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"
52 #include "unaligned.h"
54 #include "vlan-bitmap.h"
57 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
59 COVERAGE_DEFINE(ofproto_dpif_ctlr_action);
60 COVERAGE_DEFINE(ofproto_dpif_expired);
61 COVERAGE_DEFINE(ofproto_dpif_xlate);
62 COVERAGE_DEFINE(facet_changed_rule);
63 COVERAGE_DEFINE(facet_invalidated);
64 COVERAGE_DEFINE(facet_revalidate);
65 COVERAGE_DEFINE(facet_unexpected);
66 COVERAGE_DEFINE(facet_suppress);
68 /* Maximum depth of flow table recursion (due to resubmit actions) in a
69 * flow translation. */
70 #define MAX_RESUBMIT_RECURSION 32
72 /* Number of implemented OpenFlow tables. */
73 enum { N_TABLES = 255 };
74 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
75 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
85 * - Do include packets and bytes from facets that have been deleted or
86 * whose own statistics have been folded into the rule.
88 * - Do include packets and bytes sent "by hand" that were accounted to
89 * the rule without any facet being involved (this is a rare corner
90 * case in rule_execute()).
92 * - Do not include packet or bytes that can be obtained from any facet's
93 * packet_count or byte_count member or that can be obtained from the
94 * datapath by, e.g., dpif_flow_get() for any subfacet.
96 uint64_t packet_count; /* Number of packets received. */
97 uint64_t byte_count; /* Number of bytes received. */
99 tag_type tag; /* Caches rule_calculate_tag() result. */
101 struct list facets; /* List of "struct facet"s. */
104 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
106 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
109 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
110 const struct flow *);
111 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
115 static void rule_credit_stats(struct rule_dpif *,
116 const struct dpif_flow_stats *);
117 static void flow_push_stats(struct rule_dpif *, const struct flow *,
118 const struct dpif_flow_stats *);
119 static tag_type rule_calculate_tag(const struct flow *,
120 const struct flow_wildcards *,
122 static void rule_invalidate(const struct rule_dpif *);
124 #define MAX_MIRRORS 32
125 typedef uint32_t mirror_mask_t;
126 #define MIRROR_MASK_C(X) UINT32_C(X)
127 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
129 struct ofproto_dpif *ofproto; /* Owning ofproto. */
130 size_t idx; /* In ofproto's "mirrors" array. */
131 void *aux; /* Key supplied by ofproto's client. */
132 char *name; /* Identifier for log messages. */
134 /* Selection criteria. */
135 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
136 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
137 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
139 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
140 struct ofbundle *out; /* Output port or NULL. */
141 int out_vlan; /* Output VLAN or -1. */
142 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
145 int64_t packet_count; /* Number of packets sent. */
146 int64_t byte_count; /* Number of bytes sent. */
149 static void mirror_destroy(struct ofmirror *);
150 static void update_mirror_stats(struct ofproto_dpif *ofproto,
151 mirror_mask_t mirrors,
152 uint64_t packets, uint64_t bytes);
155 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
156 struct ofproto_dpif *ofproto; /* Owning ofproto. */
157 void *aux; /* Key supplied by ofproto's client. */
158 char *name; /* Identifier for log messages. */
161 struct list ports; /* Contains "struct ofport"s. */
162 enum port_vlan_mode vlan_mode; /* VLAN mode */
163 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
164 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
165 * NULL if all VLANs are trunked. */
166 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
167 struct bond *bond; /* Nonnull iff more than one port. */
168 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
171 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
173 /* Port mirroring info. */
174 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
175 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
176 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
179 static void bundle_remove(struct ofport *);
180 static void bundle_update(struct ofbundle *);
181 static void bundle_destroy(struct ofbundle *);
182 static void bundle_del_port(struct ofport_dpif *);
183 static void bundle_run(struct ofbundle *);
184 static void bundle_wait(struct ofbundle *);
185 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
186 uint16_t in_port, bool warn,
187 struct ofport_dpif **in_ofportp);
189 /* A controller may use OFPP_NONE as the ingress port to indicate that
190 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
191 * when an input bundle is needed for validation (e.g., mirroring or
192 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
193 * any 'port' structs, so care must be taken when dealing with it. */
194 static struct ofbundle ofpp_none_bundle = {
196 .vlan_mode = PORT_VLAN_TRUNK
199 static void stp_run(struct ofproto_dpif *ofproto);
200 static void stp_wait(struct ofproto_dpif *ofproto);
201 static int set_stp_port(struct ofport *,
202 const struct ofproto_port_stp_settings *);
204 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
206 struct action_xlate_ctx {
207 /* action_xlate_ctx_init() initializes these members. */
210 struct ofproto_dpif *ofproto;
212 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
213 * this flow when actions change header fields. */
216 /* The packet corresponding to 'flow', or a null pointer if we are
217 * revalidating without a packet to refer to. */
218 const struct ofpbuf *packet;
220 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
221 * actions update the flow table?
223 * We want to update these tables if we are actually processing a packet,
224 * or if we are accounting for packets that the datapath has processed, but
225 * not if we are just revalidating. */
228 /* The rule that we are currently translating, or NULL. */
229 struct rule_dpif *rule;
231 /* Union of the set of TCP flags seen so far in this flow. (Used only by
232 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
236 /* If nonnull, flow translation calls this function just before executing a
237 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
238 * when the recursion depth is exceeded.
240 * 'rule' is the rule being submitted into. It will be null if the
241 * resubmit or OFPP_TABLE action didn't find a matching rule.
243 * This is normally null so the client has to set it manually after
244 * calling action_xlate_ctx_init(). */
245 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule);
247 /* If nonnull, flow translation credits the specified statistics to each
248 * rule reached through a resubmit or OFPP_TABLE action.
250 * This is normally null so the client has to set it manually after
251 * calling action_xlate_ctx_init(). */
252 const struct dpif_flow_stats *resubmit_stats;
254 /* xlate_actions() initializes and uses these members. The client might want
255 * to look at them after it returns. */
257 struct ofpbuf *odp_actions; /* Datapath actions. */
258 tag_type tags; /* Tags associated with actions. */
259 enum slow_path_reason slow; /* 0 if fast path may be used. */
260 bool has_learn; /* Actions include NXAST_LEARN? */
261 bool has_normal; /* Actions output to OFPP_NORMAL? */
262 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
263 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
264 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
266 /* xlate_actions() initializes and uses these members, but the client has no
267 * reason to look at them. */
269 int recurse; /* Recursion level, via xlate_table_action. */
270 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
271 struct flow base_flow; /* Flow at the last commit. */
272 uint32_t orig_skb_priority; /* Priority when packet arrived. */
273 uint8_t table_id; /* OpenFlow table ID where flow was found. */
274 uint32_t sflow_n_outputs; /* Number of output ports. */
275 uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
276 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
277 bool exit; /* No further actions should be processed. */
278 struct flow orig_flow; /* Copy of original flow. */
281 static void action_xlate_ctx_init(struct action_xlate_ctx *,
282 struct ofproto_dpif *, const struct flow *,
283 ovs_be16 initial_tci, struct rule_dpif *,
284 uint8_t tcp_flags, const struct ofpbuf *);
285 static void xlate_actions(struct action_xlate_ctx *,
286 const union ofp_action *in, size_t n_in,
287 struct ofpbuf *odp_actions);
288 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
289 const union ofp_action *in,
292 static size_t put_userspace_action(const struct ofproto_dpif *,
293 struct ofpbuf *odp_actions,
295 const union user_action_cookie *);
297 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
298 enum slow_path_reason,
299 uint64_t *stub, size_t stub_size,
300 const struct nlattr **actionsp,
301 size_t *actions_lenp);
303 /* A subfacet (see "struct subfacet" below) has three possible installation
306 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
307 * case just after the subfacet is created, just before the subfacet is
308 * destroyed, or if the datapath returns an error when we try to install a
311 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
313 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
314 * ofproto_dpif is installed in the datapath.
317 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
318 SF_FAST_PATH, /* Full actions are installed. */
319 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
322 static const char *subfacet_path_to_string(enum subfacet_path);
324 /* A dpif flow and actions associated with a facet.
326 * See also the large comment on struct facet. */
329 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
330 struct list list_node; /* In struct facet's 'facets' list. */
331 struct facet *facet; /* Owning facet. */
335 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
336 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
337 * regenerate the ODP flow key from ->facet->flow. */
338 enum odp_key_fitness key_fitness;
342 long long int used; /* Time last used; time created if not used. */
344 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
345 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
349 * These should be essentially identical for every subfacet in a facet, but
350 * may differ in trivial ways due to VLAN splinters. */
351 size_t actions_len; /* Number of bytes in actions[]. */
352 struct nlattr *actions; /* Datapath actions. */
354 enum slow_path_reason slow; /* 0 if fast path may be used. */
355 enum subfacet_path path; /* Installed in datapath? */
357 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
358 * splinters can cause it to differ. This value should be removed when
359 * the VLAN splinters feature is no longer needed. */
360 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
363 static struct subfacet *subfacet_create(struct facet *, enum odp_key_fitness,
364 const struct nlattr *key,
365 size_t key_len, ovs_be16 initial_tci);
366 static struct subfacet *subfacet_find(struct ofproto_dpif *,
367 const struct nlattr *key, size_t key_len);
368 static void subfacet_destroy(struct subfacet *);
369 static void subfacet_destroy__(struct subfacet *);
370 static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *,
372 static void subfacet_reset_dp_stats(struct subfacet *,
373 struct dpif_flow_stats *);
374 static void subfacet_update_time(struct subfacet *, long long int used);
375 static void subfacet_update_stats(struct subfacet *,
376 const struct dpif_flow_stats *);
377 static void subfacet_make_actions(struct subfacet *,
378 const struct ofpbuf *packet,
379 struct ofpbuf *odp_actions);
380 static int subfacet_install(struct subfacet *,
381 const struct nlattr *actions, size_t actions_len,
382 struct dpif_flow_stats *, enum slow_path_reason);
383 static void subfacet_uninstall(struct subfacet *);
385 static enum subfacet_path subfacet_want_path(enum slow_path_reason);
387 /* An exact-match instantiation of an OpenFlow flow.
389 * A facet associates a "struct flow", which represents the Open vSwitch
390 * userspace idea of an exact-match flow, with one or more subfacets. Each
391 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
392 * the facet. When the kernel module (or other dpif implementation) and Open
393 * vSwitch userspace agree on the definition of a flow key, there is exactly
394 * one subfacet per facet. If the dpif implementation supports more-specific
395 * flow matching than userspace, however, a facet can have more than one
396 * subfacet, each of which corresponds to some distinction in flow that
397 * userspace simply doesn't understand.
399 * Flow expiration works in terms of subfacets, so a facet must have at least
400 * one subfacet or it will never expire, leaking memory. */
403 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
404 struct list list_node; /* In owning rule's 'facets' list. */
405 struct rule_dpif *rule; /* Owning rule. */
408 struct list subfacets;
409 long long int used; /* Time last used; time created if not used. */
416 * - Do include packets and bytes sent "by hand", e.g. with
419 * - Do include packets and bytes that were obtained from the datapath
420 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
421 * DPIF_FP_ZERO_STATS).
423 * - Do not include packets or bytes that can be obtained from the
424 * datapath for any existing subfacet.
426 uint64_t packet_count; /* Number of packets received. */
427 uint64_t byte_count; /* Number of bytes received. */
429 /* Resubmit statistics. */
430 uint64_t prev_packet_count; /* Number of packets from last stats push. */
431 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
432 long long int prev_used; /* Used time from last stats push. */
435 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
436 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
437 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
439 /* Properties of datapath actions.
441 * Every subfacet has its own actions because actions can differ slightly
442 * between splintered and non-splintered subfacets due to the VLAN tag
443 * being initially different (present vs. absent). All of them have these
444 * properties in common so we just store one copy of them here. */
445 bool has_learn; /* Actions include NXAST_LEARN? */
446 bool has_normal; /* Actions output to OFPP_NORMAL? */
447 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
448 tag_type tags; /* Tags that would require revalidation. */
449 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
451 /* Storage for a single subfacet, to reduce malloc() time and space
452 * overhead. (A facet always has at least one subfacet and in the common
453 * case has exactly one subfacet.) */
454 struct subfacet one_subfacet;
457 static struct facet *facet_create(struct rule_dpif *,
458 const struct flow *, uint32_t hash);
459 static void facet_remove(struct facet *);
460 static void facet_free(struct facet *);
462 static struct facet *facet_find(struct ofproto_dpif *,
463 const struct flow *, uint32_t hash);
464 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
465 const struct flow *, uint32_t hash);
466 static void facet_revalidate(struct facet *);
467 static bool facet_check_consistency(struct facet *);
469 static void facet_flush_stats(struct facet *);
471 static void facet_update_time(struct facet *, long long int used);
472 static void facet_reset_counters(struct facet *);
473 static void facet_push_stats(struct facet *);
474 static void facet_learn(struct facet *);
475 static void facet_account(struct facet *);
477 static bool facet_is_controller_flow(struct facet *);
483 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
484 struct list bundle_node; /* In struct ofbundle's "ports" list. */
485 struct cfm *cfm; /* Connectivity Fault Management, if any. */
486 tag_type tag; /* Tag associated with this port. */
487 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
488 bool may_enable; /* May be enabled in bonds. */
489 long long int carrier_seq; /* Carrier status changes. */
492 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
493 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
494 long long int stp_state_entered;
496 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
498 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
500 * This is deprecated. It is only for compatibility with broken device
501 * drivers in old versions of Linux that do not properly support VLANs when
502 * VLAN devices are not used. When broken device drivers are no longer in
503 * widespread use, we will delete these interfaces. */
504 uint16_t realdev_ofp_port;
508 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
509 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
510 * traffic egressing the 'ofport' with that priority should be marked with. */
511 struct priority_to_dscp {
512 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
513 uint32_t priority; /* Priority of this queue (see struct flow). */
515 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
518 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
520 * This is deprecated. It is only for compatibility with broken device drivers
521 * in old versions of Linux that do not properly support VLANs when VLAN
522 * devices are not used. When broken device drivers are no longer in
523 * widespread use, we will delete these interfaces. */
524 struct vlan_splinter {
525 struct hmap_node realdev_vid_node;
526 struct hmap_node vlandev_node;
527 uint16_t realdev_ofp_port;
528 uint16_t vlandev_ofp_port;
532 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
533 uint32_t realdev, ovs_be16 vlan_tci);
534 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
535 static void vsp_remove(struct ofport_dpif *);
536 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
538 static struct ofport_dpif *
539 ofport_dpif_cast(const struct ofport *ofport)
541 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
542 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
545 static void port_run(struct ofport_dpif *);
546 static void port_wait(struct ofport_dpif *);
547 static int set_cfm(struct ofport *, const struct cfm_settings *);
548 static void ofport_clear_priorities(struct ofport_dpif *);
550 struct dpif_completion {
551 struct list list_node;
552 struct ofoperation *op;
555 /* Extra information about a classifier table.
556 * Currently used just for optimized flow revalidation. */
558 /* If either of these is nonnull, then this table has a form that allows
559 * flows to be tagged to avoid revalidating most flows for the most common
560 * kinds of flow table changes. */
561 struct cls_table *catchall_table; /* Table that wildcards all fields. */
562 struct cls_table *other_table; /* Table with any other wildcard set. */
563 uint32_t basis; /* Keeps each table's tags separate. */
566 struct ofproto_dpif {
567 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
572 /* Special OpenFlow rules. */
573 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
574 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
580 struct netflow *netflow;
581 struct dpif_sflow *sflow;
582 struct hmap bundles; /* Contains "struct ofbundle"s. */
583 struct mac_learning *ml;
584 struct ofmirror *mirrors[MAX_MIRRORS];
586 bool has_bonded_bundles;
589 struct timer next_expiration;
593 struct hmap subfacets;
594 struct governor *governor;
597 struct table_dpif tables[N_TABLES];
598 bool need_revalidate;
599 struct tag_set revalidate_set;
601 /* Support for debugging async flow mods. */
602 struct list completions;
604 bool has_bundle_action; /* True when the first bundle action appears. */
605 struct netdev_stats stats; /* To account packets generated and consumed in
610 long long int stp_last_tick;
612 /* VLAN splinters. */
613 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
614 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
617 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
618 * for debugging the asynchronous flow_mod implementation.) */
621 /* All existing ofproto_dpif instances, indexed by ->up.name. */
622 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
624 static void ofproto_dpif_unixctl_init(void);
626 static struct ofproto_dpif *
627 ofproto_dpif_cast(const struct ofproto *ofproto)
629 assert(ofproto->ofproto_class == &ofproto_dpif_class);
630 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
633 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
635 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
637 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
638 const struct ofpbuf *, ovs_be16 initial_tci,
641 /* Packet processing. */
642 static void update_learning_table(struct ofproto_dpif *,
643 const struct flow *, int vlan,
646 #define FLOW_MISS_MAX_BATCH 50
647 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
649 /* Flow expiration. */
650 static int expire(struct ofproto_dpif *);
653 static void send_netflow_active_timeouts(struct ofproto_dpif *);
656 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
657 static size_t compose_sflow_action(const struct ofproto_dpif *,
658 struct ofpbuf *odp_actions,
659 const struct flow *, uint32_t odp_port);
660 static void add_mirror_actions(struct action_xlate_ctx *ctx,
661 const struct flow *flow);
662 /* Global variables. */
663 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
665 /* Factory functions. */
668 enumerate_types(struct sset *types)
670 dp_enumerate_types(types);
674 enumerate_names(const char *type, struct sset *names)
676 return dp_enumerate_names(type, names);
680 del(const char *type, const char *name)
685 error = dpif_open(name, type, &dpif);
687 error = dpif_delete(dpif);
693 /* Basic life-cycle. */
695 static int add_internal_flows(struct ofproto_dpif *);
697 static struct ofproto *
700 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
705 dealloc(struct ofproto *ofproto_)
707 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
712 construct(struct ofproto *ofproto_)
714 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
715 const char *name = ofproto->up.name;
719 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
721 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
725 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
726 ofproto->n_matches = 0;
728 dpif_flow_flush(ofproto->dpif);
729 dpif_recv_purge(ofproto->dpif);
731 error = dpif_recv_set(ofproto->dpif, true);
733 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
734 dpif_close(ofproto->dpif);
738 ofproto->netflow = NULL;
739 ofproto->sflow = NULL;
741 hmap_init(&ofproto->bundles);
742 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
743 for (i = 0; i < MAX_MIRRORS; i++) {
744 ofproto->mirrors[i] = NULL;
746 ofproto->has_bonded_bundles = false;
748 timer_set_duration(&ofproto->next_expiration, 1000);
750 hmap_init(&ofproto->facets);
751 hmap_init(&ofproto->subfacets);
752 ofproto->governor = NULL;
754 for (i = 0; i < N_TABLES; i++) {
755 struct table_dpif *table = &ofproto->tables[i];
757 table->catchall_table = NULL;
758 table->other_table = NULL;
759 table->basis = random_uint32();
761 ofproto->need_revalidate = false;
762 tag_set_init(&ofproto->revalidate_set);
764 list_init(&ofproto->completions);
766 ofproto_dpif_unixctl_init();
768 ofproto->has_mirrors = false;
769 ofproto->has_bundle_action = false;
771 hmap_init(&ofproto->vlandev_map);
772 hmap_init(&ofproto->realdev_vid_map);
774 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
775 hash_string(ofproto->up.name, 0));
776 memset(&ofproto->stats, 0, sizeof ofproto->stats);
778 ofproto_init_tables(ofproto_, N_TABLES);
779 error = add_internal_flows(ofproto);
780 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
786 add_internal_flow(struct ofproto_dpif *ofproto, int id,
787 const struct ofpbuf *actions, struct rule_dpif **rulep)
789 struct ofputil_flow_mod fm;
792 cls_rule_init_catchall(&fm.cr, 0);
793 cls_rule_set_reg(&fm.cr, 0, id);
794 fm.new_cookie = htonll(0);
795 fm.cookie = htonll(0);
796 fm.cookie_mask = htonll(0);
797 fm.table_id = TBL_INTERNAL;
798 fm.command = OFPFC_ADD;
804 fm.actions = actions->data;
805 fm.n_actions = actions->size / sizeof(union ofp_action);
807 error = ofproto_flow_mod(&ofproto->up, &fm);
809 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
810 id, ofperr_to_string(error));
814 *rulep = rule_dpif_lookup__(ofproto, &fm.cr.flow, TBL_INTERNAL);
815 assert(*rulep != NULL);
821 add_internal_flows(struct ofproto_dpif *ofproto)
823 struct nx_action_controller *nac;
824 uint64_t actions_stub[128 / 8];
825 struct ofpbuf actions;
829 ofpbuf_use_stack(&actions, actions_stub, sizeof actions_stub);
832 nac = ofputil_put_NXAST_CONTROLLER(&actions);
833 nac->max_len = htons(UINT16_MAX);
834 nac->controller_id = htons(0);
835 nac->reason = OFPR_NO_MATCH;
836 error = add_internal_flow(ofproto, id++, &actions, &ofproto->miss_rule);
841 ofpbuf_clear(&actions);
842 error = add_internal_flow(ofproto, id++, &actions,
843 &ofproto->no_packet_in_rule);
848 complete_operations(struct ofproto_dpif *ofproto)
850 struct dpif_completion *c, *next;
852 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
853 ofoperation_complete(c->op, 0);
854 list_remove(&c->list_node);
860 destruct(struct ofproto *ofproto_)
862 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
863 struct rule_dpif *rule, *next_rule;
864 struct oftable *table;
867 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
868 complete_operations(ofproto);
870 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
871 struct cls_cursor cursor;
873 cls_cursor_init(&cursor, &table->cls, NULL);
874 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
875 ofproto_rule_destroy(&rule->up);
879 for (i = 0; i < MAX_MIRRORS; i++) {
880 mirror_destroy(ofproto->mirrors[i]);
883 netflow_destroy(ofproto->netflow);
884 dpif_sflow_destroy(ofproto->sflow);
885 hmap_destroy(&ofproto->bundles);
886 mac_learning_destroy(ofproto->ml);
888 hmap_destroy(&ofproto->facets);
889 hmap_destroy(&ofproto->subfacets);
890 governor_destroy(ofproto->governor);
892 hmap_destroy(&ofproto->vlandev_map);
893 hmap_destroy(&ofproto->realdev_vid_map);
895 dpif_close(ofproto->dpif);
899 run_fast(struct ofproto *ofproto_)
901 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
904 /* Handle one or more batches of upcalls, until there's nothing left to do
905 * or until we do a fixed total amount of work.
