2 * Copyright (c) 2009, 2010, 2011, 2012 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-provider.h"
26 #include "byte-order.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
36 #include "mac-learning.h"
37 #include "meta-flow.h"
38 #include "multipath.h"
45 #include "ofp-parse.h"
46 #include "ofp-print.h"
47 #include "ofproto-dpif-governor.h"
48 #include "ofproto-dpif-sflow.h"
49 #include "poll-loop.h"
51 #include "unaligned.h"
53 #include "vlan-bitmap.h"
56 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
58 COVERAGE_DEFINE(ofproto_dpif_ctlr_action);
59 COVERAGE_DEFINE(ofproto_dpif_expired);
60 COVERAGE_DEFINE(ofproto_dpif_xlate);
61 COVERAGE_DEFINE(facet_changed_rule);
62 COVERAGE_DEFINE(facet_invalidated);
63 COVERAGE_DEFINE(facet_revalidate);
64 COVERAGE_DEFINE(facet_unexpected);
65 COVERAGE_DEFINE(facet_suppress);
67 /* Maximum depth of flow table recursion (due to resubmit actions) in a
68 * flow translation. */
69 #define MAX_RESUBMIT_RECURSION 32
71 /* Number of implemented OpenFlow tables. */
72 enum { N_TABLES = 255 };
73 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
74 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
84 * - Do include packets and bytes from facets that have been deleted or
85 * whose own statistics have been folded into the rule.
87 * - Do include packets and bytes sent "by hand" that were accounted to
88 * the rule without any facet being involved (this is a rare corner
89 * case in rule_execute()).
91 * - Do not include packet or bytes that can be obtained from any facet's
92 * packet_count or byte_count member or that can be obtained from the
93 * datapath by, e.g., dpif_flow_get() for any subfacet.
95 uint64_t packet_count; /* Number of packets received. */
96 uint64_t byte_count; /* Number of bytes received. */
98 tag_type tag; /* Caches rule_calculate_tag() result. */
100 struct list facets; /* List of "struct facet"s. */
103 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
105 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
108 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
109 const struct flow *);
110 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
114 static void rule_credit_stats(struct rule_dpif *,
115 const struct dpif_flow_stats *);
116 static void flow_push_stats(struct rule_dpif *, const struct flow *,
117 const struct dpif_flow_stats *);
118 static tag_type rule_calculate_tag(const struct flow *,
119 const struct flow_wildcards *,
121 static void rule_invalidate(const struct rule_dpif *);
123 #define MAX_MIRRORS 32
124 typedef uint32_t mirror_mask_t;
125 #define MIRROR_MASK_C(X) UINT32_C(X)
126 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
128 struct ofproto_dpif *ofproto; /* Owning ofproto. */
129 size_t idx; /* In ofproto's "mirrors" array. */
130 void *aux; /* Key supplied by ofproto's client. */
131 char *name; /* Identifier for log messages. */
133 /* Selection criteria. */
134 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
135 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
136 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
138 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
139 struct ofbundle *out; /* Output port or NULL. */
140 int out_vlan; /* Output VLAN or -1. */
141 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
144 int64_t packet_count; /* Number of packets sent. */
145 int64_t byte_count; /* Number of bytes sent. */
148 static void mirror_destroy(struct ofmirror *);
149 static void update_mirror_stats(struct ofproto_dpif *ofproto,
150 mirror_mask_t mirrors,
151 uint64_t packets, uint64_t bytes);
154 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
155 struct ofproto_dpif *ofproto; /* Owning ofproto. */
156 void *aux; /* Key supplied by ofproto's client. */
157 char *name; /* Identifier for log messages. */
160 struct list ports; /* Contains "struct ofport"s. */
161 enum port_vlan_mode vlan_mode; /* VLAN mode */
162 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
163 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
164 * NULL if all VLANs are trunked. */
165 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
166 struct bond *bond; /* Nonnull iff more than one port. */
167 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
170 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
172 /* Port mirroring info. */
173 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
174 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
175 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
178 static void bundle_remove(struct ofport *);
179 static void bundle_update(struct ofbundle *);
180 static void bundle_destroy(struct ofbundle *);
181 static void bundle_del_port(struct ofport_dpif *);
182 static void bundle_run(struct ofbundle *);
183 static void bundle_wait(struct ofbundle *);
184 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
185 uint16_t in_port, bool warn,
186 struct ofport_dpif **in_ofportp);
188 /* A controller may use OFPP_NONE as the ingress port to indicate that
189 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
190 * when an input bundle is needed for validation (e.g., mirroring or
191 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
192 * any 'port' structs, so care must be taken when dealing with it. */
193 static struct ofbundle ofpp_none_bundle = {
195 .vlan_mode = PORT_VLAN_TRUNK
198 static void stp_run(struct ofproto_dpif *ofproto);
199 static void stp_wait(struct ofproto_dpif *ofproto);
200 static int set_stp_port(struct ofport *,
201 const struct ofproto_port_stp_settings *);
203 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
205 struct action_xlate_ctx {
206 /* action_xlate_ctx_init() initializes these members. */
209 struct ofproto_dpif *ofproto;
211 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
212 * this flow when actions change header fields. */
215 /* The packet corresponding to 'flow', or a null pointer if we are
216 * revalidating without a packet to refer to. */
217 const struct ofpbuf *packet;
219 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
220 * actions update the flow table?
222 * We want to update these tables if we are actually processing a packet,
223 * or if we are accounting for packets that the datapath has processed, but
224 * not if we are just revalidating. */
227 /* The rule that we are currently translating, or NULL. */
228 struct rule_dpif *rule;
230 /* Union of the set of TCP flags seen so far in this flow. (Used only by
231 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
235 /* If nonnull, flow translation calls this function just before executing a
236 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
237 * when the recursion depth is exceeded.
239 * 'rule' is the rule being submitted into. It will be null if the
240 * resubmit or OFPP_TABLE action didn't find a matching rule.
242 * This is normally null so the client has to set it manually after
243 * calling action_xlate_ctx_init(). */
244 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule);
246 /* If nonnull, flow translation credits the specified statistics to each
247 * rule reached through a resubmit or OFPP_TABLE action.
249 * This is normally null so the client has to set it manually after
250 * calling action_xlate_ctx_init(). */
251 const struct dpif_flow_stats *resubmit_stats;
253 /* xlate_actions() initializes and uses these members. The client might want
254 * to look at them after it returns. */
256 struct ofpbuf *odp_actions; /* Datapath actions. */
257 tag_type tags; /* Tags associated with actions. */
258 enum slow_path_reason slow; /* 0 if fast path may be used. */
259 bool has_learn; /* Actions include NXAST_LEARN? */
260 bool has_normal; /* Actions output to OFPP_NORMAL? */
261 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
262 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
263 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
265 /* xlate_actions() initializes and uses these members, but the client has no
266 * reason to look at them. */
268 int recurse; /* Recursion level, via xlate_table_action. */
269 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
270 struct flow base_flow; /* Flow at the last commit. */
271 uint32_t orig_skb_priority; /* Priority when packet arrived. */
272 uint8_t table_id; /* OpenFlow table ID where flow was found. */
273 uint32_t sflow_n_outputs; /* Number of output ports. */
274 uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
275 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
276 bool exit; /* No further actions should be processed. */
277 struct flow orig_flow; /* Copy of original flow. */
280 static void action_xlate_ctx_init(struct action_xlate_ctx *,
281 struct ofproto_dpif *, const struct flow *,
282 ovs_be16 initial_tci, struct rule_dpif *,
283 uint8_t tcp_flags, const struct ofpbuf *);
284 static void xlate_actions(struct action_xlate_ctx *,
285 const union ofp_action *in, size_t n_in,
286 struct ofpbuf *odp_actions);
287 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
288 const union ofp_action *in,
291 static size_t put_userspace_action(const struct ofproto_dpif *,
292 struct ofpbuf *odp_actions,
294 const union user_action_cookie *);
296 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
297 enum slow_path_reason,
298 uint64_t *stub, size_t stub_size,
299 const struct nlattr **actionsp,
300 size_t *actions_lenp);
302 /* A subfacet (see "struct subfacet" below) has three possible installation
305 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
306 * case just after the subfacet is created, just before the subfacet is
307 * destroyed, or if the datapath returns an error when we try to install a
310 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
312 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
313 * ofproto_dpif is installed in the datapath.
316 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
317 SF_FAST_PATH, /* Full actions are installed. */
318 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
321 static const char *subfacet_path_to_string(enum subfacet_path);
323 /* A dpif flow and actions associated with a facet.
325 * See also the large comment on struct facet. */
328 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
329 struct list list_node; /* In struct facet's 'facets' list. */
330 struct facet *facet; /* Owning facet. */
334 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
335 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
336 * regenerate the ODP flow key from ->facet->flow. */
337 enum odp_key_fitness key_fitness;
341 long long int used; /* Time last used; time created if not used. */
343 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
344 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
348 * These should be essentially identical for every subfacet in a facet, but
349 * may differ in trivial ways due to VLAN splinters. */
350 size_t actions_len; /* Number of bytes in actions[]. */
351 struct nlattr *actions; /* Datapath actions. */
353 enum slow_path_reason slow; /* 0 if fast path may be used. */
354 enum subfacet_path path; /* Installed in datapath? */
356 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
357 * splinters can cause it to differ. This value should be removed when
358 * the VLAN splinters feature is no longer needed. */
359 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
362 static struct subfacet *subfacet_create(struct facet *, enum odp_key_fitness,
363 const struct nlattr *key,
364 size_t key_len, ovs_be16 initial_tci);
365 static struct subfacet *subfacet_find(struct ofproto_dpif *,
366 const struct nlattr *key, size_t key_len);
367 static void subfacet_destroy(struct subfacet *);
368 static void subfacet_destroy__(struct subfacet *);
369 static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *,
371 static void subfacet_reset_dp_stats(struct subfacet *,
372 struct dpif_flow_stats *);
373 static void subfacet_update_time(struct subfacet *, long long int used);
374 static void subfacet_update_stats(struct subfacet *,
375 const struct dpif_flow_stats *);
376 static void subfacet_make_actions(struct subfacet *,
377 const struct ofpbuf *packet,
378 struct ofpbuf *odp_actions);
379 static int subfacet_install(struct subfacet *,
380 const struct nlattr *actions, size_t actions_len,
381 struct dpif_flow_stats *, enum slow_path_reason);
382 static void subfacet_uninstall(struct subfacet *);
384 static enum subfacet_path subfacet_want_path(enum slow_path_reason);
386 /* An exact-match instantiation of an OpenFlow flow.
388 * A facet associates a "struct flow", which represents the Open vSwitch
389 * userspace idea of an exact-match flow, with one or more subfacets. Each
390 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
391 * the facet. When the kernel module (or other dpif implementation) and Open
392 * vSwitch userspace agree on the definition of a flow key, there is exactly
393 * one subfacet per facet. If the dpif implementation supports more-specific
394 * flow matching than userspace, however, a facet can have more than one
395 * subfacet, each of which corresponds to some distinction in flow that
396 * userspace simply doesn't understand.
398 * Flow expiration works in terms of subfacets, so a facet must have at least
399 * one subfacet or it will never expire, leaking memory. */
402 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
403 struct list list_node; /* In owning rule's 'facets' list. */
404 struct rule_dpif *rule; /* Owning rule. */
407 struct list subfacets;
408 long long int used; /* Time last used; time created if not used. */
415 * - Do include packets and bytes sent "by hand", e.g. with
418 * - Do include packets and bytes that were obtained from the datapath
419 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
420 * DPIF_FP_ZERO_STATS).
422 * - Do not include packets or bytes that can be obtained from the
423 * datapath for any existing subfacet.
425 uint64_t packet_count; /* Number of packets received. */
426 uint64_t byte_count; /* Number of bytes received. */
428 /* Resubmit statistics. */
429 uint64_t prev_packet_count; /* Number of packets from last stats push. */
430 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
431 long long int prev_used; /* Used time from last stats push. */
434 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
435 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
436 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
438 /* Properties of datapath actions.
