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.cookie = htonll(0);
795 fm.cookie_mask = htonll(0);
796 fm.table_id = TBL_INTERNAL;
797 fm.command = OFPFC_ADD;
803 fm.actions = actions->data;
804 fm.n_actions = actions->size / sizeof(union ofp_action);
806 error = ofproto_flow_mod(&ofproto->up, &fm);
808 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
809 id, ofperr_to_string(error));
813 *rulep = rule_dpif_lookup__(ofproto, &fm.cr.flow, TBL_INTERNAL);
814 assert(*rulep != NULL);
820 add_internal_flows(struct ofproto_dpif *ofproto)
822 struct nx_action_controller *nac;
823 uint64_t actions_stub[128 / 8];
824 struct ofpbuf actions;
828 ofpbuf_use_stack(&actions, actions_stub, sizeof actions_stub);
831 nac = ofputil_put_NXAST_CONTROLLER(&actions);
832 nac->max_len = htons(UINT16_MAX);
833 nac->controller_id = htons(0);
834 nac->reason = OFPR_NO_MATCH;
835 error = add_internal_flow(ofproto, id++, &actions, &ofproto->miss_rule);
840 ofpbuf_clear(&actions);
841 error = add_internal_flow(ofproto, id++, &actions,
842 &ofproto->no_packet_in_rule);
847 complete_operations(struct ofproto_dpif *ofproto)
849 struct dpif_completion *c, *next;
851 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
852 ofoperation_complete(c->op, 0);
853 list_remove(&c->list_node);
859 destruct(struct ofproto *ofproto_)
861 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
862 struct rule_dpif *rule, *next_rule;
863 struct oftable *table;
866 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
867 complete_operations(ofproto);
869 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
870 struct cls_cursor cursor;
872 cls_cursor_init(&cursor, &table->cls, NULL);
873 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
874 ofproto_rule_destroy(&rule->up);
878 for (i = 0; i < MAX_MIRRORS; i++) {
879 mirror_destroy(ofproto->mirrors[i]);
882 netflow_destroy(ofproto->netflow);
883 dpif_sflow_destroy(ofproto->sflow);
884 hmap_destroy(&ofproto->bundles);
885 mac_learning_destroy(ofproto->ml);
887 hmap_destroy(&ofproto->facets);
888 hmap_destroy(&ofproto->subfacets);
889 governor_destroy(ofproto->governor);
891 hmap_destroy(&ofproto->vlandev_map);
892 hmap_destroy(&ofproto->realdev_vid_map);
894 dpif_close(ofproto->dpif);
898 run_fast(struct ofproto *ofproto_)
900 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
903 /* Handle one or more batches of upcalls, until there's nothing left to do
904 * or until we do a fixed total amount of work.
906 * We do work in batches because it can be much cheaper to set up a number
907 * of flows and fire off their patches all at once. We do multiple batches
908 * because in some cases handling a packet can cause another packet to be
909 * queued almost immediately as part of the return flow. Both
910 * optimizations can make major improvements on some benchmarks and
911 * presumably for real traffic as well. */
913 while (work < FLOW_MISS_MAX_BATCH) {
914 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
924 run(struct ofproto *ofproto_)
926 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
927 struct ofport_dpif *ofport;
928 struct ofbundle *bundle;
932 complete_operations(ofproto);
934 dpif_run(ofproto->dpif);
936 error = run_fast(ofproto_);
941 if (timer_expired(&ofproto->next_expiration)) {
942 int delay = expire(ofproto);
943 timer_set_duration(&ofproto->next_expiration, delay);
946 if (ofproto->netflow) {
947 if (netflow_run(ofproto->netflow)) {
948 send_netflow_active_timeouts(ofproto);
951 if (ofproto->sflow) {
952 dpif_sflow_run(ofproto->sflow);
955 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
958 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
963 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
965 /* Now revalidate if there's anything to do. */
966 if (ofproto->need_revalidate
967 || !tag_set_is_empty(&ofproto->revalidate_set)) {
968 struct tag_set revalidate_set = ofproto->revalidate_set;
969 bool revalidate_all = ofproto->need_revalidate;
972 /* Clear the revalidation flags. */
973 tag_set_init(&ofproto->revalidate_set);
974 ofproto->need_revalidate = false;
976 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
978 || tag_set_intersects(&revalidate_set, facet->tags)) {
979 facet_revalidate(facet);
984 /* Check the consistency of a random facet, to aid debugging. */
985 if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) {
988 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
989 struct facet, hmap_node);
990 if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) {
991 if (!facet_check_consistency(facet)) {
992 ofproto->need_revalidate = true;
997 if (ofproto->governor) {
1000 governor_run(ofproto->governor);
1002 /* If the governor has shrunk to its minimum size and the number of
1003 * subfacets has dwindled, then drop the governor entirely.
1005 * For hysteresis, the number of subfacets to drop the governor is
1006 * smaller than the number needed to trigger its creation. */
1007 n_subfacets = hmap_count(&ofproto->subfacets);
1008 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1009 && governor_is_idle(ofproto->governor)) {
1010 governor_destroy(ofproto->governor);
1011 ofproto->governor = NULL;
1019 wait(struct ofproto *ofproto_)
1021 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1022 struct ofport_dpif *ofport;
1023 struct ofbundle *bundle;
1025 if (!clogged && !list_is_empty(&ofproto->completions)) {
1026 poll_immediate_wake();
1029 dpif_wait(ofproto->dpif);
1030 dpif_recv_wait(ofproto->dpif);
1031 if (ofproto->sflow) {
1032 dpif_sflow_wait(ofproto->sflow);
1034 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
1035 poll_immediate_wake();
1037 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1040 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1041 bundle_wait(bundle);
1043 if (ofproto->netflow) {
1044 netflow_wait(ofproto->netflow);
1046 mac_learning_wait(ofproto->ml);
1048 if (ofproto->need_revalidate) {
1049 /* Shouldn't happen, but if it does just go around again. */
1050 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1051 poll_immediate_wake();
1053 timer_wait(&ofproto->next_expiration);
1055 if (ofproto->governor) {
1056 governor_wait(ofproto->governor);
1061 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1063 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1065 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1066 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1070 flush(struct ofproto *ofproto_)
1072 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1073 struct facet *facet, *next_facet;
1075 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1076 /* Mark the facet as not installed so that facet_remove() doesn't
1077 * bother trying to uninstall it. There is no point in uninstalling it
1078 * individually since we are about to blow away all the facets with
1079 * dpif_flow_flush(). */
1080 struct subfacet *subfacet;
1082 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
1083 subfacet->path = SF_NOT_INSTALLED;
1084 subfacet->dp_packet_count = 0;
1085 subfacet->dp_byte_count = 0;
1087 facet_remove(facet);
1089 dpif_flow_flush(ofproto->dpif);
1093 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1094 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1096 *arp_match_ip = true;
1097 *actions = (OFPUTIL_A_OUTPUT |
1098 OFPUTIL_A_SET_VLAN_VID |
1099 OFPUTIL_A_SET_VLAN_PCP |
1100 OFPUTIL_A_STRIP_VLAN |
1101 OFPUTIL_A_SET_DL_SRC |
1102 OFPUTIL_A_SET_DL_DST |
1103 OFPUTIL_A_SET_NW_SRC |
1104 OFPUTIL_A_SET_NW_DST |
1105 OFPUTIL_A_SET_NW_TOS |
1106 OFPUTIL_A_SET_TP_SRC |
1107 OFPUTIL_A_SET_TP_DST |
1112 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
1114 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1115 struct dpif_dp_stats s;
1117 strcpy(ots->name, "classifier");
1119 dpif_get_dp_stats(ofproto->dpif, &s);
1120 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
1121 put_32aligned_be64(&ots->matched_count,
1122 htonll(s.n_hit + ofproto->n_matches));
1125 static struct ofport *
1128 struct ofport_dpif *port = xmalloc(sizeof *port);
1133 port_dealloc(struct ofport *port_)
1135 struct ofport_dpif *port = ofport_dpif_cast(port_);
1140 port_construct(struct ofport *port_)
1142 struct ofport_dpif *port = ofport_dpif_cast(port_);
1143 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1145 ofproto->need_revalidate = true;
1146 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
1147 port->bundle = NULL;
1149 port->tag = tag_create_random();
1150 port->may_enable = true;
1151 port->stp_port = NULL;
1152 port->stp_state = STP_DISABLED;
1153 hmap_init(&port->priorities);
1154 port->realdev_ofp_port = 0;
1155 port->vlandev_vid = 0;
1156 port->carrier_seq = netdev_get_carrier_resets(port->up.netdev);
1158 if (ofproto->sflow) {
1159 dpif_sflow_add_port(ofproto->sflow, port_);
1166 port_destruct(struct ofport *port_)
1168 struct ofport_dpif *port = ofport_dpif_cast(port_);
1169 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1171 ofproto->need_revalidate = true;
1172 bundle_remove(port_);
1173 set_cfm(port_, NULL);
1174 if (ofproto->sflow) {
1175 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1178 ofport_clear_priorities(port);
1179 hmap_destroy(&port->priorities);
1183 port_modified(struct ofport *port_)
1185 struct ofport_dpif *port = ofport_dpif_cast(port_);
1187 if (port->bundle && port->bundle->bond) {
1188 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1193 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1195 struct ofport_dpif *port = ofport_dpif_cast(port_);
1196 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1197 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1199 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1200 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1201 OFPUTIL_PC_NO_PACKET_IN)) {
1202 ofproto->need_revalidate = true;
1204 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1205 bundle_update(port->bundle);
1211 set_sflow(struct ofproto *ofproto_,
1212 const struct ofproto_sflow_options *sflow_options)
1214 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1215 struct dpif_sflow *ds = ofproto->sflow;
1217 if (sflow_options) {
1219 struct ofport_dpif *ofport;
1221 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1222 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1223 dpif_sflow_add_port(ds, &ofport->up);
1225 ofproto->need_revalidate = true;
1227 dpif_sflow_set_options(ds, sflow_options);
1230 dpif_sflow_destroy(ds);
1231 ofproto->need_revalidate = true;
1232 ofproto->sflow = NULL;
1239 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1241 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1248 struct ofproto_dpif *ofproto;
1250 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1251 ofproto->need_revalidate = true;
1252 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1255 if (cfm_configure(ofport->cfm, s)) {
1261 cfm_destroy(ofport->cfm);
1267 get_cfm_fault(const struct ofport *ofport_)
1269 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1271 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1275 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1278 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1281 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1289 get_cfm_health(const struct ofport *ofport_)
1291 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1293 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1296 /* Spanning Tree. */
1299 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1301 struct ofproto_dpif *ofproto = ofproto_;
1302 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1303 struct ofport_dpif *ofport;
1305 ofport = stp_port_get_aux(sp);
1307 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1308 ofproto->up.name, port_num);
1310 struct eth_header *eth = pkt->l2;
1312 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1313 if (eth_addr_is_zero(eth->eth_src)) {
1314 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1315 "with unknown MAC", ofproto->up.name, port_num);
1317 send_packet(ofport, pkt);
1323 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1325 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1327 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1329 /* Only revalidate flows if the configuration changed. */
1330 if (!s != !ofproto->stp) {
1331 ofproto->need_revalidate = true;
1335 if (!ofproto->stp) {
1336 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1337 send_bpdu_cb, ofproto);
1338 ofproto->stp_last_tick = time_msec();
1341 stp_set_bridge_id(ofproto->stp, s->system_id);
1342 stp_set_bridge_priority(ofproto->stp, s->priority);
1343 stp_set_hello_time(ofproto->stp, s->hello_time);
1344 stp_set_max_age(ofproto->stp, s->max_age);
1345 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1347 struct ofport *ofport;
1349 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1350 set_stp_port(ofport, NULL);
1353 stp_destroy(ofproto->stp);
1354 ofproto->stp = NULL;
1361 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1363 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1367 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1368 s->designated_root = stp_get_designated_root(ofproto->stp);
1369 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1378 update_stp_port_state(struct ofport_dpif *ofport)
1380 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1381 enum stp_state state;
1383 /* Figure out new state. */
1384 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1388 if (ofport->stp_state != state) {
1389 enum ofputil_port_state of_state;
1392 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1393 netdev_get_name(ofport->up.netdev),
1394 stp_state_name(ofport->stp_state),
1395 stp_state_name(state));
1396 if (stp_learn_in_state(ofport->stp_state)
1397 != stp_learn_in_state(state)) {
1398 /* xxx Learning action flows should also be flushed. */
1399 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1401 fwd_change = stp_forward_in_state(ofport->stp_state)
1402 != stp_forward_in_state(state);
1404 ofproto->need_revalidate = true;
1405 ofport->stp_state = state;
1406 ofport->stp_state_entered = time_msec();
1408 if (fwd_change && ofport->bundle) {
1409 bundle_update(ofport->bundle);
1412 /* Update the STP state bits in the OpenFlow port description. */
1413 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1414 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1415 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1416 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1417 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1419 ofproto_port_set_state(&ofport->up, of_state);
1423 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1424 * caller is responsible for assigning STP port numbers and ensuring
1425 * there are no duplicates. */
1427 set_stp_port(struct ofport *ofport_,
1428 const struct ofproto_port_stp_settings *s)
1430 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1431 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1432 struct stp_port *sp = ofport->stp_port;
1434 if (!s || !s->enable) {
1436 ofport->stp_port = NULL;
1437 stp_port_disable(sp);
1438 update_stp_port_state(ofport);
1441 } else if (sp && stp_port_no(sp) != s->port_num
1442 && ofport == stp_port_get_aux(sp)) {
1443 /* The port-id changed, so disable the old one if it's not
1444 * already in use by another port. */
1445 stp_port_disable(sp);
1448 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1449 stp_port_enable(sp);
1451 stp_port_set_aux(sp, ofport);
1452 stp_port_set_priority(sp, s->priority);
1453 stp_port_set_path_cost(sp, s->path_cost);
1455 update_stp_port_state(ofport);
1461 get_stp_port_status(struct ofport *ofport_,
1462 struct ofproto_port_stp_status *s)
1464 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1465 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1466 struct stp_port *sp = ofport->stp_port;
1468 if (!ofproto->stp || !sp) {
1474 s->port_id = stp_port_get_id(sp);
1475 s->state = stp_port_get_state(sp);
1476 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1477 s->role = stp_port_get_role(sp);
1478 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1484 stp_run(struct ofproto_dpif *ofproto)
1487 long long int now = time_msec();
1488 long long int elapsed = now - ofproto->stp_last_tick;
1489 struct stp_port *sp;
1492 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1493 ofproto->stp_last_tick = now;
1495 while (stp_get_changed_port(ofproto->stp, &sp)) {
1496 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1499 update_stp_port_state(ofport);
1503 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1504 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1510 stp_wait(struct ofproto_dpif *ofproto)
1513 poll_timer_wait(1000);
1517 /* Returns true if STP should process 'flow'. */
1519 stp_should_process_flow(const struct flow *flow)
1521 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1525 stp_process_packet(const struct ofport_dpif *ofport,
1526 const struct ofpbuf *packet)
1528 struct ofpbuf payload = *packet;
1529 struct eth_header *eth = payload.data;
1530 struct stp_port *sp = ofport->stp_port;
1532 /* Sink packets on ports that have STP disabled when the bridge has
1534 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1538 /* Trim off padding on payload. */
1539 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1540 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1543 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1544 stp_received_bpdu(sp, payload.data, payload.size);
1548 static struct priority_to_dscp *
1549 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1551 struct priority_to_dscp *pdscp;
1554 hash = hash_int(priority, 0);
1555 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1556 if (pdscp->priority == priority) {
1564 ofport_clear_priorities(struct ofport_dpif *ofport)
1566 struct priority_to_dscp *pdscp, *next;
1568 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1569 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1575 set_queues(struct ofport *ofport_,
1576 const struct ofproto_port_queue *qdscp_list,
1579 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1580 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1581 struct hmap new = HMAP_INITIALIZER(&new);
1584 for (i = 0; i < n_qdscp; i++) {
1585 struct priority_to_dscp *pdscp;
1589 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1590 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1595 pdscp = get_priority(ofport, priority);
1597 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1599 pdscp = xmalloc(sizeof *pdscp);
1600 pdscp->priority = priority;
1602 ofproto->need_revalidate = true;
1605 if (pdscp->dscp != dscp) {
1607 ofproto->need_revalidate = true;
1610 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1613 if (!hmap_is_empty(&ofport->priorities)) {
1614 ofport_clear_priorities(ofport);
1615 ofproto->need_revalidate = true;
1618 hmap_swap(&new, &ofport->priorities);
1626 /* Expires all MAC learning entries associated with 'bundle' and forces its
1627 * ofproto to revalidate every flow.