907 * We do work in batches because it can be much cheaper to set up a number
908 * of flows and fire off their patches all at once. We do multiple batches
909 * because in some cases handling a packet can cause another packet to be
910 * queued almost immediately as part of the return flow. Both
911 * optimizations can make major improvements on some benchmarks and
912 * presumably for real traffic as well. */
914 while (work < FLOW_MISS_MAX_BATCH) {
915 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
925 run(struct ofproto *ofproto_)
927 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
928 struct ofport_dpif *ofport;
929 struct ofbundle *bundle;
933 complete_operations(ofproto);
935 dpif_run(ofproto->dpif);
937 error = run_fast(ofproto_);
942 if (timer_expired(&ofproto->next_expiration)) {
943 int delay = expire(ofproto);
944 timer_set_duration(&ofproto->next_expiration, delay);
947 if (ofproto->netflow) {
948 if (netflow_run(ofproto->netflow)) {
949 send_netflow_active_timeouts(ofproto);
952 if (ofproto->sflow) {
953 dpif_sflow_run(ofproto->sflow);
956 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
959 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
964 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
966 /* Now revalidate if there's anything to do. */
967 if (ofproto->need_revalidate
968 || !tag_set_is_empty(&ofproto->revalidate_set)) {
969 struct tag_set revalidate_set = ofproto->revalidate_set;
970 bool revalidate_all = ofproto->need_revalidate;
973 /* Clear the revalidation flags. */
974 tag_set_init(&ofproto->revalidate_set);
975 ofproto->need_revalidate = false;
977 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
979 || tag_set_intersects(&revalidate_set, facet->tags)) {
980 facet_revalidate(facet);
985 /* Check the consistency of a random facet, to aid debugging. */
986 if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) {
989 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
990 struct facet, hmap_node);
991 if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) {
992 if (!facet_check_consistency(facet)) {
993 ofproto->need_revalidate = true;
998 if (ofproto->governor) {
1001 governor_run(ofproto->governor);
1003 /* If the governor has shrunk to its minimum size and the number of
1004 * subfacets has dwindled, then drop the governor entirely.
1006 * For hysteresis, the number of subfacets to drop the governor is
1007 * smaller than the number needed to trigger its creation. */
1008 n_subfacets = hmap_count(&ofproto->subfacets);
1009 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1010 && governor_is_idle(ofproto->governor)) {
1011 governor_destroy(ofproto->governor);
1012 ofproto->governor = NULL;
1020 wait(struct ofproto *ofproto_)
1022 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1023 struct ofport_dpif *ofport;
1024 struct ofbundle *bundle;
1026 if (!clogged && !list_is_empty(&ofproto->completions)) {
1027 poll_immediate_wake();
1030 dpif_wait(ofproto->dpif);
1031 dpif_recv_wait(ofproto->dpif);
1032 if (ofproto->sflow) {
1033 dpif_sflow_wait(ofproto->sflow);
1035 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
1036 poll_immediate_wake();
1038 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1041 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1042 bundle_wait(bundle);
1044 if (ofproto->netflow) {
1045 netflow_wait(ofproto->netflow);
1047 mac_learning_wait(ofproto->ml);
1049 if (ofproto->need_revalidate) {
1050 /* Shouldn't happen, but if it does just go around again. */
1051 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1052 poll_immediate_wake();
1054 timer_wait(&ofproto->next_expiration);
1056 if (ofproto->governor) {
1057 governor_wait(ofproto->governor);
1062 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1064 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1066 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1067 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1071 flush(struct ofproto *ofproto_)
1073 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1074 struct facet *facet, *next_facet;
1076 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1077 /* Mark the facet as not installed so that facet_remove() doesn't
1078 * bother trying to uninstall it. There is no point in uninstalling it
1079 * individually since we are about to blow away all the facets with
1080 * dpif_flow_flush(). */
1081 struct subfacet *subfacet;
1083 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
1084 subfacet->path = SF_NOT_INSTALLED;
1085 subfacet->dp_packet_count = 0;
1086 subfacet->dp_byte_count = 0;
1088 facet_remove(facet);
1090 dpif_flow_flush(ofproto->dpif);
1094 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1095 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1097 *arp_match_ip = true;
1098 *actions = (OFPUTIL_A_OUTPUT |
1099 OFPUTIL_A_SET_VLAN_VID |
1100 OFPUTIL_A_SET_VLAN_PCP |
1101 OFPUTIL_A_STRIP_VLAN |
1102 OFPUTIL_A_SET_DL_SRC |
1103 OFPUTIL_A_SET_DL_DST |
1104 OFPUTIL_A_SET_NW_SRC |
1105 OFPUTIL_A_SET_NW_DST |
1106 OFPUTIL_A_SET_NW_TOS |
1107 OFPUTIL_A_SET_TP_SRC |
1108 OFPUTIL_A_SET_TP_DST |
1113 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
1115 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1116 struct dpif_dp_stats s;
1118 strcpy(ots->name, "classifier");
1120 dpif_get_dp_stats(ofproto->dpif, &s);
1121 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
1122 put_32aligned_be64(&ots->matched_count,
1123 htonll(s.n_hit + ofproto->n_matches));
1126 static struct ofport *
1129 struct ofport_dpif *port = xmalloc(sizeof *port);
1134 port_dealloc(struct ofport *port_)
1136 struct ofport_dpif *port = ofport_dpif_cast(port_);
1141 port_construct(struct ofport *port_)
1143 struct ofport_dpif *port = ofport_dpif_cast(port_);
1144 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1146 ofproto->need_revalidate = true;
1147 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
1148 port->bundle = NULL;
1150 port->tag = tag_create_random();
1151 port->may_enable = true;
1152 port->stp_port = NULL;
1153 port->stp_state = STP_DISABLED;
1154 hmap_init(&port->priorities);
1155 port->realdev_ofp_port = 0;
1156 port->vlandev_vid = 0;
1157 port->carrier_seq = netdev_get_carrier_resets(port->up.netdev);
1159 if (ofproto->sflow) {
1160 dpif_sflow_add_port(ofproto->sflow, port_);
1167 port_destruct(struct ofport *port_)
1169 struct ofport_dpif *port = ofport_dpif_cast(port_);
1170 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1172 ofproto->need_revalidate = true;
1173 bundle_remove(port_);
1174 set_cfm(port_, NULL);
1175 if (ofproto->sflow) {
1176 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1179 ofport_clear_priorities(port);
1180 hmap_destroy(&port->priorities);
1184 port_modified(struct ofport *port_)
1186 struct ofport_dpif *port = ofport_dpif_cast(port_);
1188 if (port->bundle && port->bundle->bond) {
1189 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1194 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1196 struct ofport_dpif *port = ofport_dpif_cast(port_);
1197 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1198 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1200 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1201 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1202 OFPUTIL_PC_NO_PACKET_IN)) {
1203 ofproto->need_revalidate = true;
1205 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1206 bundle_update(port->bundle);
1212 set_sflow(struct ofproto *ofproto_,
1213 const struct ofproto_sflow_options *sflow_options)
1215 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1216 struct dpif_sflow *ds = ofproto->sflow;
1218 if (sflow_options) {
1220 struct ofport_dpif *ofport;
1222 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1223 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1224 dpif_sflow_add_port(ds, &ofport->up);
1226 ofproto->need_revalidate = true;
1228 dpif_sflow_set_options(ds, sflow_options);
1231 dpif_sflow_destroy(ds);
1232 ofproto->need_revalidate = true;
1233 ofproto->sflow = NULL;
1240 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1242 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1249 struct ofproto_dpif *ofproto;
1251 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1252 ofproto->need_revalidate = true;
1253 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1256 if (cfm_configure(ofport->cfm, s)) {
1262 cfm_destroy(ofport->cfm);
1268 get_cfm_fault(const struct ofport *ofport_)
1270 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1272 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1276 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1279 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1282 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1290 get_cfm_health(const struct ofport *ofport_)
1292 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1294 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1297 /* Spanning Tree. */
1300 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1302 struct ofproto_dpif *ofproto = ofproto_;
1303 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1304 struct ofport_dpif *ofport;
1306 ofport = stp_port_get_aux(sp);
1308 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1309 ofproto->up.name, port_num);
1311 struct eth_header *eth = pkt->l2;
1313 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1314 if (eth_addr_is_zero(eth->eth_src)) {
1315 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1316 "with unknown MAC", ofproto->up.name, port_num);
1318 send_packet(ofport, pkt);
1324 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1326 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1328 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1330 /* Only revalidate flows if the configuration changed. */
1331 if (!s != !ofproto->stp) {
1332 ofproto->need_revalidate = true;
1336 if (!ofproto->stp) {
1337 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1338 send_bpdu_cb, ofproto);
1339 ofproto->stp_last_tick = time_msec();
1342 stp_set_bridge_id(ofproto->stp, s->system_id);
1343 stp_set_bridge_priority(ofproto->stp, s->priority);
1344 stp_set_hello_time(ofproto->stp, s->hello_time);
1345 stp_set_max_age(ofproto->stp, s->max_age);
1346 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1348 struct ofport *ofport;
1350 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1351 set_stp_port(ofport, NULL);
1354 stp_destroy(ofproto->stp);
1355 ofproto->stp = NULL;
1362 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1364 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1368 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1369 s->designated_root = stp_get_designated_root(ofproto->stp);
1370 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1379 update_stp_port_state(struct ofport_dpif *ofport)
1381 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1382 enum stp_state state;
1384 /* Figure out new state. */
1385 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1389 if (ofport->stp_state != state) {
1390 enum ofputil_port_state of_state;
1393 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1394 netdev_get_name(ofport->up.netdev),
1395 stp_state_name(ofport->stp_state),
1396 stp_state_name(state));
1397 if (stp_learn_in_state(ofport->stp_state)
1398 != stp_learn_in_state(state)) {
1399 /* xxx Learning action flows should also be flushed. */
1400 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1402 fwd_change = stp_forward_in_state(ofport->stp_state)
1403 != stp_forward_in_state(state);
1405 ofproto->need_revalidate = true;
1406 ofport->stp_state = state;
1407 ofport->stp_state_entered = time_msec();
1409 if (fwd_change && ofport->bundle) {
1410 bundle_update(ofport->bundle);
1413 /* Update the STP state bits in the OpenFlow port description. */
1414 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1415 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1416 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1417 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1418 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1420 ofproto_port_set_state(&ofport->up, of_state);
1424 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1425 * caller is responsible for assigning STP port numbers and ensuring
1426 * there are no duplicates. */
1428 set_stp_port(struct ofport *ofport_,
1429 const struct ofproto_port_stp_settings *s)
1431 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1432 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1433 struct stp_port *sp = ofport->stp_port;
1435 if (!s || !s->enable) {
1437 ofport->stp_port = NULL;
1438 stp_port_disable(sp);
1439 update_stp_port_state(ofport);
1442 } else if (sp && stp_port_no(sp) != s->port_num
1443 && ofport == stp_port_get_aux(sp)) {
1444 /* The port-id changed, so disable the old one if it's not
1445 * already in use by another port. */
1446 stp_port_disable(sp);
1449 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1450 stp_port_enable(sp);
1452 stp_port_set_aux(sp, ofport);
1453 stp_port_set_priority(sp, s->priority);
1454 stp_port_set_path_cost(sp, s->path_cost);
1456 update_stp_port_state(ofport);
1462 get_stp_port_status(struct ofport *ofport_,
1463 struct ofproto_port_stp_status *s)
1465 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1466 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1467 struct stp_port *sp = ofport->stp_port;
1469 if (!ofproto->stp || !sp) {
1475 s->port_id = stp_port_get_id(sp);
1476 s->state = stp_port_get_state(sp);
1477 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1478 s->role = stp_port_get_role(sp);
1479 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1485 stp_run(struct ofproto_dpif *ofproto)
1488 long long int now = time_msec();
1489 long long int elapsed = now - ofproto->stp_last_tick;
1490 struct stp_port *sp;
1493 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1494 ofproto->stp_last_tick = now;
1496 while (stp_get_changed_port(ofproto->stp, &sp)) {
1497 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1500 update_stp_port_state(ofport);
1504 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1505 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1511 stp_wait(struct ofproto_dpif *ofproto)
1514 poll_timer_wait(1000);
1518 /* Returns true if STP should process 'flow'. */
1520 stp_should_process_flow(const struct flow *flow)
1522 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1526 stp_process_packet(const struct ofport_dpif *ofport,
1527 const struct ofpbuf *packet)
1529 struct ofpbuf payload = *packet;
1530 struct eth_header *eth = payload.data;
1531 struct stp_port *sp = ofport->stp_port;
1533 /* Sink packets on ports that have STP disabled when the bridge has
1535 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1539 /* Trim off padding on payload. */
1540 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1541 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1544 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1545 stp_received_bpdu(sp, payload.data, payload.size);
1549 static struct priority_to_dscp *
1550 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1552 struct priority_to_dscp *pdscp;
1555 hash = hash_int(priority, 0);
1556 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1557 if (pdscp->priority == priority) {
1565 ofport_clear_priorities(struct ofport_dpif *ofport)
1567 struct priority_to_dscp *pdscp, *next;
1569 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1570 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1576 set_queues(struct ofport *ofport_,
1577 const struct ofproto_port_queue *qdscp_list,
1580 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1581 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1582 struct hmap new = HMAP_INITIALIZER(&new);
1585 for (i = 0; i < n_qdscp; i++) {
1586 struct priority_to_dscp *pdscp;
1590 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1591 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1596 pdscp = get_priority(ofport, priority);
1598 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1600 pdscp = xmalloc(sizeof *pdscp);
1601 pdscp->priority = priority;
1603 ofproto->need_revalidate = true;
1606 if (pdscp->dscp != dscp) {
1608 ofproto->need_revalidate = true;
1611 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1614 if (!hmap_is_empty(&ofport->priorities)) {
1615 ofport_clear_priorities(ofport);
1616 ofproto->need_revalidate = true;
1619 hmap_swap(&new, &ofport->priorities);
1627 /* Expires all MAC learning entries associated with 'bundle' and forces its
1628 * ofproto to revalidate every flow.