440 * Every subfacet has its own actions because actions can differ slightly
441 * between splintered and non-splintered subfacets due to the VLAN tag
442 * being initially different (present vs. absent). All of them have these
443 * properties in common so we just store one copy of them here. */
444 bool has_learn; /* Actions include NXAST_LEARN? */
445 bool has_normal; /* Actions output to OFPP_NORMAL? */
446 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
447 tag_type tags; /* Tags that would require revalidation. */
448 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
450 /* Storage for a single subfacet, to reduce malloc() time and space
451 * overhead. (A facet always has at least one subfacet and in the common
452 * case has exactly one subfacet.) */
453 struct subfacet one_subfacet;
456 static struct facet *facet_create(struct rule_dpif *,
457 const struct flow *, uint32_t hash);
458 static void facet_remove(struct facet *);
459 static void facet_free(struct facet *);
461 static struct facet *facet_find(struct ofproto_dpif *,
462 const struct flow *, uint32_t hash);
463 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
464 const struct flow *, uint32_t hash);
465 static void facet_revalidate(struct facet *);
466 static bool facet_check_consistency(struct facet *);
468 static void facet_flush_stats(struct facet *);
470 static void facet_update_time(struct facet *, long long int used);
471 static void facet_reset_counters(struct facet *);
472 static void facet_push_stats(struct facet *);
473 static void facet_learn(struct facet *);
474 static void facet_account(struct facet *);
476 static bool facet_is_controller_flow(struct facet *);
482 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
483 struct list bundle_node; /* In struct ofbundle's "ports" list. */
484 struct cfm *cfm; /* Connectivity Fault Management, if any. */
485 tag_type tag; /* Tag associated with this port. */
486 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
487 bool may_enable; /* May be enabled in bonds. */
488 long long int carrier_seq; /* Carrier status changes. */
491 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
492 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
493 long long int stp_state_entered;
495 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
497 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
499 * This is deprecated. It is only for compatibility with broken device
500 * drivers in old versions of Linux that do not properly support VLANs when
501 * VLAN devices are not used. When broken device drivers are no longer in
502 * widespread use, we will delete these interfaces. */
503 uint16_t realdev_ofp_port;
507 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
508 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
509 * traffic egressing the 'ofport' with that priority should be marked with. */
510 struct priority_to_dscp {
511 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
512 uint32_t priority; /* Priority of this queue (see struct flow). */
514 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
517 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
519 * This is deprecated. It is only for compatibility with broken device drivers
520 * in old versions of Linux that do not properly support VLANs when VLAN
521 * devices are not used. When broken device drivers are no longer in
522 * widespread use, we will delete these interfaces. */
523 struct vlan_splinter {
524 struct hmap_node realdev_vid_node;
525 struct hmap_node vlandev_node;
526 uint16_t realdev_ofp_port;
527 uint16_t vlandev_ofp_port;
531 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
532 uint32_t realdev, ovs_be16 vlan_tci);
533 static uint16_t vsp_vlandev_to_realdev(const struct ofproto_dpif *,
534 uint16_t vlandev, int *vid);
535 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
536 static void vsp_remove(struct ofport_dpif *);
537 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
539 static struct ofport_dpif *
540 ofport_dpif_cast(const struct ofport *ofport)
542 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
543 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
546 static void port_run(struct ofport_dpif *);
547 static void port_wait(struct ofport_dpif *);
548 static int set_cfm(struct ofport *, const struct cfm_settings *);
549 static void ofport_clear_priorities(struct ofport_dpif *);
551 struct dpif_completion {
552 struct list list_node;
553 struct ofoperation *op;
556 /* Extra information about a classifier table.
557 * Currently used just for optimized flow revalidation. */
559 /* If either of these is nonnull, then this table has a form that allows
560 * flows to be tagged to avoid revalidating most flows for the most common
561 * kinds of flow table changes. */
562 struct cls_table *catchall_table; /* Table that wildcards all fields. */
563 struct cls_table *other_table; /* Table with any other wildcard set. */
564 uint32_t basis; /* Keeps each table's tags separate. */
567 struct ofproto_dpif {
568 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
573 /* Special OpenFlow rules. */
574 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
575 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
581 struct netflow *netflow;
582 struct dpif_sflow *sflow;
583 struct hmap bundles; /* Contains "struct ofbundle"s. */
584 struct mac_learning *ml;
585 struct ofmirror *mirrors[MAX_MIRRORS];
587 bool has_bonded_bundles;
590 struct timer next_expiration;
594 struct hmap subfacets;
595 struct governor *governor;
598 struct table_dpif tables[N_TABLES];
599 bool need_revalidate;
600 struct tag_set revalidate_set;
602 /* Support for debugging async flow mods. */
603 struct list completions;
605 bool has_bundle_action; /* True when the first bundle action appears. */
606 struct netdev_stats stats; /* To account packets generated and consumed in
611 long long int stp_last_tick;
613 /* VLAN splinters. */
614 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
615 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
618 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
619 * for debugging the asynchronous flow_mod implementation.) */
622 /* All existing ofproto_dpif instances, indexed by ->up.name. */
623 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
625 static void ofproto_dpif_unixctl_init(void);
627 static struct ofproto_dpif *
628 ofproto_dpif_cast(const struct ofproto *ofproto)
630 assert(ofproto->ofproto_class == &ofproto_dpif_class);
631 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
634 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
636 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
638 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
639 const struct ofpbuf *, ovs_be16 initial_tci,
642 /* Packet processing. */
643 static void update_learning_table(struct ofproto_dpif *,
644 const struct flow *, int vlan,
647 #define FLOW_MISS_MAX_BATCH 50
648 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
650 /* Flow expiration. */
651 static int expire(struct ofproto_dpif *);
654 static void send_netflow_active_timeouts(struct ofproto_dpif *);
657 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
658 static size_t compose_sflow_action(const struct ofproto_dpif *,
659 struct ofpbuf *odp_actions,
660 const struct flow *, uint32_t odp_port);
661 static void add_mirror_actions(struct action_xlate_ctx *ctx,
662 const struct flow *flow);
663 /* Global variables. */
664 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
666 /* Factory functions. */
669 enumerate_types(struct sset *types)
671 dp_enumerate_types(types);
675 enumerate_names(const char *type, struct sset *names)
677 return dp_enumerate_names(type, names);
681 del(const char *type, const char *name)
686 error = dpif_open(name, type, &dpif);
688 error = dpif_delete(dpif);
694 /* Basic life-cycle. */
696 static int add_internal_flows(struct ofproto_dpif *);
698 static struct ofproto *
701 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
706 dealloc(struct ofproto *ofproto_)
708 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
713 construct(struct ofproto *ofproto_)
715 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
716 const char *name = ofproto->up.name;
720 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
722 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
726 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
727 ofproto->n_matches = 0;
729 dpif_flow_flush(ofproto->dpif);
730 dpif_recv_purge(ofproto->dpif);
732 error = dpif_recv_set(ofproto->dpif, true);
734 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
735 dpif_close(ofproto->dpif);
739 ofproto->netflow = NULL;
740 ofproto->sflow = NULL;
742 hmap_init(&ofproto->bundles);
743 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
744 for (i = 0; i < MAX_MIRRORS; i++) {
745 ofproto->mirrors[i] = NULL;
747 ofproto->has_bonded_bundles = false;
749 timer_set_duration(&ofproto->next_expiration, 1000);
751 hmap_init(&ofproto->facets);
752 hmap_init(&ofproto->subfacets);
753 ofproto->governor = NULL;
755 for (i = 0; i < N_TABLES; i++) {
756 struct table_dpif *table = &ofproto->tables[i];
758 table->catchall_table = NULL;
759 table->other_table = NULL;
760 table->basis = random_uint32();
762 ofproto->need_revalidate = false;
763 tag_set_init(&ofproto->revalidate_set);
765 list_init(&ofproto->completions);
767 ofproto_dpif_unixctl_init();
769 ofproto->has_mirrors = false;
770 ofproto->has_bundle_action = false;
772 hmap_init(&ofproto->vlandev_map);
773 hmap_init(&ofproto->realdev_vid_map);
775 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
776 hash_string(ofproto->up.name, 0));
777 memset(&ofproto->stats, 0, sizeof ofproto->stats);
779 ofproto_init_tables(ofproto_, N_TABLES);
780 error = add_internal_flows(ofproto);
781 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
787 add_internal_flow(struct ofproto_dpif *ofproto, int id,
788 const struct ofpbuf *actions, struct rule_dpif **rulep)
790 struct ofputil_flow_mod fm;
793 cls_rule_init_catchall(&fm.cr, 0);
794 cls_rule_set_reg(&fm.cr, 0, id);
795 fm.cookie = htonll(0);
796 fm.cookie_mask = htonll(0);
797 fm.table_id = TBL_INTERNAL;
798 fm.command = OFPFC_ADD;
804 fm.actions = actions->data;
805 fm.n_actions = actions->size / sizeof(union ofp_action);
807 error = ofproto_flow_mod(&ofproto->up, &fm);
809 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
810 id, ofperr_to_string(error));
814 *rulep = rule_dpif_lookup__(ofproto, &fm.cr.flow, TBL_INTERNAL);
815 assert(*rulep != NULL);
821 add_internal_flows(struct ofproto_dpif *ofproto)
823 struct nx_action_controller *nac;
824 uint64_t actions_stub[128 / 8];
825 struct ofpbuf actions;
829 ofpbuf_use_stack(&actions, actions_stub, sizeof actions_stub);
832 nac = ofputil_put_NXAST_CONTROLLER(&actions);
833 nac->max_len = htons(UINT16_MAX);
834 nac->controller_id = htons(0);
835 nac->reason = OFPR_NO_MATCH;
836 error = add_internal_flow(ofproto, id++, &actions, &ofproto->miss_rule);
841 ofpbuf_clear(&actions);
842 error = add_internal_flow(ofproto, id++, &actions,
843 &ofproto->no_packet_in_rule);
848 complete_operations(struct ofproto_dpif *ofproto)
850 struct dpif_completion *c, *next;
852 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
853 ofoperation_complete(c->op, 0);
854 list_remove(&c->list_node);
860 destruct(struct ofproto *ofproto_)
862 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
863 struct rule_dpif *rule, *next_rule;
864 struct oftable *table;
867 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
868 complete_operations(ofproto);
870 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
871 struct cls_cursor cursor;
873 cls_cursor_init(&cursor, &table->cls, NULL);
874 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
875 ofproto_rule_destroy(&rule->up);
879 for (i = 0; i < MAX_MIRRORS; i++) {
880 mirror_destroy(ofproto->mirrors[i]);
883 netflow_destroy(ofproto->netflow);
884 dpif_sflow_destroy(ofproto->sflow);
885 hmap_destroy(&ofproto->bundles);
886 mac_learning_destroy(ofproto->ml);
888 hmap_destroy(&ofproto->facets);
889 hmap_destroy(&ofproto->subfacets);
890 governor_destroy(ofproto->governor);
892 hmap_destroy(&ofproto->vlandev_map);
893 hmap_destroy(&ofproto->realdev_vid_map);
895 dpif_close(ofproto->dpif);
899 run_fast(struct ofproto *ofproto_)
901 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
904 /* Handle one or more batches of upcalls, until there's nothing left to do
905 * or until we do a fixed total amount of work.
907 * We do work in batches because it can be much cheaper to set up a number
908 * of flows and fire off their patches all at once. We do multiple batches
909 * because in some cases handling a packet can cause another packet to be
910 * queued almost immediately as part of the return flow. Both
911 * optimizations can make major improvements on some benchmarks and
912 * presumably for real traffic as well. */
914 while (work < FLOW_MISS_MAX_BATCH) {
915 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
925 run(struct ofproto *ofproto_)
927 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
928 struct ofport_dpif *ofport;
929 struct ofbundle *bundle;
933 complete_operations(ofproto);
935 dpif_run(ofproto->dpif);
937 error = run_fast(ofproto_);
942 if (timer_expired(&ofproto->next_expiration)) {
943 int delay = expire(ofproto);
944 timer_set_duration(&ofproto->next_expiration, delay);
947 if (ofproto->netflow) {
948 if (netflow_run(ofproto->netflow)) {
949 send_netflow_active_timeouts(ofproto);
952 if (ofproto->sflow) {
953 dpif_sflow_run(ofproto->sflow);
956 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
959 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
964 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
966 /* Now revalidate if there's anything to do. */
967 if (ofproto->need_revalidate
968 || !tag_set_is_empty(&ofproto->revalidate_set)) {
969 struct tag_set revalidate_set = ofproto->revalidate_set;
970 bool revalidate_all = ofproto->need_revalidate;
973 /* Clear the revalidation flags. */
974 tag_set_init(&ofproto->revalidate_set);
975 ofproto->need_revalidate = false;
977 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
979 || tag_set_intersects(&revalidate_set, facet->tags)) {
980 facet_revalidate(facet);
985 /* Check the consistency of a random facet, to aid debugging. */
986 if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) {
989 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
990 struct facet, hmap_node);
991 if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) {
992 if (!facet_check_consistency(facet)) {
993 ofproto->need_revalidate = true;
998 if (ofproto->governor) {
1001 governor_run(ofproto->governor);
1003 /* If the governor has shrunk to its minimum size and the number of
1004 * subfacets has dwindled, then drop the governor entirely.