1629 * Normally MAC learning entries are removed only from the ofproto associated
1630 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1631 * are removed from every ofproto. When patch ports and SLB bonds are in use
1632 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1633 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1634 * with the host from which it migrated. */
1636 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1638 struct ofproto_dpif *ofproto = bundle->ofproto;
1639 struct mac_learning *ml = ofproto->ml;
1640 struct mac_entry *mac, *next_mac;
1642 ofproto->need_revalidate = true;
1643 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1644 if (mac->port.p == bundle) {
1646 struct ofproto_dpif *o;
1648 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1650 struct mac_entry *e;
1652 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1655 tag_set_add(&o->revalidate_set, e->tag);
1656 mac_learning_expire(o->ml, e);
1662 mac_learning_expire(ml, mac);
1667 static struct ofbundle *
1668 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1670 struct ofbundle *bundle;
1672 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1673 &ofproto->bundles) {
1674 if (bundle->aux == aux) {
1681 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1682 * ones that are found to 'bundles'. */
1684 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1685 void **auxes, size_t n_auxes,
1686 struct hmapx *bundles)
1690 hmapx_init(bundles);
1691 for (i = 0; i < n_auxes; i++) {
1692 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1694 hmapx_add(bundles, bundle);
1700 bundle_update(struct ofbundle *bundle)
1702 struct ofport_dpif *port;
1704 bundle->floodable = true;
1705 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1706 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1707 || !stp_forward_in_state(port->stp_state)) {
1708 bundle->floodable = false;
1715 bundle_del_port(struct ofport_dpif *port)
1717 struct ofbundle *bundle = port->bundle;
1719 bundle->ofproto->need_revalidate = true;
1721 list_remove(&port->bundle_node);
1722 port->bundle = NULL;
1725 lacp_slave_unregister(bundle->lacp, port);
1728 bond_slave_unregister(bundle->bond, port);
1731 bundle_update(bundle);
1735 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1736 struct lacp_slave_settings *lacp,
1737 uint32_t bond_stable_id)
1739 struct ofport_dpif *port;
1741 port = get_ofp_port(bundle->ofproto, ofp_port);
1746 if (port->bundle != bundle) {
1747 bundle->ofproto->need_revalidate = true;
1749 bundle_del_port(port);
1752 port->bundle = bundle;
1753 list_push_back(&bundle->ports, &port->bundle_node);
1754 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1755 || !stp_forward_in_state(port->stp_state)) {
1756 bundle->floodable = false;
1760 port->bundle->ofproto->need_revalidate = true;
1761 lacp_slave_register(bundle->lacp, port, lacp);
1764 port->bond_stable_id = bond_stable_id;
1770 bundle_destroy(struct ofbundle *bundle)
1772 struct ofproto_dpif *ofproto;
1773 struct ofport_dpif *port, *next_port;
1780 ofproto = bundle->ofproto;
1781 for (i = 0; i < MAX_MIRRORS; i++) {
1782 struct ofmirror *m = ofproto->mirrors[i];
1784 if (m->out == bundle) {
1786 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1787 || hmapx_find_and_delete(&m->dsts, bundle)) {
1788 ofproto->need_revalidate = true;
1793 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1794 bundle_del_port(port);
1797 bundle_flush_macs(bundle, true);
1798 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1800 free(bundle->trunks);
1801 lacp_destroy(bundle->lacp);
1802 bond_destroy(bundle->bond);
1807 bundle_set(struct ofproto *ofproto_, void *aux,
1808 const struct ofproto_bundle_settings *s)
1810 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1811 bool need_flush = false;
1812 struct ofport_dpif *port;
1813 struct ofbundle *bundle;
1814 unsigned long *trunks;
1820 bundle_destroy(bundle_lookup(ofproto, aux));
1824 assert(s->n_slaves == 1 || s->bond != NULL);
1825 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1827 bundle = bundle_lookup(ofproto, aux);
1829 bundle = xmalloc(sizeof *bundle);
1831 bundle->ofproto = ofproto;
1832 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1833 hash_pointer(aux, 0));
1835 bundle->name = NULL;
1837 list_init(&bundle->ports);
1838 bundle->vlan_mode = PORT_VLAN_TRUNK;
1840 bundle->trunks = NULL;
1841 bundle->use_priority_tags = s->use_priority_tags;
1842 bundle->lacp = NULL;
1843 bundle->bond = NULL;
1845 bundle->floodable = true;
1847 bundle->src_mirrors = 0;
1848 bundle->dst_mirrors = 0;
1849 bundle->mirror_out = 0;
1852 if (!bundle->name || strcmp(s->name, bundle->name)) {
1854 bundle->name = xstrdup(s->name);
1859 if (!bundle->lacp) {
1860 ofproto->need_revalidate = true;
1861 bundle->lacp = lacp_create();
1863 lacp_configure(bundle->lacp, s->lacp);
1865 lacp_destroy(bundle->lacp);
1866 bundle->lacp = NULL;
1869 /* Update set of ports. */
1871 for (i = 0; i < s->n_slaves; i++) {
1872 if (!bundle_add_port(bundle, s->slaves[i],
1873 s->lacp ? &s->lacp_slaves[i] : NULL,
1874 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1878 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1879 struct ofport_dpif *next_port;
1881 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1882 for (i = 0; i < s->n_slaves; i++) {
1883 if (s->slaves[i] == port->up.ofp_port) {
1888 bundle_del_port(port);
1892 assert(list_size(&bundle->ports) <= s->n_slaves);
1894 if (list_is_empty(&bundle->ports)) {
1895 bundle_destroy(bundle);
1899 /* Set VLAN tagging mode */
1900 if (s->vlan_mode != bundle->vlan_mode
1901 || s->use_priority_tags != bundle->use_priority_tags) {
1902 bundle->vlan_mode = s->vlan_mode;
1903 bundle->use_priority_tags = s->use_priority_tags;
1908 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1909 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1911 if (vlan != bundle->vlan) {
1912 bundle->vlan = vlan;
1916 /* Get trunked VLANs. */
1917 switch (s->vlan_mode) {
1918 case PORT_VLAN_ACCESS:
1922 case PORT_VLAN_TRUNK:
1923 trunks = (unsigned long *) s->trunks;
1926 case PORT_VLAN_NATIVE_UNTAGGED:
1927 case PORT_VLAN_NATIVE_TAGGED:
1928 if (vlan != 0 && (!s->trunks
1929 || !bitmap_is_set(s->trunks, vlan)
1930 || bitmap_is_set(s->trunks, 0))) {
1931 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1933 trunks = bitmap_clone(s->trunks, 4096);
1935 trunks = bitmap_allocate1(4096);
1937 bitmap_set1(trunks, vlan);
1938 bitmap_set0(trunks, 0);
1940 trunks = (unsigned long *) s->trunks;
1947 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1948 free(bundle->trunks);
1949 if (trunks == s->trunks) {
1950 bundle->trunks = vlan_bitmap_clone(trunks);
1952 bundle->trunks = trunks;
1957 if (trunks != s->trunks) {
1962 if (!list_is_short(&bundle->ports)) {
1963 bundle->ofproto->has_bonded_bundles = true;
1965 if (bond_reconfigure(bundle->bond, s->bond)) {
1966 ofproto->need_revalidate = true;
1969 bundle->bond = bond_create(s->bond);
1970 ofproto->need_revalidate = true;
1973 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1974 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1978 bond_destroy(bundle->bond);
1979 bundle->bond = NULL;
1982 /* If we changed something that would affect MAC learning, un-learn
1983 * everything on this port and force flow revalidation. */
1985 bundle_flush_macs(bundle, false);
1992 bundle_remove(struct ofport *port_)
1994 struct ofport_dpif *port = ofport_dpif_cast(port_);
1995 struct ofbundle *bundle = port->bundle;
1998 bundle_del_port(port);
1999 if (list_is_empty(&bundle->ports)) {
2000 bundle_destroy(bundle);
2001 } else if (list_is_short(&bundle->ports)) {
2002 bond_destroy(bundle->bond);
2003 bundle->bond = NULL;
2009 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2011 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2012 struct ofport_dpif *port = port_;
2013 uint8_t ea[ETH_ADDR_LEN];
2016 error = netdev_get_etheraddr(port->up.netdev, ea);
2018 struct ofpbuf packet;
2021 ofpbuf_init(&packet, 0);
2022 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2024 memcpy(packet_pdu, pdu, pdu_size);
2026 send_packet(port, &packet);
2027 ofpbuf_uninit(&packet);
2029 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2030 "%s (%s)", port->bundle->name,
2031 netdev_get_name(port->up.netdev), strerror(error));
2036 bundle_send_learning_packets(struct ofbundle *bundle)
2038 struct ofproto_dpif *ofproto = bundle->ofproto;
2039 int error, n_packets, n_errors;
2040 struct mac_entry *e;
2042 error = n_packets = n_errors = 0;
2043 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2044 if (e->port.p != bundle) {
2045 struct ofpbuf *learning_packet;
2046 struct ofport_dpif *port;
2050 /* The assignment to "port" is unnecessary but makes "grep"ing for
2051 * struct ofport_dpif more effective. */
2052 learning_packet = bond_compose_learning_packet(bundle->bond,
2056 ret = send_packet(port, learning_packet);
2057 ofpbuf_delete(learning_packet);
2067 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2068 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2069 "packets, last error was: %s",
2070 bundle->name, n_errors, n_packets, strerror(error));
2072 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2073 bundle->name, n_packets);
2078 bundle_run(struct ofbundle *bundle)
2081 lacp_run(bundle->lacp, send_pdu_cb);
2084 struct ofport_dpif *port;
2086 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2087 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2090 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
2091 lacp_status(bundle->lacp));
2092 if (bond_should_send_learning_packets(bundle->bond)) {
2093 bundle_send_learning_packets(bundle);
2099 bundle_wait(struct ofbundle *bundle)
2102 lacp_wait(bundle->lacp);
2105 bond_wait(bundle->bond);
2112 mirror_scan(struct ofproto_dpif *ofproto)
2116 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2117 if (!ofproto->mirrors[idx]) {
2124 static struct ofmirror *
2125 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2129 for (i = 0; i < MAX_MIRRORS; i++) {
2130 struct ofmirror *mirror = ofproto->mirrors[i];
2131 if (mirror && mirror->aux == aux) {
2139 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2141 mirror_update_dups(struct ofproto_dpif *ofproto)
2145 for (i = 0; i < MAX_MIRRORS; i++) {
2146 struct ofmirror *m = ofproto->mirrors[i];
2149 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2153 for (i = 0; i < MAX_MIRRORS; i++) {
2154 struct ofmirror *m1 = ofproto->mirrors[i];
2161 for (j = i + 1; j < MAX_MIRRORS; j++) {
2162 struct ofmirror *m2 = ofproto->mirrors[j];
2164 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2165 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2166 m2->dup_mirrors |= m1->dup_mirrors;
2173 mirror_set(struct ofproto *ofproto_, void *aux,
2174 const struct ofproto_mirror_settings *s)
2176 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2177 mirror_mask_t mirror_bit;
2178 struct ofbundle *bundle;
2179 struct ofmirror *mirror;
2180 struct ofbundle *out;
2181 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2182 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2185 mirror = mirror_lookup(ofproto, aux);
2187 mirror_destroy(mirror);
2193 idx = mirror_scan(ofproto);
2195 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2197 ofproto->up.name, MAX_MIRRORS, s->name);
2201 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2202 mirror->ofproto = ofproto;
2205 mirror->out_vlan = -1;
2206 mirror->name = NULL;
2209 if (!mirror->name || strcmp(s->name, mirror->name)) {
2211 mirror->name = xstrdup(s->name);
2214 /* Get the new configuration. */
2215 if (s->out_bundle) {
2216 out = bundle_lookup(ofproto, s->out_bundle);
2218 mirror_destroy(mirror);
2224 out_vlan = s->out_vlan;
2226 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2227 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2229 /* If the configuration has not changed, do nothing. */
2230 if (hmapx_equals(&srcs, &mirror->srcs)
2231 && hmapx_equals(&dsts, &mirror->dsts)
2232 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2233 && mirror->out == out
2234 && mirror->out_vlan == out_vlan)
2236 hmapx_destroy(&srcs);
2237 hmapx_destroy(&dsts);
2241 hmapx_swap(&srcs, &mirror->srcs);
2242 hmapx_destroy(&srcs);
2244 hmapx_swap(&dsts, &mirror->dsts);
2245 hmapx_destroy(&dsts);
2247 free(mirror->vlans);
2248 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2251 mirror->out_vlan = out_vlan;
2253 /* Update bundles. */
2254 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2255 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2256 if (hmapx_contains(&mirror->srcs, bundle)) {
2257 bundle->src_mirrors |= mirror_bit;
2259 bundle->src_mirrors &= ~mirror_bit;
2262 if (hmapx_contains(&mirror->dsts, bundle)) {
2263 bundle->dst_mirrors |= mirror_bit;
2265 bundle->dst_mirrors &= ~mirror_bit;
2268 if (mirror->out == bundle) {
2269 bundle->mirror_out |= mirror_bit;
2271 bundle->mirror_out &= ~mirror_bit;
2275 ofproto->need_revalidate = true;
2276 ofproto->has_mirrors = true;
2277 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2278 mirror_update_dups(ofproto);
2284 mirror_destroy(struct ofmirror *mirror)
2286 struct ofproto_dpif *ofproto;
2287 mirror_mask_t mirror_bit;
2288 struct ofbundle *bundle;
2295 ofproto = mirror->ofproto;
2296 ofproto->need_revalidate = true;
2297 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2299 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2300 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2301 bundle->src_mirrors &= ~mirror_bit;
2302 bundle->dst_mirrors &= ~mirror_bit;
2303 bundle->mirror_out &= ~mirror_bit;
2306 hmapx_destroy(&mirror->srcs);
2307 hmapx_destroy(&mirror->dsts);
2308 free(mirror->vlans);
2310 ofproto->mirrors[mirror->idx] = NULL;
2314 mirror_update_dups(ofproto);
2316 ofproto->has_mirrors = false;
2317 for (i = 0; i < MAX_MIRRORS; i++) {
2318 if (ofproto->mirrors[i]) {
2319 ofproto->has_mirrors = true;
2326 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2327 uint64_t *packets, uint64_t *bytes)
2329 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2330 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2333 *packets = *bytes = UINT64_MAX;
2337 *packets = mirror->packet_count;
2338 *bytes = mirror->byte_count;
2344 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2346 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2347 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2348 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2354 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2356 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2357 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2358 return bundle && bundle->mirror_out != 0;
2362 forward_bpdu_changed(struct ofproto *ofproto_)
2364 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2365 /* Revalidate cached flows whenever forward_bpdu option changes. */
2366 ofproto->need_revalidate = true;
2370 set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time)
2372 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2373 mac_learning_set_idle_time(ofproto->ml, idle_time);
2378 static struct ofport_dpif *
2379 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2381 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2382 return ofport ? ofport_dpif_cast(ofport) : NULL;
2385 static struct ofport_dpif *
2386 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2388 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2392 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2393 struct dpif_port *dpif_port)
2395 ofproto_port->name = dpif_port->name;
2396 ofproto_port->type = dpif_port->type;
2397 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2401 port_run(struct ofport_dpif *ofport)
2403 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2404 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2405 bool enable = netdev_get_carrier(ofport->up.netdev);
2407 ofport->carrier_seq = carrier_seq;
2410 cfm_run(ofport->cfm);
2412 if (cfm_should_send_ccm(ofport->cfm)) {
2413 struct ofpbuf packet;
2415 ofpbuf_init(&packet, 0);
2416 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2417 send_packet(ofport, &packet);
2418 ofpbuf_uninit(&packet);
2421 enable = enable && !cfm_get_fault(ofport->cfm)
2422 && cfm_get_opup(ofport->cfm);
2425 if (ofport->bundle) {
2426 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2427 if (carrier_changed) {
2428 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2432 if (ofport->may_enable != enable) {
2433 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2435 if (ofproto->has_bundle_action) {
2436 ofproto->need_revalidate = true;
2440 ofport->may_enable = enable;
2444 port_wait(struct ofport_dpif *ofport)
2447 cfm_wait(ofport->cfm);
2452 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2453 struct ofproto_port *ofproto_port)
2455 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2456 struct dpif_port dpif_port;
2459 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2461 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2467 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2469 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2473 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2475 *ofp_portp = odp_port_to_ofp_port(odp_port);
2481 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2483 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2486 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2488 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2490 /* The caller is going to close ofport->up.netdev. If this is a
2491 * bonded port, then the bond is using that netdev, so remove it
2492 * from the bond. The client will need to reconfigure everything
2493 * after deleting ports, so then the slave will get re-added. */
2494 bundle_remove(&ofport->up);
2501 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2503 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2506 error = netdev_get_stats(ofport->up.netdev, stats);
2508 if (!error && ofport->odp_port == OVSP_LOCAL) {
2509 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2511 /* ofproto->stats.tx_packets represents packets that we created
2512 * internally and sent to some port (e.g. packets sent with
2513 * send_packet()). Account for them as if they had come from
2514 * OFPP_LOCAL and got forwarded. */
2516 if (stats->rx_packets != UINT64_MAX) {
2517 stats->rx_packets += ofproto->stats.tx_packets;
2520 if (stats->rx_bytes != UINT64_MAX) {
2521 stats->rx_bytes += ofproto->stats.tx_bytes;
2524 /* ofproto->stats.rx_packets represents packets that were received on
2525 * some port and we processed internally and dropped (e.g. STP).
2526 * Account fro them as if they had been forwarded to OFPP_LOCAL. */
2528 if (stats->tx_packets != UINT64_MAX) {
2529 stats->tx_packets += ofproto->stats.rx_packets;
2532 if (stats->tx_bytes != UINT64_MAX) {
2533 stats->tx_bytes += ofproto->stats.rx_bytes;
2540 /* Account packets for LOCAL port. */
2542 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2543 size_t tx_size, size_t rx_size)
2545 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2548 ofproto->stats.rx_packets++;
2549 ofproto->stats.rx_bytes += rx_size;
2552 ofproto->stats.tx_packets++;
2553 ofproto->stats.tx_bytes += tx_size;
2557 struct port_dump_state {
2558 struct dpif_port_dump dump;
2563 port_dump_start(const struct ofproto *ofproto_, void **statep)
2565 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2566 struct port_dump_state *state;
2568 *statep = state = xmalloc(sizeof *state);
2569 dpif_port_dump_start(&state->dump, ofproto->dpif);
2570 state->done = false;
2575 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2576 struct ofproto_port *port)
2578 struct port_dump_state *state = state_;
2579 struct dpif_port dpif_port;
2581 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2582 ofproto_port_from_dpif_port(port, &dpif_port);
2585 int error = dpif_port_dump_done(&state->dump);
2587 return error ? error : EOF;
2592 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2594 struct port_dump_state *state = state_;
2597 dpif_port_dump_done(&state->dump);
2604 port_poll(const struct ofproto *ofproto_, char **devnamep)
2606 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2607 return dpif_port_poll(ofproto->dpif, devnamep);
2611 port_poll_wait(const struct ofproto *ofproto_)
2613 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2614 dpif_port_poll_wait(ofproto->dpif);
2618 port_is_lacp_current(const struct ofport *ofport_)
2620 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2621 return (ofport->bundle && ofport->bundle->lacp
2622 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2626 /* Upcall handling. */
2628 /* Flow miss batching.