1630 * Normally MAC learning entries are removed only from the ofproto associated
1631 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1632 * are removed from every ofproto. When patch ports and SLB bonds are in use
1633 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1634 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1635 * with the host from which it migrated. */
1637 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1639 struct ofproto_dpif *ofproto = bundle->ofproto;
1640 struct mac_learning *ml = ofproto->ml;
1641 struct mac_entry *mac, *next_mac;
1643 ofproto->need_revalidate = true;
1644 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1645 if (mac->port.p == bundle) {
1647 struct ofproto_dpif *o;
1649 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1651 struct mac_entry *e;
1653 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1656 tag_set_add(&o->revalidate_set, e->tag);
1657 mac_learning_expire(o->ml, e);
1663 mac_learning_expire(ml, mac);
1668 static struct ofbundle *
1669 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1671 struct ofbundle *bundle;
1673 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1674 &ofproto->bundles) {
1675 if (bundle->aux == aux) {
1682 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1683 * ones that are found to 'bundles'. */
1685 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1686 void **auxes, size_t n_auxes,
1687 struct hmapx *bundles)
1691 hmapx_init(bundles);
1692 for (i = 0; i < n_auxes; i++) {
1693 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1695 hmapx_add(bundles, bundle);
1701 bundle_update(struct ofbundle *bundle)
1703 struct ofport_dpif *port;
1705 bundle->floodable = true;
1706 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1707 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1708 || !stp_forward_in_state(port->stp_state)) {
1709 bundle->floodable = false;
1716 bundle_del_port(struct ofport_dpif *port)
1718 struct ofbundle *bundle = port->bundle;
1720 bundle->ofproto->need_revalidate = true;
1722 list_remove(&port->bundle_node);
1723 port->bundle = NULL;
1726 lacp_slave_unregister(bundle->lacp, port);
1729 bond_slave_unregister(bundle->bond, port);
1732 bundle_update(bundle);
1736 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1737 struct lacp_slave_settings *lacp,
1738 uint32_t bond_stable_id)
1740 struct ofport_dpif *port;
1742 port = get_ofp_port(bundle->ofproto, ofp_port);
1747 if (port->bundle != bundle) {
1748 bundle->ofproto->need_revalidate = true;
1750 bundle_del_port(port);
1753 port->bundle = bundle;
1754 list_push_back(&bundle->ports, &port->bundle_node);
1755 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1756 || !stp_forward_in_state(port->stp_state)) {
1757 bundle->floodable = false;
1761 port->bundle->ofproto->need_revalidate = true;
1762 lacp_slave_register(bundle->lacp, port, lacp);
1765 port->bond_stable_id = bond_stable_id;
1771 bundle_destroy(struct ofbundle *bundle)
1773 struct ofproto_dpif *ofproto;
1774 struct ofport_dpif *port, *next_port;
1781 ofproto = bundle->ofproto;
1782 for (i = 0; i < MAX_MIRRORS; i++) {
1783 struct ofmirror *m = ofproto->mirrors[i];
1785 if (m->out == bundle) {
1787 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1788 || hmapx_find_and_delete(&m->dsts, bundle)) {
1789 ofproto->need_revalidate = true;
1794 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1795 bundle_del_port(port);
1798 bundle_flush_macs(bundle, true);
1799 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1801 free(bundle->trunks);
1802 lacp_destroy(bundle->lacp);
1803 bond_destroy(bundle->bond);
1808 bundle_set(struct ofproto *ofproto_, void *aux,
1809 const struct ofproto_bundle_settings *s)
1811 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1812 bool need_flush = false;
1813 struct ofport_dpif *port;
1814 struct ofbundle *bundle;
1815 unsigned long *trunks;
1821 bundle_destroy(bundle_lookup(ofproto, aux));
1825 assert(s->n_slaves == 1 || s->bond != NULL);
1826 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1828 bundle = bundle_lookup(ofproto, aux);
1830 bundle = xmalloc(sizeof *bundle);
1832 bundle->ofproto = ofproto;
1833 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1834 hash_pointer(aux, 0));
1836 bundle->name = NULL;
1838 list_init(&bundle->ports);
1839 bundle->vlan_mode = PORT_VLAN_TRUNK;
1841 bundle->trunks = NULL;
1842 bundle->use_priority_tags = s->use_priority_tags;
1843 bundle->lacp = NULL;
1844 bundle->bond = NULL;
1846 bundle->floodable = true;
1848 bundle->src_mirrors = 0;
1849 bundle->dst_mirrors = 0;
1850 bundle->mirror_out = 0;
1853 if (!bundle->name || strcmp(s->name, bundle->name)) {
1855 bundle->name = xstrdup(s->name);
1860 if (!bundle->lacp) {
1861 ofproto->need_revalidate = true;
1862 bundle->lacp = lacp_create();
1864 lacp_configure(bundle->lacp, s->lacp);
1866 lacp_destroy(bundle->lacp);
1867 bundle->lacp = NULL;
1870 /* Update set of ports. */
1872 for (i = 0; i < s->n_slaves; i++) {
1873 if (!bundle_add_port(bundle, s->slaves[i],
1874 s->lacp ? &s->lacp_slaves[i] : NULL,
1875 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1879 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1880 struct ofport_dpif *next_port;
1882 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1883 for (i = 0; i < s->n_slaves; i++) {
1884 if (s->slaves[i] == port->up.ofp_port) {
1889 bundle_del_port(port);
1893 assert(list_size(&bundle->ports) <= s->n_slaves);
1895 if (list_is_empty(&bundle->ports)) {
1896 bundle_destroy(bundle);
1900 /* Set VLAN tagging mode */
1901 if (s->vlan_mode != bundle->vlan_mode
1902 || s->use_priority_tags != bundle->use_priority_tags) {
1903 bundle->vlan_mode = s->vlan_mode;
1904 bundle->use_priority_tags = s->use_priority_tags;
1909 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1910 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1912 if (vlan != bundle->vlan) {
1913 bundle->vlan = vlan;
1917 /* Get trunked VLANs. */
1918 switch (s->vlan_mode) {
1919 case PORT_VLAN_ACCESS:
1923 case PORT_VLAN_TRUNK:
1924 trunks = (unsigned long *) s->trunks;
1927 case PORT_VLAN_NATIVE_UNTAGGED:
1928 case PORT_VLAN_NATIVE_TAGGED:
1929 if (vlan != 0 && (!s->trunks
1930 || !bitmap_is_set(s->trunks, vlan)
1931 || bitmap_is_set(s->trunks, 0))) {
1932 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1934 trunks = bitmap_clone(s->trunks, 4096);
1936 trunks = bitmap_allocate1(4096);
1938 bitmap_set1(trunks, vlan);
1939 bitmap_set0(trunks, 0);
1941 trunks = (unsigned long *) s->trunks;
1948 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1949 free(bundle->trunks);
1950 if (trunks == s->trunks) {
1951 bundle->trunks = vlan_bitmap_clone(trunks);
1953 bundle->trunks = trunks;
1958 if (trunks != s->trunks) {
1963 if (!list_is_short(&bundle->ports)) {
1964 bundle->ofproto->has_bonded_bundles = true;
1966 if (bond_reconfigure(bundle->bond, s->bond)) {
1967 ofproto->need_revalidate = true;
1970 bundle->bond = bond_create(s->bond);
1971 ofproto->need_revalidate = true;
1974 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1975 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1979 bond_destroy(bundle->bond);
1980 bundle->bond = NULL;
1983 /* If we changed something that would affect MAC learning, un-learn
1984 * everything on this port and force flow revalidation. */
1986 bundle_flush_macs(bundle, false);
1993 bundle_remove(struct ofport *port_)
1995 struct ofport_dpif *port = ofport_dpif_cast(port_);
1996 struct ofbundle *bundle = port->bundle;
1999 bundle_del_port(port);
2000 if (list_is_empty(&bundle->ports)) {
2001 bundle_destroy(bundle);
2002 } else if (list_is_short(&bundle->ports)) {
2003 bond_destroy(bundle->bond);
2004 bundle->bond = NULL;
2010 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2012 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2013 struct ofport_dpif *port = port_;
2014 uint8_t ea[ETH_ADDR_LEN];
2017 error = netdev_get_etheraddr(port->up.netdev, ea);
2019 struct ofpbuf packet;
2022 ofpbuf_init(&packet, 0);
2023 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2025 memcpy(packet_pdu, pdu, pdu_size);
2027 send_packet(port, &packet);
2028 ofpbuf_uninit(&packet);
2030 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2031 "%s (%s)", port->bundle->name,
2032 netdev_get_name(port->up.netdev), strerror(error));
2037 bundle_send_learning_packets(struct ofbundle *bundle)
2039 struct ofproto_dpif *ofproto = bundle->ofproto;
2040 int error, n_packets, n_errors;
2041 struct mac_entry *e;
2043 error = n_packets = n_errors = 0;
2044 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2045 if (e->port.p != bundle) {
2046 struct ofpbuf *learning_packet;
2047 struct ofport_dpif *port;
2051 /* The assignment to "port" is unnecessary but makes "grep"ing for
2052 * struct ofport_dpif more effective. */
2053 learning_packet = bond_compose_learning_packet(bundle->bond,
2057 ret = send_packet(port, learning_packet);
2058 ofpbuf_delete(learning_packet);
2068 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2069 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2070 "packets, last error was: %s",
2071 bundle->name, n_errors, n_packets, strerror(error));
2073 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2074 bundle->name, n_packets);
2079 bundle_run(struct ofbundle *bundle)
2082 lacp_run(bundle->lacp, send_pdu_cb);
2085 struct ofport_dpif *port;
2087 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2088 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2091 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
2092 lacp_status(bundle->lacp));
2093 if (bond_should_send_learning_packets(bundle->bond)) {
2094 bundle_send_learning_packets(bundle);
2100 bundle_wait(struct ofbundle *bundle)
2103 lacp_wait(bundle->lacp);
2106 bond_wait(bundle->bond);
2113 mirror_scan(struct ofproto_dpif *ofproto)
2117 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2118 if (!ofproto->mirrors[idx]) {
2125 static struct ofmirror *
2126 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2130 for (i = 0; i < MAX_MIRRORS; i++) {
2131 struct ofmirror *mirror = ofproto->mirrors[i];
2132 if (mirror && mirror->aux == aux) {
2140 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2142 mirror_update_dups(struct ofproto_dpif *ofproto)
2146 for (i = 0; i < MAX_MIRRORS; i++) {
2147 struct ofmirror *m = ofproto->mirrors[i];
2150 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2154 for (i = 0; i < MAX_MIRRORS; i++) {
2155 struct ofmirror *m1 = ofproto->mirrors[i];
2162 for (j = i + 1; j < MAX_MIRRORS; j++) {
2163 struct ofmirror *m2 = ofproto->mirrors[j];
2165 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2166 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2167 m2->dup_mirrors |= m1->dup_mirrors;
2174 mirror_set(struct ofproto *ofproto_, void *aux,
2175 const struct ofproto_mirror_settings *s)
2177 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2178 mirror_mask_t mirror_bit;
2179 struct ofbundle *bundle;
2180 struct ofmirror *mirror;
2181 struct ofbundle *out;
2182 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2183 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2186 mirror = mirror_lookup(ofproto, aux);
2188 mirror_destroy(mirror);
2194 idx = mirror_scan(ofproto);
2196 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2198 ofproto->up.name, MAX_MIRRORS, s->name);
2202 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2203 mirror->ofproto = ofproto;
2206 mirror->out_vlan = -1;
2207 mirror->name = NULL;
2210 if (!mirror->name || strcmp(s->name, mirror->name)) {
2212 mirror->name = xstrdup(s->name);
2215 /* Get the new configuration. */
2216 if (s->out_bundle) {
2217 out = bundle_lookup(ofproto, s->out_bundle);
2219 mirror_destroy(mirror);
2225 out_vlan = s->out_vlan;
2227 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2228 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2230 /* If the configuration has not changed, do nothing. */
2231 if (hmapx_equals(&srcs, &mirror->srcs)
2232 && hmapx_equals(&dsts, &mirror->dsts)
2233 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2234 && mirror->out == out
2235 && mirror->out_vlan == out_vlan)
2237 hmapx_destroy(&srcs);
2238 hmapx_destroy(&dsts);
2242 hmapx_swap(&srcs, &mirror->srcs);
2243 hmapx_destroy(&srcs);
2245 hmapx_swap(&dsts, &mirror->dsts);
2246 hmapx_destroy(&dsts);
2248 free(mirror->vlans);
2249 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2252 mirror->out_vlan = out_vlan;
2254 /* Update bundles. */
2255 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2256 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2257 if (hmapx_contains(&mirror->srcs, bundle)) {
2258 bundle->src_mirrors |= mirror_bit;
2260 bundle->src_mirrors &= ~mirror_bit;
2263 if (hmapx_contains(&mirror->dsts, bundle)) {
2264 bundle->dst_mirrors |= mirror_bit;
2266 bundle->dst_mirrors &= ~mirror_bit;
2269 if (mirror->out == bundle) {
2270 bundle->mirror_out |= mirror_bit;
2272 bundle->mirror_out &= ~mirror_bit;
2276 ofproto->need_revalidate = true;
2277 ofproto->has_mirrors = true;
2278 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2279 mirror_update_dups(ofproto);
2285 mirror_destroy(struct ofmirror *mirror)
2287 struct ofproto_dpif *ofproto;
2288 mirror_mask_t mirror_bit;
2289 struct ofbundle *bundle;
2296 ofproto = mirror->ofproto;
2297 ofproto->need_revalidate = true;
2298 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2300 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2301 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2302 bundle->src_mirrors &= ~mirror_bit;
2303 bundle->dst_mirrors &= ~mirror_bit;
2304 bundle->mirror_out &= ~mirror_bit;
2307 hmapx_destroy(&mirror->srcs);
2308 hmapx_destroy(&mirror->dsts);
2309 free(mirror->vlans);
2311 ofproto->mirrors[mirror->idx] = NULL;
2315 mirror_update_dups(ofproto);
2317 ofproto->has_mirrors = false;
2318 for (i = 0; i < MAX_MIRRORS; i++) {
2319 if (ofproto->mirrors[i]) {
2320 ofproto->has_mirrors = true;
2327 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2328 uint64_t *packets, uint64_t *bytes)
2330 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2331 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2334 *packets = *bytes = UINT64_MAX;
2338 *packets = mirror->packet_count;
2339 *bytes = mirror->byte_count;
2345 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2347 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2348 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2349 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2355 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2357 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2358 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2359 return bundle && bundle->mirror_out != 0;
2363 forward_bpdu_changed(struct ofproto *ofproto_)
2365 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2366 /* Revalidate cached flows whenever forward_bpdu option changes. */
2367 ofproto->need_revalidate = true;
2371 set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time)
2373 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2374 mac_learning_set_idle_time(ofproto->ml, idle_time);
2379 static struct ofport_dpif *
2380 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2382 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2383 return ofport ? ofport_dpif_cast(ofport) : NULL;
2386 static struct ofport_dpif *
2387 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2389 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2393 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2394 struct dpif_port *dpif_port)
2396 ofproto_port->name = dpif_port->name;
2397 ofproto_port->type = dpif_port->type;
2398 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2402 port_run(struct ofport_dpif *ofport)
2404 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2405 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2406 bool enable = netdev_get_carrier(ofport->up.netdev);
2408 ofport->carrier_seq = carrier_seq;
2411 cfm_run(ofport->cfm);
2413 if (cfm_should_send_ccm(ofport->cfm)) {
2414 struct ofpbuf packet;
2416 ofpbuf_init(&packet, 0);
2417 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2418 send_packet(ofport, &packet);
2419 ofpbuf_uninit(&packet);
2422 enable = enable && !cfm_get_fault(ofport->cfm)
2423 && cfm_get_opup(ofport->cfm);
2426 if (ofport->bundle) {
2427 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2428 if (carrier_changed) {
2429 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2433 if (ofport->may_enable != enable) {
2434 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2436 if (ofproto->has_bundle_action) {
2437 ofproto->need_revalidate = true;
2441 ofport->may_enable = enable;
2445 port_wait(struct ofport_dpif *ofport)
2448 cfm_wait(ofport->cfm);
2453 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2454 struct ofproto_port *ofproto_port)
2456 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2457 struct dpif_port dpif_port;
2460 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2462 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2468 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2470 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2474 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2476 *ofp_portp = odp_port_to_ofp_port(odp_port);
2482 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2484 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2487 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2489 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2491 /* The caller is going to close ofport->up.netdev. If this is a
2492 * bonded port, then the bond is using that netdev, so remove it
2493 * from the bond. The client will need to reconfigure everything
2494 * after deleting ports, so then the slave will get re-added. */
2495 bundle_remove(&ofport->up);
2502 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2504 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2507 error = netdev_get_stats(ofport->up.netdev, stats);
2509 if (!error && ofport->odp_port == OVSP_LOCAL) {
2510 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2512 /* ofproto->stats.tx_packets represents packets that we created
2513 * internally and sent to some port (e.g. packets sent with
2514 * send_packet()). Account for them as if they had come from
2515 * OFPP_LOCAL and got forwarded. */
2517 if (stats->rx_packets != UINT64_MAX) {
2518 stats->rx_packets += ofproto->stats.tx_packets;
2521 if (stats->rx_bytes != UINT64_MAX) {
2522 stats->rx_bytes += ofproto->stats.tx_bytes;
2525 /* ofproto->stats.rx_packets represents packets that were received on
2526 * some port and we processed internally and dropped (e.g. STP).
2527 * Account fro them as if they had been forwarded to OFPP_LOCAL. */
2529 if (stats->tx_packets != UINT64_MAX) {
2530 stats->tx_packets += ofproto->stats.rx_packets;
2533 if (stats->tx_bytes != UINT64_MAX) {
2534 stats->tx_bytes += ofproto->stats.rx_bytes;
2541 /* Account packets for LOCAL port. */
2543 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2544 size_t tx_size, size_t rx_size)
2546 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2549 ofproto->stats.rx_packets++;
2550 ofproto->stats.rx_bytes += rx_size;
2553 ofproto->stats.tx_packets++;
2554 ofproto->stats.tx_bytes += tx_size;
2558 struct port_dump_state {
2559 struct dpif_port_dump dump;
2564 port_dump_start(const struct ofproto *ofproto_, void **statep)
2566 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2567 struct port_dump_state *state;
2569 *statep = state = xmalloc(sizeof *state);
2570 dpif_port_dump_start(&state->dump, ofproto->dpif);
2571 state->done = false;
2576 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2577 struct ofproto_port *port)
2579 struct port_dump_state *state = state_;
2580 struct dpif_port dpif_port;
2582 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2583 ofproto_port_from_dpif_port(port, &dpif_port);
2586 int error = dpif_port_dump_done(&state->dump);
2588 return error ? error : EOF;
2593 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2595 struct port_dump_state *state = state_;
2598 dpif_port_dump_done(&state->dump);
2605 port_poll(const struct ofproto *ofproto_, char **devnamep)
2607 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2608 return dpif_port_poll(ofproto->dpif, devnamep);
2612 port_poll_wait(const struct ofproto *ofproto_)
2614 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2615 dpif_port_poll_wait(ofproto->dpif);
2619 port_is_lacp_current(const struct ofport *ofport_)
2621 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2622 return (ofport->bundle && ofport->bundle->lacp
2623 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2627 /* Upcall handling. */
2629 /* Flow miss batching.
2631 * Some dpifs implement operations faster when you hand them off in a batch.
2632 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2633 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2634 * more packets, plus possibly installing the flow in the dpif.
2636 * So far we only batch the operations that affect flow setup time the most.
2637 * It's possible to batch more than that, but the benefit might be minimal. */
2639 struct hmap_node hmap_node;
2641 enum odp_key_fitness key_fitness;
2642 const struct nlattr *key;
2644 ovs_be16 initial_tci;
2645 struct list packets;
2646 enum dpif_upcall_type upcall_type;
2649 struct flow_miss_op {
2650 struct dpif_op dpif_op;
2651 struct subfacet *subfacet; /* Subfacet */
2652 void *garbage; /* Pointer to pass to free(), NULL if none. */
2653 uint64_t stub[1024 / 8]; /* Temporary buffer. */
2656 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2657 * OpenFlow controller as necessary according to their individual
2658 * configurations. */
2660 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
2661 const struct flow *flow)
2663 struct ofputil_packet_in pin;
2665 pin.packet = packet->data;
2666 pin.packet_len = packet->size;
2667 pin.reason = OFPR_NO_MATCH;
2668 pin.controller_id = 0;
2673 pin.send_len = 0; /* not used for flow table misses */
2675 flow_get_metadata(flow, &pin.fmd);
2677 /* Registers aren't meaningful on a miss. */
2678 memset(pin.fmd.reg_masks, 0, sizeof pin.fmd.reg_masks);
2680 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
2683 static enum slow_path_reason
2684 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2685 const struct ofpbuf *packet)
2687 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2693 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2695 cfm_process_heartbeat(ofport->cfm, packet);
2698 } else if (ofport->bundle && ofport->bundle->lacp
2699 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2701 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2704 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2706 stp_process_packet(ofport, packet);
2713 static struct flow_miss *
2714 flow_miss_find(struct hmap *todo, const struct flow *flow, uint32_t hash)
2716 struct flow_miss *miss;
2718 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2719 if (flow_equal(&miss->flow, flow)) {
2727 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
2728 * 'packet'. The caller must initialize op->actions and op->actions_len. If
2729 * 'miss' is associated with a subfacet the caller must also initialize the
2730 * returned op->subfacet, and if anything needs to be freed after processing
2731 * the op, the caller must initialize op->garbage also. */
2733 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
2734 struct flow_miss_op *op)
2736 if (miss->flow.vlan_tci != miss->initial_tci) {
2737 /* This packet was received on a VLAN splinter port. We
2738 * added a VLAN to the packet to make the packet resemble
2739 * the flow, but the actions were composed assuming that
2740 * the packet contained no VLAN. So, we must remove the
2741 * VLAN header from the packet before trying to execute the
2743 eth_pop_vlan(packet);
2746 op->subfacet = NULL;
2748 op->dpif_op.type = DPIF_OP_EXECUTE;
2749 op->dpif_op.u.execute.key = miss->key;
2750 op->dpif_op.u.execute.key_len = miss->key_len;
2751 op->dpif_op.u.execute.packet = packet;
2754 /* Helper for handle_flow_miss_without_facet() and
2755 * handle_flow_miss_with_facet(). */
2757 handle_flow_miss_common(struct rule_dpif *rule,
2758 struct ofpbuf *packet, const struct flow *flow)
2760 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2762 ofproto->n_matches++;
2764 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2766 * Extra-special case for fail-open mode.
2768 * We are in fail-open mode and the packet matched the fail-open
2769 * rule, but we are connected to a controller too. We should send
2770 * the packet up to the controller in the hope that it will try to
2771 * set up a flow and thereby allow us to exit fail-open.
2773 * See the top-level comment in fail-open.c for more information.