1006 * For hysteresis, the number of subfacets to drop the governor is
1007 * smaller than the number needed to trigger its creation. */
1008 n_subfacets = hmap_count(&ofproto->subfacets);
1009 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1010 && governor_is_idle(ofproto->governor)) {
1011 governor_destroy(ofproto->governor);
1012 ofproto->governor = NULL;
1020 wait(struct ofproto *ofproto_)
1022 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1023 struct ofport_dpif *ofport;
1024 struct ofbundle *bundle;
1026 if (!clogged && !list_is_empty(&ofproto->completions)) {
1027 poll_immediate_wake();
1030 dpif_wait(ofproto->dpif);
1031 dpif_recv_wait(ofproto->dpif);
1032 if (ofproto->sflow) {
1033 dpif_sflow_wait(ofproto->sflow);
1035 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
1036 poll_immediate_wake();
1038 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1041 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1042 bundle_wait(bundle);
1044 if (ofproto->netflow) {
1045 netflow_wait(ofproto->netflow);
1047 mac_learning_wait(ofproto->ml);
1049 if (ofproto->need_revalidate) {
1050 /* Shouldn't happen, but if it does just go around again. */
1051 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1052 poll_immediate_wake();
1054 timer_wait(&ofproto->next_expiration);
1056 if (ofproto->governor) {
1057 governor_wait(ofproto->governor);
1062 flush(struct ofproto *ofproto_)
1064 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1065 struct facet *facet, *next_facet;
1067 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1068 /* Mark the facet as not installed so that facet_remove() doesn't
1069 * bother trying to uninstall it. There is no point in uninstalling it
1070 * individually since we are about to blow away all the facets with
1071 * dpif_flow_flush(). */
1072 struct subfacet *subfacet;
1074 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
1075 subfacet->path = SF_NOT_INSTALLED;
1076 subfacet->dp_packet_count = 0;
1077 subfacet->dp_byte_count = 0;
1079 facet_remove(facet);
1081 dpif_flow_flush(ofproto->dpif);
1085 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1086 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1088 *arp_match_ip = true;
1089 *actions = (OFPUTIL_A_OUTPUT |
1090 OFPUTIL_A_SET_VLAN_VID |
1091 OFPUTIL_A_SET_VLAN_PCP |
1092 OFPUTIL_A_STRIP_VLAN |
1093 OFPUTIL_A_SET_DL_SRC |
1094 OFPUTIL_A_SET_DL_DST |
1095 OFPUTIL_A_SET_NW_SRC |
1096 OFPUTIL_A_SET_NW_DST |
1097 OFPUTIL_A_SET_NW_TOS |
1098 OFPUTIL_A_SET_TP_SRC |
1099 OFPUTIL_A_SET_TP_DST |
1104 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
1106 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1107 struct dpif_dp_stats s;
1109 strcpy(ots->name, "classifier");
1111 dpif_get_dp_stats(ofproto->dpif, &s);
1112 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
1113 put_32aligned_be64(&ots->matched_count,
1114 htonll(s.n_hit + ofproto->n_matches));
1117 static struct ofport *
1120 struct ofport_dpif *port = xmalloc(sizeof *port);
1125 port_dealloc(struct ofport *port_)
1127 struct ofport_dpif *port = ofport_dpif_cast(port_);
1132 port_construct(struct ofport *port_)
1134 struct ofport_dpif *port = ofport_dpif_cast(port_);
1135 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1137 ofproto->need_revalidate = true;
1138 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
1139 port->bundle = NULL;
1141 port->tag = tag_create_random();
1142 port->may_enable = true;
1143 port->stp_port = NULL;
1144 port->stp_state = STP_DISABLED;
1145 hmap_init(&port->priorities);
1146 port->realdev_ofp_port = 0;
1147 port->vlandev_vid = 0;
1148 port->carrier_seq = netdev_get_carrier_resets(port->up.netdev);
1150 if (ofproto->sflow) {
1151 dpif_sflow_add_port(ofproto->sflow, port_);
1158 port_destruct(struct ofport *port_)
1160 struct ofport_dpif *port = ofport_dpif_cast(port_);
1161 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1163 ofproto->need_revalidate = true;
1164 bundle_remove(port_);
1165 set_cfm(port_, NULL);
1166 if (ofproto->sflow) {
1167 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1170 ofport_clear_priorities(port);
1171 hmap_destroy(&port->priorities);
1175 port_modified(struct ofport *port_)
1177 struct ofport_dpif *port = ofport_dpif_cast(port_);
1179 if (port->bundle && port->bundle->bond) {
1180 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1185 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1187 struct ofport_dpif *port = ofport_dpif_cast(port_);
1188 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1189 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1191 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1192 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1193 OFPUTIL_PC_NO_PACKET_IN)) {
1194 ofproto->need_revalidate = true;
1196 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1197 bundle_update(port->bundle);
1203 set_sflow(struct ofproto *ofproto_,
1204 const struct ofproto_sflow_options *sflow_options)
1206 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1207 struct dpif_sflow *ds = ofproto->sflow;
1209 if (sflow_options) {
1211 struct ofport_dpif *ofport;
1213 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1214 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1215 dpif_sflow_add_port(ds, &ofport->up);
1217 ofproto->need_revalidate = true;
1219 dpif_sflow_set_options(ds, sflow_options);
1222 dpif_sflow_destroy(ds);
1223 ofproto->need_revalidate = true;
1224 ofproto->sflow = NULL;
1231 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1233 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1240 struct ofproto_dpif *ofproto;
1242 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1243 ofproto->need_revalidate = true;
1244 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1247 if (cfm_configure(ofport->cfm, s)) {
1253 cfm_destroy(ofport->cfm);
1259 get_cfm_fault(const struct ofport *ofport_)
1261 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1263 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1267 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1270 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1273 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1281 get_cfm_health(const struct ofport *ofport_)
1283 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1285 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1288 /* Spanning Tree. */
1291 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1293 struct ofproto_dpif *ofproto = ofproto_;
1294 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1295 struct ofport_dpif *ofport;
1297 ofport = stp_port_get_aux(sp);
1299 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1300 ofproto->up.name, port_num);
1302 struct eth_header *eth = pkt->l2;
1304 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1305 if (eth_addr_is_zero(eth->eth_src)) {
1306 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1307 "with unknown MAC", ofproto->up.name, port_num);
1309 send_packet(ofport, pkt);
1315 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1317 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1319 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1321 /* Only revalidate flows if the configuration changed. */
1322 if (!s != !ofproto->stp) {
1323 ofproto->need_revalidate = true;
1327 if (!ofproto->stp) {
1328 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1329 send_bpdu_cb, ofproto);
1330 ofproto->stp_last_tick = time_msec();
1333 stp_set_bridge_id(ofproto->stp, s->system_id);
1334 stp_set_bridge_priority(ofproto->stp, s->priority);
1335 stp_set_hello_time(ofproto->stp, s->hello_time);
1336 stp_set_max_age(ofproto->stp, s->max_age);
1337 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1339 struct ofport *ofport;
1341 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1342 set_stp_port(ofport, NULL);
1345 stp_destroy(ofproto->stp);
1346 ofproto->stp = NULL;
1353 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1355 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1359 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1360 s->designated_root = stp_get_designated_root(ofproto->stp);
1361 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1370 update_stp_port_state(struct ofport_dpif *ofport)
1372 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1373 enum stp_state state;
1375 /* Figure out new state. */
1376 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1380 if (ofport->stp_state != state) {
1381 enum ofputil_port_state of_state;
1384 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1385 netdev_get_name(ofport->up.netdev),
1386 stp_state_name(ofport->stp_state),
1387 stp_state_name(state));
1388 if (stp_learn_in_state(ofport->stp_state)
1389 != stp_learn_in_state(state)) {
1390 /* xxx Learning action flows should also be flushed. */
1391 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1393 fwd_change = stp_forward_in_state(ofport->stp_state)
1394 != stp_forward_in_state(state);
1396 ofproto->need_revalidate = true;
1397 ofport->stp_state = state;
1398 ofport->stp_state_entered = time_msec();
1400 if (fwd_change && ofport->bundle) {
1401 bundle_update(ofport->bundle);
1404 /* Update the STP state bits in the OpenFlow port description. */
1405 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1406 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1407 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1408 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1409 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1411 ofproto_port_set_state(&ofport->up, of_state);
1415 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1416 * caller is responsible for assigning STP port numbers and ensuring
1417 * there are no duplicates. */
1419 set_stp_port(struct ofport *ofport_,
1420 const struct ofproto_port_stp_settings *s)
1422 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1423 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1424 struct stp_port *sp = ofport->stp_port;
1426 if (!s || !s->enable) {
1428 ofport->stp_port = NULL;
1429 stp_port_disable(sp);
1430 update_stp_port_state(ofport);
1433 } else if (sp && stp_port_no(sp) != s->port_num
1434 && ofport == stp_port_get_aux(sp)) {
1435 /* The port-id changed, so disable the old one if it's not
1436 * already in use by another port. */
1437 stp_port_disable(sp);
1440 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1441 stp_port_enable(sp);
1443 stp_port_set_aux(sp, ofport);
1444 stp_port_set_priority(sp, s->priority);
1445 stp_port_set_path_cost(sp, s->path_cost);
1447 update_stp_port_state(ofport);
1453 get_stp_port_status(struct ofport *ofport_,
1454 struct ofproto_port_stp_status *s)
1456 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1457 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1458 struct stp_port *sp = ofport->stp_port;
1460 if (!ofproto->stp || !sp) {
1466 s->port_id = stp_port_get_id(sp);
1467 s->state = stp_port_get_state(sp);
1468 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1469 s->role = stp_port_get_role(sp);
1470 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1476 stp_run(struct ofproto_dpif *ofproto)
1479 long long int now = time_msec();
1480 long long int elapsed = now - ofproto->stp_last_tick;
1481 struct stp_port *sp;
1484 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1485 ofproto->stp_last_tick = now;
1487 while (stp_get_changed_port(ofproto->stp, &sp)) {
1488 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1491 update_stp_port_state(ofport);
1495 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1496 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1502 stp_wait(struct ofproto_dpif *ofproto)
1505 poll_timer_wait(1000);
1509 /* Returns true if STP should process 'flow'. */
1511 stp_should_process_flow(const struct flow *flow)
1513 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1517 stp_process_packet(const struct ofport_dpif *ofport,
1518 const struct ofpbuf *packet)
1520 struct ofpbuf payload = *packet;
1521 struct eth_header *eth = payload.data;
1522 struct stp_port *sp = ofport->stp_port;
1524 /* Sink packets on ports that have STP disabled when the bridge has
1526 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1530 /* Trim off padding on payload. */
1531 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1532 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1535 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1536 stp_received_bpdu(sp, payload.data, payload.size);
1540 static struct priority_to_dscp *
1541 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1543 struct priority_to_dscp *pdscp;
1546 hash = hash_int(priority, 0);
1547 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1548 if (pdscp->priority == priority) {
1556 ofport_clear_priorities(struct ofport_dpif *ofport)
1558 struct priority_to_dscp *pdscp, *next;
1560 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1561 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1567 set_queues(struct ofport *ofport_,
1568 const struct ofproto_port_queue *qdscp_list,
1571 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1572 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1573 struct hmap new = HMAP_INITIALIZER(&new);
1576 for (i = 0; i < n_qdscp; i++) {
1577 struct priority_to_dscp *pdscp;
1581 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1582 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1587 pdscp = get_priority(ofport, priority);
1589 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1591 pdscp = xmalloc(sizeof *pdscp);
1592 pdscp->priority = priority;
1594 ofproto->need_revalidate = true;
1597 if (pdscp->dscp != dscp) {
1599 ofproto->need_revalidate = true;
1602 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1605 if (!hmap_is_empty(&ofport->priorities)) {
1606 ofport_clear_priorities(ofport);
1607 ofproto->need_revalidate = true;
1610 hmap_swap(&new, &ofport->priorities);
1618 /* Expires all MAC learning entries associated with 'bundle' and forces its
1619 * ofproto to revalidate every flow.