2630 * Some dpifs implement operations faster when you hand them off in a batch.
2631 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2632 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2633 * more packets, plus possibly installing the flow in the dpif.
2635 * So far we only batch the operations that affect flow setup time the most.
2636 * It's possible to batch more than that, but the benefit might be minimal. */
2638 struct hmap_node hmap_node;
2640 enum odp_key_fitness key_fitness;
2641 const struct nlattr *key;
2643 ovs_be16 initial_tci;
2644 struct list packets;
2645 enum dpif_upcall_type upcall_type;
2648 struct flow_miss_op {
2649 struct dpif_op dpif_op;
2650 struct subfacet *subfacet; /* Subfacet */
2651 void *garbage; /* Pointer to pass to free(), NULL if none. */
2652 uint64_t stub[1024 / 8]; /* Temporary buffer. */
2655 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2656 * OpenFlow controller as necessary according to their individual
2657 * configurations. */
2659 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
2660 const struct flow *flow)
2662 struct ofputil_packet_in pin;
2664 pin.packet = packet->data;
2665 pin.packet_len = packet->size;
2666 pin.reason = OFPR_NO_MATCH;
2667 pin.controller_id = 0;
2672 pin.send_len = 0; /* not used for flow table misses */
2674 flow_get_metadata(flow, &pin.fmd);
2676 /* Registers aren't meaningful on a miss. */
2677 memset(pin.fmd.reg_masks, 0, sizeof pin.fmd.reg_masks);
2679 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
2682 static enum slow_path_reason
2683 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2684 const struct ofpbuf *packet)
2686 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2692 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2694 cfm_process_heartbeat(ofport->cfm, packet);
2697 } else if (ofport->bundle && ofport->bundle->lacp
2698 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2700 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2703 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2705 stp_process_packet(ofport, packet);
2712 static struct flow_miss *
2713 flow_miss_find(struct hmap *todo, const struct flow *flow, uint32_t hash)
2715 struct flow_miss *miss;
2717 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2718 if (flow_equal(&miss->flow, flow)) {
2726 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
2727 * 'packet'. The caller must initialize op->actions and op->actions_len. If
2728 * 'miss' is associated with a subfacet the caller must also initialize the
2729 * returned op->subfacet, and if anything needs to be freed after processing
2730 * the op, the caller must initialize op->garbage also. */
2732 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
2733 struct flow_miss_op *op)
2735 if (miss->flow.vlan_tci != miss->initial_tci) {
2736 /* This packet was received on a VLAN splinter port. We
2737 * added a VLAN to the packet to make the packet resemble
2738 * the flow, but the actions were composed assuming that
2739 * the packet contained no VLAN. So, we must remove the
2740 * VLAN header from the packet before trying to execute the
2742 eth_pop_vlan(packet);
2745 op->subfacet = NULL;
2747 op->dpif_op.type = DPIF_OP_EXECUTE;
2748 op->dpif_op.u.execute.key = miss->key;
2749 op->dpif_op.u.execute.key_len = miss->key_len;
2750 op->dpif_op.u.execute.packet = packet;
2753 /* Helper for handle_flow_miss_without_facet() and
2754 * handle_flow_miss_with_facet(). */
2756 handle_flow_miss_common(struct rule_dpif *rule,
2757 struct ofpbuf *packet, const struct flow *flow)
2759 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2761 ofproto->n_matches++;
2763 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2765 * Extra-special case for fail-open mode.
2767 * We are in fail-open mode and the packet matched the fail-open
2768 * rule, but we are connected to a controller too. We should send
2769 * the packet up to the controller in the hope that it will try to
2770 * set up a flow and thereby allow us to exit fail-open.
2772 * See the top-level comment in fail-open.c for more information.
2774 send_packet_in_miss(ofproto, packet, flow);
2778 /* Figures out whether a flow that missed in 'ofproto', whose details are in
2779 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
2780 * installing a datapath flow. The answer is usually "yes" (a return value of
2781 * true). However, for short flows the cost of bookkeeping is much higher than
2782 * the benefits, so when the datapath holds a large number of flows we impose
2783 * some heuristics to decide which flows are likely to be worth tracking. */
2785 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
2786 struct flow_miss *miss, uint32_t hash)
2788 if (!ofproto->governor) {
2791 n_subfacets = hmap_count(&ofproto->subfacets);
2792 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
2796 ofproto->governor = governor_create(ofproto->up.name);
2799 return governor_should_install_flow(ofproto->governor, hash,
2800 list_size(&miss->packets));
2803 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
2804 * or creating any datapath flow. May add an "execute" operation to 'ops' and
2805 * increment '*n_ops'. */
2807 handle_flow_miss_without_facet(struct flow_miss *miss,
2808 struct rule_dpif *rule,
2809 struct flow_miss_op *ops, size_t *n_ops)
2811 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2812 struct action_xlate_ctx ctx;
2813 struct ofpbuf *packet;
2815 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2816 struct flow_miss_op *op = &ops[*n_ops];
2817 struct dpif_flow_stats stats;
2818 struct ofpbuf odp_actions;
2820 COVERAGE_INC(facet_suppress);
2822 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
2824 dpif_flow_stats_extract(&miss->flow, packet, &stats);
2825 rule_credit_stats(rule, &stats);
2827 action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci,
2829 ctx.resubmit_stats = &stats;
2830 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions,
2833 if (odp_actions.size) {
2834 struct dpif_execute *execute = &op->dpif_op.u.execute;
2836 init_flow_miss_execute_op(miss, packet, op);
2837 execute->actions = odp_actions.data;
2838 execute->actions_len = odp_actions.size;
2839 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
2843 ofpbuf_uninit(&odp_actions);
2848 /* Handles 'miss', which matches 'facet'. May add any required datapath
2849 * operations to 'ops', incrementing '*n_ops' for each new op. */
2851 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
2852 struct flow_miss_op *ops, size_t *n_ops)
2854 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
2855 enum subfacet_path want_path;
2856 struct subfacet *subfacet;
2857 struct ofpbuf *packet;
2859 subfacet = subfacet_create(facet,
2860 miss->key_fitness, miss->key, miss->key_len,
2863 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2864 struct flow_miss_op *op = &ops[*n_ops];
2865 struct dpif_flow_stats stats;
2866 struct ofpbuf odp_actions;
2868 handle_flow_miss_common(facet->rule, packet, &miss->flow);
2870 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
2871 if (!subfacet->actions || subfacet->slow) {
2872 subfacet_make_actions(subfacet, packet, &odp_actions);
2875 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2876 subfacet_update_stats(subfacet, &stats);
2878 if (subfacet->actions_len) {
2879 struct dpif_execute *execute = &op->dpif_op.u.execute;
2881 init_flow_miss_execute_op(miss, packet, op);
2882 op->subfacet = subfacet;
2883 if (!subfacet->slow) {
2884 execute->actions = subfacet->actions;
2885 execute->actions_len = subfacet->actions_len;
2886 ofpbuf_uninit(&odp_actions);
2888 execute->actions = odp_actions.data;
2889 execute->actions_len = odp_actions.size;
2890 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
2895 ofpbuf_uninit(&odp_actions);
2899 want_path = subfacet_want_path(subfacet->slow);
2900 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
2901 struct flow_miss_op *op = &ops[(*n_ops)++];
2902 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
2904 op->subfacet = subfacet;
2906 op->dpif_op.type = DPIF_OP_FLOW_PUT;
2907 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2908 put->key = miss->key;
2909 put->key_len = miss->key_len;
2910 if (want_path == SF_FAST_PATH) {
2911 put->actions = subfacet->actions;
2912 put->actions_len = subfacet->actions_len;
2914 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
2915 op->stub, sizeof op->stub,
2916 &put->actions, &put->actions_len);
2922 /* Handles flow miss 'miss' on 'ofproto'. May add any required datapath
2923 * operations to 'ops', incrementing '*n_ops' for each new op. */
2925 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2926 struct flow_miss_op *ops, size_t *n_ops)
2928 struct facet *facet;
2931 /* The caller must ensure that miss->hmap_node.hash contains
2932 * flow_hash(miss->flow, 0). */
2933 hash = miss->hmap_node.hash;
2935 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
2937 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
2939 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
2940 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
2944 facet = facet_create(rule, &miss->flow, hash);
2946 handle_flow_miss_with_facet(miss, facet, ops, n_ops);
2949 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
2950 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
2951 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
2952 * what a flow key should contain.
2954 * This function also includes some logic to help make VLAN splinters
2955 * transparent to the rest of the upcall processing logic. In particular, if
2956 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
2957 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
2958 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
2960 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
2961 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
2962 * (This differs from the value returned in flow->vlan_tci only for packets
2963 * received on VLAN splinters.)
2965 static enum odp_key_fitness
2966 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
2967 const struct nlattr *key, size_t key_len,
2968 struct flow *flow, ovs_be16 *initial_tci,
2969 struct ofpbuf *packet)
2971 enum odp_key_fitness fitness;
2973 fitness = odp_flow_key_to_flow(key, key_len, flow);
2974 if (fitness == ODP_FIT_ERROR) {
2977 *initial_tci = flow->vlan_tci;
2979 if (vsp_adjust_flow(ofproto, flow)) {
2981 /* Make the packet resemble the flow, so that it gets sent to an
2982 * OpenFlow controller properly, so that it looks correct for
2983 * sFlow, and so that flow_extract() will get the correct vlan_tci
2984 * if it is called on 'packet'.
2986 * The allocated space inside 'packet' probably also contains
2987 * 'key', that is, both 'packet' and 'key' are probably part of a
2988 * struct dpif_upcall (see the large comment on that structure
2989 * definition), so pushing data on 'packet' is in general not a
2990 * good idea since it could overwrite 'key' or free it as a side
2991 * effect. However, it's OK in this special case because we know
2992 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
2993 * will just overwrite the 4-byte "struct nlattr", which is fine
2994 * since we don't need that header anymore. */
2995 eth_push_vlan(packet, flow->vlan_tci);
2998 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
2999 if (fitness == ODP_FIT_PERFECT) {
3000 fitness = ODP_FIT_TOO_MUCH;
3008 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
3011 struct dpif_upcall *upcall;
3012 struct flow_miss *miss;
3013 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3014 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3015 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3025 /* Construct the to-do list.
3027 * This just amounts to extracting the flow from each packet and sticking
3028 * the packets that have the same flow in the same "flow_miss" structure so
3029 * that we can process them together. */
3032 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3033 struct flow_miss *miss = &misses[n_misses];
3034 struct flow_miss *existing_miss;
3037 /* Obtain metadata and check userspace/kernel agreement on flow match,
3038 * then set 'flow''s header pointers. */
3039 miss->key_fitness = ofproto_dpif_extract_flow_key(
3040 ofproto, upcall->key, upcall->key_len,
3041 &miss->flow, &miss->initial_tci, upcall->packet);
3042 if (miss->key_fitness == ODP_FIT_ERROR) {
3045 flow_extract(upcall->packet, miss->flow.skb_priority,
3046 miss->flow.tun_id, miss->flow.in_port, &miss->flow);
3048 /* Add other packets to a to-do list. */
3049 hash = flow_hash(&miss->flow, 0);
3050 existing_miss = flow_miss_find(&todo, &miss->flow, hash);
3051 if (!existing_miss) {
3052 hmap_insert(&todo, &miss->hmap_node, hash);
3053 miss->key = upcall->key;
3054 miss->key_len = upcall->key_len;
3055 miss->upcall_type = upcall->type;
3056 list_init(&miss->packets);
3060 miss = existing_miss;
3062 list_push_back(&miss->packets, &upcall->packet->list_node);
3065 /* Process each element in the to-do list, constructing the set of
3066 * operations to batch. */
3068 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3069 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
3071 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3073 /* Execute batch. */
3074 for (i = 0; i < n_ops; i++) {
3075 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3077 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
3079 /* Free memory and update facets. */
3080 for (i = 0; i < n_ops; i++) {
3081 struct flow_miss_op *op = &flow_miss_ops[i];
3083 switch (op->dpif_op.type) {
3084 case DPIF_OP_EXECUTE:
3087 case DPIF_OP_FLOW_PUT:
3088 if (!op->dpif_op.error) {
3089 op->subfacet->path = subfacet_want_path(op->subfacet->slow);
3093 case DPIF_OP_FLOW_DEL:
3099 hmap_destroy(&todo);
3102 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
3103 classify_upcall(const struct dpif_upcall *upcall)
3105 union user_action_cookie cookie;
3107 /* First look at the upcall type. */
3108 switch (upcall->type) {
3109 case DPIF_UC_ACTION:
3115 case DPIF_N_UC_TYPES:
3117 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3121 /* "action" upcalls need a closer look. */
3122 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3123 switch (cookie.type) {
3124 case USER_ACTION_COOKIE_SFLOW:
3125 return SFLOW_UPCALL;
3127 case USER_ACTION_COOKIE_SLOW_PATH:
3130 case USER_ACTION_COOKIE_UNSPEC:
3132 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
3138 handle_sflow_upcall(struct ofproto_dpif *ofproto,
3139 const struct dpif_upcall *upcall)
3141 union user_action_cookie cookie;
3142 enum odp_key_fitness fitness;
3143 ovs_be16 initial_tci;
3146 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
3147 upcall->key_len, &flow,
3148 &initial_tci, upcall->packet);
3149 if (fitness == ODP_FIT_ERROR) {
3153 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3154 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, &cookie);
3158 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
3160 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3161 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3162 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3167 assert(max_batch <= FLOW_MISS_MAX_BATCH);
3170 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3171 struct dpif_upcall *upcall = &misses[n_misses];
3172 struct ofpbuf *buf = &miss_bufs[n_misses];
3175 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3176 sizeof miss_buf_stubs[n_misses]);
3177 error = dpif_recv(ofproto->dpif, upcall, buf);
3183 switch (classify_upcall(upcall)) {
3185 /* Handle it later. */
3190 if (ofproto->sflow) {
3191 handle_sflow_upcall(ofproto, upcall);
3202 /* Handle deferred MISS_UPCALL processing. */
3203 handle_miss_upcalls(ofproto, misses, n_misses);
3204 for (i = 0; i < n_misses; i++) {
3205 ofpbuf_uninit(&miss_bufs[i]);
3211 /* Flow expiration. */
3213 static int subfacet_max_idle(const struct ofproto_dpif *);
3214 static void update_stats(struct ofproto_dpif *);
3215 static void rule_expire(struct rule_dpif *);
3216 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
3218 /* This function is called periodically by run(). Its job is to collect
3219 * updates for the flows that have been installed into the datapath, most
3220 * importantly when they last were used, and then use that information to
3221 * expire flows that have not been used recently.
3223 * Returns the number of milliseconds after which it should be called again. */
3225 expire(struct ofproto_dpif *ofproto)
3227 struct rule_dpif *rule, *next_rule;
3228 struct oftable *table;
3231 /* Update stats for each flow in the datapath. */
3232 update_stats(ofproto);
3234 /* Expire subfacets that have been idle too long. */
3235 dp_max_idle = subfacet_max_idle(ofproto);
3236 expire_subfacets(ofproto, dp_max_idle);
3238 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
3239 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
3240 struct cls_cursor cursor;
3242 cls_cursor_init(&cursor, &table->cls, NULL);
3243 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
3248 /* All outstanding data in existing flows has been accounted, so it's a
3249 * good time to do bond rebalancing. */
3250 if (ofproto->has_bonded_bundles) {
3251 struct ofbundle *bundle;
3253 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3255 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
3260 return MIN(dp_max_idle, 1000);
3263 /* Updates flow table statistics given that the datapath just reported 'stats'
3264 * as 'subfacet''s statistics. */
3266 update_subfacet_stats(struct subfacet *subfacet,
3267 const struct dpif_flow_stats *stats)
3269 struct facet *facet = subfacet->facet;
3271 if (stats->n_packets >= subfacet->dp_packet_count) {
3272 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
3273 facet->packet_count += extra;
3275 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3278 if (stats->n_bytes >= subfacet->dp_byte_count) {
3279 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
3281 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3284 subfacet->dp_packet_count = stats->n_packets;
3285 subfacet->dp_byte_count = stats->n_bytes;
3287 facet->tcp_flags |= stats->tcp_flags;
3289 subfacet_update_time(subfacet, stats->used);
3290 if (facet->accounted_bytes < facet->byte_count) {
3292 facet_account(facet);
3293 facet->accounted_bytes = facet->byte_count;
3295 facet_push_stats(facet);
3298 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3299 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3301 delete_unexpected_flow(struct dpif *dpif,
3302 const struct nlattr *key, size_t key_len)
3304 if (!VLOG_DROP_WARN(&rl)) {
3308 odp_flow_key_format(key, key_len, &s);
3309 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
3313 COVERAGE_INC(facet_unexpected);
3314 dpif_flow_del(dpif, key, key_len, NULL);
3317 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3319 * This function also pushes statistics updates to rules which each facet
3320 * resubmits into. Generally these statistics will be accurate. However, if a
3321 * facet changes the rule it resubmits into at some time in between
3322 * update_stats() runs, it is possible that statistics accrued to the
3323 * old rule will be incorrectly attributed to the new rule. This could be
3324 * avoided by calling update_stats() whenever rules are created or
3325 * deleted. However, the performance impact of making so many calls to the
3326 * datapath do not justify the benefit of having perfectly accurate statistics.
3329 update_stats(struct ofproto_dpif *p)
3331 const struct dpif_flow_stats *stats;
3332 struct dpif_flow_dump dump;
3333 const struct nlattr *key;
3336 dpif_flow_dump_start(&dump, p->dpif);
3337 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
3338 struct subfacet *subfacet;
3340 subfacet = subfacet_find(p, key, key_len);
3341 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3343 update_subfacet_stats(subfacet, stats);
3347 /* Stats are updated per-packet. */
3350 case SF_NOT_INSTALLED:
3352 delete_unexpected_flow(p->dpif, key, key_len);
3356 dpif_flow_dump_done(&dump);
3359 /* Calculates and returns the number of milliseconds of idle time after which
3360 * subfacets should expire from the datapath. When a subfacet expires, we fold
3361 * its statistics into its facet, and when a facet's last subfacet expires, we
3362 * fold its statistic into its rule. */
3364 subfacet_max_idle(const struct ofproto_dpif *ofproto)
3367 * Idle time histogram.