2775 send_packet_in_miss(ofproto, packet, flow);
2779 /* Figures out whether a flow that missed in 'ofproto', whose details are in
2780 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
2781 * installing a datapath flow. The answer is usually "yes" (a return value of
2782 * true). However, for short flows the cost of bookkeeping is much higher than
2783 * the benefits, so when the datapath holds a large number of flows we impose
2784 * some heuristics to decide which flows are likely to be worth tracking. */
2786 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
2787 struct flow_miss *miss, uint32_t hash)
2789 if (!ofproto->governor) {
2792 n_subfacets = hmap_count(&ofproto->subfacets);
2793 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
2797 ofproto->governor = governor_create(ofproto->up.name);
2800 return governor_should_install_flow(ofproto->governor, hash,
2801 list_size(&miss->packets));
2804 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
2805 * or creating any datapath flow. May add an "execute" operation to 'ops' and
2806 * increment '*n_ops'. */
2808 handle_flow_miss_without_facet(struct flow_miss *miss,
2809 struct rule_dpif *rule,
2810 struct flow_miss_op *ops, size_t *n_ops)
2812 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2813 struct action_xlate_ctx ctx;
2814 struct ofpbuf *packet;
2816 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2817 struct flow_miss_op *op = &ops[*n_ops];
2818 struct dpif_flow_stats stats;
2819 struct ofpbuf odp_actions;
2821 COVERAGE_INC(facet_suppress);
2823 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
2825 dpif_flow_stats_extract(&miss->flow, packet, &stats);
2826 rule_credit_stats(rule, &stats);
2828 action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci,
2830 ctx.resubmit_stats = &stats;
2831 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions,
2834 if (odp_actions.size) {
2835 struct dpif_execute *execute = &op->dpif_op.u.execute;
2837 init_flow_miss_execute_op(miss, packet, op);
2838 execute->actions = odp_actions.data;
2839 execute->actions_len = odp_actions.size;
2840 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
2844 ofpbuf_uninit(&odp_actions);
2849 /* Handles 'miss', which matches 'facet'. May add any required datapath
2850 * operations to 'ops', incrementing '*n_ops' for each new op. */
2852 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
2853 struct flow_miss_op *ops, size_t *n_ops)
2855 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
2856 enum subfacet_path want_path;
2857 struct subfacet *subfacet;
2858 struct ofpbuf *packet;
2860 subfacet = subfacet_create(facet,
2861 miss->key_fitness, miss->key, miss->key_len,
2864 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2865 struct flow_miss_op *op = &ops[*n_ops];
2866 struct dpif_flow_stats stats;
2867 struct ofpbuf odp_actions;
2869 handle_flow_miss_common(facet->rule, packet, &miss->flow);
2871 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
2872 if (!subfacet->actions || subfacet->slow) {
2873 subfacet_make_actions(subfacet, packet, &odp_actions);
2876 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2877 subfacet_update_stats(subfacet, &stats);
2879 if (subfacet->actions_len) {
2880 struct dpif_execute *execute = &op->dpif_op.u.execute;
2882 init_flow_miss_execute_op(miss, packet, op);
2883 op->subfacet = subfacet;
2884 if (!subfacet->slow) {
2885 execute->actions = subfacet->actions;
2886 execute->actions_len = subfacet->actions_len;
2887 ofpbuf_uninit(&odp_actions);
2889 execute->actions = odp_actions.data;
2890 execute->actions_len = odp_actions.size;
2891 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
2896 ofpbuf_uninit(&odp_actions);
2900 want_path = subfacet_want_path(subfacet->slow);
2901 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
2902 struct flow_miss_op *op = &ops[(*n_ops)++];
2903 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
2905 op->subfacet = subfacet;
2907 op->dpif_op.type = DPIF_OP_FLOW_PUT;
2908 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2909 put->key = miss->key;
2910 put->key_len = miss->key_len;
2911 if (want_path == SF_FAST_PATH) {
2912 put->actions = subfacet->actions;
2913 put->actions_len = subfacet->actions_len;
2915 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
2916 op->stub, sizeof op->stub,
2917 &put->actions, &put->actions_len);
2923 /* Handles flow miss 'miss' on 'ofproto'. May add any required datapath
2924 * operations to 'ops', incrementing '*n_ops' for each new op. */
2926 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2927 struct flow_miss_op *ops, size_t *n_ops)
2929 struct facet *facet;
2932 /* The caller must ensure that miss->hmap_node.hash contains
2933 * flow_hash(miss->flow, 0). */
2934 hash = miss->hmap_node.hash;
2936 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
2938 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
2940 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
2941 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
2945 facet = facet_create(rule, &miss->flow, hash);
2947 handle_flow_miss_with_facet(miss, facet, ops, n_ops);
2950 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
2951 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
2952 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
2953 * what a flow key should contain.
2955 * This function also includes some logic to help make VLAN splinters
2956 * transparent to the rest of the upcall processing logic. In particular, if
2957 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
2958 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
2959 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
2961 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
2962 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
2963 * (This differs from the value returned in flow->vlan_tci only for packets
2964 * received on VLAN splinters.)
2966 static enum odp_key_fitness
2967 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
2968 const struct nlattr *key, size_t key_len,
2969 struct flow *flow, ovs_be16 *initial_tci,
2970 struct ofpbuf *packet)
2972 enum odp_key_fitness fitness;
2974 fitness = odp_flow_key_to_flow(key, key_len, flow);
2975 if (fitness == ODP_FIT_ERROR) {
2978 *initial_tci = flow->vlan_tci;
2980 if (vsp_adjust_flow(ofproto, flow)) {
2982 /* Make the packet resemble the flow, so that it gets sent to an
2983 * OpenFlow controller properly, so that it looks correct for
2984 * sFlow, and so that flow_extract() will get the correct vlan_tci
2985 * if it is called on 'packet'.
2987 * The allocated space inside 'packet' probably also contains
2988 * 'key', that is, both 'packet' and 'key' are probably part of a
2989 * struct dpif_upcall (see the large comment on that structure
2990 * definition), so pushing data on 'packet' is in general not a
2991 * good idea since it could overwrite 'key' or free it as a side
2992 * effect. However, it's OK in this special case because we know
2993 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
2994 * will just overwrite the 4-byte "struct nlattr", which is fine
2995 * since we don't need that header anymore. */
2996 eth_push_vlan(packet, flow->vlan_tci);
2999 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
3000 if (fitness == ODP_FIT_PERFECT) {
3001 fitness = ODP_FIT_TOO_MUCH;
3009 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
3012 struct dpif_upcall *upcall;
3013 struct flow_miss *miss;
3014 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3015 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3016 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3026 /* Construct the to-do list.
3028 * This just amounts to extracting the flow from each packet and sticking
3029 * the packets that have the same flow in the same "flow_miss" structure so
3030 * that we can process them together. */
3033 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3034 struct flow_miss *miss = &misses[n_misses];
3035 struct flow_miss *existing_miss;
3038 /* Obtain metadata and check userspace/kernel agreement on flow match,
3039 * then set 'flow''s header pointers. */
3040 miss->key_fitness = ofproto_dpif_extract_flow_key(
3041 ofproto, upcall->key, upcall->key_len,
3042 &miss->flow, &miss->initial_tci, upcall->packet);
3043 if (miss->key_fitness == ODP_FIT_ERROR) {
3046 flow_extract(upcall->packet, miss->flow.skb_priority,
3047 miss->flow.tun_id, miss->flow.in_port, &miss->flow);
3049 /* Add other packets to a to-do list. */
3050 hash = flow_hash(&miss->flow, 0);
3051 existing_miss = flow_miss_find(&todo, &miss->flow, hash);
3052 if (!existing_miss) {
3053 hmap_insert(&todo, &miss->hmap_node, hash);
3054 miss->key = upcall->key;
3055 miss->key_len = upcall->key_len;
3056 miss->upcall_type = upcall->type;
3057 list_init(&miss->packets);
3061 miss = existing_miss;
3063 list_push_back(&miss->packets, &upcall->packet->list_node);
3066 /* Process each element in the to-do list, constructing the set of
3067 * operations to batch. */
3069 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3070 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
3072 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3074 /* Execute batch. */
3075 for (i = 0; i < n_ops; i++) {
3076 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3078 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
3080 /* Free memory and update facets. */
3081 for (i = 0; i < n_ops; i++) {
3082 struct flow_miss_op *op = &flow_miss_ops[i];
3084 switch (op->dpif_op.type) {
3085 case DPIF_OP_EXECUTE:
3088 case DPIF_OP_FLOW_PUT:
3089 if (!op->dpif_op.error) {
3090 op->subfacet->path = subfacet_want_path(op->subfacet->slow);
3094 case DPIF_OP_FLOW_DEL:
3100 hmap_destroy(&todo);
3103 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
3104 classify_upcall(const struct dpif_upcall *upcall)
3106 union user_action_cookie cookie;
3108 /* First look at the upcall type. */
3109 switch (upcall->type) {
3110 case DPIF_UC_ACTION:
3116 case DPIF_N_UC_TYPES:
3118 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3122 /* "action" upcalls need a closer look. */
3123 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3124 switch (cookie.type) {
3125 case USER_ACTION_COOKIE_SFLOW:
3126 return SFLOW_UPCALL;
3128 case USER_ACTION_COOKIE_SLOW_PATH:
3131 case USER_ACTION_COOKIE_UNSPEC:
3133 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
3139 handle_sflow_upcall(struct ofproto_dpif *ofproto,
3140 const struct dpif_upcall *upcall)
3142 union user_action_cookie cookie;
3143 enum odp_key_fitness fitness;
3144 ovs_be16 initial_tci;
3147 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
3148 upcall->key_len, &flow,
3149 &initial_tci, upcall->packet);
3150 if (fitness == ODP_FIT_ERROR) {
3154 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3155 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, &cookie);
3159 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
3161 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3162 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3163 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3168 assert(max_batch <= FLOW_MISS_MAX_BATCH);
3171 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3172 struct dpif_upcall *upcall = &misses[n_misses];
3173 struct ofpbuf *buf = &miss_bufs[n_misses];
3176 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3177 sizeof miss_buf_stubs[n_misses]);
3178 error = dpif_recv(ofproto->dpif, upcall, buf);
3184 switch (classify_upcall(upcall)) {
3186 /* Handle it later. */
3191 if (ofproto->sflow) {
3192 handle_sflow_upcall(ofproto, upcall);
3203 /* Handle deferred MISS_UPCALL processing. */
3204 handle_miss_upcalls(ofproto, misses, n_misses);
3205 for (i = 0; i < n_misses; i++) {
3206 ofpbuf_uninit(&miss_bufs[i]);
3212 /* Flow expiration. */
3214 static int subfacet_max_idle(const struct ofproto_dpif *);
3215 static void update_stats(struct ofproto_dpif *);
3216 static void rule_expire(struct rule_dpif *);
3217 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
3219 /* This function is called periodically by run(). Its job is to collect
3220 * updates for the flows that have been installed into the datapath, most
3221 * importantly when they last were used, and then use that information to
3222 * expire flows that have not been used recently.
3224 * Returns the number of milliseconds after which it should be called again. */
3226 expire(struct ofproto_dpif *ofproto)
3228 struct rule_dpif *rule, *next_rule;
3229 struct oftable *table;
3232 /* Update stats for each flow in the datapath. */
3233 update_stats(ofproto);
3235 /* Expire subfacets that have been idle too long. */
3236 dp_max_idle = subfacet_max_idle(ofproto);
3237 expire_subfacets(ofproto, dp_max_idle);
3239 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
3240 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
3241 struct cls_cursor cursor;
3243 cls_cursor_init(&cursor, &table->cls, NULL);
3244 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
3249 /* All outstanding data in existing flows has been accounted, so it's a
3250 * good time to do bond rebalancing. */
3251 if (ofproto->has_bonded_bundles) {
3252 struct ofbundle *bundle;
3254 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3256 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
3261 return MIN(dp_max_idle, 1000);
3264 /* Updates flow table statistics given that the datapath just reported 'stats'
3265 * as 'subfacet''s statistics. */
3267 update_subfacet_stats(struct subfacet *subfacet,
3268 const struct dpif_flow_stats *stats)
3270 struct facet *facet = subfacet->facet;
3272 if (stats->n_packets >= subfacet->dp_packet_count) {
3273 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
3274 facet->packet_count += extra;
3276 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3279 if (stats->n_bytes >= subfacet->dp_byte_count) {
3280 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
3282 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3285 subfacet->dp_packet_count = stats->n_packets;
3286 subfacet->dp_byte_count = stats->n_bytes;
3288 facet->tcp_flags |= stats->tcp_flags;
3290 subfacet_update_time(subfacet, stats->used);
3291 if (facet->accounted_bytes < facet->byte_count) {
3293 facet_account(facet);
3294 facet->accounted_bytes = facet->byte_count;
3296 facet_push_stats(facet);
3299 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3300 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3302 delete_unexpected_flow(struct dpif *dpif,
3303 const struct nlattr *key, size_t key_len)
3305 if (!VLOG_DROP_WARN(&rl)) {
3309 odp_flow_key_format(key, key_len, &s);
3310 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
3314 COVERAGE_INC(facet_unexpected);
3315 dpif_flow_del(dpif, key, key_len, NULL);
3318 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3320 * This function also pushes statistics updates to rules which each facet
3321 * resubmits into. Generally these statistics will be accurate. However, if a
3322 * facet changes the rule it resubmits into at some time in between
3323 * update_stats() runs, it is possible that statistics accrued to the
3324 * old rule will be incorrectly attributed to the new rule. This could be
3325 * avoided by calling update_stats() whenever rules are created or
3326 * deleted. However, the performance impact of making so many calls to the
3327 * datapath do not justify the benefit of having perfectly accurate statistics.
3330 update_stats(struct ofproto_dpif *p)
3332 const struct dpif_flow_stats *stats;
3333 struct dpif_flow_dump dump;
3334 const struct nlattr *key;
3337 dpif_flow_dump_start(&dump, p->dpif);
3338 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
3339 struct subfacet *subfacet;
3341 subfacet = subfacet_find(p, key, key_len);
3342 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3344 update_subfacet_stats(subfacet, stats);
3348 /* Stats are updated per-packet. */
3351 case SF_NOT_INSTALLED:
3353 delete_unexpected_flow(p->dpif, key, key_len);
3357 dpif_flow_dump_done(&dump);
3360 /* Calculates and returns the number of milliseconds of idle time after which
3361 * subfacets should expire from the datapath. When a subfacet expires, we fold
3362 * its statistics into its facet, and when a facet's last subfacet expires, we
3363 * fold its statistic into its rule. */
3365 subfacet_max_idle(const struct ofproto_dpif *ofproto)
3368 * Idle time histogram.
3370 * Most of the time a switch has a relatively small number of subfacets.
3371 * When this is the case we might as well keep statistics for all of them
3372 * in userspace and to cache them in the kernel datapath for performance as
3375 * As the number of subfacets increases, the memory required to maintain
3376 * statistics about them in userspace and in the kernel becomes
3377 * significant. However, with a large number of subfacets it is likely
3378 * that only a few of them are "heavy hitters" that consume a large amount
3379 * of bandwidth. At this point, only heavy hitters are worth caching in
3380 * the kernel and maintaining in userspaces; other subfacets we can
3383 * The technique used to compute the idle time is to build a histogram with
3384 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3385 * that is installed in the kernel gets dropped in the appropriate bucket.
3386 * After the histogram has been built, we compute the cutoff so that only
3387 * the most-recently-used 1% of subfacets (but at least
3388 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3389 * the most-recently-used bucket of subfacets is kept, so actually an
3390 * arbitrary number of subfacets can be kept in any given expiration run
3391 * (though the next run will delete most of those unless they receive
3394 * This requires a second pass through the subfacets, in addition to the
3395 * pass made by update_stats(), because the former function never looks at
3396 * uninstallable subfacets.
3398 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3399 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3400 int buckets[N_BUCKETS] = { 0 };
3401 int total, subtotal, bucket;
3402 struct subfacet *subfacet;
3406 total = hmap_count(&ofproto->subfacets);
3407 if (total <= ofproto->up.flow_eviction_threshold) {
3408 return N_BUCKETS * BUCKET_WIDTH;
3411 /* Build histogram. */
3413 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3414 long long int idle = now - subfacet->used;
3415 int bucket = (idle <= 0 ? 0
3416 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3417 : (unsigned int) idle / BUCKET_WIDTH);
3421 /* Find the first bucket whose flows should be expired. */
3422 subtotal = bucket = 0;
3424 subtotal += buckets[bucket++];
3425 } while (bucket < N_BUCKETS &&
3426 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3428 if (VLOG_IS_DBG_ENABLED()) {
3432 ds_put_cstr(&s, "keep");
3433 for (i = 0; i < N_BUCKETS; i++) {
3435 ds_put_cstr(&s, ", drop");
3438 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3441 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3445 return bucket * BUCKET_WIDTH;
3448 enum { EXPIRE_MAX_BATCH = 50 };
3451 expire_batch(struct ofproto_dpif *ofproto, struct subfacet **subfacets, int n)
3453 struct odputil_keybuf keybufs[EXPIRE_MAX_BATCH];
3454 struct dpif_op ops[EXPIRE_MAX_BATCH];
3455 struct dpif_op *opsp[EXPIRE_MAX_BATCH];
3456 struct ofpbuf keys[EXPIRE_MAX_BATCH];
3457 struct dpif_flow_stats stats[EXPIRE_MAX_BATCH];
3460 for (i = 0; i < n; i++) {
3461 ops[i].type = DPIF_OP_FLOW_DEL;
3462 subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]);
3463 ops[i].u.flow_del.key = keys[i].data;
3464 ops[i].u.flow_del.key_len = keys[i].size;
3465 ops[i].u.flow_del.stats = &stats[i];
3469 dpif_operate(ofproto->dpif, opsp, n);
3470 for (i = 0; i < n; i++) {
3471 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
3472 subfacets[i]->path = SF_NOT_INSTALLED;
3473 subfacet_destroy(subfacets[i]);
3478 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3480 /* Cutoff time for most flows. */
3481 long long int normal_cutoff = time_msec() - dp_max_idle;
3483 /* We really want to keep flows for special protocols around, so use a more
3484 * conservative cutoff. */
3485 long long int special_cutoff = time_msec() - 10000;
3487 struct subfacet *subfacet, *next_subfacet;
3488 struct subfacet *batch[EXPIRE_MAX_BATCH];
3492 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3493 &ofproto->subfacets) {
3494 long long int cutoff;
3496 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
3499 if (subfacet->used < cutoff) {
3500 if (subfacet->path != SF_NOT_INSTALLED) {
3501 batch[n_batch++] = subfacet;
3502 if (n_batch >= EXPIRE_MAX_BATCH) {
3503 expire_batch(ofproto, batch, n_batch);
3507 subfacet_destroy(subfacet);
3513 expire_batch(ofproto, batch, n_batch);
3517 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3518 * then delete it entirely. */
3520 rule_expire(struct rule_dpif *rule)
3522 struct facet *facet, *next_facet;
3526 /* Has 'rule' expired? */
3528 if (rule->up.hard_timeout
3529 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3530 reason = OFPRR_HARD_TIMEOUT;
3531 } else if (rule->up.idle_timeout
3532 && now > rule->up.used + rule->up.idle_timeout * 1000) {
3533 reason = OFPRR_IDLE_TIMEOUT;
3538 COVERAGE_INC(ofproto_dpif_expired);
3540 /* Update stats. (This is a no-op if the rule expired due to an idle
3541 * timeout, because that only happens when the rule has no facets left.) */
3542 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3543 facet_remove(facet);
3546 /* Get rid of the rule. */
3547 ofproto_rule_expire(&rule->up, reason);
3552 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3554 * The caller must already have determined that no facet with an identical
3555 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3556 * the ofproto's classifier table.
3558 * 'hash' must be the return value of flow_hash(flow, 0).
3560 * The facet will initially have no subfacets. The caller should create (at
3561 * least) one subfacet with subfacet_create(). */
3562 static struct facet *
3563 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
3565 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3566 struct facet *facet;
3568 facet = xzalloc(sizeof *facet);
3569 facet->used = time_msec();
3570 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
3571 list_push_back(&rule->facets, &facet->list_node);
3573 facet->flow = *flow;
3574 list_init(&facet->subfacets);
3575 netflow_flow_init(&facet->nf_flow);
3576 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3582 facet_free(struct facet *facet)
3587 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3588 * 'packet', which arrived on 'in_port'.
3590 * Takes ownership of 'packet'. */
3592 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3593 const struct nlattr *odp_actions, size_t actions_len,
3594 struct ofpbuf *packet)
3596 struct odputil_keybuf keybuf;
3600 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3601 odp_flow_key_from_flow(&key, flow);
3603 error = dpif_execute(ofproto->dpif, key.data, key.size,
3604 odp_actions, actions_len, packet);
3606 ofpbuf_delete(packet);
3610 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3612 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3613 * rule's statistics, via subfacet_uninstall().
3615 * - Removes 'facet' from its rule and from ofproto->facets.
3618 facet_remove(struct facet *facet)
3620 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3621 struct subfacet *subfacet, *next_subfacet;
3623 assert(!list_is_empty(&facet->subfacets));
3625 /* First uninstall all of the subfacets to get final statistics. */
3626 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3627 subfacet_uninstall(subfacet);
3630 /* Flush the final stats to the rule.
