multiple(result, '\t', level);
6591 ds_put_format(result, "%s: ", title);
6592 format_odp_actions(result, odp_actions->data, odp_actions->size);
6593 ds_put_char(result, '\n');
6597 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
6599 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
6600 struct ds *result = trace->result;
6602 ds_put_char(result, '\n');
6603 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
6604 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
6605 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
6606 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
6610 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
6611 void *aux OVS_UNUSED)
6613 const char *dpname = argv[1];
6614 struct ofproto_dpif *ofproto;
6615 struct ofpbuf odp_key;
6616 struct ofpbuf *packet;
6617 ovs_be16 initial_tci;
6623 ofpbuf_init(&odp_key, 0);
6626 ofproto = ofproto_dpif_lookup(dpname);
6628 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
6632 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
6633 /* ofproto/trace dpname flow [-generate] */
6634 const char *flow_s = argv[2];
6635 const char *generate_s = argv[3];
6637 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
6638 * flow. We guess which type it is based on whether 'flow_s' contains
6639 * an '(', since a datapath flow always contains '(') but an
6640 * OpenFlow-like flow should not (in fact it's allowed but I believe
6641 * that's not documented anywhere).
6643 * An alternative would be to try to parse 'flow_s' both ways, but then
6644 * it would be tricky giving a sensible error message. After all, do
6645 * you just say "syntax error" or do you present both error messages?
6646 * Both choices seem lousy. */
6647 if (strchr(flow_s, '(')) {
6650 /* Convert string to datapath key. */
6651 ofpbuf_init(&odp_key, 0);
6652 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
6654 unixctl_command_reply_error(conn, "Bad flow syntax");
6658 /* Convert odp_key to flow. */
6659 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
6660 odp_key.size, &flow,
6661 &initial_tci, NULL);
6662 if (error == ODP_FIT_ERROR) {
6663 unixctl_command_reply_error(conn, "Invalid flow");
6669 error_s = parse_ofp_exact_flow(&flow, argv[2]);
6671 unixctl_command_reply_error(conn, error_s);
6676 initial_tci = flow.vlan_tci;
6677 vsp_adjust_flow(ofproto, &flow);
6680 /* Generate a packet, if requested. */
6682 packet = ofpbuf_new(0);
6683 flow_compose(packet, &flow);
6685 } else if (argc == 6) {
6686 /* ofproto/trace dpname priority tun_id in_port packet */
6687 const char *priority_s = argv[2];
6688 const char *tun_id_s = argv[3];
6689 const char *in_port_s = argv[4];
6690 const char *packet_s = argv[5];
6691 uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s));
6692 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
6693 uint32_t priority = atoi(priority_s);
6696 msg = eth_from_hex(packet_s, &packet);
6698 unixctl_command_reply_error(conn, msg);
6702 ds_put_cstr(&result, "Packet: ");
6703 s = ofp_packet_to_string(packet->data, packet->size);
6704 ds_put_cstr(&result, s);
6707 flow_extract(packet, priority, tun_id, in_port, &flow);
6708 initial_tci = flow.vlan_tci;
6710 unixctl_command_reply_error(conn, "Bad command syntax");
6714 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
6715 unixctl_command_reply(conn, ds_cstr(&result));
6718 ds_destroy(&result);
6719 ofpbuf_delete(packet);
6720 ofpbuf_uninit(&odp_key);
6724 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
6725 const struct ofpbuf *packet, ovs_be16 initial_tci,
6728 struct rule_dpif *rule;
6730 ds_put_cstr(ds, "Flow: ");
6731 flow_format(ds, flow);
6732 ds_put_char(ds, '\n');
6734 rule = rule_dpif_lookup(ofproto, flow);
6736 trace_format_rule(ds, 0, 0, rule);
6737 if (rule == ofproto->miss_rule) {
6738 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
6739 } else if (rule == ofproto->no_packet_in_rule) {
6740 ds_put_cstr(ds, "\nNo match, packets dropped because "
6741 "OFPPC_NO_PACKET_IN is set on in_port.\n");
6745 uint64_t odp_actions_stub[1024 / 8];
6746 struct ofpbuf odp_actions;
6748 struct trace_ctx trace;
6751 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
6754 ofpbuf_use_stub(&odp_actions,
6755 odp_actions_stub, sizeof odp_actions_stub);
6756 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
6757 rule, tcp_flags, packet);
6758 trace.ctx.resubmit_hook = trace_resubmit;
6759 xlate_actions(&trace.ctx, rule->up.actions, rule->up.n_actions,
6762 ds_put_char(ds, '\n');
6763 trace_format_flow(ds, 0, "Final flow", &trace);
6764 ds_put_cstr(ds, "Datapath actions: ");
6765 format_odp_actions(ds, odp_actions.data, odp_actions.size);
6766 ofpbuf_uninit(&odp_actions);
6768 if (trace.ctx.slow) {
6769 enum slow_path_reason slow;
6771 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
6772 "slow path because it:");
6773 for (slow = trace.ctx.slow; slow; ) {
6774 enum slow_path_reason bit = rightmost_1bit(slow);
6778 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
6781 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
6784 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
6787 ds_put_cstr(ds, "\n\t- Needs in-band special case "
6790 ds_put_cstr(ds, "\n\t (The datapath actions are "
6791 "incomplete--for complete actions, "
6792 "please supply a packet.)");
6795 case SLOW_CONTROLLER:
6796 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
6797 "to the OpenFlow controller.");
6800 ds_put_cstr(ds, "\n\t- Needs more specific matching "
6801 "than the datapath supports.");
6808 if (slow & ~SLOW_MATCH) {
6809 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
6810 "the special slow-path processing.");
6817 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6818 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6821 unixctl_command_reply(conn, NULL);
6825 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6826 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6829 unixctl_command_reply(conn, NULL);
6832 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
6833 * 'reply' describing the results. */
6835 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
6837 struct facet *facet;
6841 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6842 if (!facet_check_consistency(facet)) {
6847 ofproto->need_revalidate = true;
6851 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
6852 ofproto->up.name, errors);