1621 * Normally MAC learning entries are removed only from the ofproto associated
1622 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1623 * are removed from every ofproto. When patch ports and SLB bonds are in use
1624 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1625 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1626 * with the host from which it migrated. */
1628 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1630 struct ofproto_dpif *ofproto = bundle->ofproto;
1631 struct mac_learning *ml = ofproto->ml;
1632 struct mac_entry *mac, *next_mac;
1634 ofproto->need_revalidate = true;
1635 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1636 if (mac->port.p == bundle) {
1638 struct ofproto_dpif *o;
1640 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1642 struct mac_entry *e;
1644 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1647 tag_set_add(&o->revalidate_set, e->tag);
1648 mac_learning_expire(o->ml, e);
1654 mac_learning_expire(ml, mac);
1659 static struct ofbundle *
1660 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1662 struct ofbundle *bundle;
1664 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1665 &ofproto->bundles) {
1666 if (bundle->aux == aux) {
1673 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1674 * ones that are found to 'bundles'. */
1676 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1677 void **auxes, size_t n_auxes,
1678 struct hmapx *bundles)
1682 hmapx_init(bundles);
1683 for (i = 0; i < n_auxes; i++) {
1684 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1686 hmapx_add(bundles, bundle);
1692 bundle_update(struct ofbundle *bundle)
1694 struct ofport_dpif *port;
1696 bundle->floodable = true;
1697 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1698 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1699 || !stp_forward_in_state(port->stp_state)) {
1700 bundle->floodable = false;
1707 bundle_del_port(struct ofport_dpif *port)
1709 struct ofbundle *bundle = port->bundle;
1711 bundle->ofproto->need_revalidate = true;
1713 list_remove(&port->bundle_node);
1714 port->bundle = NULL;
1717 lacp_slave_unregister(bundle->lacp, port);
1720 bond_slave_unregister(bundle->bond, port);
1723 bundle_update(bundle);
1727 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1728 struct lacp_slave_settings *lacp,
1729 uint32_t bond_stable_id)
1731 struct ofport_dpif *port;
1733 port = get_ofp_port(bundle->ofproto, ofp_port);
1738 if (port->bundle != bundle) {
1739 bundle->ofproto->need_revalidate = true;
1741 bundle_del_port(port);
1744 port->bundle = bundle;
1745 list_push_back(&bundle->ports, &port->bundle_node);
1746 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1747 || !stp_forward_in_state(port->stp_state)) {
1748 bundle->floodable = false;
1752 port->bundle->ofproto->need_revalidate = true;
1753 lacp_slave_register(bundle->lacp, port, lacp);
1756 port->bond_stable_id = bond_stable_id;
1762 bundle_destroy(struct ofbundle *bundle)
1764 struct ofproto_dpif *ofproto;
1765 struct ofport_dpif *port, *next_port;
1772 ofproto = bundle->ofproto;
1773 for (i = 0; i < MAX_MIRRORS; i++) {
1774 struct ofmirror *m = ofproto->mirrors[i];
1776 if (m->out == bundle) {
1778 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1779 || hmapx_find_and_delete(&m->dsts, bundle)) {
1780 ofproto->need_revalidate = true;
1785 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1786 bundle_del_port(port);
1789 bundle_flush_macs(bundle, true);
1790 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1792 free(bundle->trunks);
1793 lacp_destroy(bundle->lacp);
1794 bond_destroy(bundle->bond);
1799 bundle_set(struct ofproto *ofproto_, void *aux,
1800 const struct ofproto_bundle_settings *s)
1802 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1803 bool need_flush = false;
1804 struct ofport_dpif *port;
1805 struct ofbundle *bundle;
1806 unsigned long *trunks;
1812 bundle_destroy(bundle_lookup(ofproto, aux));
1816 assert(s->n_slaves == 1 || s->bond != NULL);
1817 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1819 bundle = bundle_lookup(ofproto, aux);
1821 bundle = xmalloc(sizeof *bundle);
1823 bundle->ofproto = ofproto;
1824 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1825 hash_pointer(aux, 0));
1827 bundle->name = NULL;
1829 list_init(&bundle->ports);
1830 bundle->vlan_mode = PORT_VLAN_TRUNK;
1832 bundle->trunks = NULL;
1833 bundle->use_priority_tags = s->use_priority_tags;
1834 bundle->lacp = NULL;
1835 bundle->bond = NULL;
1837 bundle->floodable = true;
1839 bundle->src_mirrors = 0;
1840 bundle->dst_mirrors = 0;
1841 bundle->mirror_out = 0;
1844 if (!bundle->name || strcmp(s->name, bundle->name)) {
1846 bundle->name = xstrdup(s->name);
1851 if (!bundle->lacp) {
1852 ofproto->need_revalidate = true;
1853 bundle->lacp = lacp_create();
1855 lacp_configure(bundle->lacp, s->lacp);
1857 lacp_destroy(bundle->lacp);
1858 bundle->lacp = NULL;
1861 /* Update set of ports. */
1863 for (i = 0; i < s->n_slaves; i++) {
1864 if (!bundle_add_port(bundle, s->slaves[i],
1865 s->lacp ? &s->lacp_slaves[i] : NULL,
1866 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1870 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1871 struct ofport_dpif *next_port;
1873 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1874 for (i = 0; i < s->n_slaves; i++) {
1875 if (s->slaves[i] == port->up.ofp_port) {
1880 bundle_del_port(port);
1884 assert(list_size(&bundle->ports) <= s->n_slaves);
1886 if (list_is_empty(&bundle->ports)) {
1887 bundle_destroy(bundle);
1891 /* Set VLAN tagging mode */
1892 if (s->vlan_mode != bundle->vlan_mode
1893 || s->use_priority_tags != bundle->use_priority_tags) {
1894 bundle->vlan_mode = s->vlan_mode;
1895 bundle->use_priority_tags = s->use_priority_tags;
1900 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1901 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1903 if (vlan != bundle->vlan) {
1904 bundle->vlan = vlan;
1908 /* Get trunked VLANs. */
1909 switch (s->vlan_mode) {
1910 case PORT_VLAN_ACCESS:
1914 case PORT_VLAN_TRUNK:
1915 trunks = (unsigned long *) s->trunks;
1918 case PORT_VLAN_NATIVE_UNTAGGED:
1919 case PORT_VLAN_NATIVE_TAGGED:
1920 if (vlan != 0 && (!s->trunks
1921 || !bitmap_is_set(s->trunks, vlan)
1922 || bitmap_is_set(s->trunks, 0))) {
1923 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1925 trunks = bitmap_clone(s->trunks, 4096);
1927 trunks = bitmap_allocate1(4096);
1929 bitmap_set1(trunks, vlan);
1930 bitmap_set0(trunks, 0);
1932 trunks = (unsigned long *) s->trunks;
1939 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1940 free(bundle->trunks);
1941 if (trunks == s->trunks) {
1942 bundle->trunks = vlan_bitmap_clone(trunks);
1944 bundle->trunks = trunks;
1949 if (trunks != s->trunks) {
1954 if (!list_is_short(&bundle->ports)) {
1955 bundle->ofproto->has_bonded_bundles = true;
1957 if (bond_reconfigure(bundle->bond, s->bond)) {
1958 ofproto->need_revalidate = true;
1961 bundle->bond = bond_create(s->bond);
1962 ofproto->need_revalidate = true;
1965 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1966 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1970 bond_destroy(bundle->bond);
1971 bundle->bond = NULL;
1974 /* If we changed something that would affect MAC learning, un-learn
1975 * everything on this port and force flow revalidation. */
1977 bundle_flush_macs(bundle, false);
1984 bundle_remove(struct ofport *port_)
1986 struct ofport_dpif *port = ofport_dpif_cast(port_);
1987 struct ofbundle *bundle = port->bundle;
1990 bundle_del_port(port);
1991 if (list_is_empty(&bundle->ports)) {
1992 bundle_destroy(bundle);
1993 } else if (list_is_short(&bundle->ports)) {
1994 bond_destroy(bundle->bond);
1995 bundle->bond = NULL;
2001 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2003 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2004 struct ofport_dpif *port = port_;
2005 uint8_t ea[ETH_ADDR_LEN];
2008 error = netdev_get_etheraddr(port->up.netdev, ea);
2010 struct ofpbuf packet;
2013 ofpbuf_init(&packet, 0);
2014 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2016 memcpy(packet_pdu, pdu, pdu_size);
2018 send_packet(port, &packet);
2019 ofpbuf_uninit(&packet);
2021 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2022 "%s (%s)", port->bundle->name,
2023 netdev_get_name(port->up.netdev), strerror(error));
2028 bundle_send_learning_packets(struct ofbundle *bundle)
2030 struct ofproto_dpif *ofproto = bundle->ofproto;
2031 int error, n_packets, n_errors;
2032 struct mac_entry *e;
2034 error = n_packets = n_errors = 0;
2035 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2036 if (e->port.p != bundle) {
2037 struct ofpbuf *learning_packet;
2038 struct ofport_dpif *port;
2042 /* The assignment to "port" is unnecessary but makes "grep"ing for
2043 * struct ofport_dpif more effective. */
2044 learning_packet = bond_compose_learning_packet(bundle->bond,
2048 ret = send_packet(port, learning_packet);
2049 ofpbuf_delete(learning_packet);
2059 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2060 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2061 "packets, last error was: %s",
2062 bundle->name, n_errors, n_packets, strerror(error));
2064 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2065 bundle->name, n_packets);
2070 bundle_run(struct ofbundle *bundle)
2073 lacp_run(bundle->lacp, send_pdu_cb);
2076 struct ofport_dpif *port;
2078 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2079 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2082 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
2083 lacp_status(bundle->lacp));
2084 if (bond_should_send_learning_packets(bundle->bond)) {
2085 bundle_send_learning_packets(bundle);
2091 bundle_wait(struct ofbundle *bundle)
2094 lacp_wait(bundle->lacp);
2097 bond_wait(bundle->bond);
2104 mirror_scan(struct ofproto_dpif *ofproto)
2108 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2109 if (!ofproto->mirrors[idx]) {
2116 static struct ofmirror *
2117 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2121 for (i = 0; i < MAX_MIRRORS; i++) {
2122 struct ofmirror *mirror = ofproto->mirrors[i];
2123 if (mirror && mirror->aux == aux) {
2131 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2133 mirror_update_dups(struct ofproto_dpif *ofproto)
2137 for (i = 0; i < MAX_MIRRORS; i++) {
2138 struct ofmirror *m = ofproto->mirrors[i];
2141 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2145 for (i = 0; i < MAX_MIRRORS; i++) {
2146 struct ofmirror *m1 = ofproto->mirrors[i];
2153 for (j = i + 1; j < MAX_MIRRORS; j++) {
2154 struct ofmirror *m2 = ofproto->mirrors[j];
2156 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2157 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2158 m2->dup_mirrors |= m1->dup_mirrors;
2165 mirror_set(struct ofproto *ofproto_, void *aux,
2166 const struct ofproto_mirror_settings *s)
2168 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2169 mirror_mask_t mirror_bit;
2170 struct ofbundle *bundle;
2171 struct ofmirror *mirror;
2172 struct ofbundle *out;
2173 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2174 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2177 mirror = mirror_lookup(ofproto, aux);
2179 mirror_destroy(mirror);
2185 idx = mirror_scan(ofproto);
2187 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2189 ofproto->up.name, MAX_MIRRORS, s->name);
2193 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2194 mirror->ofproto = ofproto;
2197 mirror->out_vlan = -1;
2198 mirror->name = NULL;
2201 if (!mirror->name || strcmp(s->name, mirror->name)) {
2203 mirror->name = xstrdup(s->name);
2206 /* Get the new configuration. */
2207 if (s->out_bundle) {
2208 out = bundle_lookup(ofproto, s->out_bundle);
2210 mirror_destroy(mirror);
2216 out_vlan = s->out_vlan;
2218 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2219 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2221 /* If the configuration has not changed, do nothing. */
2222 if (hmapx_equals(&srcs, &mirror->srcs)
2223 && hmapx_equals(&dsts, &mirror->dsts)
2224 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2225 && mirror->out == out
2226 && mirror->out_vlan == out_vlan)
2228 hmapx_destroy(&srcs);
2229 hmapx_destroy(&dsts);
2233 hmapx_swap(&srcs, &mirror->srcs);
2234 hmapx_destroy(&srcs);
2236 hmapx_swap(&dsts, &mirror->dsts);
2237 hmapx_destroy(&dsts);
2239 free(mirror->vlans);
2240 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2243 mirror->out_vlan = out_vlan;
2245 /* Update bundles. */
2246 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2247 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2248 if (hmapx_contains(&mirror->srcs, bundle)) {
2249 bundle->src_mirrors |= mirror_bit;
2251 bundle->src_mirrors &= ~mirror_bit;
2254 if (hmapx_contains(&mirror->dsts, bundle)) {
2255 bundle->dst_mirrors |= mirror_bit;
2257 bundle->dst_mirrors &= ~mirror_bit;
2260 if (mirror->out == bundle) {
2261 bundle->mirror_out |= mirror_bit;
2263 bundle->mirror_out &= ~mirror_bit;
2267 ofproto->need_revalidate = true;
2268 ofproto->has_mirrors = true;
2269 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2270 mirror_update_dups(ofproto);
2276 mirror_destroy(struct ofmirror *mirror)
2278 struct ofproto_dpif *ofproto;
2279 mirror_mask_t mirror_bit;
2280 struct ofbundle *bundle;
2287 ofproto = mirror->ofproto;
2288 ofproto->need_revalidate = true;
2289 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2291 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2292 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2293 bundle->src_mirrors &= ~mirror_bit;
2294 bundle->dst_mirrors &= ~mirror_bit;
2295 bundle->mirror_out &= ~mirror_bit;
2298 hmapx_destroy(&mirror->srcs);
2299 hmapx_destroy(&mirror->dsts);
2300 free(mirror->vlans);
2302 ofproto->mirrors[mirror->idx] = NULL;
2306 mirror_update_dups(ofproto);
2308 ofproto->has_mirrors = false;
2309 for (i = 0; i < MAX_MIRRORS; i++) {
2310 if (ofproto->mirrors[i]) {
2311 ofproto->has_mirrors = true;
2318 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2319 uint64_t *packets, uint64_t *bytes)
2321 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2322 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2325 *packets = *bytes = UINT64_MAX;
2329 *packets = mirror->packet_count;
2330 *bytes = mirror->byte_count;
2336 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2338 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2339 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2340 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2346 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2348 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2349 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2350 return bundle && bundle->mirror_out != 0;
2354 forward_bpdu_changed(struct ofproto *ofproto_)
2356 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2357 /* Revalidate cached flows whenever forward_bpdu option changes. */
2358 ofproto->need_revalidate = true;
2362 set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time)
2364 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2365 mac_learning_set_idle_time(ofproto->ml, idle_time);
2370 static struct ofport_dpif *
2371 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2373 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2374 return ofport ? ofport_dpif_cast(ofport) : NULL;
2377 static struct ofport_dpif *
2378 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2380 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2384 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2385 struct dpif_port *dpif_port)
2387 ofproto_port->name = dpif_port->name;
2388 ofproto_port->type = dpif_port->type;
2389 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2393 port_run(struct ofport_dpif *ofport)
2395 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2396 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2397 bool enable = netdev_get_carrier(ofport->up.netdev);
2399 ofport->carrier_seq = carrier_seq;
2402 cfm_run(ofport->cfm);
2404 if (cfm_should_send_ccm(ofport->cfm)) {
2405 struct ofpbuf packet;
2407 ofpbuf_init(&packet, 0);
2408 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2409 send_packet(ofport, &packet);
2410 ofpbuf_uninit(&packet);
2413 enable = enable && !cfm_get_fault(ofport->cfm)
2414 && cfm_get_opup(ofport->cfm);
2417 if (ofport->bundle) {
2418 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2419 if (carrier_changed) {
2420 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2424 if (ofport->may_enable != enable) {
2425 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2427 if (ofproto->has_bundle_action) {
2428 ofproto->need_revalidate = true;
2432 ofport->may_enable = enable;
2436 port_wait(struct ofport_dpif *ofport)
2439 cfm_wait(ofport->cfm);
2444 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2445 struct ofproto_port *ofproto_port)
2447 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2448 struct dpif_port dpif_port;
2451 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2453 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2459 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2461 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2465 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2467 *ofp_portp = odp_port_to_ofp_port(odp_port);
2473 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2475 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2478 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2480 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2482 /* The caller is going to close ofport->up.netdev. If this is a
2483 * bonded port, then the bond is using that netdev, so remove it
2484 * from the bond. The client will need to reconfigure everything
2485 * after deleting ports, so then the slave will get re-added. */
2486 bundle_remove(&ofport->up);
2493 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2495 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2498 error = netdev_get_stats(ofport->up.netdev, stats);
2500 if (!error && ofport->odp_port == OVSP_LOCAL) {
2501 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2503 /* ofproto->stats.tx_packets represents packets that we created
2504 * internally and sent to some port (e.g. packets sent with
2505 * send_packet()). Account for them as if they had come from
2506 * OFPP_LOCAL and got forwarded. */
2508 if (stats->rx_packets != UINT64_MAX) {
2509 stats->rx_packets += ofproto->stats.tx_packets;
2512 if (stats->rx_bytes != UINT64_MAX) {
2513 stats->rx_bytes += ofproto->stats.tx_bytes;
2516 /* ofproto->stats.rx_packets represents packets that were received on
2517 * some port and we processed internally and dropped (e.g. STP).