3632 * This might require us to have at least one subfacet around so that we
3633 * can use its actions for accounting in facet_account(), which is why we
3634 * have uninstalled but not yet destroyed the subfacets. */
3635 facet_flush_stats(facet);
3637 /* Now we're really all done so destroy everything. */
3638 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3639 &facet->subfacets) {
3640 subfacet_destroy__(subfacet);
3642 hmap_remove(&ofproto->facets, &facet->hmap_node);
3643 list_remove(&facet->list_node);
3647 /* Feed information from 'facet' back into the learning table to keep it in
3648 * sync with what is actually flowing through the datapath. */
3650 facet_learn(struct facet *facet)
3652 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3653 struct action_xlate_ctx ctx;
3655 if (!facet->has_learn
3656 && !facet->has_normal
3657 && (!facet->has_fin_timeout
3658 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
3662 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3663 facet->flow.vlan_tci,
3664 facet->rule, facet->tcp_flags, NULL);
3665 ctx.may_learn = true;
3666 xlate_actions_for_side_effects(&ctx, facet->rule->up.actions,
3667 facet->rule->up.n_actions);
3671 facet_account(struct facet *facet)
3673 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3674 struct subfacet *subfacet;
3675 const struct nlattr *a;
3680 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3683 n_bytes = facet->byte_count - facet->accounted_bytes;
3685 /* This loop feeds byte counters to bond_account() for rebalancing to use
3686 * as a basis. We also need to track the actual VLAN on which the packet
3687 * is going to be sent to ensure that it matches the one passed to
3688 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3691 * We use the actions from an arbitrary subfacet because they should all
3692 * be equally valid for our purpose. */
3693 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3694 struct subfacet, list_node);
3695 vlan_tci = facet->flow.vlan_tci;
3696 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3697 subfacet->actions, subfacet->actions_len) {
3698 const struct ovs_action_push_vlan *vlan;
3699 struct ofport_dpif *port;
3701 switch (nl_attr_type(a)) {
3702 case OVS_ACTION_ATTR_OUTPUT:
3703 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3704 if (port && port->bundle && port->bundle->bond) {
3705 bond_account(port->bundle->bond, &facet->flow,
3706 vlan_tci_to_vid(vlan_tci), n_bytes);
3710 case OVS_ACTION_ATTR_POP_VLAN:
3711 vlan_tci = htons(0);
3714 case OVS_ACTION_ATTR_PUSH_VLAN:
3715 vlan = nl_attr_get(a);
3716 vlan_tci = vlan->vlan_tci;
3722 /* Returns true if the only action for 'facet' is to send to the controller.
3723 * (We don't report NetFlow expiration messages for such facets because they
3724 * are just part of the control logic for the network, not real traffic). */
3726 facet_is_controller_flow(struct facet *facet)
3729 && facet->rule->up.n_actions == 1
3730 && action_outputs_to_port(&facet->rule->up.actions[0],
3731 htons(OFPP_CONTROLLER)));
3734 /* Folds all of 'facet''s statistics into its rule. Also updates the
3735 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3736 * 'facet''s statistics in the datapath should have been zeroed and folded into
3737 * its packet and byte counts before this function is called. */
3739 facet_flush_stats(struct facet *facet)
3741 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3742 struct subfacet *subfacet;
3744 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3745 assert(!subfacet->dp_byte_count);
3746 assert(!subfacet->dp_packet_count);
3749 facet_push_stats(facet);
3750 if (facet->accounted_bytes < facet->byte_count) {
3751 facet_account(facet);
3752 facet->accounted_bytes = facet->byte_count;
3755 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3756 struct ofexpired expired;
3757 expired.flow = facet->flow;
3758 expired.packet_count = facet->packet_count;
3759 expired.byte_count = facet->byte_count;
3760 expired.used = facet->used;
3761 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3764 facet->rule->packet_count += facet->packet_count;
3765 facet->rule->byte_count += facet->byte_count;
3767 /* Reset counters to prevent double counting if 'facet' ever gets
3769 facet_reset_counters(facet);
3771 netflow_flow_clear(&facet->nf_flow);
3772 facet->tcp_flags = 0;
3775 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3776 * Returns it if found, otherwise a null pointer.
3778 * 'hash' must be the return value of flow_hash(flow, 0).
3780 * The returned facet might need revalidation; use facet_lookup_valid()
3781 * instead if that is important. */
3782 static struct facet *
3783 facet_find(struct ofproto_dpif *ofproto,
3784 const struct flow *flow, uint32_t hash)
3786 struct facet *facet;
3788 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
3789 if (flow_equal(flow, &facet->flow)) {
3797 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3798 * Returns it if found, otherwise a null pointer.
3800 * 'hash' must be the return value of flow_hash(flow, 0).
3802 * The returned facet is guaranteed to be valid. */
3803 static struct facet *
3804 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
3807 struct facet *facet;
3809 facet = facet_find(ofproto, flow, hash);
3811 && (ofproto->need_revalidate
3812 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))) {
3813 facet_revalidate(facet);
3820 subfacet_path_to_string(enum subfacet_path path)
3823 case SF_NOT_INSTALLED:
3824 return "not installed";
3826 return "in fast path";
3828 return "in slow path";
3834 /* Returns the path in which a subfacet should be installed if its 'slow'
3835 * member has the specified value. */
3836 static enum subfacet_path
3837 subfacet_want_path(enum slow_path_reason slow)
3839 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
3842 /* Returns true if 'subfacet' needs to have its datapath flow updated,
3843 * supposing that its actions have been recalculated as 'want_actions' and that
3844 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
3846 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
3847 const struct ofpbuf *want_actions)
3849 enum subfacet_path want_path = subfacet_want_path(slow);
3850 return (want_path != subfacet->path
3851 || (want_path == SF_FAST_PATH
3852 && (subfacet->actions_len != want_actions->size
3853 || memcmp(subfacet->actions, want_actions->data,
3854 subfacet->actions_len))));
3858 facet_check_consistency(struct facet *facet)
3860 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3862 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3864 uint64_t odp_actions_stub[1024 / 8];
3865 struct ofpbuf odp_actions;
3867 struct rule_dpif *rule;
3868 struct subfacet *subfacet;
3869 bool may_log = false;
3872 /* Check the rule for consistency. */
3873 rule = rule_dpif_lookup(ofproto, &facet->flow);
3874 ok = rule == facet->rule;
3876 may_log = !VLOG_DROP_WARN(&rl);
3881 flow_format(&s, &facet->flow);
3882 ds_put_format(&s, ": facet associated with wrong rule (was "
3883 "table=%"PRIu8",", facet->rule->up.table_id);
3884 cls_rule_format(&facet->rule->up.cr, &s);
3885 ds_put_format(&s, ") (should have been table=%"PRIu8",",
3887 cls_rule_format(&rule->up.cr, &s);
3888 ds_put_char(&s, ')');
3890 VLOG_WARN("%s", ds_cstr(&s));
3895 /* Check the datapath actions for consistency. */
3896 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
3897 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3898 enum subfacet_path want_path;
3899 struct odputil_keybuf keybuf;
3900 struct action_xlate_ctx ctx;
3904 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3905 subfacet->initial_tci, rule, 0, NULL);
3906 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions,
3909 if (subfacet->path == SF_NOT_INSTALLED) {
3910 /* This only happens if the datapath reported an error when we
3911 * tried to install the flow. Don't flag another error here. */
3915 want_path = subfacet_want_path(subfacet->slow);
3916 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
3917 /* The actions for slow-path flows may legitimately vary from one
3918 * packet to the next. We're done. */
3922 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
3926 /* Inconsistency! */
3928 may_log = !VLOG_DROP_WARN(&rl);
3932 /* Rate-limited, skip reporting. */
3937 subfacet_get_key(subfacet, &keybuf, &key);
3938 odp_flow_key_format(key.data, key.size, &s);
3940 ds_put_cstr(&s, ": inconsistency in subfacet");
3941 if (want_path != subfacet->path) {
3942 enum odp_key_fitness fitness = subfacet->key_fitness;
3944 ds_put_format(&s, " (%s, fitness=%s)",
3945 subfacet_path_to_string(subfacet->path),
3946 odp_key_fitness_to_string(fitness));
3947 ds_put_format(&s, " (should have been %s)",
3948 subfacet_path_to_string(want_path));
3949 } else if (want_path == SF_FAST_PATH) {
3950 ds_put_cstr(&s, " (actions were: ");
3951 format_odp_actions(&s, subfacet->actions,
3952 subfacet->actions_len);
3953 ds_put_cstr(&s, ") (correct actions: ");
3954 format_odp_actions(&s, odp_actions.data, odp_actions.size);
3955 ds_put_char(&s, ')');
3957 ds_put_cstr(&s, " (actions: ");
3958 format_odp_actions(&s, subfacet->actions,
3959 subfacet->actions_len);
3960 ds_put_char(&s, ')');
3962 VLOG_WARN("%s", ds_cstr(&s));
3965 ofpbuf_uninit(&odp_actions);
3970 /* Re-searches the classifier for 'facet':
3972 * - If the rule found is different from 'facet''s current rule, moves
3973 * 'facet' to the new rule and recompiles its actions.
3975 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3976 * where it is and recompiles its actions anyway. */
3978 facet_revalidate(struct facet *facet)
3980 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3982 struct nlattr *odp_actions;
3985 struct actions *new_actions;
3987 struct action_xlate_ctx ctx;
3988 uint64_t odp_actions_stub[1024 / 8];
3989 struct ofpbuf odp_actions;
3991 struct rule_dpif *new_rule;
3992 struct subfacet *subfacet;
3995 COVERAGE_INC(facet_revalidate);
3997 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
3999 /* Calculate new datapath actions.
4001 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4002 * emit a NetFlow expiration and, if so, we need to have the old state
4003 * around to properly compose it. */
4005 /* If the datapath actions changed or the installability changed,
4006 * then we need to talk to the datapath. */
4009 memset(&ctx, 0, sizeof ctx);
4010 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4011 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4012 enum slow_path_reason slow;
4014 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4015 subfacet->initial_tci, new_rule, 0, NULL);
4016 xlate_actions(&ctx, new_rule->up.actions, new_rule->up.n_actions,
4019 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4020 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
4021 struct dpif_flow_stats stats;
4023 subfacet_install(subfacet,
4024 odp_actions.data, odp_actions.size, &stats, slow);
4025 subfacet_update_stats(subfacet, &stats);
4028 new_actions = xcalloc(list_size(&facet->subfacets),
4029 sizeof *new_actions);
4031 new_actions[i].odp_actions = xmemdup(odp_actions.data,
4033 new_actions[i].actions_len = odp_actions.size;
4038 ofpbuf_uninit(&odp_actions);
4041 facet_flush_stats(facet);
4044 /* Update 'facet' now that we've taken care of all the old state. */
4045 facet->tags = ctx.tags;
4046 facet->nf_flow.output_iface = ctx.nf_output_iface;
4047 facet->has_learn = ctx.has_learn;
4048 facet->has_normal = ctx.has_normal;
4049 facet->has_fin_timeout = ctx.has_fin_timeout;
4050 facet->mirrors = ctx.mirrors;
4053 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4054 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4056 if (new_actions && new_actions[i].odp_actions) {
4057 free(subfacet->actions);
4058 subfacet->actions = new_actions[i].odp_actions;
4059 subfacet->actions_len = new_actions[i].actions_len;
4065 if (facet->rule != new_rule) {
4066 COVERAGE_INC(facet_changed_rule);
4067 list_remove(&facet->list_node);
4068 list_push_back(&new_rule->facets, &facet->list_node);
4069 facet->rule = new_rule;
4070 facet->used = new_rule->up.created;
4071 facet->prev_used = facet->used;
4075 /* Updates 'facet''s used time. Caller is responsible for calling
4076 * facet_push_stats() to update the flows which 'facet' resubmits into. */
4078 facet_update_time(struct facet *facet, long long int used)
4080 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4081 if (used > facet->used) {
4083 ofproto_rule_update_used(&facet->rule->up, used);
4084 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
4089 facet_reset_counters(struct facet *facet)
4091 facet->packet_count = 0;
4092 facet->byte_count = 0;
4093 facet->prev_packet_count = 0;
4094 facet->prev_byte_count = 0;
4095 facet->accounted_bytes = 0;
4099 facet_push_stats(struct facet *facet)
4101 struct dpif_flow_stats stats;
4103 assert(facet->packet_count >= facet->prev_packet_count);
4104 assert(facet->byte_count >= facet->prev_byte_count);
4105 assert(facet->used >= facet->prev_used);
4107 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4108 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4109 stats.used = facet->used;
4110 stats.tcp_flags = 0;
4112 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
4113 facet->prev_packet_count = facet->packet_count;
4114 facet->prev_byte_count = facet->byte_count;
4115 facet->prev_used = facet->used;
4117 flow_push_stats(facet->rule, &facet->flow, &stats);
4119 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
4120 facet->mirrors, stats.n_packets, stats.n_bytes);
4125 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4127 rule->packet_count += stats->n_packets;
4128 rule->byte_count += stats->n_bytes;
4129 ofproto_rule_update_used(&rule->up, stats->used);
4132 /* Pushes flow statistics to the rules which 'flow' resubmits into given
4133 * 'rule''s actions and mirrors. */
4135 flow_push_stats(struct rule_dpif *rule,
4136 const struct flow *flow, const struct dpif_flow_stats *stats)
4138 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4139 struct action_xlate_ctx ctx;
4141 ofproto_rule_update_used(&rule->up, stats->used);
4143 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule,
4145 ctx.resubmit_stats = stats;
4146 xlate_actions_for_side_effects(&ctx, rule->up.actions, rule->up.n_actions);
4151 static struct subfacet *
4152 subfacet_find__(struct ofproto_dpif *ofproto,
4153 const struct nlattr *key, size_t key_len, uint32_t key_hash,
4154 const struct flow *flow)
4156 struct subfacet *subfacet;
4158 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4159 &ofproto->subfacets) {
4161 ? (subfacet->key_len == key_len
4162 && !memcmp(key, subfacet->key, key_len))
4163 : flow_equal(flow, &subfacet->facet->flow)) {
4171 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4172 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
4173 * there is one, otherwise creates and returns a new subfacet.
4175 * If the returned subfacet is new, then subfacet->actions will be NULL, in
4176 * which case the caller must populate the actions with
4177 * subfacet_make_actions(). */
4178 static struct subfacet *
4179 subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness,
4180 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
4182 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4183 uint32_t key_hash = odp_flow_key_hash(key, key_len);
4184 struct subfacet *subfacet;
4186 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
4188 if (subfacet->facet == facet) {
4192 /* This shouldn't happen. */
4193 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4194 subfacet_destroy(subfacet);
4197 subfacet = (list_is_empty(&facet->subfacets)
4198 ? &facet->one_subfacet
4199 : xmalloc(sizeof *subfacet));
4200 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
4201 list_push_back(&facet->subfacets, &subfacet->list_node);
4202 subfacet->facet = facet;
4203 subfacet->key_fitness = key_fitness;
4204 if (key_fitness != ODP_FIT_PERFECT) {
4205 subfacet->key = xmemdup(key, key_len);
4206 subfacet->key_len = key_len;
4208 subfacet->key = NULL;
4209 subfacet->key_len = 0;
4211 subfacet->used = time_msec();
4212 subfacet->dp_packet_count = 0;
4213 subfacet->dp_byte_count = 0;
4214 subfacet->actions_len = 0;
4215 subfacet->actions = NULL;
4216 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
4219 subfacet->path = SF_NOT_INSTALLED;
4220 subfacet->initial_tci = initial_tci;
4225 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
4226 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
4227 static struct subfacet *
4228 subfacet_find(struct ofproto_dpif *ofproto,
4229 const struct nlattr *key, size_t key_len)
4231 uint32_t key_hash = odp_flow_key_hash(key, key_len);
4232 enum odp_key_fitness fitness;
4235 fitness = odp_flow_key_to_flow(key, key_len, &flow);
4236 if (fitness == ODP_FIT_ERROR) {
4240 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
4243 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4244 * its facet within 'ofproto', and frees it. */
4246 subfacet_destroy__(struct subfacet *subfacet)
4248 struct facet *facet = subfacet->facet;
4249 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4251 subfacet_uninstall(subfacet);
4252 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
4253 list_remove(&subfacet->list_node);
4254 free(subfacet->key);
4255 free(subfacet->actions);
4256 if (subfacet != &facet->one_subfacet) {
4261 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4262 * last remaining subfacet in its facet destroys the facet too. */
4264 subfacet_destroy(struct subfacet *subfacet)
4266 struct facet *facet = subfacet->facet;
4268 if (list_is_singleton(&facet->subfacets)) {
4269 /* facet_remove() needs at least one subfacet (it will remove it). */
4270 facet_remove(facet);
4272 subfacet_destroy__(subfacet);
4276 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
4277 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
4278 * for use as temporary storage. */
4280 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
4283 if (!subfacet->key) {
4284 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
4285 odp_flow_key_from_flow(key, &subfacet->facet->flow);
4287 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
4291 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
4292 * Translates the actions into 'odp_actions', which the caller must have
4293 * initialized and is responsible for uninitializing. */
4295 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
4296 struct ofpbuf *odp_actions)
4298 struct facet *facet = subfacet->facet;
4299 struct rule_dpif *rule = facet->rule;
4300 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4302 struct action_xlate_ctx ctx;
4304 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
4306 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions, odp_actions);
4307 facet->tags = ctx.tags;
4308 facet->has_learn = ctx.has_learn;
4309 facet->has_normal = ctx.has_normal;
4310 facet->has_fin_timeout = ctx.has_fin_timeout;
4311 facet->nf_flow.output_iface = ctx.nf_output_iface;
4312 facet->mirrors = ctx.mirrors;
4314 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4315 if (subfacet->actions_len != odp_actions->size
4316 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
4317 free(subfacet->actions);
4318 subfacet->actions_len = odp_actions->size;
4319 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
4323 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4324 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4325 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4326 * since 'subfacet' was last updated.