3369 * Most of the time a switch has a relatively small number of subfacets.
3370 * When this is the case we might as well keep statistics for all of them
3371 * in userspace and to cache them in the kernel datapath for performance as
3374 * As the number of subfacets increases, the memory required to maintain
3375 * statistics about them in userspace and in the kernel becomes
3376 * significant. However, with a large number of subfacets it is likely
3377 * that only a few of them are "heavy hitters" that consume a large amount
3378 * of bandwidth. At this point, only heavy hitters are worth caching in
3379 * the kernel and maintaining in userspaces; other subfacets we can
3382 * The technique used to compute the idle time is to build a histogram with
3383 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3384 * that is installed in the kernel gets dropped in the appropriate bucket.
3385 * After the histogram has been built, we compute the cutoff so that only
3386 * the most-recently-used 1% of subfacets (but at least
3387 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3388 * the most-recently-used bucket of subfacets is kept, so actually an
3389 * arbitrary number of subfacets can be kept in any given expiration run
3390 * (though the next run will delete most of those unless they receive
3393 * This requires a second pass through the subfacets, in addition to the
3394 * pass made by update_stats(), because the former function never looks at
3395 * uninstallable subfacets.
3397 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3398 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3399 int buckets[N_BUCKETS] = { 0 };
3400 int total, subtotal, bucket;
3401 struct subfacet *subfacet;
3405 total = hmap_count(&ofproto->subfacets);
3406 if (total <= ofproto->up.flow_eviction_threshold) {
3407 return N_BUCKETS * BUCKET_WIDTH;
3410 /* Build histogram. */
3412 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3413 long long int idle = now - subfacet->used;
3414 int bucket = (idle <= 0 ? 0
3415 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3416 : (unsigned int) idle / BUCKET_WIDTH);
3420 /* Find the first bucket whose flows should be expired. */
3421 subtotal = bucket = 0;
3423 subtotal += buckets[bucket++];
3424 } while (bucket < N_BUCKETS &&
3425 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3427 if (VLOG_IS_DBG_ENABLED()) {
3431 ds_put_cstr(&s, "keep");
3432 for (i = 0; i < N_BUCKETS; i++) {
3434 ds_put_cstr(&s, ", drop");
3437 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3440 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3444 return bucket * BUCKET_WIDTH;
3447 enum { EXPIRE_MAX_BATCH = 50 };
3450 expire_batch(struct ofproto_dpif *ofproto, struct subfacet **subfacets, int n)
3452 struct odputil_keybuf keybufs[EXPIRE_MAX_BATCH];
3453 struct dpif_op ops[EXPIRE_MAX_BATCH];
3454 struct dpif_op *opsp[EXPIRE_MAX_BATCH];
3455 struct ofpbuf keys[EXPIRE_MAX_BATCH];
3456 struct dpif_flow_stats stats[EXPIRE_MAX_BATCH];
3459 for (i = 0; i < n; i++) {
3460 ops[i].type = DPIF_OP_FLOW_DEL;
3461 subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]);
3462 ops[i].u.flow_del.key = keys[i].data;
3463 ops[i].u.flow_del.key_len = keys[i].size;
3464 ops[i].u.flow_del.stats = &stats[i];
3468 dpif_operate(ofproto->dpif, opsp, n);
3469 for (i = 0; i < n; i++) {
3470 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
3471 subfacets[i]->path = SF_NOT_INSTALLED;
3472 subfacet_destroy(subfacets[i]);
3477 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3479 /* Cutoff time for most flows. */
3480 long long int normal_cutoff = time_msec() - dp_max_idle;
3482 /* We really want to keep flows for special protocols around, so use a more
3483 * conservative cutoff. */
3484 long long int special_cutoff = time_msec() - 10000;
3486 struct subfacet *subfacet, *next_subfacet;
3487 struct subfacet *batch[EXPIRE_MAX_BATCH];
3491 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3492 &ofproto->subfacets) {
3493 long long int cutoff;
3495 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
3498 if (subfacet->used < cutoff) {
3499 if (subfacet->path != SF_NOT_INSTALLED) {
3500 batch[n_batch++] = subfacet;
3501 if (n_batch >= EXPIRE_MAX_BATCH) {
3502 expire_batch(ofproto, batch, n_batch);
3506 subfacet_destroy(subfacet);
3512 expire_batch(ofproto, batch, n_batch);
3516 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3517 * then delete it entirely. */
3519 rule_expire(struct rule_dpif *rule)
3521 struct facet *facet, *next_facet;
3525 /* Has 'rule' expired? */
3527 if (rule->up.hard_timeout
3528 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3529 reason = OFPRR_HARD_TIMEOUT;
3530 } else if (rule->up.idle_timeout
3531 && now > rule->up.used + rule->up.idle_timeout * 1000) {
3532 reason = OFPRR_IDLE_TIMEOUT;
3537 COVERAGE_INC(ofproto_dpif_expired);
3539 /* Update stats. (This is a no-op if the rule expired due to an idle
3540 * timeout, because that only happens when the rule has no facets left.) */
3541 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3542 facet_remove(facet);
3545 /* Get rid of the rule. */
3546 ofproto_rule_expire(&rule->up, reason);
3551 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3553 * The caller must already have determined that no facet with an identical
3554 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3555 * the ofproto's classifier table.
3557 * 'hash' must be the return value of flow_hash(flow, 0).
3559 * The facet will initially have no subfacets. The caller should create (at
3560 * least) one subfacet with subfacet_create(). */
3561 static struct facet *
3562 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
3564 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3565 struct facet *facet;
3567 facet = xzalloc(sizeof *facet);
3568 facet->used = time_msec();
3569 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
3570 list_push_back(&rule->facets, &facet->list_node);
3572 facet->flow = *flow;
3573 list_init(&facet->subfacets);
3574 netflow_flow_init(&facet->nf_flow);
3575 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3581 facet_free(struct facet *facet)
3586 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3587 * 'packet', which arrived on 'in_port'.
3589 * Takes ownership of 'packet'. */
3591 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3592 const struct nlattr *odp_actions, size_t actions_len,
3593 struct ofpbuf *packet)
3595 struct odputil_keybuf keybuf;
3599 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3600 odp_flow_key_from_flow(&key, flow);
3602 error = dpif_execute(ofproto->dpif, key.data, key.size,
3603 odp_actions, actions_len, packet);
3605 ofpbuf_delete(packet);
3609 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3611 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3612 * rule's statistics, via subfacet_uninstall().
3614 * - Removes 'facet' from its rule and from ofproto->facets.
3617 facet_remove(struct facet *facet)
3619 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3620 struct subfacet *subfacet, *next_subfacet;
3622 assert(!list_is_empty(&facet->subfacets));
3624 /* First uninstall all of the subfacets to get final statistics. */
3625 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3626 subfacet_uninstall(subfacet);
3629 /* Flush the final stats to the rule.
3631 * This might require us to have at least one subfacet around so that we
3632 * can use its actions for accounting in facet_account(), which is why we
3633 * have uninstalled but not yet destroyed the subfacets. */
3634 facet_flush_stats(facet);
3636 /* Now we're really all done so destroy everything. */
3637 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3638 &facet->subfacets) {
3639 subfacet_destroy__(subfacet);
3641 hmap_remove(&ofproto->facets, &facet->hmap_node);
3642 list_remove(&facet->list_node);
3646 /* Feed information from 'facet' back into the learning table to keep it in
3647 * sync with what is actually flowing through the datapath. */
3649 facet_learn(struct facet *facet)
3651 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3652 struct action_xlate_ctx ctx;
3654 if (!facet->has_learn
3655 && !facet->has_normal
3656 && (!facet->has_fin_timeout
3657 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
3661 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3662 facet->flow.vlan_tci,
3663 facet->rule, facet->tcp_flags, NULL);
3664 ctx.may_learn = true;
3665 xlate_actions_for_side_effects(&ctx, facet->rule->up.actions,
3666 facet->rule->up.n_actions);
3670 facet_account(struct facet *facet)
3672 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3673 struct subfacet *subfacet;
3674 const struct nlattr *a;
3679 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3682 n_bytes = facet->byte_count - facet->accounted_bytes;
3684 /* This loop feeds byte counters to bond_account() for rebalancing to use
3685 * as a basis. We also need to track the actual VLAN on which the packet
3686 * is going to be sent to ensure that it matches the one passed to
3687 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3690 * We use the actions from an arbitrary subfacet because they should all
3691 * be equally valid for our purpose. */
3692 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3693 struct subfacet, list_node);
3694 vlan_tci = facet->flow.vlan_tci;
3695 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3696 subfacet->actions, subfacet->actions_len) {
3697 const struct ovs_action_push_vlan *vlan;
3698 struct ofport_dpif *port;
3700 switch (nl_attr_type(a)) {
3701 case OVS_ACTION_ATTR_OUTPUT:
3702 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3703 if (port && port->bundle && port->bundle->bond) {
3704 bond_account(port->bundle->bond, &facet->flow,
3705 vlan_tci_to_vid(vlan_tci), n_bytes);
3709 case OVS_ACTION_ATTR_POP_VLAN:
3710 vlan_tci = htons(0);
3713 case OVS_ACTION_ATTR_PUSH_VLAN:
3714 vlan = nl_attr_get(a);
3715 vlan_tci = vlan->vlan_tci;
3721 /* Returns true if the only action for 'facet' is to send to the controller.
3722 * (We don't report NetFlow expiration messages for such facets because they
3723 * are just part of the control logic for the network, not real traffic). */
3725 facet_is_controller_flow(struct facet *facet)
3728 && facet->rule->up.n_actions == 1
3729 && action_outputs_to_port(&facet->rule->up.actions[0],
3730 htons(OFPP_CONTROLLER)));
3733 /* Folds all of 'facet''s statistics into its rule. Also updates the
3734 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3735 * 'facet''s statistics in the datapath should have been zeroed and folded into
3736 * its packet and byte counts before this function is called. */
3738 facet_flush_stats(struct facet *facet)
3740 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3741 struct subfacet *subfacet;
3743 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3744 assert(!subfacet->dp_byte_count);
3745 assert(!subfacet->dp_packet_count);
3748 facet_push_stats(facet);
3749 if (facet->accounted_bytes < facet->byte_count) {
3750 facet_account(facet);
3751 facet->accounted_bytes = facet->byte_count;
3754 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3755 struct ofexpired expired;
3756 expired.flow = facet->flow;
3757 expired.packet_count = facet->packet_count;
3758 expired.byte_count = facet->byte_count;
3759 expired.used = facet->used;
3760 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3763 facet->rule->packet_count += facet->packet_count;
3764 facet->rule->byte_count += facet->byte_count;
3766 /* Reset counters to prevent double counting if 'facet' ever gets
3768 facet_reset_counters(facet);
3770 netflow_flow_clear(&facet->nf_flow);
3771 facet->tcp_flags = 0;
3774 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3775 * Returns it if found, otherwise a null pointer.
3777 * 'hash' must be the return value of flow_hash(flow, 0).
3779 * The returned facet might need revalidation; use facet_lookup_valid()
3780 * instead if that is important. */
3781 static struct facet *
3782 facet_find(struct ofproto_dpif *ofproto,
3783 const struct flow *flow, uint32_t hash)
3785 struct facet *facet;
3787 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
3788 if (flow_equal(flow, &facet->flow)) {
3796 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3797 * Returns it if found, otherwise a null pointer.
3799 * 'hash' must be the return value of flow_hash(flow, 0).
3801 * The returned facet is guaranteed to be valid. */
3802 static struct facet *
3803 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
3806 struct facet *facet;
3808 facet = facet_find(ofproto, flow, hash);
3810 && (ofproto->need_revalidate
3811 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))) {
3812 facet_revalidate(facet);
3819 subfacet_path_to_string(enum subfacet_path path)
3822 case SF_NOT_INSTALLED:
3823 return "not installed";
3825 return "in fast path";
3827 return "in slow path";
3833 /* Returns the path in which a subfacet should be installed if its 'slow'
3834 * member has the specified value. */
3835 static enum subfacet_path
3836 subfacet_want_path(enum slow_path_reason slow)
3838 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
3841 /* Returns true if 'subfacet' needs to have its datapath flow updated,
3842 * supposing that its actions have been recalculated as 'want_actions' and that
3843 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
3845 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
3846 const struct ofpbuf *want_actions)
3848 enum subfacet_path want_path = subfacet_want_path(slow);
3849 return (want_path != subfacet->path
3850 || (want_path == SF_FAST_PATH
3851 && (subfacet->actions_len != want_actions->size
3852 || memcmp(subfacet->actions, want_actions->data,
3853 subfacet->actions_len))));
3857 facet_check_consistency(struct facet *facet)
3859 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3861 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3863 uint64_t odp_actions_stub[1024 / 8];
3864 struct ofpbuf odp_actions;
3866 struct rule_dpif *rule;
3867 struct subfacet *subfacet;
3868 bool may_log = false;
3871 /* Check the rule for consistency. */
3872 rule = rule_dpif_lookup(ofproto, &facet->flow);
3873 ok = rule == facet->rule;
3875 may_log = !VLOG_DROP_WARN(&rl);
3880 flow_format(&s, &facet->flow);
3881 ds_put_format(&s, ": facet associated with wrong rule (was "
3882 "table=%"PRIu8",", facet->rule->up.table_id);
3883 cls_rule_format(&facet->rule->up.cr, &s);
3884 ds_put_format(&s, ") (should have been table=%"PRIu8",",
3886 cls_rule_format(&rule->up.cr, &s);
3887 ds_put_char(&s, ')');
3889 VLOG_WARN("%s", ds_cstr(&s));
3894 /* Check the datapath actions for consistency. */
3895 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
3896 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3897 enum subfacet_path want_path;
3898 struct odputil_keybuf keybuf;
3899 struct action_xlate_ctx ctx;
3903 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3904 subfacet->initial_tci, rule, 0, NULL);
3905 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions,
3908 if (subfacet->path == SF_NOT_INSTALLED) {
3909 /* This only happens if the datapath reported an error when we
3910 * tried to install the flow. Don't flag another error here. */
3914 want_path = subfacet_want_path(subfacet->slow);
3915 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
3916 /* The actions for slow-path flows may legitimately vary from one
3917 * packet to the next. We're done. */
3921 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
3925 /* Inconsistency! */
3927 may_log = !VLOG_DROP_WARN(&rl);
3931 /* Rate-limited, skip reporting. */
3936 subfacet_get_key(subfacet, &keybuf, &key);
3937 odp_flow_key_format(key.data, key.size, &s);
3939 ds_put_cstr(&s, ": inconsistency in subfacet");
3940 if (want_path != subfacet->path) {
3941 enum odp_key_fitness fitness = subfacet->key_fitness;
3943 ds_put_format(&s, " (%s, fitness=%s)",
3944 subfacet_path_to_string(subfacet->path),
3945 odp_key_fitness_to_string(fitness));
3946 ds_put_format(&s, " (should have been %s)",
3947 subfacet_path_to_string(want_path));
3948 } else if (want_path == SF_FAST_PATH) {
3949 ds_put_cstr(&s, " (actions were: ");
3950 format_odp_actions(&s, subfacet->actions,
3951 subfacet->actions_len);
3952 ds_put_cstr(&s, ") (correct actions: ");
3953 format_odp_actions(&s, odp_actions.data, odp_actions.size);
3954 ds_put_char(&s, ')');
3956 ds_put_cstr(&s, " (actions: ");
3957 format_odp_actions(&s, subfacet->actions,
3958 subfacet->actions_len);
3959 ds_put_char(&s, ')');
3961 VLOG_WARN("%s", ds_cstr(&s));
3964 ofpbuf_uninit(&odp_actions);
3969 /* Re-searches the classifier for 'facet':
3971 * - If the rule found is different from 'facet''s current rule, moves
3972 * 'facet' to the new rule and recompiles its actions.
3974 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3975 * where it is and recompiles its actions anyway. */
3977 facet_revalidate(struct facet *facet)
3979 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3981 struct nlattr *odp_actions;
3984 struct actions *new_actions;
3986 struct action_xlate_ctx ctx;
3987 uint64_t odp_actions_stub[1024 / 8];
3988 struct ofpbuf odp_actions;
3990 struct rule_dpif *new_rule;
3991 struct subfacet *subfacet;
3994 COVERAGE_INC(facet_revalidate);
3996 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
3998 /* Calculate new datapath actions.