2518 * Account fro them as if they had been forwarded to OFPP_LOCAL. */
2520 if (stats->tx_packets != UINT64_MAX) {
2521 stats->tx_packets += ofproto->stats.rx_packets;
2524 if (stats->tx_bytes != UINT64_MAX) {
2525 stats->tx_bytes += ofproto->stats.rx_bytes;
2532 /* Account packets for LOCAL port. */
2534 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2535 size_t tx_size, size_t rx_size)
2537 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2540 ofproto->stats.rx_packets++;
2541 ofproto->stats.rx_bytes += rx_size;
2544 ofproto->stats.tx_packets++;
2545 ofproto->stats.tx_bytes += tx_size;
2549 struct port_dump_state {
2550 struct dpif_port_dump dump;
2555 port_dump_start(const struct ofproto *ofproto_, void **statep)
2557 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2558 struct port_dump_state *state;
2560 *statep = state = xmalloc(sizeof *state);
2561 dpif_port_dump_start(&state->dump, ofproto->dpif);
2562 state->done = false;
2567 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2568 struct ofproto_port *port)
2570 struct port_dump_state *state = state_;
2571 struct dpif_port dpif_port;
2573 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2574 ofproto_port_from_dpif_port(port, &dpif_port);
2577 int error = dpif_port_dump_done(&state->dump);
2579 return error ? error : EOF;
2584 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2586 struct port_dump_state *state = state_;
2589 dpif_port_dump_done(&state->dump);
2596 port_poll(const struct ofproto *ofproto_, char **devnamep)
2598 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2599 return dpif_port_poll(ofproto->dpif, devnamep);
2603 port_poll_wait(const struct ofproto *ofproto_)
2605 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2606 dpif_port_poll_wait(ofproto->dpif);
2610 port_is_lacp_current(const struct ofport *ofport_)
2612 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2613 return (ofport->bundle && ofport->bundle->lacp
2614 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2618 /* Upcall handling. */
2620 /* Flow miss batching.
2622 * Some dpifs implement operations faster when you hand them off in a batch.
2623 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2624 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2625 * more packets, plus possibly installing the flow in the dpif.
2627 * So far we only batch the operations that affect flow setup time the most.
2628 * It's possible to batch more than that, but the benefit might be minimal. */
2630 struct hmap_node hmap_node;
2632 enum odp_key_fitness key_fitness;
2633 const struct nlattr *key;
2635 ovs_be16 initial_tci;
2636 struct list packets;
2637 enum dpif_upcall_type upcall_type;
2640 struct flow_miss_op {
2641 struct dpif_op dpif_op;
2642 struct subfacet *subfacet; /* Subfacet */
2643 void *garbage; /* Pointer to pass to free(), NULL if none. */
2644 uint64_t stub[1024 / 8]; /* Temporary buffer. */
2647 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2648 * OpenFlow controller as necessary according to their individual
2649 * configurations. */
2651 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
2652 const struct flow *flow)
2654 struct ofputil_packet_in pin;
2656 pin.packet = packet->data;
2657 pin.packet_len = packet->size;
2658 pin.reason = OFPR_NO_MATCH;
2659 pin.controller_id = 0;
2664 pin.send_len = 0; /* not used for flow table misses */
2666 flow_get_metadata(flow, &pin.fmd);
2668 /* Registers aren't meaningful on a miss. */
2669 memset(pin.fmd.reg_masks, 0, sizeof pin.fmd.reg_masks);
2671 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
2674 static enum slow_path_reason
2675 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2676 const struct ofpbuf *packet)
2678 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2684 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2686 cfm_process_heartbeat(ofport->cfm, packet);
2689 } else if (ofport->bundle && ofport->bundle->lacp
2690 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2692 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2695 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2697 stp_process_packet(ofport, packet);
2704 static struct flow_miss *
2705 flow_miss_find(struct hmap *todo, const struct flow *flow, uint32_t hash)
2707 struct flow_miss *miss;
2709 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2710 if (flow_equal(&miss->flow, flow)) {
2718 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
2719 * 'packet'. The caller must initialize op->actions and op->actions_len. If
2720 * 'miss' is associated with a subfacet the caller must also initialize the
2721 * returned op->subfacet, and if anything needs to be freed after processing
2722 * the op, the caller must initialize op->garbage also. */
2724 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
2725 struct flow_miss_op *op)
2727 if (miss->flow.vlan_tci != miss->initial_tci) {
2728 /* This packet was received on a VLAN splinter port. We
2729 * added a VLAN to the packet to make the packet resemble
2730 * the flow, but the actions were composed assuming that
2731 * the packet contained no VLAN. So, we must remove the
2732 * VLAN header from the packet before trying to execute the
2734 eth_pop_vlan(packet);
2737 op->subfacet = NULL;
2739 op->dpif_op.type = DPIF_OP_EXECUTE;
2740 op->dpif_op.u.execute.key = miss->key;
2741 op->dpif_op.u.execute.key_len = miss->key_len;
2742 op->dpif_op.u.execute.packet = packet;
2745 /* Helper for handle_flow_miss_without_facet() and
2746 * handle_flow_miss_with_facet(). */
2748 handle_flow_miss_common(struct rule_dpif *rule,
2749 struct ofpbuf *packet, const struct flow *flow)
2751 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2753 ofproto->n_matches++;
2755 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2757 * Extra-special case for fail-open mode.
2759 * We are in fail-open mode and the packet matched the fail-open
2760 * rule, but we are connected to a controller too. We should send
2761 * the packet up to the controller in the hope that it will try to
2762 * set up a flow and thereby allow us to exit fail-open.
2764 * See the top-level comment in fail-open.c for more information.
2766 send_packet_in_miss(ofproto, packet, flow);
2770 /* Figures out whether a flow that missed in 'ofproto', whose details are in
2771 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
2772 * installing a datapath flow. The answer is usually "yes" (a return value of
2773 * true). However, for short flows the cost of bookkeeping is much higher than
2774 * the benefits, so when the datapath holds a large number of flows we impose
2775 * some heuristics to decide which flows are likely to be worth tracking. */
2777 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
2778 struct flow_miss *miss, uint32_t hash)
2780 if (!ofproto->governor) {
2783 n_subfacets = hmap_count(&ofproto->subfacets);
2784 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
2788 ofproto->governor = governor_create(ofproto->up.name);
2791 return governor_should_install_flow(ofproto->governor, hash,
2792 list_size(&miss->packets));
2795 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
2796 * or creating any datapath flow. May add an "execute" operation to 'ops' and
2797 * increment '*n_ops'. */
2799 handle_flow_miss_without_facet(struct flow_miss *miss,
2800 struct rule_dpif *rule,
2801 struct flow_miss_op *ops, size_t *n_ops)
2803 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2804 struct action_xlate_ctx ctx;
2805 struct ofpbuf *packet;
2807 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2808 struct flow_miss_op *op = &ops[*n_ops];
2809 struct dpif_flow_stats stats;
2810 struct ofpbuf odp_actions;
2812 COVERAGE_INC(facet_suppress);
2814 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
2816 dpif_flow_stats_extract(&miss->flow, packet, &stats);
2817 rule_credit_stats(rule, &stats);
2819 action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci,
2821 ctx.resubmit_stats = &stats;
2822 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions,
2825 if (odp_actions.size) {
2826 struct dpif_execute *execute = &op->dpif_op.u.execute;
2828 init_flow_miss_execute_op(miss, packet, op);
2829 execute->actions = odp_actions.data;
2830 execute->actions_len = odp_actions.size;
2831 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
2835 ofpbuf_uninit(&odp_actions);
2840 /* Handles 'miss', which matches 'facet'. May add any required datapath
2841 * operations to 'ops', incrementing '*n_ops' for each new op. */
2843 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
2844 struct flow_miss_op *ops, size_t *n_ops)
2846 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
2847 enum subfacet_path want_path;
2848 struct subfacet *subfacet;
2849 struct ofpbuf *packet;
2851 subfacet = subfacet_create(facet,
2852 miss->key_fitness, miss->key, miss->key_len,
2855 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2856 struct flow_miss_op *op = &ops[*n_ops];
2857 struct dpif_flow_stats stats;
2858 struct ofpbuf odp_actions;
2860 handle_flow_miss_common(facet->rule, packet, &miss->flow);
2862 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
2863 if (!subfacet->actions || subfacet->slow) {
2864 subfacet_make_actions(subfacet, packet, &odp_actions);
2867 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2868 subfacet_update_stats(subfacet, &stats);
2870 if (subfacet->actions_len) {
2871 struct dpif_execute *execute = &op->dpif_op.u.execute;
2873 init_flow_miss_execute_op(miss, packet, op);
2874 op->subfacet = subfacet;
2875 if (!subfacet->slow) {
2876 execute->actions = subfacet->actions;
2877 execute->actions_len = subfacet->actions_len;
2878 ofpbuf_uninit(&odp_actions);
2880 execute->actions = odp_actions.data;
2881 execute->actions_len = odp_actions.size;
2882 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
2887 ofpbuf_uninit(&odp_actions);
2891 want_path = subfacet_want_path(subfacet->slow);
2892 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
2893 struct flow_miss_op *op = &ops[(*n_ops)++];
2894 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
2896 op->subfacet = subfacet;
2898 op->dpif_op.type = DPIF_OP_FLOW_PUT;
2899 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2900 put->key = miss->key;
2901 put->key_len = miss->key_len;
2902 if (want_path == SF_FAST_PATH) {
2903 put->actions = subfacet->actions;
2904 put->actions_len = subfacet->actions_len;
2906 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
2907 op->stub, sizeof op->stub,
2908 &put->actions, &put->actions_len);
2914 /* Handles flow miss 'miss' on 'ofproto'. May add any required datapath
2915 * operations to 'ops', incrementing '*n_ops' for each new op. */
2917 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2918 struct flow_miss_op *ops, size_t *n_ops)
2920 struct facet *facet;
2923 /* The caller must ensure that miss->hmap_node.hash contains
2924 * flow_hash(miss->flow, 0). */
2925 hash = miss->hmap_node.hash;
2927 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
2929 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
2931 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
2932 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
2936 facet = facet_create(rule, &miss->flow, hash);
2938 handle_flow_miss_with_facet(miss, facet, ops, n_ops);
2941 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
2942 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
2943 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
2944 * what a flow key should contain.