4328 * Returns 0 if successful, otherwise a positive errno value. */
4330 subfacet_install(struct subfacet *subfacet,
4331 const struct nlattr *actions, size_t actions_len,
4332 struct dpif_flow_stats *stats,
4333 enum slow_path_reason slow)
4335 struct facet *facet = subfacet->facet;
4336 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4337 enum subfacet_path path = subfacet_want_path(slow);
4338 uint64_t slow_path_stub[128 / 8];
4339 struct odputil_keybuf keybuf;
4340 enum dpif_flow_put_flags flags;
4344 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
4346 flags |= DPIF_FP_ZERO_STATS;
4349 if (path == SF_SLOW_PATH) {
4350 compose_slow_path(ofproto, &facet->flow, slow,
4351 slow_path_stub, sizeof slow_path_stub,
4352 &actions, &actions_len);
4355 subfacet_get_key(subfacet, &keybuf, &key);
4356 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
4357 actions, actions_len, stats);
4360 subfacet_reset_dp_stats(subfacet, stats);
4364 subfacet->path = path;
4370 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
4372 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
4373 stats, subfacet->slow);
4376 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4378 subfacet_uninstall(struct subfacet *subfacet)
4380 if (subfacet->path != SF_NOT_INSTALLED) {
4381 struct rule_dpif *rule = subfacet->facet->rule;
4382 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4383 struct odputil_keybuf keybuf;
4384 struct dpif_flow_stats stats;
4388 subfacet_get_key(subfacet, &keybuf, &key);
4389 error = dpif_flow_del(ofproto->dpif, key.data, key.size, &stats);
4390 subfacet_reset_dp_stats(subfacet, &stats);
4392 subfacet_update_stats(subfacet, &stats);
4394 subfacet->path = SF_NOT_INSTALLED;
4396 assert(subfacet->dp_packet_count == 0);
4397 assert(subfacet->dp_byte_count == 0);
4401 /* Resets 'subfacet''s datapath statistics counters. This should be called
4402 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4403 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4404 * was reset in the datapath. 'stats' will be modified to include only
4405 * statistics new since 'subfacet' was last updated. */
4407 subfacet_reset_dp_stats(struct subfacet *subfacet,
4408 struct dpif_flow_stats *stats)
4411 && subfacet->dp_packet_count <= stats->n_packets
4412 && subfacet->dp_byte_count <= stats->n_bytes) {
4413 stats->n_packets -= subfacet->dp_packet_count;
4414 stats->n_bytes -= subfacet->dp_byte_count;
4417 subfacet->dp_packet_count = 0;
4418 subfacet->dp_byte_count = 0;
4421 /* Updates 'subfacet''s used time. The caller is responsible for calling
4422 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
4424 subfacet_update_time(struct subfacet *subfacet, long long int used)
4426 if (used > subfacet->used) {
4427 subfacet->used = used;
4428 facet_update_time(subfacet->facet, used);
4432 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4434 * Because of the meaning of a subfacet's counters, it only makes sense to do
4435 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4436 * represents a packet that was sent by hand or if it represents statistics
4437 * that have been cleared out of the datapath. */
4439 subfacet_update_stats(struct subfacet *subfacet,
4440 const struct dpif_flow_stats *stats)
4442 if (stats->n_packets || stats->used > subfacet->used) {
4443 struct facet *facet = subfacet->facet;
4445 subfacet_update_time(subfacet, stats->used);
4446 facet->packet_count += stats->n_packets;
4447 facet->byte_count += stats->n_bytes;
4448 facet->tcp_flags |= stats->tcp_flags;
4449 facet_push_stats(facet);
4450 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
4456 static struct rule_dpif *
4457 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
4459 struct ofport_dpif *port;
4460 struct rule_dpif *rule;
4462 rule = rule_dpif_lookup__(ofproto, flow, 0);
4467 port = get_ofp_port(ofproto, flow->in_port);
4469 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
4470 return ofproto->miss_rule;
4473 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
4474 return ofproto->no_packet_in_rule;
4476 return ofproto->miss_rule;
4479 static struct rule_dpif *
4480 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
4483 struct cls_rule *cls_rule;
4484 struct classifier *cls;
4486 if (table_id >= N_TABLES) {
4490 cls = &ofproto->up.tables[table_id].cls;
4491 if (flow->nw_frag & FLOW_NW_FRAG_ANY
4492 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4493 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
4494 * are unavailable. */
4495 struct flow ofpc_normal_flow = *flow;
4496 ofpc_normal_flow.tp_src = htons(0);
4497 ofpc_normal_flow.tp_dst = htons(0);
4498 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
4500 cls_rule = classifier_lookup(cls, flow);
4502 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
4506 complete_operation(struct rule_dpif *rule)
4508 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4510 rule_invalidate(rule);
4512 struct dpif_completion *c = xmalloc(sizeof *c);
4513 c->op = rule->up.pending;
4514 list_push_back(&ofproto->completions, &c->list_node);
4516 ofoperation_complete(rule->up.pending, 0);
4520 static struct rule *
4523 struct rule_dpif *rule = xmalloc(sizeof *rule);
4528 rule_dealloc(struct rule *rule_)
4530 struct rule_dpif *rule = rule_dpif_cast(rule_);
4535 rule_construct(struct rule *rule_)
4537 struct rule_dpif *rule = rule_dpif_cast(rule_);
4538 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4539 struct rule_dpif *victim;
4543 error = validate_actions(rule->up.actions, rule->up.n_actions,
4544 &rule->up.cr.flow, ofproto->max_ports);
4549 rule->packet_count = 0;
4550 rule->byte_count = 0;
4552 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
4553 if (victim && !list_is_empty(&victim->facets)) {
4554 struct facet *facet;
4556 rule->facets = victim->facets;
4557 list_moved(&rule->facets);
4558 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4559 /* XXX: We're only clearing our local counters here. It's possible
4560 * that quite a few packets are unaccounted for in the datapath
4561 * statistics. These will be accounted to the new rule instead of
4562 * cleared as required. This could be fixed by clearing out the
4563 * datapath statistics for this facet, but currently it doesn't
4565 facet_reset_counters(facet);
4569 /* Must avoid list_moved() in this case. */
4570 list_init(&rule->facets);
4573 table_id = rule->up.table_id;
4574 rule->tag = (victim ? victim->tag
4576 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
4577 ofproto->tables[table_id].basis));
4579 complete_operation(rule);
4584 rule_destruct(struct rule *rule_)
4586 struct rule_dpif *rule = rule_dpif_cast(rule_);
4587 struct facet *facet, *next_facet;
4589 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4590 facet_revalidate(facet);
4593 complete_operation(rule);
4597 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4599 struct rule_dpif *rule = rule_dpif_cast(rule_);
4600 struct facet *facet;
4602 /* Start from historical data for 'rule' itself that are no longer tracked
4603 * in facets. This counts, for example, facets that have expired. */
4604 *packets = rule->packet_count;
4605 *bytes = rule->byte_count;
4607 /* Add any statistics that are tracked by facets. This includes
4608 * statistical data recently updated by ofproto_update_stats() as well as
4609 * stats for packets that were executed "by hand" via dpif_execute(). */
4610 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4611 *packets += facet->packet_count;
4612 *bytes += facet->byte_count;
4617 rule_execute(struct rule *rule_, const struct flow *flow,
4618 struct ofpbuf *packet)
4620 struct rule_dpif *rule = rule_dpif_cast(rule_);
4621 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4623 struct dpif_flow_stats stats;
4625 struct action_xlate_ctx ctx;
4626 uint64_t odp_actions_stub[1024 / 8];
4627 struct ofpbuf odp_actions;
4629 dpif_flow_stats_extract(flow, packet, &stats);
4630 rule_credit_stats(rule, &stats);
4632 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4633 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
4634 rule, stats.tcp_flags, packet);
4635 ctx.resubmit_stats = &stats;
4636 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions, &odp_actions);
4638 execute_odp_actions(ofproto, flow, odp_actions.data,
4639 odp_actions.size, packet);
4641 ofpbuf_uninit(&odp_actions);
4647 rule_modify_actions(struct rule *rule_)
4649 struct rule_dpif *rule = rule_dpif_cast(rule_);
4650 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4653 error = validate_actions(rule->up.actions, rule->up.n_actions,
4654 &rule->up.cr.flow, ofproto->max_ports);
4656 ofoperation_complete(rule->up.pending, error);
4660 complete_operation(rule);
4663 /* Sends 'packet' out 'ofport'.
4664 * May modify 'packet'.
4665 * Returns 0 if successful, otherwise a positive errno value. */
4667 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4669 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4670 struct ofpbuf key, odp_actions;
4671 struct odputil_keybuf keybuf;
4676 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
4677 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4679 if (odp_port != ofport->odp_port) {
4680 eth_pop_vlan(packet);
4681 flow.vlan_tci = htons(0);
4684 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4685 odp_flow_key_from_flow(&key, &flow);
4687 ofpbuf_init(&odp_actions, 32);
4688 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4690 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4691 error = dpif_execute(ofproto->dpif,
4693 odp_actions.data, odp_actions.size,
4695 ofpbuf_uninit(&odp_actions);
4698 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4699 ofproto->up.name, odp_port, strerror(error));
4701 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
4705 /* OpenFlow to datapath action translation. */
4707 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
4708 struct action_xlate_ctx *ctx);
4709 static void xlate_normal(struct action_xlate_ctx *);
4711 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
4712 * The action will state 'slow' as the reason that the action is in the slow
4713 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
4714 * dump-flows" output to see why a flow is in the slow path.)
4716 * The 'stub_size' bytes in 'stub' will be used to store the action.
4717 * 'stub_size' must be large enough for the action.
4719 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
4722 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
4723 enum slow_path_reason slow,
4724 uint64_t *stub, size_t stub_size,
4725 const struct nlattr **actionsp, size_t *actions_lenp)
4727 union user_action_cookie cookie;
4730 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
4731 cookie.slow_path.unused = 0;
4732 cookie.slow_path.reason = slow;
4734 ofpbuf_use_stack(&buf, stub, stub_size);
4735 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
4736 uint32_t pid = dpif_port_get_pid(ofproto->dpif, UINT16_MAX);
4737 odp_put_userspace_action(pid, &cookie, &buf);
4739 put_userspace_action(ofproto, &buf, flow, &cookie);
4741 *actionsp = buf.data;
4742 *actions_lenp = buf.size;
4746 put_userspace_action(const struct ofproto_dpif *ofproto,
4747 struct ofpbuf *odp_actions,
4748 const struct flow *flow,
4749 const union user_action_cookie *cookie)
4753 pid = dpif_port_get_pid(ofproto->dpif,
4754 ofp_port_to_odp_port(flow->in_port));
4756 return odp_put_userspace_action(pid, cookie, odp_actions);
4760 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
4761 ovs_be16 vlan_tci, uint32_t odp_port,
4762 unsigned int n_outputs, union user_action_cookie *cookie)
4766 cookie->type = USER_ACTION_COOKIE_SFLOW;
4767 cookie->sflow.vlan_tci = vlan_tci;
4769 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
4770 * port information") for the interpretation of cookie->output. */
4771 switch (n_outputs) {
4773 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
4774 cookie->sflow.output = 0x40000000 | 256;
4778 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4780 cookie->sflow.output = ifindex;
4785 /* 0x80000000 means "multiple output ports. */
4786 cookie->sflow.output = 0x80000000 | n_outputs;
4791 /* Compose SAMPLE action for sFlow. */
4793 compose_sflow_action(const struct ofproto_dpif *ofproto,
4794 struct ofpbuf *odp_actions,
4795 const struct flow *flow,
4798 uint32_t probability;
4799 union user_action_cookie cookie;
4800 size_t sample_offset, actions_offset;
4803 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4807 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4809 /* Number of packets out of UINT_MAX to sample. */
4810 probability = dpif_sflow_get_probability(ofproto->sflow);
4811 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4813 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4814 compose_sflow_cookie(ofproto, htons(0), odp_port,
4815 odp_port == OVSP_NONE ? 0 : 1, &cookie);
4816 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4818 nl_msg_end_nested(odp_actions, actions_offset);
4819 nl_msg_end_nested(odp_actions, sample_offset);
4820 return cookie_offset;
4823 /* SAMPLE action must be first action in any given list of actions.
4824 * At this point we do not have all information required to build it. So try to
4825 * build sample action as complete as possible. */
4827 add_sflow_action(struct action_xlate_ctx *ctx)
4829 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4831 &ctx->flow, OVSP_NONE);
4832 ctx->sflow_odp_port = 0;
4833 ctx->sflow_n_outputs = 0;
4836 /* Fix SAMPLE action according to data collected while composing ODP actions.
4837 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4838 * USERSPACE action's user-cookie which is required for sflow. */
4840 fix_sflow_action(struct action_xlate_ctx *ctx)
4842 const struct flow *base = &ctx->base_flow;
4843 union user_action_cookie *cookie;
4845 if (!ctx->user_cookie_offset) {
4849 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4851 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4853 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
4854 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
4858 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4861 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4862 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4863 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4864 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4868 struct priority_to_dscp *pdscp;
4870 if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD
4871 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4875 pdscp = get_priority(ofport, ctx->flow.skb_priority);
4877 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4878 ctx->flow.nw_tos |= pdscp->dscp;
4881 /* We may not have an ofport record for this port, but it doesn't hurt
4882 * to allow forwarding to it anyhow. Maybe such a port will appear
4883 * later and we're pre-populating the flow table. */
4886 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4887 ctx->flow.vlan_tci);
4888 if (out_port != odp_port) {
4889 ctx->flow.vlan_tci = htons(0);
4891 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4892 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4894 ctx->sflow_odp_port = odp_port;
4895 ctx->sflow_n_outputs++;
4896 ctx->nf_output_iface = ofp_port;
4897 ctx->flow.vlan_tci = flow_vlan_tci;
4898 ctx->flow.nw_tos = flow_nw_tos;
4902 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4904 compose_output_action__(ctx, ofp_port, true);
4908 xlate_table_action(struct action_xlate_ctx *ctx,
4909 uint16_t in_port, uint8_t table_id)
4911 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4912 struct ofproto_dpif *ofproto = ctx->ofproto;
4913 struct rule_dpif *rule;
4914 uint16_t old_in_port;
4915 uint8_t old_table_id;
4917 old_table_id = ctx->table_id;
4918 ctx->table_id = table_id;
4920 /* Look up a flow with 'in_port' as the input port. */
4921 old_in_port = ctx->flow.in_port;
4922 ctx->flow.in_port = in_port;
4923 rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id);
4926 if (table_id > 0 && table_id < N_TABLES) {
4927 struct table_dpif *table = &ofproto->tables[table_id];
4928 if (table->other_table) {
4929 ctx->tags |= (rule && rule->tag
4931 : rule_calculate_tag(&ctx->flow,
4932 &table->other_table->wc,
4937 /* Restore the original input port. Otherwise OFPP_NORMAL and
4938 * OFPP_IN_PORT will have surprising behavior. */
4939 ctx->flow.in_port = old_in_port;
4941 if (ctx->resubmit_hook) {
4942 ctx->resubmit_hook(ctx, rule);
4946 struct rule_dpif *old_rule = ctx->rule;
4948 if (ctx->resubmit_stats) {
4949 rule_credit_stats(rule, ctx->resubmit_stats);
4954 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
4955 ctx->rule = old_rule;
4959 ctx->table_id = old_table_id;
4961 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4963 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
4964 MAX_RESUBMIT_RECURSION);
4965 ctx->max_resubmit_trigger = true;
4970 xlate_resubmit_table(struct action_xlate_ctx *ctx,
4971 const struct nx_action_resubmit *nar)
4976 in_port = (nar->in_port == htons(OFPP_IN_PORT)
4978 : ntohs(nar->in_port));
4979 table_id = nar->table == 255 ? ctx->table_id : nar->table;
4981 xlate_table_action(ctx, in_port, table_id);
4985 flood_packets(struct action_xlate_ctx *ctx, bool all)
4987 struct ofport_dpif *ofport;
4989 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
4990 uint16_t ofp_port = ofport->up.ofp_port;
4992 if (ofp_port == ctx->flow.in_port) {
4997 compose_output_action__(ctx, ofp_port, false);
4998 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
4999 compose_output_action(ctx, ofp_port);
5003 ctx->nf_output_iface = NF_OUT_FLOOD;
5007 execute_controller_action(struct action_xlate_ctx *ctx, int len,
5008 enum ofp_packet_in_reason reason,
5009 uint16_t controller_id)
5011 struct ofputil_packet_in pin;
5012 struct ofpbuf *packet;
5014 ctx->slow |= SLOW_CONTROLLER;
5019 packet = ofpbuf_clone(ctx->packet);
5021 if (packet->l2 && packet->l3) {
5022 struct eth_header *eh;
5024 eth_pop_vlan(packet);
5027 /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
5028 * LLC frame. Calculating the Ethernet type of these frames is more
5029 * trouble than seems appropriate for a simple assertion. */
5030 assert(ntohs(eh->eth_type) < ETH_TYPE_MIN
5031 || eh->eth_type == ctx->flow.dl_type);
5033 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
5034 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
5036 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
5037 eth_push_vlan(packet, ctx->flow.vlan_tci);
5041 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5042 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
5043 ctx->flow.nw_tos, ctx->flow.nw_ttl);
5047 if (ctx->flow.nw_proto == IPPROTO_TCP) {
5048 packet_set_tcp_port(packet, ctx->flow.tp_src,
5050 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
5051 packet_set_udp_port(packet, ctx->flow.tp_src,
5058 pin.packet = packet->data;
5059 pin.packet_len = packet->size;
5060 pin.reason = reason;
5061 pin.controller_id = controller_id;
5062 pin.table_id = ctx->table_id;
5063 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
5066 flow_get_metadata(&ctx->flow, &pin.fmd);
5068 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
5069 ofpbuf_delete(packet);
5073 compose_dec_ttl(struct action_xlate_ctx *ctx)
5075 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
5076 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
5080 if (ctx->flow.nw_ttl > 1) {
5084 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
5086 /* Stop processing for current table. */
5092 xlate_output_action__(struct action_xlate_ctx *ctx,
5093 uint16_t port, uint16_t max_len)
5095 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
5097 ctx->nf_output_iface = NF_OUT_DROP;
5101 compose_output_action(ctx, ctx->flow.in_port);
5104 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
5110 flood_packets(ctx, false);
5113 flood_packets(ctx, true);
5115 case OFPP_CONTROLLER:
5116 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
5122 if (port != ctx->flow.in_port) {
5123 compose_output_action(ctx, port);
5128 if (prev_nf_output_iface == NF_OUT_FLOOD) {
5129 ctx->nf_output_iface = NF_OUT_FLOOD;
5130 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
5131 ctx->nf_output_iface = prev_nf_output_iface;
5132 } else if (prev_nf_output_iface != NF_OUT_DROP &&
5133 ctx->nf_output_iface != NF_OUT_FLOOD) {
5134 ctx->nf_output_iface = NF_OUT_MULTI;
5139 xlate_output_reg_action(struct action_xlate_ctx *ctx,
5140 const struct nx_action_output_reg *naor)
5142 struct mf_subfield src;
5145 nxm_decode(&src, naor->src, naor->ofs_nbits);
5146 ofp_port = mf_get_subfield(&src, &ctx->flow);
5148 if (ofp_port <= UINT16_MAX) {
5149 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
5154 xlate_output_action(struct action_xlate_ctx *ctx,
5155 const struct ofp_action_output *oao)
5157 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
5161 xlate_enqueue_action(struct action_xlate_ctx *ctx,
5162 const struct ofp_action_enqueue *oae)
5165 uint32_t flow_priority, priority;
5168 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
5171 /* Fall back to ordinary output action. */
5172 xlate_output_action__(ctx, ntohs(oae->port), 0);
5176 /* Figure out datapath output port. */
5177 ofp_port = ntohs(oae->port);
5178 if (ofp_port == OFPP_IN_PORT) {
5179 ofp_port = ctx->flow.in_port;
5180 } else if (ofp_port == ctx->flow.in_port) {
5184 /* Add datapath actions. */
5185 flow_priority = ctx->flow.skb_priority;
5186 ctx->flow.skb_priority = priority;
5187 compose_output_action(ctx, ofp_port);
5188 ctx->flow.skb_priority = flow_priority;
5190 /* Update NetFlow output port. */
5191 if (ctx->nf_output_iface == NF_OUT_DROP) {
5192 ctx->nf_output_iface = ofp_port;
5193 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
5194 ctx->nf_output_iface = NF_OUT_MULTI;
5199 xlate_set_queue_action(struct action_xlate_ctx *ctx,
5200 const struct nx_action_set_queue *nasq)
5205 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
5208 /* Couldn't translate queue to a priority, so ignore. A warning
5209 * has already been logged. */
5213 ctx->flow.skb_priority = priority;
5216 struct xlate_reg_state {
5222 xlate_autopath(struct action_xlate_ctx *ctx,
5223 const struct nx_action_autopath *naa)
5225 uint16_t ofp_port = ntohl(naa->id);
5226 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
5228 if (!port || !port->bundle) {
5229 ofp_port = OFPP_NONE;
5230 } else if (port->bundle->bond) {
5231 /* Autopath does not support VLAN hashing. */
5232 struct ofport_dpif *slave = bond_choose_output_slave(
5233 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
5235 ofp_port = slave->up.ofp_port;
5238 autopath_execute(naa, &ctx->flow, ofp_port);
5242 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