4000 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4001 * emit a NetFlow expiration and, if so, we need to have the old state
4002 * around to properly compose it. */
4004 /* If the datapath actions changed or the installability changed,
4005 * then we need to talk to the datapath. */
4008 memset(&ctx, 0, sizeof ctx);
4009 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4010 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4011 enum slow_path_reason slow;
4013 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4014 subfacet->initial_tci, new_rule, 0, NULL);
4015 xlate_actions(&ctx, new_rule->up.actions, new_rule->up.n_actions,
4018 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4019 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
4020 struct dpif_flow_stats stats;
4022 subfacet_install(subfacet,
4023 odp_actions.data, odp_actions.size, &stats, slow);
4024 subfacet_update_stats(subfacet, &stats);
4027 new_actions = xcalloc(list_size(&facet->subfacets),
4028 sizeof *new_actions);
4030 new_actions[i].odp_actions = xmemdup(odp_actions.data,
4032 new_actions[i].actions_len = odp_actions.size;
4037 ofpbuf_uninit(&odp_actions);
4040 facet_flush_stats(facet);
4043 /* Update 'facet' now that we've taken care of all the old state. */
4044 facet->tags = ctx.tags;
4045 facet->nf_flow.output_iface = ctx.nf_output_iface;
4046 facet->has_learn = ctx.has_learn;
4047 facet->has_normal = ctx.has_normal;
4048 facet->has_fin_timeout = ctx.has_fin_timeout;
4049 facet->mirrors = ctx.mirrors;
4052 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4053 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4055 if (new_actions && new_actions[i].odp_actions) {
4056 free(subfacet->actions);
4057 subfacet->actions = new_actions[i].odp_actions;
4058 subfacet->actions_len = new_actions[i].actions_len;
4064 if (facet->rule != new_rule) {
4065 COVERAGE_INC(facet_changed_rule);
4066 list_remove(&facet->list_node);
4067 list_push_back(&new_rule->facets, &facet->list_node);
4068 facet->rule = new_rule;
4069 facet->used = new_rule->up.created;
4070 facet->prev_used = facet->used;
4074 /* Updates 'facet''s used time. Caller is responsible for calling
4075 * facet_push_stats() to update the flows which 'facet' resubmits into. */
4077 facet_update_time(struct facet *facet, long long int used)
4079 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4080 if (used > facet->used) {
4082 ofproto_rule_update_used(&facet->rule->up, used);
4083 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
4088 facet_reset_counters(struct facet *facet)
4090 facet->packet_count = 0;
4091 facet->byte_count = 0;
4092 facet->prev_packet_count = 0;
4093 facet->prev_byte_count = 0;
4094 facet->accounted_bytes = 0;
4098 facet_push_stats(struct facet *facet)
4100 struct dpif_flow_stats stats;
4102 assert(facet->packet_count >= facet->prev_packet_count);
4103 assert(facet->byte_count >= facet->prev_byte_count);
4104 assert(facet->used >= facet->prev_used);
4106 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4107 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4108 stats.used = facet->used;
4109 stats.tcp_flags = 0;
4111 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
4112 facet->prev_packet_count = facet->packet_count;
4113 facet->prev_byte_count = facet->byte_count;
4114 facet->prev_used = facet->used;
4116 flow_push_stats(facet->rule, &facet->flow, &stats);
4118 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
4119 facet->mirrors, stats.n_packets, stats.n_bytes);
4124 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4126 rule->packet_count += stats->n_packets;
4127 rule->byte_count += stats->n_bytes;
4128 ofproto_rule_update_used(&rule->up, stats->used);
4131 /* Pushes flow statistics to the rules which 'flow' resubmits into given
4132 * 'rule''s actions and mirrors. */
4134 flow_push_stats(struct rule_dpif *rule,
4135 const struct flow *flow, const struct dpif_flow_stats *stats)
4137 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4138 struct action_xlate_ctx ctx;
4140 ofproto_rule_update_used(&rule->up, stats->used);
4142 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule,
4144 ctx.resubmit_stats = stats;
4145 xlate_actions_for_side_effects(&ctx, rule->up.actions, rule->up.n_actions);
4150 static struct subfacet *
4151 subfacet_find__(struct ofproto_dpif *ofproto,
4152 const struct nlattr *key, size_t key_len, uint32_t key_hash,
4153 const struct flow *flow)
4155 struct subfacet *subfacet;
4157 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4158 &ofproto->subfacets) {
4160 ? (subfacet->key_len == key_len
4161 && !memcmp(key, subfacet->key, key_len))
4162 : flow_equal(flow, &subfacet->facet->flow)) {
4170 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4171 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
4172 * there is one, otherwise creates and returns a new subfacet.
4174 * If the returned subfacet is new, then subfacet->actions will be NULL, in
4175 * which case the caller must populate the actions with
4176 * subfacet_make_actions(). */
4177 static struct subfacet *
4178 subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness,
4179 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
4181 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4182 uint32_t key_hash = odp_flow_key_hash(key, key_len);
4183 struct subfacet *subfacet;
4185 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
4187 if (subfacet->facet == facet) {
4191 /* This shouldn't happen. */
4192 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4193 subfacet_destroy(subfacet);
4196 subfacet = (list_is_empty(&facet->subfacets)
4197 ? &facet->one_subfacet
4198 : xmalloc(sizeof *subfacet));
4199 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
4200 list_push_back(&facet->subfacets, &subfacet->list_node);
4201 subfacet->facet = facet;
4202 subfacet->key_fitness = key_fitness;
4203 if (key_fitness != ODP_FIT_PERFECT) {
4204 subfacet->key = xmemdup(key, key_len);
4205 subfacet->key_len = key_len;
4207 subfacet->key = NULL;
4208 subfacet->key_len = 0;
4210 subfacet->used = time_msec();
4211 subfacet->dp_packet_count = 0;
4212 subfacet->dp_byte_count = 0;
4213 subfacet->actions_len = 0;
4214 subfacet->actions = NULL;
4215 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
4218 subfacet->path = SF_NOT_INSTALLED;
4219 subfacet->initial_tci = initial_tci;
4224 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
4225 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
4226 static struct subfacet *
4227 subfacet_find(struct ofproto_dpif *ofproto,
4228 const struct nlattr *key, size_t key_len)
4230 uint32_t key_hash = odp_flow_key_hash(key, key_len);
4231 enum odp_key_fitness fitness;
4234 fitness = odp_flow_key_to_flow(key, key_len, &flow);
4235 if (fitness == ODP_FIT_ERROR) {
4239 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
4242 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4243 * its facet within 'ofproto', and frees it. */
4245 subfacet_destroy__(struct subfacet *subfacet)
4247 struct facet *facet = subfacet->facet;
4248 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4250 subfacet_uninstall(subfacet);
4251 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
4252 list_remove(&subfacet->list_node);
4253 free(subfacet->key);
4254 free(subfacet->actions);
4255 if (subfacet != &facet->one_subfacet) {
4260 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4261 * last remaining subfacet in its facet destroys the facet too. */
4263 subfacet_destroy(struct subfacet *subfacet)
4265 struct facet *facet = subfacet->facet;
4267 if (list_is_singleton(&facet->subfacets)) {
4268 /* facet_remove() needs at least one subfacet (it will remove it). */
4269 facet_remove(facet);
4271 subfacet_destroy__(subfacet);
4275 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
4276 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
4277 * for use as temporary storage. */
4279 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
4282 if (!subfacet->key) {
4283 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
4284 odp_flow_key_from_flow(key, &subfacet->facet->flow);
4286 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
4290 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
4291 * Translates the actions into 'odp_actions', which the caller must have
4292 * initialized and is responsible for uninitializing. */
4294 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
4295 struct ofpbuf *odp_actions)
4297 struct facet *facet = subfacet->facet;
4298 struct rule_dpif *rule = facet->rule;
4299 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4301 struct action_xlate_ctx ctx;
4303 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
4305 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions, odp_actions);
4306 facet->tags = ctx.tags;
4307 facet->has_learn = ctx.has_learn;
4308 facet->has_normal = ctx.has_normal;
4309 facet->has_fin_timeout = ctx.has_fin_timeout;
4310 facet->nf_flow.output_iface = ctx.nf_output_iface;
4311 facet->mirrors = ctx.mirrors;
4313 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4314 if (subfacet->actions_len != odp_actions->size
4315 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
4316 free(subfacet->actions);
4317 subfacet->actions_len = odp_actions->size;
4318 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
4322 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4323 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4324 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4325 * since 'subfacet' was last updated.
4327 * Returns 0 if successful, otherwise a positive errno value. */
4329 subfacet_install(struct subfacet *subfacet,
4330 const struct nlattr *actions, size_t actions_len,
4331 struct dpif_flow_stats *stats,
4332 enum slow_path_reason slow)
4334 struct facet *facet = subfacet->facet;
4335 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4336 enum subfacet_path path = subfacet_want_path(slow);
4337 uint64_t slow_path_stub[128 / 8];
4338 struct odputil_keybuf keybuf;
4339 enum dpif_flow_put_flags flags;
4343 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
4345 flags |= DPIF_FP_ZERO_STATS;
4348 if (path == SF_SLOW_PATH) {
4349 compose_slow_path(ofproto, &facet->flow, slow,
4350 slow_path_stub, sizeof slow_path_stub,
4351 &actions, &actions_len);
4354 subfacet_get_key(subfacet, &keybuf, &key);
4355 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
4356 actions, actions_len, stats);
4359 subfacet_reset_dp_stats(subfacet, stats);
4363 subfacet->path = path;
4369 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
4371 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
4372 stats, subfacet->slow);
4375 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4377 subfacet_uninstall(struct subfacet *subfacet)
4379 if (subfacet->path != SF_NOT_INSTALLED) {
4380 struct rule_dpif *rule = subfacet->facet->rule;
4381 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4382 struct odputil_keybuf keybuf;
4383 struct dpif_flow_stats stats;
4387 subfacet_get_key(subfacet, &keybuf, &key);
4388 error = dpif_flow_del(ofproto->dpif, key.data, key.size, &stats);
4389 subfacet_reset_dp_stats(subfacet, &stats);
4391 subfacet_update_stats(subfacet, &stats);
4393 subfacet->path = SF_NOT_INSTALLED;
4395 assert(subfacet->dp_packet_count == 0);
4396 assert(subfacet->dp_byte_count == 0);
4400 /* Resets 'subfacet''s datapath statistics counters. This should be called
4401 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4402 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4403 * was reset in the datapath. 'stats' will be modified to include only
4404 * statistics new since 'subfacet' was last updated. */
4406 subfacet_reset_dp_stats(struct subfacet *subfacet,
4407 struct dpif_flow_stats *stats)
4410 && subfacet->dp_packet_count <= stats->n_packets
4411 && subfacet->dp_byte_count <= stats->n_bytes) {
4412 stats->n_packets -= subfacet->dp_packet_count;
4413 stats->n_bytes -= subfacet->dp_byte_count;
4416 subfacet->dp_packet_count = 0;
4417 subfacet->dp_byte_count = 0;
4420 /* Updates 'subfacet''s used time. The caller is responsible for calling
4421 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
4423 subfacet_update_time(struct subfacet *subfacet, long long int used)
4425 if (used > subfacet->used) {
4426 subfacet->used = used;
4427 facet_update_time(subfacet->facet, used);
4431 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4433 * Because of the meaning of a subfacet's counters, it only makes sense to do
4434 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4435 * represents a packet that was sent by hand or if it represents statistics
4436 * that have been cleared out of the datapath. */
4438 subfacet_update_stats(struct subfacet *subfacet,
4439 const struct dpif_flow_stats *stats)
4441 if (stats->n_packets || stats->used > subfacet->used) {
4442 struct facet *facet = subfacet->facet;
4444 subfacet_update_time(subfacet, stats->used);
4445 facet->packet_count += stats->n_packets;
4446 facet->byte_count += stats->n_bytes;
4447 facet->tcp_flags |= stats->tcp_flags;
4448 facet_push_stats(facet);
4449 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
4455 static struct rule_dpif *
4456 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
4458 struct ofport_dpif *port;
4459 struct rule_dpif *rule;
4461 rule = rule_dpif_lookup__(ofproto, flow, 0);
4466 port = get_ofp_port(ofproto, flow->in_port);
4468 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
4469 return ofproto->miss_rule;
4472 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
4473 return ofproto->no_packet_in_rule;
4475 return ofproto->miss_rule;
4478 static struct rule_dpif *
4479 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
4482 struct cls_rule *cls_rule;
4483 struct classifier *cls;
4485 if (table_id >= N_TABLES) {
4489 cls = &ofproto->up.tables[table_id].cls;
4490 if (flow->nw_frag & FLOW_NW_FRAG_ANY
4491 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4492 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
4493 * are unavailable. */
4494 struct flow ofpc_normal_flow = *flow;
4495 ofpc_normal_flow.tp_src = htons(0);
4496 ofpc_normal_flow.tp_dst = htons(0);
4497 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
4499 cls_rule = classifier_lookup(cls, flow);
4501 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
4505 complete_operation(struct rule_dpif *rule)
4507 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4509 rule_invalidate(rule);
4511 struct dpif_completion *c = xmalloc(sizeof *c);
4512 c->op = rule->up.pending;
4513 list_push_back(&ofproto->completions, &c->list_node);
4515 ofoperation_complete(rule->up.pending, 0);
4519 static struct rule *
4522 struct rule_dpif *rule = xmalloc(sizeof *rule);
4527 rule_dealloc(struct rule *rule_)
4529 struct rule_dpif *rule = rule_dpif_cast(rule_);
4534 rule_construct(struct rule *rule_)
4536 struct rule_dpif *rule = rule_dpif_cast(rule_);
4537 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4538 struct rule_dpif *victim;
4542 error = validate_actions(rule->up.actions, rule->up.n_actions,
4543 &rule->up.cr.flow, ofproto->max_ports);
4548 rule->packet_count = 0;
4549 rule->byte_count = 0;
4551 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
4552 if (victim && !list_is_empty(&victim->facets)) {
4553 struct facet *facet;
4555 rule->facets = victim->facets;
4556 list_moved(&rule->facets);
4557 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4558 /* XXX: We're only clearing our local counters here. It's possible
4559 * that quite a few packets are unaccounted for in the datapath
4560 * statistics. These will be accounted to the new rule instead of
4561 * cleared as required. This could be fixed by clearing out the
4562 * datapath statistics for this facet, but currently it doesn't
4564 facet_reset_counters(facet);
4568 /* Must avoid list_moved() in this case. */
4569 list_init(&rule->facets);
4572 table_id = rule->up.table_id;
4573 rule->tag = (victim ? victim->tag
4575 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
4576 ofproto->tables[table_id].basis));
4578 complete_operation(rule);
4583 rule_destruct(struct rule *rule_)
4585 struct rule_dpif *rule = rule_dpif_cast(rule_);
4586 struct facet *facet, *next_facet;
4588 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4589 facet_revalidate(facet);
4592 complete_operation(rule);
4596 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4598 struct rule_dpif *rule = rule_dpif_cast(rule_);
4599 struct facet *facet;
4601 /* Start from historical data for 'rule' itself that are no longer tracked
4602 * in facets. This counts, for example, facets that have expired. */
4603 *packets = rule->packet_count;
4604 *bytes = rule->byte_count;
4606 /* Add any statistics that are tracked by facets. This includes
4607 * statistical data recently updated by ofproto_update_stats() as well as
4608 * stats for packets that were executed "by hand" via dpif_execute(). */
4609 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4610 *packets += facet->packet_count;
4611 *bytes += facet->byte_count;
4616 rule_execute(struct rule *rule_, const struct flow *flow,
4617 struct ofpbuf *packet)
4619 struct rule_dpif *rule = rule_dpif_cast(rule_);
4620 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4622 struct dpif_flow_stats stats;
4624 struct action_xlate_ctx ctx;
4625 uint64_t odp_actions_stub[1024 / 8];
4626 struct ofpbuf odp_actions;
4628 dpif_flow_stats_extract(flow, packet, &stats);
4629 rule_credit_stats(rule, &stats);
4631 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4632 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
4633 rule, stats.tcp_flags, packet);
4634 ctx.resubmit_stats = &stats;
4635 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions, &odp_actions);
4637 execute_odp_actions(ofproto, flow, odp_actions.data,
4638 odp_actions.size, packet);
4640 ofpbuf_uninit(&odp_actions);
4646 rule_modify_actions(struct rule *rule_)
4648 struct rule_dpif *rule = rule_dpif_cast(rule_);
4649 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4652 error = validate_actions(rule->up.actions, rule->up.n_actions,
4653 &rule->up.cr.flow, ofproto->max_ports);
4655 ofoperation_complete(rule->up.pending, error);
4659 complete_operation(rule);
4662 /* Sends 'packet' out 'ofport'.
4663 * May modify 'packet'.
4664 * Returns 0 if successful, otherwise a positive errno value. */
4666 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4668 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4669 struct ofpbuf key, odp_actions;
4670 struct odputil_keybuf keybuf;
4675 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
4676 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4678 if (odp_port != ofport->odp_port) {
4679 eth_pop_vlan(packet);
4680 flow.vlan_tci = htons(0);
4683 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4684 odp_flow_key_from_flow(&key, &flow);
4686 ofpbuf_init(&odp_actions, 32);
4687 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4689 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4690 error = dpif_execute(ofproto->dpif,
4692 odp_actions.data, odp_actions.size,
4694 ofpbuf_uninit(&odp_actions);
4697 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4698 ofproto->up.name, odp_port, strerror(error));
4700 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
4704 /* OpenFlow to datapath action translation. */
4706 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
4707 struct action_xlate_ctx *ctx);
4708 static void xlate_normal(struct action_xlate_ctx *);
4710 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
4711 * The action will state 'slow' as the reason that the action is in the slow
4712 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
4713 * dump-flows" output to see why a flow is in the slow path.)
4715 * The 'stub_size' bytes in 'stub' will be used to store the action.
4716 * 'stub_size' must be large enough for the action.
4718 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
4721 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
4722 enum slow_path_reason slow,
4723 uint64_t *stub, size_t stub_size,
4724 const struct nlattr **actionsp, size_t *actions_lenp)
4726 union user_action_cookie cookie;
4729 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
4730 cookie.slow_path.unused = 0;
4731 cookie.slow_path.reason = slow;
4733 ofpbuf_use_stack(&buf, stub, stub_size);
4734 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
4735 uint32_t pid = dpif_port_get_pid(ofproto->dpif, UINT16_MAX);
4736 odp_put_userspace_action(pid, &cookie, &buf);
4738 put_userspace_action(ofproto, &buf, flow, &cookie);
4740 *actionsp = buf.data;
4741 *actions_lenp = buf.size;
4745 put_userspace_action(const struct ofproto_dpif *ofproto,
4746 struct ofpbuf *odp_actions,
4747 const struct flow *flow,
4748 const union user_action_cookie *cookie)
4752 pid = dpif_port_get_pid(ofproto->dpif,
4753 ofp_port_to_odp_port(flow->in_port));
4755 return odp_put_userspace_action(pid, cookie, odp_actions);
4759 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
4760 ovs_be16 vlan_tci, uint32_t odp_port,
4761 unsigned int n_outputs, union user_action_cookie *cookie)
4765 cookie->type = USER_ACTION_COOKIE_SFLOW;
4766 cookie->sflow.vlan_tci = vlan_tci;
4768 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
4769 * port information") for the interpretation of cookie->output. */
4770 switch (n_outputs) {
4772 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
4773 cookie->sflow.output = 0x40000000 | 256;
4777 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4779 cookie->sflow.output = ifindex;
4784 /* 0x80000000 means "multiple output ports. */
4785 cookie->sflow.output = 0x80000000 | n_outputs;
4790 /* Compose SAMPLE action for sFlow. */
4792 compose_sflow_action(const struct ofproto_dpif *ofproto,
4793 struct ofpbuf *odp_actions,
4794 const struct flow *flow,
4797 uint32_t probability;
4798 union user_action_cookie cookie;
4799 size_t sample_offset, actions_offset;
4802 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4806 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4808 /* Number of packets out of UINT_MAX to sample. */
4809 probability = dpif_sflow_get_probability(ofproto->sflow);
4810 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4812 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4813 compose_sflow_cookie(ofproto, htons(0), odp_port,
4814 odp_port == OVSP_NONE ? 0 : 1, &cookie);
4815 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4817 nl_msg_end_nested(odp_actions, actions_offset);
4818 nl_msg_end_nested(odp_actions, sample_offset);
4819 return cookie_offset;
4822 /* SAMPLE action must be first action in any given list of actions.