2946 * This function also includes some logic to help make VLAN splinters
2947 * transparent to the rest of the upcall processing logic. In particular, if
2948 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
2949 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
2950 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
2952 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
2953 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
2954 * (This differs from the value returned in flow->vlan_tci only for packets
2955 * received on VLAN splinters.)
2957 static enum odp_key_fitness
2958 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
2959 const struct nlattr *key, size_t key_len,
2960 struct flow *flow, ovs_be16 *initial_tci,
2961 struct ofpbuf *packet)
2963 enum odp_key_fitness fitness;
2965 fitness = odp_flow_key_to_flow(key, key_len, flow);
2966 if (fitness == ODP_FIT_ERROR) {
2969 *initial_tci = flow->vlan_tci;
2971 if (vsp_adjust_flow(ofproto, flow)) {
2973 /* Make the packet resemble the flow, so that it gets sent to an
2974 * OpenFlow controller properly, so that it looks correct for
2975 * sFlow, and so that flow_extract() will get the correct vlan_tci
2976 * if it is called on 'packet'.
2978 * The allocated space inside 'packet' probably also contains
2979 * 'key', that is, both 'packet' and 'key' are probably part of a
2980 * struct dpif_upcall (see the large comment on that structure
2981 * definition), so pushing data on 'packet' is in general not a
2982 * good idea since it could overwrite 'key' or free it as a side
2983 * effect. However, it's OK in this special case because we know
2984 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
2985 * will just overwrite the 4-byte "struct nlattr", which is fine
2986 * since we don't need that header anymore. */
2987 eth_push_vlan(packet, flow->vlan_tci);
2990 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
2991 if (fitness == ODP_FIT_PERFECT) {
2992 fitness = ODP_FIT_TOO_MUCH;
3000 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
3003 struct dpif_upcall *upcall;
3004 struct flow_miss *miss;
3005 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3006 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3007 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3017 /* Construct the to-do list.
3019 * This just amounts to extracting the flow from each packet and sticking
3020 * the packets that have the same flow in the same "flow_miss" structure so
3021 * that we can process them together. */
3024 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3025 struct flow_miss *miss = &misses[n_misses];
3026 struct flow_miss *existing_miss;
3029 /* Obtain metadata and check userspace/kernel agreement on flow match,
3030 * then set 'flow''s header pointers. */
3031 miss->key_fitness = ofproto_dpif_extract_flow_key(
3032 ofproto, upcall->key, upcall->key_len,
3033 &miss->flow, &miss->initial_tci, upcall->packet);
3034 if (miss->key_fitness == ODP_FIT_ERROR) {
3037 flow_extract(upcall->packet, miss->flow.skb_priority,
3038 miss->flow.tun_id, miss->flow.in_port, &miss->flow);
3040 /* Add other packets to a to-do list. */
3041 hash = flow_hash(&miss->flow, 0);
3042 existing_miss = flow_miss_find(&todo, &miss->flow, hash);
3043 if (!existing_miss) {
3044 hmap_insert(&todo, &miss->hmap_node, hash);
3045 miss->key = upcall->key;
3046 miss->key_len = upcall->key_len;
3047 miss->upcall_type = upcall->type;
3048 list_init(&miss->packets);
3052 miss = existing_miss;
3054 list_push_back(&miss->packets, &upcall->packet->list_node);
3057 /* Process each element in the to-do list, constructing the set of
3058 * operations to batch. */
3060 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3061 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
3063 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3065 /* Execute batch. */
3066 for (i = 0; i < n_ops; i++) {
3067 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3069 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
3071 /* Free memory and update facets. */
3072 for (i = 0; i < n_ops; i++) {
3073 struct flow_miss_op *op = &flow_miss_ops[i];
3075 switch (op->dpif_op.type) {
3076 case DPIF_OP_EXECUTE:
3079 case DPIF_OP_FLOW_PUT:
3080 if (!op->dpif_op.error) {
3081 op->subfacet->path = subfacet_want_path(op->subfacet->slow);
3085 case DPIF_OP_FLOW_DEL:
3091 hmap_destroy(&todo);
3094 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
3095 classify_upcall(const struct dpif_upcall *upcall)
3097 union user_action_cookie cookie;
3099 /* First look at the upcall type. */
3100 switch (upcall->type) {
3101 case DPIF_UC_ACTION:
3107 case DPIF_N_UC_TYPES:
3109 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3113 /* "action" upcalls need a closer look. */
3114 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3115 switch (cookie.type) {
3116 case USER_ACTION_COOKIE_SFLOW:
3117 return SFLOW_UPCALL;
3119 case USER_ACTION_COOKIE_SLOW_PATH:
3122 case USER_ACTION_COOKIE_UNSPEC:
3124 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
3130 handle_sflow_upcall(struct ofproto_dpif *ofproto,
3131 const struct dpif_upcall *upcall)
3133 union user_action_cookie cookie;
3134 enum odp_key_fitness fitness;
3135 ovs_be16 initial_tci;
3138 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
3139 upcall->key_len, &flow,
3140 &initial_tci, upcall->packet);
3141 if (fitness == ODP_FIT_ERROR) {
3145 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3146 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, &cookie);
3150 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
3152 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3153 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3154 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3159 assert(max_batch <= FLOW_MISS_MAX_BATCH);
3163 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3164 struct dpif_upcall *upcall = &misses[n_misses];
3165 struct ofpbuf *buf = &miss_bufs[n_misses];
3168 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3169 sizeof miss_buf_stubs[n_misses]);
3170 error = dpif_recv(ofproto->dpif, upcall, buf);
3176 switch (classify_upcall(upcall)) {
3178 /* Handle it later. */
3183 if (ofproto->sflow) {
3184 handle_sflow_upcall(ofproto, upcall);
3195 /* Handle deferred MISS_UPCALL processing. */
3196 handle_miss_upcalls(ofproto, misses, n_misses);
3197 for (i = 0; i < n_misses; i++) {
3198 ofpbuf_uninit(&miss_bufs[i]);
3204 /* Flow expiration. */
3206 static int subfacet_max_idle(const struct ofproto_dpif *);
3207 static void update_stats(struct ofproto_dpif *);
3208 static void rule_expire(struct rule_dpif *);
3209 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
3211 /* This function is called periodically by run(). Its job is to collect
3212 * updates for the flows that have been installed into the datapath, most
3213 * importantly when they last were used, and then use that information to
3214 * expire flows that have not been used recently.
3216 * Returns the number of milliseconds after which it should be called again. */
3218 expire(struct ofproto_dpif *ofproto)
3220 struct rule_dpif *rule, *next_rule;
3221 struct oftable *table;
3224 /* Update stats for each flow in the datapath. */
3225 update_stats(ofproto);
3227 /* Expire subfacets that have been idle too long. */
3228 dp_max_idle = subfacet_max_idle(ofproto);
3229 expire_subfacets(ofproto, dp_max_idle);
3231 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
3232 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
3233 struct cls_cursor cursor;
3235 cls_cursor_init(&cursor, &table->cls, NULL);
3236 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
3241 /* All outstanding data in existing flows has been accounted, so it's a
3242 * good time to do bond rebalancing. */
3243 if (ofproto->has_bonded_bundles) {
3244 struct ofbundle *bundle;
3246 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3248 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
3253 return MIN(dp_max_idle, 1000);
3256 /* Updates flow table statistics given that the datapath just reported 'stats'
3257 * as 'subfacet''s statistics. */
3259 update_subfacet_stats(struct subfacet *subfacet,
3260 const struct dpif_flow_stats *stats)
3262 struct facet *facet = subfacet->facet;
3264 if (stats->n_packets >= subfacet->dp_packet_count) {
3265 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
3266 facet->packet_count += extra;
3268 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3271 if (stats->n_bytes >= subfacet->dp_byte_count) {
3272 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
3274 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3277 subfacet->dp_packet_count = stats->n_packets;
3278 subfacet->dp_byte_count = stats->n_bytes;
3280 facet->tcp_flags |= stats->tcp_flags;
3282 subfacet_update_time(subfacet, stats->used);
3283 if (facet->accounted_bytes < facet->byte_count) {
3285 facet_account(facet);
3286 facet->accounted_bytes = facet->byte_count;
3288 facet_push_stats(facet);
3291 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3292 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3294 delete_unexpected_flow(struct dpif *dpif,
3295 const struct nlattr *key, size_t key_len)
3297 if (!VLOG_DROP_WARN(&rl)) {
3301 odp_flow_key_format(key, key_len, &s);
3302 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
3306 COVERAGE_INC(facet_unexpected);
3307 dpif_flow_del(dpif, key, key_len, NULL);
3310 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3312 * This function also pushes statistics updates to rules which each facet
3313 * resubmits into. Generally these statistics will be accurate. However, if a
3314 * facet changes the rule it resubmits into at some time in between
3315 * update_stats() runs, it is possible that statistics accrued to the
3316 * old rule will be incorrectly attributed to the new rule. This could be
3317 * avoided by calling update_stats() whenever rules are created or
3318 * deleted. However, the performance impact of making so many calls to the
3319 * datapath do not justify the benefit of having perfectly accurate statistics.
3322 update_stats(struct ofproto_dpif *p)
3324 const struct dpif_flow_stats *stats;
3325 struct dpif_flow_dump dump;
3326 const struct nlattr *key;
3329 dpif_flow_dump_start(&dump, p->dpif);
3330 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
3331 struct subfacet *subfacet;
3333 subfacet = subfacet_find(p, key, key_len);
3334 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3336 update_subfacet_stats(subfacet, stats);
3340 /* Stats are updated per-packet. */
3343 case SF_NOT_INSTALLED:
3345 delete_unexpected_flow(p->dpif, key, key_len);
3349 dpif_flow_dump_done(&dump);
3352 /* Calculates and returns the number of milliseconds of idle time after which
3353 * subfacets should expire from the datapath. When a subfacet expires, we fold
3354 * its statistics into its facet, and when a facet's last subfacet expires, we
3355 * fold its statistic into its rule. */
3357 subfacet_max_idle(const struct ofproto_dpif *ofproto)
3360 * Idle time histogram.
3362 * Most of the time a switch has a relatively small number of subfacets.
3363 * When this is the case we might as well keep statistics for all of them
3364 * in userspace and to cache them in the kernel datapath for performance as
3367 * As the number of subfacets increases, the memory required to maintain
3368 * statistics about them in userspace and in the kernel becomes
3369 * significant. However, with a large number of subfacets it is likely
3370 * that only a few of them are "heavy hitters" that consume a large amount
3371 * of bandwidth. At this point, only heavy hitters are worth caching in
3372 * the kernel and maintaining in userspaces; other subfacets we can
3375 * The technique used to compute the idle time is to build a histogram with
3376 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3377 * that is installed in the kernel gets dropped in the appropriate bucket.
3378 * After the histogram has been built, we compute the cutoff so that only
3379 * the most-recently-used 1% of subfacets (but at least
3380 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3381 * the most-recently-used bucket of subfacets is kept, so actually an
3382 * arbitrary number of subfacets can be kept in any given expiration run
3383 * (though the next run will delete most of those unless they receive
3386 * This requires a second pass through the subfacets, in addition to the
3387 * pass made by update_stats(), because the former function never looks at
3388 * uninstallable subfacets.