6854 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
6859 ofproto_dpif_self_check(struct unixctl_conn *conn,
6860 int argc, const char *argv[], void *aux OVS_UNUSED)
6862 struct ds reply = DS_EMPTY_INITIALIZER;
6863 struct ofproto_dpif *ofproto;
6866 ofproto = ofproto_dpif_lookup(argv[1]);
6868 unixctl_command_reply_error(conn, "Unknown ofproto (use "
6869 "ofproto/list for help)");
6872 ofproto_dpif_self_check__(ofproto, &reply);
6874 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6875 ofproto_dpif_self_check__(ofproto, &reply);
6879 unixctl_command_reply(conn, ds_cstr(&reply));
6884 ofproto_dpif_unixctl_init(void)
6886 static bool registered;
6892 unixctl_command_register(
6894 "bridge {tun_id in_port packet | odp_flow [-generate]}",
6895 2, 5, ofproto_unixctl_trace, NULL);
6896 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6897 ofproto_unixctl_fdb_flush, NULL);
6898 unixctl_command_register("fdb/show", "bridge", 1, 1,
6899 ofproto_unixctl_fdb_show, NULL);
6900 unixctl_command_register("ofproto/clog", "", 0, 0,
6901 ofproto_dpif_clog, NULL);
6902 unixctl_command_register("ofproto/unclog", "", 0, 0,
6903 ofproto_dpif_unclog, NULL);
6904 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6905 ofproto_dpif_self_check, NULL);
6908 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6910 * This is deprecated. It is only for compatibility with broken device drivers
6911 * in old versions of Linux that do not properly support VLANs when VLAN
6912 * devices are not used. When broken device drivers are no longer in
6913 * widespread use, we will delete these interfaces. */
6916 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
6918 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6919 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6921 if (realdev_ofp_port == ofport->realdev_ofp_port
6922 && vid == ofport->vlandev_vid) {
6926 ofproto->need_revalidate = true;
6928 if (ofport->realdev_ofp_port) {
6931 if (realdev_ofp_port && ofport->bundle) {
6932 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6933 * themselves be part of a bundle. */
6934 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6937 ofport->realdev_ofp_port = realdev_ofp_port;
6938 ofport->vlandev_vid = vid;
6940 if (realdev_ofp_port) {
6941 vsp_add(ofport, realdev_ofp_port, vid);
6948 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
6950 return hash_2words(realdev_ofp_port, vid);
6953 /* Returns the ODP port number of the Linux VLAN device that corresponds to
6954 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
6955 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
6956 * it would return the port number of eth0.9.
6958 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
6959 * function just returns its 'realdev_odp_port' argument. */
6961 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6962 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
6964 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6965 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
6966 int vid = vlan_tci_to_vid(vlan_tci);
6967 const struct vlan_splinter *vsp;
6969 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6970 hash_realdev_vid(realdev_ofp_port, vid),
6971 &ofproto->realdev_vid_map) {
6972 if (vsp->realdev_ofp_port == realdev_ofp_port
6973 && vsp->vid == vid) {
6974 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
6978 return realdev_odp_port;
6981 static struct vlan_splinter *
6982 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
6984 struct vlan_splinter *vsp;
6986 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
6987 &ofproto->vlandev_map) {
6988 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6996 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6997 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6998 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6999 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
7000 * eth0 and store 9 in '*vid'.
7002 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
7003 * VLAN device. Unless VLAN splinters are enabled, this is what this function
7006 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
7007 uint16_t vlandev_ofp_port, int *vid)
7009 if (!hmap_is_empty(&ofproto->vlandev_map)) {
7010 const struct vlan_splinter *vsp;
7012 vsp = vlandev_find(ofproto, vlandev_ofp_port);
7017 return vsp->realdev_ofp_port;
7023 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
7024 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
7025 * 'flow->in_port' to the "real" device backing the VLAN device, sets
7026 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
7027 * always the case unless VLAN splinters are enabled), returns false without
7028 * making any changes. */
7030 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
7035 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
7040 /* Cause the flow to be processed as if it came in on the real device with
7041 * the VLAN device's VLAN ID. */
7042 flow->in_port = realdev;
7043 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
7048 vsp_remove(struct ofport_dpif *port)
7050 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7051 struct vlan_splinter *vsp;
7053 vsp = vlandev_find(ofproto, port->up.ofp_port);
7055 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
7056 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
7059 port->realdev_ofp_port = 0;
7061 VLOG_ERR("missing vlan device record");
7066 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
7068 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7070 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
7071 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
7072 == realdev_ofp_port)) {
7073 struct vlan_splinter *vsp;
7075 vsp = xmalloc(sizeof *vsp);
7076 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
7077 hash_int(port->up.ofp_port, 0));
7078 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
7079 hash_realdev_vid(realdev_ofp_port, vid));
7080 vsp->realdev_ofp_port = realdev_ofp_port;
7081 vsp->vlandev_ofp_port = port->up.ofp_port;
7084 port->realdev_ofp_port = realdev_ofp_port;
7086 VLOG_ERR("duplicate vlan device record");
7090 const struct ofproto_class ofproto_dpif_class = {
7119 port_is_lacp_current,
7120 NULL, /* rule_choose_table */
7127 rule_modify_actions,
7135 get_cfm_remote_mpids,
7140 get_stp_port_status,
7147 is_mirror_output_bundle,
7148 forward_bpdu_changed,