4823 * At this point we do not have all information required to build it. So try to
4824 * build sample action as complete as possible. */
4826 add_sflow_action(struct action_xlate_ctx *ctx)
4828 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4830 &ctx->flow, OVSP_NONE);
4831 ctx->sflow_odp_port = 0;
4832 ctx->sflow_n_outputs = 0;
4835 /* Fix SAMPLE action according to data collected while composing ODP actions.
4836 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4837 * USERSPACE action's user-cookie which is required for sflow. */
4839 fix_sflow_action(struct action_xlate_ctx *ctx)
4841 const struct flow *base = &ctx->base_flow;
4842 union user_action_cookie *cookie;
4844 if (!ctx->user_cookie_offset) {
4848 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4850 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4852 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
4853 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
4857 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4860 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4861 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4862 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4863 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4867 struct priority_to_dscp *pdscp;
4869 if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD
4870 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4874 pdscp = get_priority(ofport, ctx->flow.skb_priority);
4876 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4877 ctx->flow.nw_tos |= pdscp->dscp;
4880 /* We may not have an ofport record for this port, but it doesn't hurt
4881 * to allow forwarding to it anyhow. Maybe such a port will appear
4882 * later and we're pre-populating the flow table. */
4885 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4886 ctx->flow.vlan_tci);
4887 if (out_port != odp_port) {
4888 ctx->flow.vlan_tci = htons(0);
4890 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4891 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4893 ctx->sflow_odp_port = odp_port;
4894 ctx->sflow_n_outputs++;
4895 ctx->nf_output_iface = ofp_port;
4896 ctx->flow.vlan_tci = flow_vlan_tci;
4897 ctx->flow.nw_tos = flow_nw_tos;
4901 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4903 compose_output_action__(ctx, ofp_port, true);
4907 xlate_table_action(struct action_xlate_ctx *ctx,
4908 uint16_t in_port, uint8_t table_id)
4910 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4911 struct ofproto_dpif *ofproto = ctx->ofproto;
4912 struct rule_dpif *rule;
4913 uint16_t old_in_port;
4914 uint8_t old_table_id;
4916 old_table_id = ctx->table_id;
4917 ctx->table_id = table_id;
4919 /* Look up a flow with 'in_port' as the input port. */
4920 old_in_port = ctx->flow.in_port;
4921 ctx->flow.in_port = in_port;
4922 rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id);
4925 if (table_id > 0 && table_id < N_TABLES) {
4926 struct table_dpif *table = &ofproto->tables[table_id];
4927 if (table->other_table) {
4928 ctx->tags |= (rule && rule->tag
4930 : rule_calculate_tag(&ctx->flow,
4931 &table->other_table->wc,
4936 /* Restore the original input port. Otherwise OFPP_NORMAL and
4937 * OFPP_IN_PORT will have surprising behavior. */
4938 ctx->flow.in_port = old_in_port;
4940 if (ctx->resubmit_hook) {
4941 ctx->resubmit_hook(ctx, rule);
4945 struct rule_dpif *old_rule = ctx->rule;
4947 if (ctx->resubmit_stats) {
4948 rule_credit_stats(rule, ctx->resubmit_stats);
4953 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
4954 ctx->rule = old_rule;
4958 ctx->table_id = old_table_id;
4960 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4962 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
4963 MAX_RESUBMIT_RECURSION);
4964 ctx->max_resubmit_trigger = true;
4969 xlate_resubmit_table(struct action_xlate_ctx *ctx,
4970 const struct nx_action_resubmit *nar)
4975 in_port = (nar->in_port == htons(OFPP_IN_PORT)
4977 : ntohs(nar->in_port));
4978 table_id = nar->table == 255 ? ctx->table_id : nar->table;
4980 xlate_table_action(ctx, in_port, table_id);
4984 flood_packets(struct action_xlate_ctx *ctx, bool all)
4986 struct ofport_dpif *ofport;
4988 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
4989 uint16_t ofp_port = ofport->up.ofp_port;
4991 if (ofp_port == ctx->flow.in_port) {
4996 compose_output_action__(ctx, ofp_port, false);
4997 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
4998 compose_output_action(ctx, ofp_port);
5002 ctx->nf_output_iface = NF_OUT_FLOOD;
5006 execute_controller_action(struct action_xlate_ctx *ctx, int len,
5007 enum ofp_packet_in_reason reason,
5008 uint16_t controller_id)
5010 struct ofputil_packet_in pin;
5011 struct ofpbuf *packet;
5013 ctx->slow |= SLOW_CONTROLLER;
5018 packet = ofpbuf_clone(ctx->packet);
5020 if (packet->l2 && packet->l3) {
5021 struct eth_header *eh;
5023 eth_pop_vlan(packet);
5026 /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
5027 * LLC frame. Calculating the Ethernet type of these frames is more
5028 * trouble than seems appropriate for a simple assertion. */
5029 assert(ntohs(eh->eth_type) < ETH_TYPE_MIN
5030 || eh->eth_type == ctx->flow.dl_type);
5032 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
5033 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
5035 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
5036 eth_push_vlan(packet, ctx->flow.vlan_tci);
5040 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5041 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
5042 ctx->flow.nw_tos, ctx->flow.nw_ttl);
5046 if (ctx->flow.nw_proto == IPPROTO_TCP) {
5047 packet_set_tcp_port(packet, ctx->flow.tp_src,
5049 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
5050 packet_set_udp_port(packet, ctx->flow.tp_src,
5057 pin.packet = packet->data;
5058 pin.packet_len = packet->size;
5059 pin.reason = reason;
5060 pin.controller_id = controller_id;
5061 pin.table_id = ctx->table_id;
5062 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
5065 flow_get_metadata(&ctx->flow, &pin.fmd);
5067 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
5068 ofpbuf_delete(packet);
5072 compose_dec_ttl(struct action_xlate_ctx *ctx)
5074 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
5075 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
5079 if (ctx->flow.nw_ttl > 1) {
5083 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
5085 /* Stop processing for current table. */
5091 xlate_output_action__(struct action_xlate_ctx *ctx,
5092 uint16_t port, uint16_t max_len)
5094 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
5096 ctx->nf_output_iface = NF_OUT_DROP;
5100 compose_output_action(ctx, ctx->flow.in_port);
5103 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
5109 flood_packets(ctx, false);
5112 flood_packets(ctx, true);
5114 case OFPP_CONTROLLER:
5115 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
5121 if (port != ctx->flow.in_port) {
5122 compose_output_action(ctx, port);
5127 if (prev_nf_output_iface == NF_OUT_FLOOD) {
5128 ctx->nf_output_iface = NF_OUT_FLOOD;
5129 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
5130 ctx->nf_output_iface = prev_nf_output_iface;
5131 } else if (prev_nf_output_iface != NF_OUT_DROP &&
5132 ctx->nf_output_iface != NF_OUT_FLOOD) {
5133 ctx->nf_output_iface = NF_OUT_MULTI;
5138 xlate_output_reg_action(struct action_xlate_ctx *ctx,
5139 const struct nx_action_output_reg *naor)
5141 struct mf_subfield src;
5144 nxm_decode(&src, naor->src, naor->ofs_nbits);
5145 ofp_port = mf_get_subfield(&src, &ctx->flow);
5147 if (ofp_port <= UINT16_MAX) {
5148 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
5153 xlate_output_action(struct action_xlate_ctx *ctx,
5154 const struct ofp_action_output *oao)
5156 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
5160 xlate_enqueue_action(struct action_xlate_ctx *ctx,
5161 const struct ofp_action_enqueue *oae)
5164 uint32_t flow_priority, priority;
5167 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
5170 /* Fall back to ordinary output action. */
5171 xlate_output_action__(ctx, ntohs(oae->port), 0);
5175 /* Figure out datapath output port. */
5176 ofp_port = ntohs(oae->port);
5177 if (ofp_port == OFPP_IN_PORT) {
5178 ofp_port = ctx->flow.in_port;
5179 } else if (ofp_port == ctx->flow.in_port) {
5183 /* Add datapath actions. */
5184 flow_priority = ctx->flow.skb_priority;
5185 ctx->flow.skb_priority = priority;
5186 compose_output_action(ctx, ofp_port);
5187 ctx->flow.skb_priority = flow_priority;
5189 /* Update NetFlow output port. */
5190 if (ctx->nf_output_iface == NF_OUT_DROP) {
5191 ctx->nf_output_iface = ofp_port;
5192 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
5193 ctx->nf_output_iface = NF_OUT_MULTI;
5198 xlate_set_queue_action(struct action_xlate_ctx *ctx,
5199 const struct nx_action_set_queue *nasq)
5204 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
5207 /* Couldn't translate queue to a priority, so ignore. A warning
5208 * has already been logged. */
5212 ctx->flow.skb_priority = priority;
5215 struct xlate_reg_state {
5221 xlate_autopath(struct action_xlate_ctx *ctx,
5222 const struct nx_action_autopath *naa)
5224 uint16_t ofp_port = ntohl(naa->id);
5225 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
5227 if (!port || !port->bundle) {
5228 ofp_port = OFPP_NONE;
5229 } else if (port->bundle->bond) {
5230 /* Autopath does not support VLAN hashing. */
5231 struct ofport_dpif *slave = bond_choose_output_slave(
5232 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
5234 ofp_port = slave->up.ofp_port;
5237 autopath_execute(naa, &ctx->flow, ofp_port);
5241 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
5243 struct ofproto_dpif *ofproto = ofproto_;
5244 struct ofport_dpif *port;
5254 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
5257 port = get_ofp_port(ofproto, ofp_port);
5258 return port ? port->may_enable : false;
5263 xlate_learn_action(struct action_xlate_ctx *ctx,
5264 const struct nx_action_learn *learn)
5266 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
5267 struct ofputil_flow_mod fm;
5270 learn_execute(learn, &ctx->flow, &fm);
5272 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
5273 if (error && !VLOG_DROP_WARN(&rl)) {
5274 VLOG_WARN("learning action failed to modify flow table (%s)",
5275 ofperr_get_name(error));
5281 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
5282 * means "infinite". */
5284 reduce_timeout(uint16_t max, uint16_t *timeout)
5286 if (max && (!*timeout || *timeout > max)) {
5292 xlate_fin_timeout(struct action_xlate_ctx *ctx,
5293 const struct nx_action_fin_timeout *naft)
5295 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
5296 struct rule_dpif *rule = ctx->rule;
5298 reduce_timeout(ntohs(naft->fin_idle_timeout), &rule->up.idle_timeout);
5299 reduce_timeout(ntohs(naft->fin_hard_timeout), &rule->up.hard_timeout);
5304 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
5306 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
5307 ? OFPUTIL_PC_NO_RECV_STP
5308 : OFPUTIL_PC_NO_RECV)) {
5312 /* Only drop packets here if both forwarding and learning are
5313 * disabled. If just learning is enabled, we need to have
5314 * OFPP_NORMAL and the learning action have a look at the packet
5315 * before we can drop it. */
5316 if (!stp_forward_in_state(port->stp_state)
5317 && !stp_learn_in_state(port->stp_state)) {
5325 do_xlate_actions(const union ofp_action *in, size_t n_in,
5326 struct action_xlate_ctx *ctx)
5328 const struct ofport_dpif *port;
5329 const union ofp_action *ia;
5330 bool was_evictable = true;
5333 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5334 if (port && !may_receive(port, ctx)) {
5335 /* Drop this flow. */
5340 /* Don't let the rule we're working on get evicted underneath us. */
5341 was_evictable = ctx->rule->up.evictable;
5342 ctx->rule->up.evictable = false;
5344 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
5345 const struct ofp_action_dl_addr *oada;
5346 const struct nx_action_resubmit *nar;
5347 const struct nx_action_set_tunnel *nast;
5348 const struct nx_action_set_queue *nasq;
5349 const struct nx_action_multipath *nam;
5350 const struct nx_action_autopath *naa;
5351 const struct nx_action_bundle *nab;
5352 const struct nx_action_output_reg *naor;
5353 const struct nx_action_controller *nac;
5354 enum ofputil_action_code code;
5361 code = ofputil_decode_action_unsafe(ia);
5363 case OFPUTIL_OFPAT10_OUTPUT:
5364 xlate_output_action(ctx, &ia->output);
5367 case OFPUTIL_OFPAT10_SET_VLAN_VID:
5368 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
5369 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
5372 case OFPUTIL_OFPAT10_SET_VLAN_PCP:
5373 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
5374 ctx->flow.vlan_tci |= htons(
5375 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
5378 case OFPUTIL_OFPAT10_STRIP_VLAN:
5379 ctx->flow.vlan_tci = htons(0);
5382 case OFPUTIL_OFPAT10_SET_DL_SRC:
5383 oada = ((struct ofp_action_dl_addr *) ia);
5384 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
5387 case OFPUTIL_OFPAT10_SET_DL_DST:
5388 oada = ((struct ofp_action_dl_addr *) ia);
5389 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
5392 case OFPUTIL_OFPAT10_SET_NW_SRC:
5393 ctx->flow.nw_src = ia->nw_addr.nw_addr;
5396 case OFPUTIL_OFPAT10_SET_NW_DST:
5397 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
5400 case OFPUTIL_OFPAT10_SET_NW_TOS:
5401 /* OpenFlow 1.0 only supports IPv4. */
5402 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5403 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5404 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
5408 case OFPUTIL_OFPAT10_SET_TP_SRC:
5409 ctx->flow.tp_src = ia->tp_port.tp_port;
5412 case OFPUTIL_OFPAT10_SET_TP_DST:
5413 ctx->flow.tp_dst = ia->tp_port.tp_port;
5416 case OFPUTIL_OFPAT10_ENQUEUE:
5417 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
5420 case OFPUTIL_NXAST_RESUBMIT:
5421 nar = (const struct nx_action_resubmit *) ia;
5422 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
5425 case OFPUTIL_NXAST_RESUBMIT_TABLE:
5426 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
5429 case OFPUTIL_NXAST_SET_TUNNEL:
5430 nast = (const struct nx_action_set_tunnel *) ia;
5431 tun_id = htonll(ntohl(nast->tun_id));
5432 ctx->flow.tun_id = tun_id;
5435 case OFPUTIL_NXAST_SET_QUEUE:
5436 nasq = (const struct nx_action_set_queue *) ia;
5437 xlate_set_queue_action(ctx, nasq);
5440 case OFPUTIL_NXAST_POP_QUEUE:
5441 ctx->flow.skb_priority = ctx->orig_skb_priority;
5444 case OFPUTIL_NXAST_REG_MOVE:
5445 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
5449 case OFPUTIL_NXAST_REG_LOAD:
5450 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
5454 case OFPUTIL_NXAST_NOTE:
5455 /* Nothing to do. */
5458 case OFPUTIL_NXAST_SET_TUNNEL64:
5459 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
5460 ctx->flow.tun_id = tun_id;
5463 case OFPUTIL_NXAST_MULTIPATH:
5464 nam = (const struct nx_action_multipath *) ia;
5465 multipath_execute(nam, &ctx->flow);
5468 case OFPUTIL_NXAST_AUTOPATH:
5469 naa = (const struct nx_action_autopath *) ia;
5470 xlate_autopath(ctx, naa);
5473 case OFPUTIL_NXAST_BUNDLE:
5474 ctx->ofproto->has_bundle_action = true;
5475 nab = (const struct nx_action_bundle *) ia;
5476 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
5481 case OFPUTIL_NXAST_BUNDLE_LOAD:
5482 ctx->ofproto->has_bundle_action = true;
5483 nab = (const struct nx_action_bundle *) ia;
5484 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
5488 case OFPUTIL_NXAST_OUTPUT_REG:
5489 naor = (const struct nx_action_output_reg *) ia;
5490 xlate_output_reg_action(ctx, naor);
5493 case OFPUTIL_NXAST_LEARN:
5494 ctx->has_learn = true;
5495 if (ctx->may_learn) {
5496 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
5500 case OFPUTIL_NXAST_DEC_TTL:
5501 if (compose_dec_ttl(ctx)) {
5506 case OFPUTIL_NXAST_EXIT:
5510 case OFPUTIL_NXAST_FIN_TIMEOUT:
5511 ctx->has_fin_timeout = true;
5512 xlate_fin_timeout(ctx, (const struct nx_action_fin_timeout *) ia);
5515 case OFPUTIL_NXAST_CONTROLLER:
5516 nac = (const struct nx_action_controller *) ia;
5517 execute_controller_action(ctx, ntohs(nac->max_len), nac->reason,
5518 ntohs(nac->controller_id));
5524 /* We've let OFPP_NORMAL and the learning action look at the packet,
5525 * so drop it now if forwarding is disabled. */
5526 if (port && !stp_forward_in_state(port->stp_state)) {
5527 ofpbuf_clear(ctx->odp_actions);
5528 add_sflow_action(ctx);
5531 ctx->rule->up.evictable = was_evictable;
5536 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
5537 struct ofproto_dpif *ofproto, const struct flow *flow,
5538 ovs_be16 initial_tci, struct rule_dpif *rule,
5539 uint8_t tcp_flags, const struct ofpbuf *packet)
5541 ctx->ofproto = ofproto;
5543 ctx->base_flow = ctx->flow;
5544 ctx->base_flow.tun_id = 0;
5545 ctx->base_flow.vlan_tci = initial_tci;
5547 ctx->packet = packet;
5548 ctx->may_learn = packet != NULL;
5549 ctx->tcp_flags = tcp_flags;
5550 ctx->resubmit_hook = NULL;
5551 ctx->resubmit_stats = NULL;
5554 /* Translates the 'n_in' "union ofp_action"s in 'in' into datapath actions in
5555 * 'odp_actions', using 'ctx'. */
5557 xlate_actions(struct action_xlate_ctx *ctx,
5558 const union ofp_action *in, size_t n_in,
5559 struct ofpbuf *odp_actions)
5561 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
5562 * that in the future we always keep a copy of the original flow for
5563 * tracing purposes. */
5564 static bool hit_resubmit_limit;
5566 enum slow_path_reason special;
5568 COVERAGE_INC(ofproto_dpif_xlate);
5570 ofpbuf_clear(odp_actions);
5571 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
5573 ctx->odp_actions = odp_actions;
5576 ctx->has_learn = false;
5577 ctx->has_normal = false;
5578 ctx->has_fin_timeout = false;
5579 ctx->nf_output_iface = NF_OUT_DROP;
5582 ctx->max_resubmit_trigger = false;
5583 ctx->orig_skb_priority = ctx->flow.skb_priority;
5587 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
5588 /* Do this conditionally because the copy is expensive enough that it
5589 * shows up in profiles.