3390 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3391 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3392 int buckets[N_BUCKETS] = { 0 };
3393 int total, subtotal, bucket;
3394 struct subfacet *subfacet;
3398 total = hmap_count(&ofproto->subfacets);
3399 if (total <= ofproto->up.flow_eviction_threshold) {
3400 return N_BUCKETS * BUCKET_WIDTH;
3403 /* Build histogram. */
3405 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3406 long long int idle = now - subfacet->used;
3407 int bucket = (idle <= 0 ? 0
3408 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3409 : (unsigned int) idle / BUCKET_WIDTH);
3413 /* Find the first bucket whose flows should be expired. */
3414 subtotal = bucket = 0;
3416 subtotal += buckets[bucket++];
3417 } while (bucket < N_BUCKETS &&
3418 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3420 if (VLOG_IS_DBG_ENABLED()) {
3424 ds_put_cstr(&s, "keep");
3425 for (i = 0; i < N_BUCKETS; i++) {
3427 ds_put_cstr(&s, ", drop");
3430 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3433 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3437 return bucket * BUCKET_WIDTH;
3440 enum { EXPIRE_MAX_BATCH = 50 };
3443 expire_batch(struct ofproto_dpif *ofproto, struct subfacet **subfacets, int n)
3445 struct odputil_keybuf keybufs[EXPIRE_MAX_BATCH];
3446 struct dpif_op ops[EXPIRE_MAX_BATCH];
3447 struct dpif_op *opsp[EXPIRE_MAX_BATCH];
3448 struct ofpbuf keys[EXPIRE_MAX_BATCH];
3449 struct dpif_flow_stats stats[EXPIRE_MAX_BATCH];
3452 for (i = 0; i < n; i++) {
3453 ops[i].type = DPIF_OP_FLOW_DEL;
3454 subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]);
3455 ops[i].u.flow_del.key = keys[i].data;
3456 ops[i].u.flow_del.key_len = keys[i].size;
3457 ops[i].u.flow_del.stats = &stats[i];
3461 dpif_operate(ofproto->dpif, opsp, n);
3462 for (i = 0; i < n; i++) {
3463 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
3464 subfacets[i]->path = SF_NOT_INSTALLED;
3465 subfacet_destroy(subfacets[i]);
3470 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3472 /* Cutoff time for most flows. */
3473 long long int normal_cutoff = time_msec() - dp_max_idle;
3475 /* We really want to keep flows for special protocols around, so use a more
3476 * conservative cutoff. */
3477 long long int special_cutoff = time_msec() - 10000;
3479 struct subfacet *subfacet, *next_subfacet;
3480 struct subfacet *batch[EXPIRE_MAX_BATCH];
3484 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3485 &ofproto->subfacets) {
3486 long long int cutoff;
3488 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
3491 if (subfacet->used < cutoff) {
3492 if (subfacet->path != SF_NOT_INSTALLED) {
3493 batch[n_batch++] = subfacet;
3494 if (n_batch >= EXPIRE_MAX_BATCH) {
3495 expire_batch(ofproto, batch, n_batch);
3499 subfacet_destroy(subfacet);
3505 expire_batch(ofproto, batch, n_batch);
3509 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3510 * then delete it entirely. */
3512 rule_expire(struct rule_dpif *rule)
3514 struct facet *facet, *next_facet;
3518 /* Has 'rule' expired? */
3520 if (rule->up.hard_timeout
3521 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3522 reason = OFPRR_HARD_TIMEOUT;
3523 } else if (rule->up.idle_timeout
3524 && now > rule->up.used + rule->up.idle_timeout * 1000) {
3525 reason = OFPRR_IDLE_TIMEOUT;
3530 COVERAGE_INC(ofproto_dpif_expired);
3532 /* Update stats. (This is a no-op if the rule expired due to an idle
3533 * timeout, because that only happens when the rule has no facets left.) */
3534 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3535 facet_remove(facet);
3538 /* Get rid of the rule. */
3539 ofproto_rule_expire(&rule->up, reason);
3544 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3546 * The caller must already have determined that no facet with an identical
3547 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3548 * the ofproto's classifier table.
3550 * 'hash' must be the return value of flow_hash(flow, 0).
3552 * The facet will initially have no subfacets. The caller should create (at
3553 * least) one subfacet with subfacet_create(). */
3554 static struct facet *
3555 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
3557 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3558 struct facet *facet;
3560 facet = xzalloc(sizeof *facet);
3561 facet->used = time_msec();
3562 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
3563 list_push_back(&rule->facets, &facet->list_node);
3565 facet->flow = *flow;
3566 list_init(&facet->subfacets);
3567 netflow_flow_init(&facet->nf_flow);
3568 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3574 facet_free(struct facet *facet)
3579 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3580 * 'packet', which arrived on 'in_port'.
3582 * Takes ownership of 'packet'. */
3584 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3585 const struct nlattr *odp_actions, size_t actions_len,
3586 struct ofpbuf *packet)
3588 struct odputil_keybuf keybuf;
3592 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3593 odp_flow_key_from_flow(&key, flow);
3595 error = dpif_execute(ofproto->dpif, key.data, key.size,
3596 odp_actions, actions_len, packet);
3598 ofpbuf_delete(packet);
3602 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3604 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3605 * rule's statistics, via subfacet_uninstall().
3607 * - Removes 'facet' from its rule and from ofproto->facets.
3610 facet_remove(struct facet *facet)
3612 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3613 struct subfacet *subfacet, *next_subfacet;
3615 assert(!list_is_empty(&facet->subfacets));
3617 /* First uninstall all of the subfacets to get final statistics. */
3618 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3619 subfacet_uninstall(subfacet);
3622 /* Flush the final stats to the rule.
3624 * This might require us to have at least one subfacet around so that we
3625 * can use its actions for accounting in facet_account(), which is why we
3626 * have uninstalled but not yet destroyed the subfacets. */
3627 facet_flush_stats(facet);
3629 /* Now we're really all done so destroy everything. */
3630 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3631 &facet->subfacets) {
3632 subfacet_destroy__(subfacet);
3634 hmap_remove(&ofproto->facets, &facet->hmap_node);
3635 list_remove(&facet->list_node);
3639 /* Feed information from 'facet' back into the learning table to keep it in
3640 * sync with what is actually flowing through the datapath. */
3642 facet_learn(struct facet *facet)
3644 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3645 struct action_xlate_ctx ctx;
3647 if (!facet->has_learn
3648 && !facet->has_normal
3649 && (!facet->has_fin_timeout
3650 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
3654 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3655 facet->flow.vlan_tci,
3656 facet->rule, facet->tcp_flags, NULL);
3657 ctx.may_learn = true;
3658 xlate_actions_for_side_effects(&ctx, facet->rule->up.actions,
3659 facet->rule->up.n_actions);
3663 facet_account(struct facet *facet)
3665 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3666 struct subfacet *subfacet;
3667 const struct nlattr *a;
3672 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3675 n_bytes = facet->byte_count - facet->accounted_bytes;
3677 /* This loop feeds byte counters to bond_account() for rebalancing to use
3678 * as a basis. We also need to track the actual VLAN on which the packet
3679 * is going to be sent to ensure that it matches the one passed to
3680 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3683 * We use the actions from an arbitrary subfacet because they should all
3684 * be equally valid for our purpose. */
3685 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3686 struct subfacet, list_node);
3687 vlan_tci = facet->flow.vlan_tci;
3688 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3689 subfacet->actions, subfacet->actions_len) {
3690 const struct ovs_action_push_vlan *vlan;
3691 struct ofport_dpif *port;
3693 switch (nl_attr_type(a)) {
3694 case OVS_ACTION_ATTR_OUTPUT:
3695 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3696 if (port && port->bundle && port->bundle->bond) {
3697 bond_account(port->bundle->bond, &facet->flow,
3698 vlan_tci_to_vid(vlan_tci), n_bytes);
3702 case OVS_ACTION_ATTR_POP_VLAN:
3703 vlan_tci = htons(0);
3706 case OVS_ACTION_ATTR_PUSH_VLAN:
3707 vlan = nl_attr_get(a);
3708 vlan_tci = vlan->vlan_tci;
3714 /* Returns true if the only action for 'facet' is to send to the controller.
3715 * (We don't report NetFlow expiration messages for such facets because they
3716 * are just part of the control logic for the network, not real traffic). */
3718 facet_is_controller_flow(struct facet *facet)
3721 && facet->rule->up.n_actions == 1
3722 && action_outputs_to_port(&facet->rule->up.actions[0],
3723 htons(OFPP_CONTROLLER)));
3726 /* Folds all of 'facet''s statistics into its rule. Also updates the
3727 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3728 * 'facet''s statistics in the datapath should have been zeroed and folded into
3729 * its packet and byte counts before this function is called. */
3731 facet_flush_stats(struct facet *facet)
3733 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3734 struct subfacet *subfacet;
3736 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3737 assert(!subfacet->dp_byte_count);
3738 assert(!subfacet->dp_packet_count);
3741 facet_push_stats(facet);
3742 if (facet->accounted_bytes < facet->byte_count) {
3743 facet_account(facet);
3744 facet->accounted_bytes = facet->byte_count;
3747 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3748 struct ofexpired expired;
3749 expired.flow = facet->flow;
3750 expired.packet_count = facet->packet_count;
3751 expired.byte_count = facet->byte_count;
3752 expired.used = facet->used;
3753 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3756 facet->rule->packet_count += facet->packet_count;
3757 facet->rule->byte_count += facet->byte_count;
3759 /* Reset counters to prevent double counting if 'facet' ever gets
3761 facet_reset_counters(facet);
3763 netflow_flow_clear(&facet->nf_flow);
3764 facet->tcp_flags = 0;
3767 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3768 * Returns it if found, otherwise a null pointer.
3770 * 'hash' must be the return value of flow_hash(flow, 0).
3772 * The returned facet might need revalidation; use facet_lookup_valid()
3773 * instead if that is important. */
3774 static struct facet *
3775 facet_find(struct ofproto_dpif *ofproto,
3776 const struct flow *flow, uint32_t hash)
3778 struct facet *facet;
3780 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
3781 if (flow_equal(flow, &facet->flow)) {
3789 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3790 * Returns it if found, otherwise a null pointer.
3792 * 'hash' must be the return value of flow_hash(flow, 0).
3794 * The returned facet is guaranteed to be valid. */
3795 static struct facet *
3796 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
3799 struct facet *facet;
3801 facet = facet_find(ofproto, flow, hash);
3803 && (ofproto->need_revalidate
3804 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))) {
3805 facet_revalidate(facet);
3812 subfacet_path_to_string(enum subfacet_path path)
3815 case SF_NOT_INSTALLED:
3816 return "not installed";
3818 return "in fast path";
3820 return "in slow path";
3826 /* Returns the path in which a subfacet should be installed if its 'slow'
3827 * member has the specified value. */
3828 static enum subfacet_path
3829 subfacet_want_path(enum slow_path_reason slow)
3831 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
3834 /* Returns true if 'subfacet' needs to have its datapath flow updated,
3835 * supposing that its actions have been recalculated as 'want_actions' and that
3836 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
3838 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
3839 const struct ofpbuf *want_actions)
3841 enum subfacet_path want_path = subfacet_want_path(slow);
3842 return (want_path != subfacet->path
3843 || (want_path == SF_FAST_PATH
3844 && (subfacet->actions_len != want_actions->size
3845 || memcmp(subfacet->actions, want_actions->data,
3846 subfacet->actions_len))));
3850 facet_check_consistency(struct facet *facet)
3852 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3854 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3856 uint64_t odp_actions_stub[1024 / 8];
3857 struct ofpbuf odp_actions;
3859 struct rule_dpif *rule;
3860 struct subfacet *subfacet;
3861 bool may_log = false;
3864 /* Check the rule for consistency. */
3865 rule = rule_dpif_lookup(ofproto, &facet->flow);
3866 ok = rule == facet->rule;
3868 may_log = !VLOG_DROP_WARN(&rl);
3873 flow_format(&s, &facet->flow);
3874 ds_put_format(&s, ": facet associated with wrong rule (was "
3875 "table=%"PRIu8",", facet->rule->up.table_id);
3876 cls_rule_format(&facet->rule->up.cr, &s);
3877 ds_put_format(&s, ") (should have been table=%"PRIu8",",
3879 cls_rule_format(&rule->up.cr, &s);
3880 ds_put_char(&s, ')');
3882 VLOG_WARN("%s", ds_cstr(&s));
3887 /* Check the datapath actions for consistency. */
3888 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
3889 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3890 enum subfacet_path want_path;
3891 struct odputil_keybuf keybuf;
3892 struct action_xlate_ctx ctx;
3896 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3897 subfacet->initial_tci, rule, 0, NULL);
3898 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions,
3901 if (subfacet->path == SF_NOT_INSTALLED) {
3902 /* This only happens if the datapath reported an error when we
3903 * tried to install the flow. Don't flag another error here. */
3907 want_path = subfacet_want_path(subfacet->slow);
3908 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
3909 /* The actions for slow-path flows may legitimately vary from one
3910 * packet to the next. We're done. */
3914 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
3918 /* Inconsistency! */
3920 may_log = !VLOG_DROP_WARN(&rl);
3924 /* Rate-limited, skip reporting. */
3929 subfacet_get_key(subfacet, &keybuf, &key);
3930 odp_flow_key_format(key.data, key.size, &s);
3932 ds_put_cstr(&s, ": inconsistency in subfacet");
3933 if (want_path != subfacet->path) {
3934 enum odp_key_fitness fitness = subfacet->key_fitness;
3936 ds_put_format(&s, " (%s, fitness=%s)",
3937 subfacet_path_to_string(subfacet->path),
3938 odp_key_fitness_to_string(fitness));
3939 ds_put_format(&s, " (should have been %s)",
3940 subfacet_path_to_string(want_path));
3941 } else if (want_path == SF_FAST_PATH) {
3942 ds_put_cstr(&s, " (actions were: ");
3943 format_odp_actions(&s, subfacet->actions,
3944 subfacet->actions_len);
3945 ds_put_cstr(&s, ") (correct actions: ");
3946 format_odp_actions(&s, odp_actions.data, odp_actions.size);
3947 ds_put_char(&s, ')');
3949 ds_put_cstr(&s, " (actions: ");
3950 format_odp_actions(&s, subfacet->actions,
3951 subfacet->actions_len);
3952 ds_put_char(&s, ')');
3954 VLOG_WARN("%s", ds_cstr(&s));
3957 ofpbuf_uninit(&odp_actions);
3962 /* Re-searches the classifier for 'facet':
3964 * - If the rule found is different from 'facet''s current rule, moves
3965 * 'facet' to the new rule and recompiles its actions.