5244 struct ofproto_dpif *ofproto = ofproto_;
5245 struct ofport_dpif *port;
5255 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
5258 port = get_ofp_port(ofproto, ofp_port);
5259 return port ? port->may_enable : false;
5264 xlate_learn_action(struct action_xlate_ctx *ctx,
5265 const struct nx_action_learn *learn)
5267 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
5268 struct ofputil_flow_mod fm;
5271 learn_execute(learn, &ctx->flow, &fm);
5273 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
5274 if (error && !VLOG_DROP_WARN(&rl)) {
5275 VLOG_WARN("learning action failed to modify flow table (%s)",
5276 ofperr_get_name(error));
5282 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
5283 * means "infinite". */
5285 reduce_timeout(uint16_t max, uint16_t *timeout)
5287 if (max && (!*timeout || *timeout > max)) {
5293 xlate_fin_timeout(struct action_xlate_ctx *ctx,
5294 const struct nx_action_fin_timeout *naft)
5296 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
5297 struct rule_dpif *rule = ctx->rule;
5299 reduce_timeout(ntohs(naft->fin_idle_timeout), &rule->up.idle_timeout);
5300 reduce_timeout(ntohs(naft->fin_hard_timeout), &rule->up.hard_timeout);
5305 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
5307 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
5308 ? OFPUTIL_PC_NO_RECV_STP
5309 : OFPUTIL_PC_NO_RECV)) {
5313 /* Only drop packets here if both forwarding and learning are
5314 * disabled. If just learning is enabled, we need to have
5315 * OFPP_NORMAL and the learning action have a look at the packet
5316 * before we can drop it. */
5317 if (!stp_forward_in_state(port->stp_state)
5318 && !stp_learn_in_state(port->stp_state)) {
5326 do_xlate_actions(const union ofp_action *in, size_t n_in,
5327 struct action_xlate_ctx *ctx)
5329 const struct ofport_dpif *port;
5330 const union ofp_action *ia;
5331 bool was_evictable = true;
5334 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5335 if (port && !may_receive(port, ctx)) {
5336 /* Drop this flow. */
5341 /* Don't let the rule we're working on get evicted underneath us. */
5342 was_evictable = ctx->rule->up.evictable;
5343 ctx->rule->up.evictable = false;
5345 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
5346 const struct ofp_action_dl_addr *oada;
5347 const struct nx_action_resubmit *nar;
5348 const struct nx_action_set_tunnel *nast;
5349 const struct nx_action_set_queue *nasq;
5350 const struct nx_action_multipath *nam;
5351 const struct nx_action_autopath *naa;
5352 const struct nx_action_bundle *nab;
5353 const struct nx_action_output_reg *naor;
5354 const struct nx_action_controller *nac;
5355 enum ofputil_action_code code;
5362 code = ofputil_decode_action_unsafe(ia);
5364 case OFPUTIL_OFPAT10_OUTPUT:
5365 xlate_output_action(ctx, &ia->output);
5368 case OFPUTIL_OFPAT10_SET_VLAN_VID:
5369 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
5370 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
5373 case OFPUTIL_OFPAT10_SET_VLAN_PCP:
5374 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
5375 ctx->flow.vlan_tci |= htons(
5376 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
5379 case OFPUTIL_OFPAT10_STRIP_VLAN:
5380 ctx->flow.vlan_tci = htons(0);
5383 case OFPUTIL_OFPAT10_SET_DL_SRC:
5384 oada = ((struct ofp_action_dl_addr *) ia);
5385 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
5388 case OFPUTIL_OFPAT10_SET_DL_DST:
5389 oada = ((struct ofp_action_dl_addr *) ia);
5390 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
5393 case OFPUTIL_OFPAT10_SET_NW_SRC:
5394 ctx->flow.nw_src = ia->nw_addr.nw_addr;
5397 case OFPUTIL_OFPAT10_SET_NW_DST:
5398 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
5401 case OFPUTIL_OFPAT10_SET_NW_TOS:
5402 /* OpenFlow 1.0 only supports IPv4. */
5403 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5404 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5405 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
5409 case OFPUTIL_OFPAT10_SET_TP_SRC:
5410 ctx->flow.tp_src = ia->tp_port.tp_port;
5413 case OFPUTIL_OFPAT10_SET_TP_DST:
5414 ctx->flow.tp_dst = ia->tp_port.tp_port;
5417 case OFPUTIL_OFPAT10_ENQUEUE:
5418 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
5421 case OFPUTIL_NXAST_RESUBMIT:
5422 nar = (const struct nx_action_resubmit *) ia;
5423 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
5426 case OFPUTIL_NXAST_RESUBMIT_TABLE:
5427 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
5430 case OFPUTIL_NXAST_SET_TUNNEL:
5431 nast = (const struct nx_action_set_tunnel *) ia;
5432 tun_id = htonll(ntohl(nast->tun_id));
5433 ctx->flow.tun_id = tun_id;
5436 case OFPUTIL_NXAST_SET_QUEUE:
5437 nasq = (const struct nx_action_set_queue *) ia;
5438 xlate_set_queue_action(ctx, nasq);
5441 case OFPUTIL_NXAST_POP_QUEUE:
5442 ctx->flow.skb_priority = ctx->orig_skb_priority;
5445 case OFPUTIL_NXAST_REG_MOVE:
5446 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
5450 case OFPUTIL_NXAST_REG_LOAD:
5451 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
5455 case OFPUTIL_NXAST_NOTE:
5456 /* Nothing to do. */
5459 case OFPUTIL_NXAST_SET_TUNNEL64:
5460 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
5461 ctx->flow.tun_id = tun_id;
5464 case OFPUTIL_NXAST_MULTIPATH:
5465 nam = (const struct nx_action_multipath *) ia;
5466 multipath_execute(nam, &ctx->flow);
5469 case OFPUTIL_NXAST_AUTOPATH:
5470 naa = (const struct nx_action_autopath *) ia;
5471 xlate_autopath(ctx, naa);
5474 case OFPUTIL_NXAST_BUNDLE:
5475 ctx->ofproto->has_bundle_action = true;
5476 nab = (const struct nx_action_bundle *) ia;
5477 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
5482 case OFPUTIL_NXAST_BUNDLE_LOAD:
5483 ctx->ofproto->has_bundle_action = true;
5484 nab = (const struct nx_action_bundle *) ia;
5485 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
5489 case OFPUTIL_NXAST_OUTPUT_REG:
5490 naor = (const struct nx_action_output_reg *) ia;
5491 xlate_output_reg_action(ctx, naor);
5494 case OFPUTIL_NXAST_LEARN:
5495 ctx->has_learn = true;
5496 if (ctx->may_learn) {
5497 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
5501 case OFPUTIL_NXAST_DEC_TTL:
5502 if (compose_dec_ttl(ctx)) {
5507 case OFPUTIL_NXAST_EXIT:
5511 case OFPUTIL_NXAST_FIN_TIMEOUT:
5512 ctx->has_fin_timeout = true;
5513 xlate_fin_timeout(ctx, (const struct nx_action_fin_timeout *) ia);
5516 case OFPUTIL_NXAST_CONTROLLER:
5517 nac = (const struct nx_action_controller *) ia;
5518 execute_controller_action(ctx, ntohs(nac->max_len), nac->reason,
5519 ntohs(nac->controller_id));
5525 /* We've let OFPP_NORMAL and the learning action look at the packet,
5526 * so drop it now if forwarding is disabled. */
5527 if (port && !stp_forward_in_state(port->stp_state)) {
5528 ofpbuf_clear(ctx->odp_actions);
5529 add_sflow_action(ctx);
5532 ctx->rule->up.evictable = was_evictable;
5537 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
5538 struct ofproto_dpif *ofproto, const struct flow *flow,
5539 ovs_be16 initial_tci, struct rule_dpif *rule,
5540 uint8_t tcp_flags, const struct ofpbuf *packet)
5542 ctx->ofproto = ofproto;
5544 ctx->base_flow = ctx->flow;
5545 ctx->base_flow.tun_id = 0;
5546 ctx->base_flow.vlan_tci = initial_tci;
5548 ctx->packet = packet;
5549 ctx->may_learn = packet != NULL;
5550 ctx->tcp_flags = tcp_flags;
5551 ctx->resubmit_hook = NULL;
5552 ctx->resubmit_stats = NULL;
5555 /* Translates the 'n_in' "union ofp_action"s in 'in' into datapath actions in
5556 * 'odp_actions', using 'ctx'. */
5558 xlate_actions(struct action_xlate_ctx *ctx,
5559 const union ofp_action *in, size_t n_in,
5560 struct ofpbuf *odp_actions)
5562 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
5563 * that in the future we always keep a copy of the original flow for
5564 * tracing purposes. */
5565 static bool hit_resubmit_limit;
5567 enum slow_path_reason special;
5569 COVERAGE_INC(ofproto_dpif_xlate);
5571 ofpbuf_clear(odp_actions);
5572 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
5574 ctx->odp_actions = odp_actions;
5577 ctx->has_learn = false;
5578 ctx->has_normal = false;
5579 ctx->has_fin_timeout = false;
5580 ctx->nf_output_iface = NF_OUT_DROP;
5583 ctx->max_resubmit_trigger = false;
5584 ctx->orig_skb_priority = ctx->flow.skb_priority;
5588 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
5589 /* Do this conditionally because the copy is expensive enough that it
5590 * shows up in profiles.
5592 * We keep orig_flow in 'ctx' only because I couldn't make GCC 4.4
5593 * believe that I wasn't using it without initializing it if I kept it
5594 * in a local variable. */
5595 ctx->orig_flow = ctx->flow;
5598 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
5599 switch (ctx->ofproto->up.frag_handling) {
5600 case OFPC_FRAG_NORMAL:
5601 /* We must pretend that transport ports are unavailable. */
5602 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
5603 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
5606 case OFPC_FRAG_DROP:
5609 case OFPC_FRAG_REASM:
5612 case OFPC_FRAG_NX_MATCH:
5613 /* Nothing to do. */
5616 case OFPC_INVALID_TTL_TO_CONTROLLER:
5621 special = process_special(ctx->ofproto, &ctx->flow, ctx->packet);
5623 ctx->slow |= special;
5625 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5626 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
5628 add_sflow_action(ctx);
5629 do_xlate_actions(in, n_in, ctx);
5631 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
5632 if (!hit_resubmit_limit) {
5633 /* We didn't record the original flow. Make sure we do from
5635 hit_resubmit_limit = true;
5636 } else if (!VLOG_DROP_ERR(&trace_rl)) {
5637 struct ds ds = DS_EMPTY_INITIALIZER;
5639 ofproto_trace(ctx->ofproto, &ctx->orig_flow, ctx->packet,
5641 VLOG_ERR("Trace triggered by excessive resubmit "
5642 "recursion:\n%s", ds_cstr(&ds));
5647 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
5648 ctx->odp_actions->data,
5649 ctx->odp_actions->size)) {
5650 ctx->slow |= SLOW_IN_BAND;
5652 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
5654 compose_output_action(ctx, OFPP_LOCAL);
5657 if (ctx->ofproto->has_mirrors) {
5658 add_mirror_actions(ctx, &ctx->orig_flow);
5660 fix_sflow_action(ctx);
5664 /* Translates the 'n_in' "union ofp_action"s in 'in' into datapath actions,
5665 * using 'ctx', and discards the datapath actions. */
5667 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
5668 const union ofp_action *in, size_t n_in)
5670 uint64_t odp_actions_stub[1024 / 8];
5671 struct ofpbuf odp_actions;
5673 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5674 xlate_actions(ctx, in, n_in, &odp_actions);
5675 ofpbuf_uninit(&odp_actions);
5678 /* OFPP_NORMAL implementation. */
5680 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
5682 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
5683 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
5684 * the bundle on which the packet was received, returns the VLAN to which the
5687 * Both 'vid' and the return value are in the range 0...4095. */
5689 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
5691 switch (in_bundle->vlan_mode) {
5692 case PORT_VLAN_ACCESS:
5693 return in_bundle->vlan;
5696 case PORT_VLAN_TRUNK:
5699 case PORT_VLAN_NATIVE_UNTAGGED:
5700 case PORT_VLAN_NATIVE_TAGGED:
5701 return vid ? vid : in_bundle->vlan;
5708 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
5709 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
5712 * 'vid' should be the VID obtained from the 802.1Q header that was received as
5713 * part of a packet (specify 0 if there was no 802.1Q header), in the range
5716 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
5718 /* Allow any VID on the OFPP_NONE port. */
5719 if (in_bundle == &ofpp_none_bundle) {
5723 switch (in_bundle->vlan_mode) {
5724 case PORT_VLAN_ACCESS:
5727 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5728 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
5729 "packet received on port %s configured as VLAN "
5730 "%"PRIu16" access port",
5731 in_bundle->ofproto->up.name, vid,
5732 in_bundle->name, in_bundle->vlan);
5738 case PORT_VLAN_NATIVE_UNTAGGED:
5739 case PORT_VLAN_NATIVE_TAGGED:
5741 /* Port must always carry its native VLAN. */
5745 case PORT_VLAN_TRUNK:
5746 if (!ofbundle_includes_vlan(in_bundle, vid)) {
5748 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5749 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
5750 "received on port %s not configured for trunking "
5752 in_bundle->ofproto->up.name, vid,
5753 in_bundle->name, vid);
5765 /* Given 'vlan', the VLAN that a packet belongs to, and
5766 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
5767 * that should be included in the 802.1Q header. (If the return value is 0,
5768 * then the 802.1Q header should only be included in the packet if there is a
5771 * Both 'vlan' and the return value are in the range 0...4095. */
5773 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
5775 switch (out_bundle->vlan_mode) {
5776 case PORT_VLAN_ACCESS:
5779 case PORT_VLAN_TRUNK:
5780 case PORT_VLAN_NATIVE_TAGGED:
5783 case PORT_VLAN_NATIVE_UNTAGGED:
5784 return vlan == out_bundle->vlan ? 0 : vlan;
5792 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
5795 struct ofport_dpif *port;
5797 ovs_be16 tci, old_tci;
5799 vid = output_vlan_to_vid(out_bundle, vlan);
5800 if (!out_bundle->bond) {
5801 port = ofbundle_get_a_port(out_bundle);
5803 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
5806 /* No slaves enabled, so drop packet. */
5811 old_tci = ctx->flow.vlan_tci;
5813 if (tci || out_bundle->use_priority_tags) {
5814 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
5816 tci |= htons(VLAN_CFI);
5819 ctx->flow.vlan_tci = tci;
5821 compose_output_action(ctx, port->up.ofp_port);
5822 ctx->flow.vlan_tci = old_tci;
5826 mirror_mask_ffs(mirror_mask_t mask)
5828 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
5833 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
5835 return (bundle->vlan_mode != PORT_VLAN_ACCESS
5836 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
5840 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
5842 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
5845 /* Returns an arbitrary interface within 'bundle'. */
5846 static struct ofport_dpif *
5847 ofbundle_get_a_port(const struct ofbundle *bundle)
5849 return CONTAINER_OF(list_front(&bundle->ports),
5850 struct ofport_dpif, bundle_node);
5854 vlan_is_mirrored(const struct ofmirror *m, int vlan)
5856 return !m->vlans || bitmap_is_set(m->vlans, vlan);
5860 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5862 struct ofproto_dpif *ofproto = ctx->ofproto;
5863 mirror_mask_t mirrors;
5864 struct ofbundle *in_bundle;
5867 const struct nlattr *a;
5870 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5871 ctx->packet != NULL, NULL);
5875 mirrors = in_bundle->src_mirrors;
5877 /* Drop frames on bundles reserved for mirroring. */
5878 if (in_bundle->mirror_out) {
5879 if (ctx->packet != NULL) {
5880 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5881 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5882 "%s, which is reserved exclusively for mirroring",
5883 ctx->ofproto->up.name, in_bundle->name);
5889 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5890 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5893 vlan = input_vid_to_vlan(in_bundle, vid);
5895 /* Look at the output ports to check for destination selections. */
5897 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5898 ctx->odp_actions->size) {
5899 enum ovs_action_attr type = nl_attr_type(a);
5900 struct ofport_dpif *ofport;
5902 if (type != OVS_ACTION_ATTR_OUTPUT) {
5906 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5907 if (ofport && ofport->bundle) {
5908 mirrors |= ofport->bundle->dst_mirrors;
5916 /* Restore the original packet before adding the mirror actions. */
5917 ctx->flow = *orig_flow;
5922 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5924 if (!vlan_is_mirrored(m, vlan)) {
5925 mirrors &= mirrors - 1;
5929 mirrors &= ~m->dup_mirrors;
5930 ctx->mirrors |= m->dup_mirrors;
5932 output_normal(ctx, m->out, vlan);
5933 } else if (vlan != m->out_vlan
5934 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
5935 struct ofbundle *bundle;
5937 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5938 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5939 && !bundle->mirror_out) {
5940 output_normal(ctx, bundle, m->out_vlan);
5948 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5949 uint64_t packets, uint64_t bytes)
5955 for (; mirrors; mirrors &= mirrors - 1) {
5958 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5961 /* In normal circumstances 'm' will not be NULL. However,
5962 * if mirrors are reconfigured, we can temporarily get out
5963 * of sync in facet_revalidate(). We could "correct" the
5964 * mirror list before reaching here, but doing that would
5965 * not properly account the traffic stats we've currently
5966 * accumulated for previous mirror configuration. */
5970 m->packet_count += packets;
5971 m->byte_count += bytes;
5975 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
5976 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
5977 * indicate this; newer upstream kernels use gratuitous ARP requests. */
5979 is_gratuitous_arp(const struct flow *flow)
5981 return (flow->dl_type == htons(ETH_TYPE_ARP)
5982 && eth_addr_is_broadcast(flow->dl_dst)
5983 && (flow->nw_proto == ARP_OP_REPLY
5984 || (flow->nw_proto == ARP_OP_REQUEST
5985 && flow->nw_src == flow->nw_dst)));
5989 update_learning_table(struct ofproto_dpif *ofproto,
5990 const struct flow *flow, int vlan,
5991 struct ofbundle *in_bundle)
5993 struct mac_entry *mac;
5995 /* Don't learn the OFPP_NONE port. */
5996 if (in_bundle == &ofpp_none_bundle) {
6000 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
6004 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
6005 if (is_gratuitous_arp(flow)) {
6006 /* We don't want to learn from gratuitous ARP packets that are
6007 * reflected back over bond slaves so we lock the learning table. */
6008 if (!in_bundle->bond) {
6009 mac_entry_set_grat_arp_lock(mac);
6010 } else if (mac_entry_is_grat_arp_locked(mac)) {
6015 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
6016 /* The log messages here could actually be useful in debugging,
6017 * so keep the rate limit relatively high. */
6018 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
6019 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
6020 "on port %s in VLAN %d",
6021 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
6022 in_bundle->name, vlan);
6024 mac->port.p = in_bundle;
6025 tag_set_add(&ofproto->revalidate_set,
6026 mac_learning_changed(ofproto->ml, mac));
6030 static struct ofbundle *
6031 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
6032 bool warn, struct ofport_dpif **in_ofportp)
6034 struct ofport_dpif *ofport;
6036 /* Find the port and bundle for the received packet. */
6037 ofport = get_ofp_port(ofproto, in_port);
6039 *in_ofportp = ofport;
6041 if (ofport && ofport->bundle) {
6042 return ofport->bundle;
6045 /* Special-case OFPP_NONE, which a controller may use as the ingress
6046 * port for traffic that it is sourcing. */
6047 if (in_port == OFPP_NONE) {
6048 return &ofpp_none_bundle;
6051 /* Odd. A few possible reasons here:
6053 * - We deleted a port but there are still a few packets queued up
6056 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
6057 * we don't know about.
6059 * - The ofproto client didn't configure the port as part of a bundle.
6062 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6064 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
6065 "port %"PRIu16, ofproto->up.name, in_port);
6070 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
6071 * dropped. Returns true if they may be forwarded, false if they should be
6074 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
6075 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
6077 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
6078 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
6079 * checked by input_vid_is_valid().
6081 * May also add tags to '*tags', although the current implementation only does
6082 * so in one special case.
6085 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
6086 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
6088 struct ofbundle *in_bundle = in_port->bundle;
6090 /* Drop frames for reserved multicast addresses
6091 * only if forward_bpdu option is absent. */
6092 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
6096 if (in_bundle->bond) {
6097 struct mac_entry *mac;
6099 switch (bond_check_admissibility(in_bundle->bond, in_port,
6100 flow->dl_dst, tags)) {
6107 case BV_DROP_IF_MOVED:
6108 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
6109 if (mac && mac->port.p != in_bundle &&
6110 (!is_gratuitous_arp(flow)
6111 || mac_entry_is_grat_arp_locked(mac))) {
6122 xlate_normal(struct action_xlate_ctx *ctx)
6124 struct ofport_dpif *in_port;
6125 struct ofbundle *in_bundle;
6126 struct mac_entry *mac;
6130 ctx->has_normal = true;
6132 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
6133 ctx->packet != NULL, &in_port);
6138 /* Drop malformed frames. */
6139 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
6140 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
6141 if (ctx->packet != NULL) {
6142 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6143 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
6144 "VLAN tag received on port %s",
6145 ctx->ofproto->up.name, in_bundle->name);
6150 /* Drop frames on bundles reserved for mirroring. */
6151 if (in_bundle->mirror_out) {
6152 if (ctx->packet != NULL) {
6153 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6154 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6155 "%s, which is reserved exclusively for mirroring",
6156 ctx->ofproto->up.name, in_bundle->name);
6162 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
6163 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6166 vlan = input_vid_to_vlan(in_bundle, vid);
6168 /* Check other admissibility requirements. */
6170 !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
6174 /* Learn source MAC. */
6175 if (ctx->may_learn) {
6176 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
6179 /* Determine output bundle. */
6180 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
6183 if (mac->port.p != in_bundle) {
6184 output_normal(ctx, mac->port.p, vlan);
6187 struct ofbundle *bundle;
6189 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
6190 if (bundle != in_bundle
6191 && ofbundle_includes_vlan(bundle, vlan)
6192 && bundle->floodable
6193 && !bundle->mirror_out) {
6194 output_normal(ctx, bundle, vlan);
6197 ctx->nf_output_iface = NF_OUT_FLOOD;
6201 /* Optimized flow revalidation.