5591 * We keep orig_flow in 'ctx' only because I couldn't make GCC 4.4
5592 * believe that I wasn't using it without initializing it if I kept it
5593 * in a local variable. */
5594 ctx->orig_flow = ctx->flow;
5597 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
5598 switch (ctx->ofproto->up.frag_handling) {
5599 case OFPC_FRAG_NORMAL:
5600 /* We must pretend that transport ports are unavailable. */
5601 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
5602 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
5605 case OFPC_FRAG_DROP:
5608 case OFPC_FRAG_REASM:
5611 case OFPC_FRAG_NX_MATCH:
5612 /* Nothing to do. */
5615 case OFPC_INVALID_TTL_TO_CONTROLLER:
5620 special = process_special(ctx->ofproto, &ctx->flow, ctx->packet);
5622 ctx->slow |= special;
5624 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5625 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
5627 add_sflow_action(ctx);
5628 do_xlate_actions(in, n_in, ctx);
5630 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
5631 if (!hit_resubmit_limit) {
5632 /* We didn't record the original flow. Make sure we do from
5634 hit_resubmit_limit = true;
5635 } else if (!VLOG_DROP_ERR(&trace_rl)) {
5636 struct ds ds = DS_EMPTY_INITIALIZER;
5638 ofproto_trace(ctx->ofproto, &ctx->orig_flow, ctx->packet,
5640 VLOG_ERR("Trace triggered by excessive resubmit "
5641 "recursion:\n%s", ds_cstr(&ds));
5646 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
5647 ctx->odp_actions->data,
5648 ctx->odp_actions->size)) {
5649 ctx->slow |= SLOW_IN_BAND;
5651 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
5653 compose_output_action(ctx, OFPP_LOCAL);
5656 if (ctx->ofproto->has_mirrors) {
5657 add_mirror_actions(ctx, &ctx->orig_flow);
5659 fix_sflow_action(ctx);
5663 /* Translates the 'n_in' "union ofp_action"s in 'in' into datapath actions,
5664 * using 'ctx', and discards the datapath actions. */
5666 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
5667 const union ofp_action *in, size_t n_in)
5669 uint64_t odp_actions_stub[1024 / 8];
5670 struct ofpbuf odp_actions;
5672 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5673 xlate_actions(ctx, in, n_in, &odp_actions);
5674 ofpbuf_uninit(&odp_actions);
5677 /* OFPP_NORMAL implementation. */
5679 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
5681 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
5682 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
5683 * the bundle on which the packet was received, returns the VLAN to which the
5686 * Both 'vid' and the return value are in the range 0...4095. */
5688 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
5690 switch (in_bundle->vlan_mode) {
5691 case PORT_VLAN_ACCESS:
5692 return in_bundle->vlan;
5695 case PORT_VLAN_TRUNK:
5698 case PORT_VLAN_NATIVE_UNTAGGED:
5699 case PORT_VLAN_NATIVE_TAGGED:
5700 return vid ? vid : in_bundle->vlan;
5707 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
5708 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
5711 * 'vid' should be the VID obtained from the 802.1Q header that was received as
5712 * part of a packet (specify 0 if there was no 802.1Q header), in the range
5715 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
5717 /* Allow any VID on the OFPP_NONE port. */
5718 if (in_bundle == &ofpp_none_bundle) {
5722 switch (in_bundle->vlan_mode) {
5723 case PORT_VLAN_ACCESS:
5726 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5727 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
5728 "packet received on port %s configured as VLAN "
5729 "%"PRIu16" access port",
5730 in_bundle->ofproto->up.name, vid,
5731 in_bundle->name, in_bundle->vlan);
5737 case PORT_VLAN_NATIVE_UNTAGGED:
5738 case PORT_VLAN_NATIVE_TAGGED:
5740 /* Port must always carry its native VLAN. */
5744 case PORT_VLAN_TRUNK:
5745 if (!ofbundle_includes_vlan(in_bundle, vid)) {
5747 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5748 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
5749 "received on port %s not configured for trunking "
5751 in_bundle->ofproto->up.name, vid,
5752 in_bundle->name, vid);
5764 /* Given 'vlan', the VLAN that a packet belongs to, and
5765 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
5766 * that should be included in the 802.1Q header. (If the return value is 0,
5767 * then the 802.1Q header should only be included in the packet if there is a
5770 * Both 'vlan' and the return value are in the range 0...4095. */
5772 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
5774 switch (out_bundle->vlan_mode) {
5775 case PORT_VLAN_ACCESS:
5778 case PORT_VLAN_TRUNK:
5779 case PORT_VLAN_NATIVE_TAGGED:
5782 case PORT_VLAN_NATIVE_UNTAGGED:
5783 return vlan == out_bundle->vlan ? 0 : vlan;
5791 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
5794 struct ofport_dpif *port;
5796 ovs_be16 tci, old_tci;
5798 vid = output_vlan_to_vid(out_bundle, vlan);
5799 if (!out_bundle->bond) {
5800 port = ofbundle_get_a_port(out_bundle);
5802 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
5805 /* No slaves enabled, so drop packet. */
5810 old_tci = ctx->flow.vlan_tci;
5812 if (tci || out_bundle->use_priority_tags) {
5813 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
5815 tci |= htons(VLAN_CFI);
5818 ctx->flow.vlan_tci = tci;
5820 compose_output_action(ctx, port->up.ofp_port);
5821 ctx->flow.vlan_tci = old_tci;
5825 mirror_mask_ffs(mirror_mask_t mask)
5827 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
5832 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
5834 return (bundle->vlan_mode != PORT_VLAN_ACCESS
5835 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
5839 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
5841 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
5844 /* Returns an arbitrary interface within 'bundle'. */
5845 static struct ofport_dpif *
5846 ofbundle_get_a_port(const struct ofbundle *bundle)
5848 return CONTAINER_OF(list_front(&bundle->ports),
5849 struct ofport_dpif, bundle_node);
5853 vlan_is_mirrored(const struct ofmirror *m, int vlan)
5855 return !m->vlans || bitmap_is_set(m->vlans, vlan);
5858 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
5859 * to a VLAN. In general most packets may be mirrored but we want to drop
5860 * protocols that may confuse switches. */
5862 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
5864 /* If you change this function's behavior, please update corresponding
5865 * documentation in vswitch.xml at the same time. */
5866 if (dst[0] != 0x01) {
5867 /* All the currently banned MACs happen to start with 01 currently, so
5868 * this is a quick way to eliminate most of the good ones. */
5870 if (eth_addr_is_reserved(dst)) {
5871 /* Drop STP, IEEE pause frames, and other reserved protocols
5872 * (01-80-c2-00-00-0x). */
5876 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
5878 if ((dst[3] & 0xfe) == 0xcc &&
5879 (dst[4] & 0xfe) == 0xcc &&
5880 (dst[5] & 0xfe) == 0xcc) {
5881 /* Drop the following protocols plus others following the same
5884 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
5885 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
5886 STP Uplink Fast (01-00-0c-cd-cd-cd) */
5890 if (!(dst[3] | dst[4] | dst[5])) {
5891 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
5900 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5902 struct ofproto_dpif *ofproto = ctx->ofproto;
5903 mirror_mask_t mirrors;
5904 struct ofbundle *in_bundle;
5907 const struct nlattr *a;
5910 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5911 ctx->packet != NULL, NULL);
5915 mirrors = in_bundle->src_mirrors;
5917 /* Drop frames on bundles reserved for mirroring. */
5918 if (in_bundle->mirror_out) {
5919 if (ctx->packet != NULL) {
5920 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5921 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5922 "%s, which is reserved exclusively for mirroring",
5923 ctx->ofproto->up.name, in_bundle->name);
5929 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5930 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5933 vlan = input_vid_to_vlan(in_bundle, vid);
5935 /* Look at the output ports to check for destination selections. */
5937 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5938 ctx->odp_actions->size) {
5939 enum ovs_action_attr type = nl_attr_type(a);
5940 struct ofport_dpif *ofport;
5942 if (type != OVS_ACTION_ATTR_OUTPUT) {
5946 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5947 if (ofport && ofport->bundle) {
5948 mirrors |= ofport->bundle->dst_mirrors;
5956 /* Restore the original packet before adding the mirror actions. */
5957 ctx->flow = *orig_flow;
5962 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5964 if (!vlan_is_mirrored(m, vlan)) {
5965 mirrors &= mirrors - 1;
5969 mirrors &= ~m->dup_mirrors;
5970 ctx->mirrors |= m->dup_mirrors;
5972 output_normal(ctx, m->out, vlan);
5973 } else if (eth_dst_may_rspan(orig_flow->dl_dst)
5974 && vlan != m->out_vlan) {
5975 struct ofbundle *bundle;
5977 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5978 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5979 && !bundle->mirror_out) {
5980 output_normal(ctx, bundle, m->out_vlan);
5988 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5989 uint64_t packets, uint64_t bytes)
5995 for (; mirrors; mirrors &= mirrors - 1) {
5998 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6001 /* In normal circumstances 'm' will not be NULL. However,
6002 * if mirrors are reconfigured, we can temporarily get out
6003 * of sync in facet_revalidate(). We could "correct" the
6004 * mirror list before reaching here, but doing that would
6005 * not properly account the traffic stats we've currently
6006 * accumulated for previous mirror configuration. */
6010 m->packet_count += packets;
6011 m->byte_count += bytes;
6015 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
6016 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
6017 * indicate this; newer upstream kernels use gratuitous ARP requests. */
6019 is_gratuitous_arp(const struct flow *flow)
6021 return (flow->dl_type == htons(ETH_TYPE_ARP)
6022 && eth_addr_is_broadcast(flow->dl_dst)
6023 && (flow->nw_proto == ARP_OP_REPLY
6024 || (flow->nw_proto == ARP_OP_REQUEST
6025 && flow->nw_src == flow->nw_dst)));
6029 update_learning_table(struct ofproto_dpif *ofproto,
6030 const struct flow *flow, int vlan,
6031 struct ofbundle *in_bundle)
6033 struct mac_entry *mac;
6035 /* Don't learn the OFPP_NONE port. */
6036 if (in_bundle == &ofpp_none_bundle) {
6040 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
6044 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
6045 if (is_gratuitous_arp(flow)) {
6046 /* We don't want to learn from gratuitous ARP packets that are
6047 * reflected back over bond slaves so we lock the learning table. */
6048 if (!in_bundle->bond) {
6049 mac_entry_set_grat_arp_lock(mac);
6050 } else if (mac_entry_is_grat_arp_locked(mac)) {
6055 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
6056 /* The log messages here could actually be useful in debugging,
6057 * so keep the rate limit relatively high. */
6058 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
6059 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
6060 "on port %s in VLAN %d",
6061 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
6062 in_bundle->name, vlan);
6064 mac->port.p = in_bundle;
6065 tag_set_add(&ofproto->revalidate_set,
6066 mac_learning_changed(ofproto->ml, mac));
6070 static struct ofbundle *
6071 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
6072 bool warn, struct ofport_dpif **in_ofportp)
6074 struct ofport_dpif *ofport;
6076 /* Find the port and bundle for the received packet. */
6077 ofport = get_ofp_port(ofproto, in_port);
6079 *in_ofportp = ofport;
6081 if (ofport && ofport->bundle) {
6082 return ofport->bundle;
6085 /* Special-case OFPP_NONE, which a controller may use as the ingress
6086 * port for traffic that it is sourcing. */
6087 if (in_port == OFPP_NONE) {
6088 return &ofpp_none_bundle;
6091 /* Odd. A few possible reasons here:
6093 * - We deleted a port but there are still a few packets queued up
6096 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
6097 * we don't know about.
6099 * - The ofproto client didn't configure the port as part of a bundle.
6102 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6104 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
6105 "port %"PRIu16, ofproto->up.name, in_port);
6110 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
6111 * dropped. Returns true if they may be forwarded, false if they should be
6114 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
6115 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
6117 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
6118 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
6119 * checked by input_vid_is_valid().
6121 * May also add tags to '*tags', although the current implementation only does
6122 * so in one special case.
6125 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
6126 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
6128 struct ofbundle *in_bundle = in_port->bundle;
6130 /* Drop frames for reserved multicast addresses
6131 * only if forward_bpdu option is absent. */
6132 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
6136 if (in_bundle->bond) {
6137 struct mac_entry *mac;
6139 switch (bond_check_admissibility(in_bundle->bond, in_port,
6140 flow->dl_dst, tags)) {
6147 case BV_DROP_IF_MOVED:
6148 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
6149 if (mac && mac->port.p != in_bundle &&
6150 (!is_gratuitous_arp(flow)
6151 || mac_entry_is_grat_arp_locked(mac))) {
6162 xlate_normal(struct action_xlate_ctx *ctx)
6164 struct ofport_dpif *in_port;
6165 struct ofbundle *in_bundle;
6166 struct mac_entry *mac;
6170 ctx->has_normal = true;
6172 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
6173 ctx->packet != NULL, &in_port);
6178 /* Drop malformed frames. */
6179 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
6180 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
6181 if (ctx->packet != NULL) {
6182 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6183 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
6184 "VLAN tag received on port %s",
6185 ctx->ofproto->up.name, in_bundle->name);
6190 /* Drop frames on bundles reserved for mirroring. */
6191 if (in_bundle->mirror_out) {
6192 if (ctx->packet != NULL) {
6193 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6194 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6195 "%s, which is reserved exclusively for mirroring",
6196 ctx->ofproto->up.name, in_bundle->name);
6202 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
6203 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6206 vlan = input_vid_to_vlan(in_bundle, vid);
6208 /* Check other admissibility requirements. */
6210 !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
6214 /* Learn source MAC. */
6215 if (ctx->may_learn) {
6216 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
6219 /* Determine output bundle. */
6220 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
6223 if (mac->port.p != in_bundle) {
6224 output_normal(ctx, mac->port.p, vlan);
6227 struct ofbundle *bundle;
6229 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
6230 if (bundle != in_bundle
6231 && ofbundle_includes_vlan(bundle, vlan)
6232 && bundle->floodable
6233 && !bundle->mirror_out) {
6234 output_normal(ctx, bundle, vlan);
6237 ctx->nf_output_iface = NF_OUT_FLOOD;
6241 /* Optimized flow revalidation.
6243 * It's a difficult problem, in general, to tell which facets need to have
6244 * their actions recalculated whenever the OpenFlow flow table changes. We
6245 * don't try to solve that general problem: for most kinds of OpenFlow flow
6246 * table changes, we recalculate the actions for every facet. This is
6247 * relatively expensive, but it's good enough if the OpenFlow flow table
6248 * doesn't change very often.
6250 * However, we can expect one particular kind of OpenFlow flow table change to
6251 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
6252 * of CPU on revalidating every facet whenever MAC learning modifies the flow
6253 * table, we add a special case that applies to flow tables in which every rule
6254 * has the same form (that is, the same wildcards), except that the table is
6255 * also allowed to have a single "catch-all" flow that matches all packets. We
6256 * optimize this case by tagging all of the facets that resubmit into the table
6257 * and invalidating the same tag whenever a flow changes in that table. The
6258 * end result is that we revalidate just the facets that need it (and sometimes
6259 * a few more, but not all of the facets or even all of the facets that
6260 * resubmit to the table modified by MAC learning). */
6262 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
6263 * into an OpenFlow table with the given 'basis'. */
6265 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
6268 if (flow_wildcards_is_catchall(wc)) {
6271 struct flow tag_flow = *flow;
6272 flow_zero_wildcards(&tag_flow, wc);
6273 return tag_create_deterministic(flow_hash(&tag_flow, secret));
6277 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
6278 * taggability of that table.
6280 * This function must be called after *each* change to a flow table. If you
6281 * skip calling it on some changes then the pointer comparisons at the end can
6282 * be invalid if you get unlucky. For example, if a flow removal causes a
6283 * cls_table to be destroyed and then a flow insertion causes a cls_table with
6284 * different wildcards to be created with the same address, then this function
6285 * will incorrectly skip revalidation. */
6287 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
6289 struct table_dpif *table = &ofproto->tables[table_id];
6290 const struct oftable *oftable = &ofproto->up.tables[table_id];
6291 struct cls_table *catchall, *other;
6292 struct cls_table *t;
6294 catchall = other = NULL;
6296 switch (hmap_count(&oftable->cls.tables)) {
6298 /* We could tag this OpenFlow table but it would make the logic a
6299 * little harder and it's a corner case that doesn't seem worth it
6305 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
6306 if (cls_table_is_catchall(t)) {
6308 } else if (!other) {
6311 /* Indicate that we can't tag this by setting both tables to
6312 * NULL. (We know that 'catchall' is already NULL.) */
6319 /* Can't tag this table. */
6323 if (table->catchall_table != catchall || table->other_table != other) {
6324 table->catchall_table = catchall;
6325 table->other_table = other;
6326 ofproto->need_revalidate = true;
6330 /* Given 'rule' that has changed in some way (either it is a rule being
6331 * inserted, a rule being deleted, or a rule whose actions are being
6332 * modified), marks facets for revalidation to ensure that packets will be
6333 * forwarded correctly according to the new state of the flow table.
6335 * This function must be called after *each* change to a flow table. See
6336 * the comment on table_update_taggable() for more information. */
6338 rule_invalidate(const struct rule_dpif *rule)
6340 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
6342 table_update_taggable(ofproto, rule->up.table_id);
6344 if (!ofproto->need_revalidate) {
6345 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
6347 if (table->other_table && rule->tag) {
6348 tag_set_add(&ofproto->revalidate_set, rule->tag);
6350 ofproto->need_revalidate = true;
6356 set_frag_handling(struct ofproto *ofproto_,
6357 enum ofp_config_flags frag_handling)
6359 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6361 if (frag_handling != OFPC_FRAG_REASM) {
6362 ofproto->need_revalidate = true;
6370 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
6371 const struct flow *flow,
6372 const union ofp_action *ofp_actions, size_t n_ofp_actions)
6374 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6377 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
6378 return OFPERR_NXBRC_BAD_IN_PORT;
6381 error = validate_actions(ofp_actions, n_ofp_actions, flow,
6382 ofproto->max_ports);
6384 struct odputil_keybuf keybuf;
6385 struct dpif_flow_stats stats;
6389 struct action_xlate_ctx ctx;
6390 uint64_t odp_actions_stub[1024 / 8];
6391 struct ofpbuf odp_actions;
6393 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
6394 odp_flow_key_from_flow(&key, flow);
6396 dpif_flow_stats_extract(flow, packet, &stats);
6398 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL,
6399 packet_get_tcp_flags(packet, flow), packet);
6400 ctx.resubmit_stats = &stats;
6402 ofpbuf_use_stub(&odp_actions,
6403 odp_actions_stub, sizeof odp_actions_stub);
6404 xlate_actions(&ctx, ofp_actions, n_ofp_actions, &odp_actions);
6405 dpif_execute(ofproto->dpif, key.data, key.size,
6406 odp_actions.data, odp_actions.size, packet);
6407 ofpbuf_uninit(&odp_actions);
6415 set_netflow(struct ofproto *ofproto_,
6416 const struct netflow_options *netflow_options)
6418 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6420 if (netflow_options) {
6421 if (!ofproto->netflow) {
6422 ofproto->netflow = netflow_create();
6424 return netflow_set_options(ofproto->netflow, netflow_options);
6426 netflow_destroy(ofproto->netflow);
6427 ofproto->netflow = NULL;
6433 get_netflow_ids(const struct ofproto *ofproto_,
6434 uint8_t *engine_type, uint8_t *engine_id)
6436 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6438 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
6442 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
6444 if (!facet_is_controller_flow(facet) &&
6445 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
6446 struct subfacet *subfacet;
6447 struct ofexpired expired;
6449 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
6450 if (subfacet->path == SF_FAST_PATH) {
6451 struct dpif_flow_stats stats;
6453 subfacet_reinstall(subfacet, &stats);
6454 subfacet_update_stats(subfacet, &stats);
6458 expired.flow = facet->flow;
6459 expired.packet_count = facet->packet_count;
6460 expired.byte_count = facet->byte_count;
6461 expired.used = facet->used;
6462 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
6467 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
6469 struct facet *facet;
6471 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6472 send_active_timeout(ofproto, facet);
6476 static struct ofproto_dpif *
6477 ofproto_dpif_lookup(const char *name)
6479 struct ofproto_dpif *ofproto;
6481 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
6482 hash_string(name, 0), &all_ofproto_dpifs) {
6483 if (!strcmp(ofproto->up.name, name)) {
6491 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
6492 const char *argv[], void *aux OVS_UNUSED)
6494 struct ofproto_dpif *ofproto;
6497 ofproto = ofproto_dpif_lookup(argv[1]);
6499 unixctl_command_reply_error(conn, "no such bridge");
6502 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6504 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6505 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6509 unixctl_command_reply(conn, "table successfully flushed");
6513 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6514 const char *argv[], void *aux OVS_UNUSED)
6516 struct ds ds = DS_EMPTY_INITIALIZER;
6517 const struct ofproto_dpif *ofproto;
6518 const struct mac_entry *e;
6520 ofproto = ofproto_dpif_lookup(argv[1]);
6522 unixctl_command_reply_error(conn, "no such bridge");
6526 ds_put_cstr(&ds, " port VLAN MAC Age\n");
6527 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
6528 struct ofbundle *bundle = e->port.p;
6529 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
6530 ofbundle_get_a_port(bundle)->odp_port,
6531 e->vlan, ETH_ADDR_ARGS(e->mac),
6532 mac_entry_age(ofproto->ml, e));
6534 unixctl_command_reply(conn, ds_cstr(&ds));
6539 struct action_xlate_ctx ctx;
6545 trace_format_rule(struct ds *result, uint8_t table_id, int level,
6546 const struct rule_dpif *rule)
6548 ds_put_char_multiple(result, '\t', level);
6550 ds_put_cstr(result, "No match\n");
6554 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
6555 table_id, ntohll(rule->up.flow_cookie));
6556 cls_rule_format(&rule->up.cr, result);
6557 ds_put_char(result, '\n');
6559 ds_put_char_multiple(result, '\t', level);
6560 ds_put_cstr(result, "OpenFlow ");
6561 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
6562 ds_put_char(result, '\n');
6566 trace_format_flow(struct ds *result, int level, const char *title,
6567 struct trace_ctx *trace)
6569 ds_put_char_multiple(result, '\t', level);
6570 ds_put_format(result, "%s: ", title);
6571 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
6572 ds_put_cstr(result, "unchanged");
6574 flow_format(result, &trace->ctx.flow);
6575 trace->flow = trace->ctx.flow;
6577 ds_put_char(result, '\n');
6581 trace_format_regs(struct ds *result, int level, const char *title,
6582 struct trace_ctx *trace)
6586 ds_put_char_multiple(result, '\t', level);
6587 ds_put_format(result, "%s:", title);
6588 for (i = 0; i < FLOW_N_REGS; i++) {
6589 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
6591 ds_put_char(result, '\n');
6595 trace_format_odp(struct ds *result, int level, const char *title,
6596 struct trace_ctx *trace)
6598 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
6600 ds_put_char_multiple(result, '\t', level);
6601 ds_put_format(result, "%s: ", title);
6602 format_odp_actions(result, odp_actions->data, odp_actions->size);
6603 ds_put_char(result, '\n');
6607 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
6609 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
6610 struct ds *result = trace->result;
6612 ds_put_char(result, '\n');
6613 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
6614 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
6615 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
6616 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
6620 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
6621 void *aux OVS_UNUSED)
6623 const char *dpname = argv[1];
6624 struct ofproto_dpif *ofproto;
6625 struct ofpbuf odp_key;
6626 struct ofpbuf *packet;
6627 ovs_be16 initial_tci;
6633 ofpbuf_init(&odp_key, 0);
6636 ofproto = ofproto_dpif_lookup(dpname);
6638 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
6642 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
6643 /* ofproto/trace dpname flow [-generate] */
6644 const char *flow_s = argv[2];
6645 const char *generate_s = argv[3];
6647 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
6648 * flow. We guess which type it is based on whether 'flow_s' contains
6649 * an '(', since a datapath flow always contains '(') but an
6650 * OpenFlow-like flow should not (in fact it's allowed but I believe
6651 * that's not documented anywhere).
6653 * An alternative would be to try to parse 'flow_s' both ways, but then
6654 * it would be tricky giving a sensible error message. After all, do
6655 * you just say "syntax error" or do you present both error messages?
6656 * Both choices seem lousy. */
6657 if (strchr(flow_s, '(')) {
6660 /* Convert string to datapath key. */
6661 ofpbuf_init(&odp_key, 0);
6662 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
6664 unixctl_command_reply_error(conn, "Bad flow syntax");
6668 /* Convert odp_key to flow. */
6669 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
6670 odp_key.size, &flow,
6671 &initial_tci, NULL);
6672 if (error == ODP_FIT_ERROR) {
6673 unixctl_command_reply_error(conn, "Invalid flow");
6679 error_s = parse_ofp_exact_flow(&flow, argv[2]);
6681 unixctl_command_reply_error(conn, error_s);
6686 initial_tci = flow.vlan_tci;
6687 vsp_adjust_flow(ofproto, &flow);
6690 /* Generate a packet, if requested. */
6692 packet = ofpbuf_new(0);
6693 flow_compose(packet, &flow);
6695 } else if (argc == 6) {
6696 /* ofproto/trace dpname priority tun_id in_port packet */
6697 const char *priority_s = argv[2];
6698 const char *tun_id_s = argv[3];
6699 const char *in_port_s = argv[4];
6700 const char *packet_s = argv[5];
6701 uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s));
6702 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
6703 uint32_t priority = atoi(priority_s);
6706 msg = eth_from_hex(packet_s, &packet);
6708 unixctl_command_reply_error(conn, msg);
6712 ds_put_cstr(&result, "Packet: ");
6713 s = ofp_packet_to_string(packet->data, packet->size);
6714 ds_put_cstr(&result, s);
6717 flow_extract(packet, priority, tun_id, in_port, &flow);
6718 initial_tci = flow.vlan_tci;
6720 unixctl_command_reply_error(conn, "Bad command syntax");
6724 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
6725 unixctl_command_reply(conn, ds_cstr(&result));
6728 ds_destroy(&result);
6729 ofpbuf_delete(packet);
6730 ofpbuf_uninit(&odp_key);
6734 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
6735 const struct ofpbuf *packet, ovs_be16 initial_tci,
6738 struct rule_dpif *rule;
6740 ds_put_cstr(ds, "Flow: ");
6741 flow_format(ds, flow);
6742 ds_put_char(ds, '\n');
6744 rule = rule_dpif_lookup(ofproto, flow);
6746 trace_format_rule(ds, 0, 0, rule);
6747 if (rule == ofproto->miss_rule) {
6748 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
6749 } else if (rule == ofproto->no_packet_in_rule) {
6750 ds_put_cstr(ds, "\nNo match, packets dropped because "
6751 "OFPPC_NO_PACKET_IN is set on in_port.\n");
6755 uint64_t odp_actions_stub[1024 / 8];
6756 struct ofpbuf odp_actions;
6758 struct trace_ctx trace;
6761 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
6764 ofpbuf_use_stub(&odp_actions,
6765 odp_actions_stub, sizeof odp_actions_stub);
6766 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
6767 rule, tcp_flags, packet);
6768 trace.ctx.resubmit_hook = trace_resubmit;
6769 xlate_actions(&trace.ctx, rule->up.actions, rule->up.n_actions,
6772 ds_put_char(ds, '\n');
6773 trace_format_flow(ds, 0, "Final flow", &trace);
6774 ds_put_cstr(ds, "Datapath actions: ");
6775 format_odp_actions(ds, odp_actions.data, odp_actions.size);
6776 ofpbuf_uninit(&odp_actions);
6778 if (trace.ctx.slow) {
6779 enum slow_path_reason slow;
6781 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
6782 "slow path because it:");
6783 for (slow = trace.ctx.slow; slow; ) {
6784 enum slow_path_reason bit = rightmost_1bit(slow);
6788 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
6791 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
6794 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
6797 ds_put_cstr(ds, "\n\t- Needs in-band special case "
6800 ds_put_cstr(ds, "\n\t (The datapath actions are "
6801 "incomplete--for complete actions, "
6802 "please supply a packet.)");
6805 case SLOW_CONTROLLER:
6806 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
6807 "to the OpenFlow controller.");
6810 ds_put_cstr(ds, "\n\t- Needs more specific matching "
6811 "than the datapath supports.");
6818 if (slow & ~SLOW_MATCH) {
6819 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
6820 "the special slow-path processing.");
6827 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6828 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6831 unixctl_command_reply(conn, NULL);
6835 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6836 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6839 unixctl_command_reply(conn, NULL);
6842 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
6843 * 'reply' describing the results. */
6845 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
6847 struct facet *facet;
6851 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6852 if (!facet_check_consistency(facet)) {
6857 ofproto->need_revalidate = true;
6861 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
6862 ofproto->up.name, errors);
6864 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
6869 ofproto_dpif_self_check(struct unixctl_conn *conn,
6870 int argc, const char *argv[], void *aux OVS_UNUSED)
6872 struct ds reply = DS_EMPTY_INITIALIZER;
6873 struct ofproto_dpif *ofproto;
6876 ofproto = ofproto_dpif_lookup(argv[1]);
6878 unixctl_command_reply_error(conn, "Unknown ofproto (use "
6879 "ofproto/list for help)");
6882 ofproto_dpif_self_check__(ofproto, &reply);
6884 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6885 ofproto_dpif_self_check__(ofproto, &reply);
6889 unixctl_command_reply(conn, ds_cstr(&reply));
6894 ofproto_dpif_unixctl_init(void)
6896 static bool registered;
6902 unixctl_command_register(
6904 "bridge {tun_id in_port packet | odp_flow [-generate]}",
6905 2, 5, ofproto_unixctl_trace, NULL);
6906 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6907 ofproto_unixctl_fdb_flush, NULL);
6908 unixctl_command_register("fdb/show", "bridge", 1, 1,
6909 ofproto_unixctl_fdb_show, NULL);
6910 unixctl_command_register("ofproto/clog", "", 0, 0,
6911 ofproto_dpif_clog, NULL);
6912 unixctl_command_register("ofproto/unclog", "", 0, 0,
6913 ofproto_dpif_unclog, NULL);
6914 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6915 ofproto_dpif_self_check, NULL);
6918 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6920 * This is deprecated. It is only for compatibility with broken device drivers
6921 * in old versions of Linux that do not properly support VLANs when VLAN
6922 * devices are not used. When broken device drivers are no longer in
6923 * widespread use, we will delete these interfaces. */
6926 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
6928 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6929 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6931 if (realdev_ofp_port == ofport->realdev_ofp_port
6932 && vid == ofport->vlandev_vid) {
6936 ofproto->need_revalidate = true;
6938 if (ofport->realdev_ofp_port) {
6941 if (realdev_ofp_port && ofport->bundle) {
6942 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6943 * themselves be part of a bundle. */
6944 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6947 ofport->realdev_ofp_port = realdev_ofp_port;
6948 ofport->vlandev_vid = vid;
6950 if (realdev_ofp_port) {
6951 vsp_add(ofport, realdev_ofp_port, vid);
6958 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
6960 return hash_2words(realdev_ofp_port, vid);
6963 /* Returns the ODP port number of the Linux VLAN device that corresponds to
6964 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
6965 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
6966 * it would return the port number of eth0.9.
6968 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
6969 * function just returns its 'realdev_odp_port' argument. */
6971 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6972 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
6974 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6975 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
6976 int vid = vlan_tci_to_vid(vlan_tci);
6977 const struct vlan_splinter *vsp;
6979 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6980 hash_realdev_vid(realdev_ofp_port, vid),
6981 &ofproto->realdev_vid_map) {
6982 if (vsp->realdev_ofp_port == realdev_ofp_port
6983 && vsp->vid == vid) {
6984 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
6988 return realdev_odp_port;
6991 static struct vlan_splinter *
6992 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
6994 struct vlan_splinter *vsp;
6996 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
6997 &ofproto->vlandev_map) {
6998 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
7006 /* Returns the OpenFlow port number of the "real" device underlying the Linux
7007 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
7008 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
7009 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
7010 * eth0 and store 9 in '*vid'.
7012 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
7013 * VLAN device. Unless VLAN splinters are enabled, this is what this function
7016 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
7017 uint16_t vlandev_ofp_port, int *vid)
7019 if (!hmap_is_empty(&ofproto->vlandev_map)) {
7020 const struct vlan_splinter *vsp;
7022 vsp = vlandev_find(ofproto, vlandev_ofp_port);
7027 return vsp->realdev_ofp_port;
7033 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
7034 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
7035 * 'flow->in_port' to the "real" device backing the VLAN device, sets
7036 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
7037 * always the case unless VLAN splinters are enabled), returns false without
7038 * making any changes. */
7040 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
7045 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
7050 /* Cause the flow to be processed as if it came in on the real device with
7051 * the VLAN device's VLAN ID. */
7052 flow->in_port = realdev;
7053 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
7058 vsp_remove(struct ofport_dpif *port)
7060 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7061 struct vlan_splinter *vsp;
7063 vsp = vlandev_find(ofproto, port->up.ofp_port);
7065 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
7066 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
7069 port->realdev_ofp_port = 0;
7071 VLOG_ERR("missing vlan device record");
7076 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
7078 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7080 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
7081 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
7082 == realdev_ofp_port)) {
7083 struct vlan_splinter *vsp;
7085 vsp = xmalloc(sizeof *vsp);
7086 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
7087 hash_int(port->up.ofp_port, 0));
7088 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
7089 hash_realdev_vid(realdev_ofp_port, vid));
7090 vsp->realdev_ofp_port = realdev_ofp_port;
7091 vsp->vlandev_ofp_port = port->up.ofp_port;
7094 port->realdev_ofp_port = realdev_ofp_port;
7096 VLOG_ERR("duplicate vlan device record");
7100 const struct ofproto_class ofproto_dpif_class = {
7130 port_is_lacp_current,
7131 NULL, /* rule_choose_table */
7138 rule_modify_actions,
7146 get_cfm_remote_mpids,
7151 get_stp_port_status,
7158 is_mirror_output_bundle,
7159 forward_bpdu_changed,