6203 * It's a difficult problem, in general, to tell which facets need to have
6204 * their actions recalculated whenever the OpenFlow flow table changes. We
6205 * don't try to solve that general problem: for most kinds of OpenFlow flow
6206 * table changes, we recalculate the actions for every facet. This is
6207 * relatively expensive, but it's good enough if the OpenFlow flow table
6208 * doesn't change very often.
6210 * However, we can expect one particular kind of OpenFlow flow table change to
6211 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
6212 * of CPU on revalidating every facet whenever MAC learning modifies the flow
6213 * table, we add a special case that applies to flow tables in which every rule
6214 * has the same form (that is, the same wildcards), except that the table is
6215 * also allowed to have a single "catch-all" flow that matches all packets. We
6216 * optimize this case by tagging all of the facets that resubmit into the table
6217 * and invalidating the same tag whenever a flow changes in that table. The
6218 * end result is that we revalidate just the facets that need it (and sometimes
6219 * a few more, but not all of the facets or even all of the facets that
6220 * resubmit to the table modified by MAC learning). */
6222 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
6223 * into an OpenFlow table with the given 'basis'. */
6225 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
6228 if (flow_wildcards_is_catchall(wc)) {
6231 struct flow tag_flow = *flow;
6232 flow_zero_wildcards(&tag_flow, wc);
6233 return tag_create_deterministic(flow_hash(&tag_flow, secret));
6237 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
6238 * taggability of that table.
6240 * This function must be called after *each* change to a flow table. If you
6241 * skip calling it on some changes then the pointer comparisons at the end can
6242 * be invalid if you get unlucky. For example, if a flow removal causes a
6243 * cls_table to be destroyed and then a flow insertion causes a cls_table with
6244 * different wildcards to be created with the same address, then this function
6245 * will incorrectly skip revalidation. */
6247 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
6249 struct table_dpif *table = &ofproto->tables[table_id];
6250 const struct oftable *oftable = &ofproto->up.tables[table_id];
6251 struct cls_table *catchall, *other;
6252 struct cls_table *t;
6254 catchall = other = NULL;
6256 switch (hmap_count(&oftable->cls.tables)) {
6258 /* We could tag this OpenFlow table but it would make the logic a
6259 * little harder and it's a corner case that doesn't seem worth it
6265 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
6266 if (cls_table_is_catchall(t)) {
6268 } else if (!other) {
6271 /* Indicate that we can't tag this by setting both tables to
6272 * NULL. (We know that 'catchall' is already NULL.) */
6279 /* Can't tag this table. */
6283 if (table->catchall_table != catchall || table->other_table != other) {
6284 table->catchall_table = catchall;
6285 table->other_table = other;
6286 ofproto->need_revalidate = true;
6290 /* Given 'rule' that has changed in some way (either it is a rule being
6291 * inserted, a rule being deleted, or a rule whose actions are being
6292 * modified), marks facets for revalidation to ensure that packets will be
6293 * forwarded correctly according to the new state of the flow table.
6295 * This function must be called after *each* change to a flow table. See
6296 * the comment on table_update_taggable() for more information. */
6298 rule_invalidate(const struct rule_dpif *rule)
6300 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
6302 table_update_taggable(ofproto, rule->up.table_id);
6304 if (!ofproto->need_revalidate) {
6305 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
6307 if (table->other_table && rule->tag) {
6308 tag_set_add(&ofproto->revalidate_set, rule->tag);
6310 ofproto->need_revalidate = true;
6316 set_frag_handling(struct ofproto *ofproto_,
6317 enum ofp_config_flags frag_handling)
6319 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6321 if (frag_handling != OFPC_FRAG_REASM) {
6322 ofproto->need_revalidate = true;
6330 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
6331 const struct flow *flow,
6332 const union ofp_action *ofp_actions, size_t n_ofp_actions)
6334 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6337 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
6338 return OFPERR_NXBRC_BAD_IN_PORT;
6341 error = validate_actions(ofp_actions, n_ofp_actions, flow,
6342 ofproto->max_ports);
6344 struct odputil_keybuf keybuf;
6345 struct dpif_flow_stats stats;
6349 struct action_xlate_ctx ctx;
6350 uint64_t odp_actions_stub[1024 / 8];
6351 struct ofpbuf odp_actions;
6353 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
6354 odp_flow_key_from_flow(&key, flow);
6356 dpif_flow_stats_extract(flow, packet, &stats);
6358 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL,
6359 packet_get_tcp_flags(packet, flow), packet);
6360 ctx.resubmit_stats = &stats;
6362 ofpbuf_use_stub(&odp_actions,
6363 odp_actions_stub, sizeof odp_actions_stub);
6364 xlate_actions(&ctx, ofp_actions, n_ofp_actions, &odp_actions);
6365 dpif_execute(ofproto->dpif, key.data, key.size,
6366 odp_actions.data, odp_actions.size, packet);
6367 ofpbuf_uninit(&odp_actions);
6375 set_netflow(struct ofproto *ofproto_,
6376 const struct netflow_options *netflow_options)
6378 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6380 if (netflow_options) {
6381 if (!ofproto->netflow) {
6382 ofproto->netflow = netflow_create();
6384 return netflow_set_options(ofproto->netflow, netflow_options);
6386 netflow_destroy(ofproto->netflow);
6387 ofproto->netflow = NULL;
6393 get_netflow_ids(const struct ofproto *ofproto_,
6394 uint8_t *engine_type, uint8_t *engine_id)
6396 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6398 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
6402 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
6404 if (!facet_is_controller_flow(facet) &&
6405 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
6406 struct subfacet *subfacet;
6407 struct ofexpired expired;
6409 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
6410 if (subfacet->path == SF_FAST_PATH) {
6411 struct dpif_flow_stats stats;
6413 subfacet_reinstall(subfacet, &stats);
6414 subfacet_update_stats(subfacet, &stats);
6418 expired.flow = facet->flow;
6419 expired.packet_count = facet->packet_count;
6420 expired.byte_count = facet->byte_count;
6421 expired.used = facet->used;
6422 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
6427 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
6429 struct facet *facet;
6431 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6432 send_active_timeout(ofproto, facet);
6436 static struct ofproto_dpif *
6437 ofproto_dpif_lookup(const char *name)
6439 struct ofproto_dpif *ofproto;
6441 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
6442 hash_string(name, 0), &all_ofproto_dpifs) {
6443 if (!strcmp(ofproto->up.name, name)) {
6451 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
6452 const char *argv[], void *aux OVS_UNUSED)
6454 struct ofproto_dpif *ofproto;
6457 ofproto = ofproto_dpif_lookup(argv[1]);
6459 unixctl_command_reply_error(conn, "no such bridge");
6462 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6464 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6465 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6469 unixctl_command_reply(conn, "table successfully flushed");
6473 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6474 const char *argv[], void *aux OVS_UNUSED)
6476 struct ds ds = DS_EMPTY_INITIALIZER;
6477 const struct ofproto_dpif *ofproto;
6478 const struct mac_entry *e;
6480 ofproto = ofproto_dpif_lookup(argv[1]);
6482 unixctl_command_reply_error(conn, "no such bridge");
6486 ds_put_cstr(&ds, " port VLAN MAC Age\n");
6487 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
6488 struct ofbundle *bundle = e->port.p;
6489 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
6490 ofbundle_get_a_port(bundle)->odp_port,
6491 e->vlan, ETH_ADDR_ARGS(e->mac),
6492 mac_entry_age(ofproto->ml, e));
6494 unixctl_command_reply(conn, ds_cstr(&ds));
6499 struct action_xlate_ctx ctx;
6505 trace_format_rule(struct ds *result, uint8_t table_id, int level,
6506 const struct rule_dpif *rule)
6508 ds_put_char_multiple(result, '\t', level);
6510 ds_put_cstr(result, "No match\n");
6514 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
6515 table_id, ntohll(rule->up.flow_cookie));
6516 cls_rule_format(&rule->up.cr, result);
6517 ds_put_char(result, '\n');
6519 ds_put_char_multiple(result, '\t', level);
6520 ds_put_cstr(result, "OpenFlow ");
6521 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
6522 ds_put_char(result, '\n');
6526 trace_format_flow(struct ds *result, int level, const char *title,
6527 struct trace_ctx *trace)
6529 ds_put_char_multiple(result, '\t', level);
6530 ds_put_format(result, "%s: ", title);
6531 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
6532 ds_put_cstr(result, "unchanged");
6534 flow_format(result, &trace->ctx.flow);
6535 trace->flow = trace->ctx.flow;
6537 ds_put_char(result, '\n');
6541 trace_format_regs(struct ds *result, int level, const char *title,
6542 struct trace_ctx *trace)
6546 ds_put_char_multiple(result, '\t', level);
6547 ds_put_format(result, "%s:", title);
6548 for (i = 0; i < FLOW_N_REGS; i++) {
6549 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
6551 ds_put_char(result, '\n');
6555 trace_format_odp(struct ds *result, int level, const char *title,
6556 struct trace_ctx *trace)
6558 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
6560 ds_put_char_multiple(result, '\t', level);
6561 ds_put_format(result, "%s: ", title);
6562 format_odp_actions(result, odp_actions->data, odp_actions->size);
6563 ds_put_char(result, '\n');
6567 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
6569 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
6570 struct ds *result = trace->result;
6572 ds_put_char(result, '\n');
6573 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
6574 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
6575 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
6576 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
6580 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
6581 void *aux OVS_UNUSED)
6583 const char *dpname = argv[1];
6584 struct ofproto_dpif *ofproto;
6585 struct ofpbuf odp_key;
6586 struct ofpbuf *packet;
6587 ovs_be16 initial_tci;
6593 ofpbuf_init(&odp_key, 0);
6596 ofproto = ofproto_dpif_lookup(dpname);
6598 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
6602 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
6603 /* ofproto/trace dpname flow [-generate] */
6604 const char *flow_s = argv[2];
6605 const char *generate_s = argv[3];
6607 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
6608 * flow. We guess which type it is based on whether 'flow_s' contains
6609 * an '(', since a datapath flow always contains '(') but an
6610 * OpenFlow-like flow should not (in fact it's allowed but I believe
6611 * that's not documented anywhere).
6613 * An alternative would be to try to parse 'flow_s' both ways, but then
6614 * it would be tricky giving a sensible error message. After all, do
6615 * you just say "syntax error" or do you present both error messages?
6616 * Both choices seem lousy. */
6617 if (strchr(flow_s, '(')) {
6620 /* Convert string to datapath key. */
6621 ofpbuf_init(&odp_key, 0);
6622 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
6624 unixctl_command_reply_error(conn, "Bad flow syntax");
6628 /* Convert odp_key to flow. */
6629 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
6630 odp_key.size, &flow,
6631 &initial_tci, NULL);
6632 if (error == ODP_FIT_ERROR) {
6633 unixctl_command_reply_error(conn, "Invalid flow");
6639 error_s = parse_ofp_exact_flow(&flow, argv[2]);
6641 unixctl_command_reply_error(conn, error_s);
6646 initial_tci = flow.vlan_tci;
6647 vsp_adjust_flow(ofproto, &flow);
6650 /* Generate a packet, if requested. */
6652 packet = ofpbuf_new(0);
6653 flow_compose(packet, &flow);
6655 } else if (argc == 6) {
6656 /* ofproto/trace dpname priority tun_id in_port packet */
6657 const char *priority_s = argv[2];
6658 const char *tun_id_s = argv[3];
6659 const char *in_port_s = argv[4];
6660 const char *packet_s = argv[5];
6661 uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s));
6662 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
6663 uint32_t priority = atoi(priority_s);
6666 msg = eth_from_hex(packet_s, &packet);
6668 unixctl_command_reply_error(conn, msg);
6672 ds_put_cstr(&result, "Packet: ");
6673 s = ofp_packet_to_string(packet->data, packet->size);
6674 ds_put_cstr(&result, s);
6677 flow_extract(packet, priority, tun_id, in_port, &flow);
6678 initial_tci = flow.vlan_tci;
6680 unixctl_command_reply_error(conn, "Bad command syntax");
6684 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
6685 unixctl_command_reply(conn, ds_cstr(&result));
6688 ds_destroy(&result);
6689 ofpbuf_delete(packet);
6690 ofpbuf_uninit(&odp_key);
6694 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
6695 const struct ofpbuf *packet, ovs_be16 initial_tci,
6698 struct rule_dpif *rule;
6700 ds_put_cstr(ds, "Flow: ");
6701 flow_format(ds, flow);
6702 ds_put_char(ds, '\n');
6704 rule = rule_dpif_lookup(ofproto, flow);
6706 trace_format_rule(ds, 0, 0, rule);
6707 if (rule == ofproto->miss_rule) {
6708 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
6709 } else if (rule == ofproto->no_packet_in_rule) {
6710 ds_put_cstr(ds, "\nNo match, packets dropped because "
6711 "OFPPC_NO_PACKET_IN is set on in_port.\n");
6715 uint64_t odp_actions_stub[1024 / 8];
6716 struct ofpbuf odp_actions;
6718 struct trace_ctx trace;
6721 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
6724 ofpbuf_use_stub(&odp_actions,
6725 odp_actions_stub, sizeof odp_actions_stub);
6726 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
6727 rule, tcp_flags, packet);
6728 trace.ctx.resubmit_hook = trace_resubmit;
6729 xlate_actions(&trace.ctx, rule->up.actions, rule->up.n_actions,
6732 ds_put_char(ds, '\n');
6733 trace_format_flow(ds, 0, "Final flow", &trace);
6734 ds_put_cstr(ds, "Datapath actions: ");
6735 format_odp_actions(ds, odp_actions.data, odp_actions.size);
6736 ofpbuf_uninit(&odp_actions);
6738 if (trace.ctx.slow) {
6739 enum slow_path_reason slow;
6741 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
6742 "slow path because it:");
6743 for (slow = trace.ctx.slow; slow; ) {
6744 enum slow_path_reason bit = rightmost_1bit(slow);
6748 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
6751 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
6754 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
6757 ds_put_cstr(ds, "\n\t- Needs in-band special case "
6760 ds_put_cstr(ds, "\n\t (The datapath actions are "
6761 "incomplete--for complete actions, "
6762 "please supply a packet.)");
6765 case SLOW_CONTROLLER:
6766 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
6767 "to the OpenFlow controller.");
6770 ds_put_cstr(ds, "\n\t- Needs more specific matching "
6771 "than the datapath supports.");
6778 if (slow & ~SLOW_MATCH) {
6779 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
6780 "the special slow-path processing.");
6787 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6788 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6791 unixctl_command_reply(conn, NULL);
6795 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6796 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6799 unixctl_command_reply(conn, NULL);
6802 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
6803 * 'reply' describing the results. */
6805 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
6807 struct facet *facet;
6811 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6812 if (!facet_check_consistency(facet)) {
6817 ofproto->need_revalidate = true;
6821 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
6822 ofproto->up.name, errors);
6824 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
6829 ofproto_dpif_self_check(struct unixctl_conn *conn,
6830 int argc, const char *argv[], void *aux OVS_UNUSED)
6832 struct ds reply = DS_EMPTY_INITIALIZER;
6833 struct ofproto_dpif *ofproto;
6836 ofproto = ofproto_dpif_lookup(argv[1]);
6838 unixctl_command_reply_error(conn, "Unknown ofproto (use "
6839 "ofproto/list for help)");
6842 ofproto_dpif_self_check__(ofproto, &reply);
6844 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6845 ofproto_dpif_self_check__(ofproto, &reply);
6849 unixctl_command_reply(conn, ds_cstr(&reply));
6854 ofproto_dpif_unixctl_init(void)
6856 static bool registered;
6862 unixctl_command_register(
6864 "bridge {tun_id in_port packet | odp_flow [-generate]}",
6865 2, 5, ofproto_unixctl_trace, NULL);
6866 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6867 ofproto_unixctl_fdb_flush, NULL);
6868 unixctl_command_register("fdb/show", "bridge", 1, 1,
6869 ofproto_unixctl_fdb_show, NULL);
6870 unixctl_command_register("ofproto/clog", "", 0, 0,
6871 ofproto_dpif_clog, NULL);
6872 unixctl_command_register("ofproto/unclog", "", 0, 0,
6873 ofproto_dpif_unclog, NULL);
6874 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6875 ofproto_dpif_self_check, NULL);
6878 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6880 * This is deprecated. It is only for compatibility with broken device drivers
6881 * in old versions of Linux that do not properly support VLANs when VLAN
6882 * devices are not used. When broken device drivers are no longer in
6883 * widespread use, we will delete these interfaces. */
6886 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
6888 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6889 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6891 if (realdev_ofp_port == ofport->realdev_ofp_port
6892 && vid == ofport->vlandev_vid) {
6896 ofproto->need_revalidate = true;
6898 if (ofport->realdev_ofp_port) {
6901 if (realdev_ofp_port && ofport->bundle) {
6902 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6903 * themselves be part of a bundle. */
6904 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6907 ofport->realdev_ofp_port = realdev_ofp_port;
6908 ofport->vlandev_vid = vid;
6910 if (realdev_ofp_port) {
6911 vsp_add(ofport, realdev_ofp_port, vid);
6918 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
6920 return hash_2words(realdev_ofp_port, vid);
6923 /* Returns the ODP port number of the Linux VLAN device that corresponds to
6924 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
6925 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
6926 * it would return the port number of eth0.9.
6928 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
6929 * function just returns its 'realdev_odp_port' argument. */
6931 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6932 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
6934 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6935 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
6936 int vid = vlan_tci_to_vid(vlan_tci);
6937 const struct vlan_splinter *vsp;
6939 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6940 hash_realdev_vid(realdev_ofp_port, vid),
6941 &ofproto->realdev_vid_map) {
6942 if (vsp->realdev_ofp_port == realdev_ofp_port
6943 && vsp->vid == vid) {
6944 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
6948 return realdev_odp_port;
6951 static struct vlan_splinter *
6952 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
6954 struct vlan_splinter *vsp;
6956 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
6957 &ofproto->vlandev_map) {
6958 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6966 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6967 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6968 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6969 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6970 * eth0 and store 9 in '*vid'.
6972 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6973 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6976 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6977 uint16_t vlandev_ofp_port, int *vid)
6979 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6980 const struct vlan_splinter *vsp;
6982 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6987 return vsp->realdev_ofp_port;
6993 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6994 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
6995 * 'flow->in_port' to the "real" device backing the VLAN device, sets
6996 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
6997 * always the case unless VLAN splinters are enabled), returns false without
6998 * making any changes. */
7000 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
7005 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
7010 /* Cause the flow to be processed as if it came in on the real device with
7011 * the VLAN device's VLAN ID. */
7012 flow->in_port = realdev;
7013 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
7018 vsp_remove(struct ofport_dpif *port)
7020 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7021 struct vlan_splinter *vsp;
7023 vsp = vlandev_find(ofproto, port->up.ofp_port);
7025 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
7026 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
7029 port->realdev_ofp_port = 0;
7031 VLOG_ERR("missing vlan device record");
7036 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
7038 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7040 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
7041 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
7042 == realdev_ofp_port)) {
7043 struct vlan_splinter *vsp;
7045 vsp = xmalloc(sizeof *vsp);
7046 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
7047 hash_int(port->up.ofp_port, 0));
7048 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
7049 hash_realdev_vid(realdev_ofp_port, vid));
7050 vsp->realdev_ofp_port = realdev_ofp_port;
7051 vsp->vlandev_ofp_port = port->up.ofp_port;
7054 port->realdev_ofp_port = realdev_ofp_port;
7056 VLOG_ERR("duplicate vlan device record");
7060 const struct ofproto_class ofproto_dpif_class = {
7090 port_is_lacp_current,
7091 NULL, /* rule_choose_table */
7098 rule_modify_actions,
7106 get_cfm_remote_mpids,
7111 get_stp_port_status,
7118 is_mirror_output_bundle,
7119 forward_bpdu_changed,