3967 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3968 * where it is and recompiles its actions anyway. */
3970 facet_revalidate(struct facet *facet)
3972 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3974 struct nlattr *odp_actions;
3977 struct actions *new_actions;
3979 struct action_xlate_ctx ctx;
3980 uint64_t odp_actions_stub[1024 / 8];
3981 struct ofpbuf odp_actions;
3983 struct rule_dpif *new_rule;
3984 struct subfacet *subfacet;
3987 COVERAGE_INC(facet_revalidate);
3989 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
3991 /* Calculate new datapath actions.
3993 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3994 * emit a NetFlow expiration and, if so, we need to have the old state
3995 * around to properly compose it. */
3997 /* If the datapath actions changed or the installability changed,
3998 * then we need to talk to the datapath. */
4001 memset(&ctx, 0, sizeof ctx);
4002 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4003 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4004 enum slow_path_reason slow;
4006 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4007 subfacet->initial_tci, new_rule, 0, NULL);
4008 xlate_actions(&ctx, new_rule->up.actions, new_rule->up.n_actions,
4011 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4012 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
4013 struct dpif_flow_stats stats;
4015 subfacet_install(subfacet,
4016 odp_actions.data, odp_actions.size, &stats, slow);
4017 subfacet_update_stats(subfacet, &stats);
4020 new_actions = xcalloc(list_size(&facet->subfacets),
4021 sizeof *new_actions);
4023 new_actions[i].odp_actions = xmemdup(odp_actions.data,
4025 new_actions[i].actions_len = odp_actions.size;
4030 ofpbuf_uninit(&odp_actions);
4033 facet_flush_stats(facet);
4036 /* Update 'facet' now that we've taken care of all the old state. */
4037 facet->tags = ctx.tags;
4038 facet->nf_flow.output_iface = ctx.nf_output_iface;
4039 facet->has_learn = ctx.has_learn;
4040 facet->has_normal = ctx.has_normal;
4041 facet->has_fin_timeout = ctx.has_fin_timeout;
4042 facet->mirrors = ctx.mirrors;
4045 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4046 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4048 if (new_actions && new_actions[i].odp_actions) {
4049 free(subfacet->actions);
4050 subfacet->actions = new_actions[i].odp_actions;
4051 subfacet->actions_len = new_actions[i].actions_len;
4057 if (facet->rule != new_rule) {
4058 COVERAGE_INC(facet_changed_rule);
4059 list_remove(&facet->list_node);
4060 list_push_back(&new_rule->facets, &facet->list_node);
4061 facet->rule = new_rule;
4062 facet->used = new_rule->up.created;
4063 facet->prev_used = facet->used;
4067 /* Updates 'facet''s used time. Caller is responsible for calling
4068 * facet_push_stats() to update the flows which 'facet' resubmits into. */
4070 facet_update_time(struct facet *facet, long long int used)
4072 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4073 if (used > facet->used) {
4075 ofproto_rule_update_used(&facet->rule->up, used);
4076 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
4081 facet_reset_counters(struct facet *facet)
4083 facet->packet_count = 0;
4084 facet->byte_count = 0;
4085 facet->prev_packet_count = 0;
4086 facet->prev_byte_count = 0;
4087 facet->accounted_bytes = 0;
4091 facet_push_stats(struct facet *facet)
4093 struct dpif_flow_stats stats;
4095 assert(facet->packet_count >= facet->prev_packet_count);
4096 assert(facet->byte_count >= facet->prev_byte_count);
4097 assert(facet->used >= facet->prev_used);
4099 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4100 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4101 stats.used = facet->used;
4102 stats.tcp_flags = 0;
4104 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
4105 facet->prev_packet_count = facet->packet_count;
4106 facet->prev_byte_count = facet->byte_count;
4107 facet->prev_used = facet->used;
4109 flow_push_stats(facet->rule, &facet->flow, &stats);
4111 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
4112 facet->mirrors, stats.n_packets, stats.n_bytes);
4117 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4119 rule->packet_count += stats->n_packets;
4120 rule->byte_count += stats->n_bytes;
4121 ofproto_rule_update_used(&rule->up, stats->used);
4124 /* Pushes flow statistics to the rules which 'flow' resubmits into given
4125 * 'rule''s actions and mirrors. */
4127 flow_push_stats(struct rule_dpif *rule,
4128 const struct flow *flow, const struct dpif_flow_stats *stats)
4130 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4131 struct action_xlate_ctx ctx;
4133 ofproto_rule_update_used(&rule->up, stats->used);
4135 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule,
4137 ctx.resubmit_stats = stats;
4138 xlate_actions_for_side_effects(&ctx, rule->up.actions, rule->up.n_actions);
4143 static struct subfacet *
4144 subfacet_find__(struct ofproto_dpif *ofproto,
4145 const struct nlattr *key, size_t key_len, uint32_t key_hash,
4146 const struct flow *flow)
4148 struct subfacet *subfacet;
4150 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4151 &ofproto->subfacets) {
4153 ? (subfacet->key_len == key_len
4154 && !memcmp(key, subfacet->key, key_len))
4155 : flow_equal(flow, &subfacet->facet->flow)) {
4163 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4164 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
4165 * there is one, otherwise creates and returns a new subfacet.
4167 * If the returned subfacet is new, then subfacet->actions will be NULL, in
4168 * which case the caller must populate the actions with
4169 * subfacet_make_actions(). */
4170 static struct subfacet *
4171 subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness,
4172 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
4174 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4175 uint32_t key_hash = odp_flow_key_hash(key, key_len);
4176 struct subfacet *subfacet;
4178 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
4180 if (subfacet->facet == facet) {
4184 /* This shouldn't happen. */
4185 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4186 subfacet_destroy(subfacet);
4189 subfacet = (list_is_empty(&facet->subfacets)
4190 ? &facet->one_subfacet
4191 : xmalloc(sizeof *subfacet));
4192 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
4193 list_push_back(&facet->subfacets, &subfacet->list_node);
4194 subfacet->facet = facet;
4195 subfacet->key_fitness = key_fitness;
4196 if (key_fitness != ODP_FIT_PERFECT) {
4197 subfacet->key = xmemdup(key, key_len);
4198 subfacet->key_len = key_len;
4200 subfacet->key = NULL;
4201 subfacet->key_len = 0;
4203 subfacet->used = time_msec();
4204 subfacet->dp_packet_count = 0;
4205 subfacet->dp_byte_count = 0;
4206 subfacet->actions_len = 0;
4207 subfacet->actions = NULL;
4208 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
4211 subfacet->path = SF_NOT_INSTALLED;
4212 subfacet->initial_tci = initial_tci;
4217 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
4218 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
4219 static struct subfacet *
4220 subfacet_find(struct ofproto_dpif *ofproto,
4221 const struct nlattr *key, size_t key_len)
4223 uint32_t key_hash = odp_flow_key_hash(key, key_len);
4224 enum odp_key_fitness fitness;
4227 fitness = odp_flow_key_to_flow(key, key_len, &flow);
4228 if (fitness == ODP_FIT_ERROR) {
4232 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
4235 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4236 * its facet within 'ofproto', and frees it. */
4238 subfacet_destroy__(struct subfacet *subfacet)
4240 struct facet *facet = subfacet->facet;
4241 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4243 subfacet_uninstall(subfacet);
4244 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
4245 list_remove(&subfacet->list_node);
4246 free(subfacet->key);
4247 free(subfacet->actions);
4248 if (subfacet != &facet->one_subfacet) {
4253 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4254 * last remaining subfacet in its facet destroys the facet too. */
4256 subfacet_destroy(struct subfacet *subfacet)
4258 struct facet *facet = subfacet->facet;
4260 if (list_is_singleton(&facet->subfacets)) {
4261 /* facet_remove() needs at least one subfacet (it will remove it). */
4262 facet_remove(facet);
4264 subfacet_destroy__(subfacet);
4268 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
4269 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
4270 * for use as temporary storage. */
4272 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
4275 if (!subfacet->key) {
4276 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
4277 odp_flow_key_from_flow(key, &subfacet->facet->flow);
4279 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
4283 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
4284 * Translates the actions into 'odp_actions', which the caller must have
4285 * initialized and is responsible for uninitializing. */
4287 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
4288 struct ofpbuf *odp_actions)
4290 struct facet *facet = subfacet->facet;
4291 struct rule_dpif *rule = facet->rule;
4292 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4294 struct action_xlate_ctx ctx;
4296 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
4298 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions, odp_actions);
4299 facet->tags = ctx.tags;
4300 facet->has_learn = ctx.has_learn;
4301 facet->has_normal = ctx.has_normal;
4302 facet->has_fin_timeout = ctx.has_fin_timeout;
4303 facet->nf_flow.output_iface = ctx.nf_output_iface;
4304 facet->mirrors = ctx.mirrors;
4306 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4307 if (subfacet->actions_len != odp_actions->size
4308 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
4309 free(subfacet->actions);
4310 subfacet->actions_len = odp_actions->size;
4311 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
4315 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4316 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4317 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4318 * since 'subfacet' was last updated.
4320 * Returns 0 if successful, otherwise a positive errno value. */
4322 subfacet_install(struct subfacet *subfacet,
4323 const struct nlattr *actions, size_t actions_len,
4324 struct dpif_flow_stats *stats,
4325 enum slow_path_reason slow)
4327 struct facet *facet = subfacet->facet;
4328 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4329 enum subfacet_path path = subfacet_want_path(slow);
4330 uint64_t slow_path_stub[128 / 8];
4331 struct odputil_keybuf keybuf;
4332 enum dpif_flow_put_flags flags;
4336 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
4338 flags |= DPIF_FP_ZERO_STATS;
4341 if (path == SF_SLOW_PATH) {
4342 compose_slow_path(ofproto, &facet->flow, slow,
4343 slow_path_stub, sizeof slow_path_stub,
4344 &actions, &actions_len);
4347 subfacet_get_key(subfacet, &keybuf, &key);
4348 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
4349 actions, actions_len, stats);
4352 subfacet_reset_dp_stats(subfacet, stats);
4356 subfacet->path = path;
4362 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
4364 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
4365 stats, subfacet->slow);
4368 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4370 subfacet_uninstall(struct subfacet *subfacet)
4372 if (subfacet->path != SF_NOT_INSTALLED) {
4373 struct rule_dpif *rule = subfacet->facet->rule;
4374 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4375 struct odputil_keybuf keybuf;
4376 struct dpif_flow_stats stats;
4380 subfacet_get_key(subfacet, &keybuf, &key);
4381 error = dpif_flow_del(ofproto->dpif, key.data, key.size, &stats);
4382 subfacet_reset_dp_stats(subfacet, &stats);
4384 subfacet_update_stats(subfacet, &stats);
4386 subfacet->path = SF_NOT_INSTALLED;
4388 assert(subfacet->dp_packet_count == 0);
4389 assert(subfacet->dp_byte_count == 0);
4393 /* Resets 'subfacet''s datapath statistics counters. This should be called
4394 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4395 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4396 * was reset in the datapath. 'stats' will be modified to include only
4397 * statistics new since 'subfacet' was last updated. */
4399 subfacet_reset_dp_stats(struct subfacet *subfacet,
4400 struct dpif_flow_stats *stats)
4403 && subfacet->dp_packet_count <= stats->n_packets
4404 && subfacet->dp_byte_count <= stats->n_bytes) {
4405 stats->n_packets -= subfacet->dp_packet_count;
4406 stats->n_bytes -= subfacet->dp_byte_count;
4409 subfacet->dp_packet_count = 0;
4410 subfacet->dp_byte_count = 0;
4413 /* Updates 'subfacet''s used time. The caller is responsible for calling
4414 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
4416 subfacet_update_time(struct subfacet *subfacet, long long int used)
4418 if (used > subfacet->used) {
4419 subfacet->used = used;
4420 facet_update_time(subfacet->facet, used);
4424 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4426 * Because of the meaning of a subfacet's counters, it only makes sense to do
4427 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4428 * represents a packet that was sent by hand or if it represents statistics
4429 * that have been cleared out of the datapath. */
4431 subfacet_update_stats(struct subfacet *subfacet,
4432 const struct dpif_flow_stats *stats)
4434 if (stats->n_packets || stats->used > subfacet->used) {
4435 struct facet *facet = subfacet->facet;
4437 subfacet_update_time(subfacet, stats->used);
4438 facet->packet_count += stats->n_packets;
4439 facet->byte_count += stats->n_bytes;
4440 facet->tcp_flags |= stats->tcp_flags;
4441 facet_push_stats(facet);
4442 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
4448 static struct rule_dpif *
4449 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
4451 struct ofport_dpif *port;
4452 struct rule_dpif *rule;
4454 rule = rule_dpif_lookup__(ofproto, flow, 0);
4459 port = get_ofp_port(ofproto, flow->in_port);
4461 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
4462 return ofproto->miss_rule;
4465 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
4466 return ofproto->no_packet_in_rule;
4468 return ofproto->miss_rule;
4471 static struct rule_dpif *
4472 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
4475 struct cls_rule *cls_rule;
4476 struct classifier *cls;
4478 if (table_id >= N_TABLES) {
4482 cls = &ofproto->up.tables[table_id].cls;
4483 if (flow->nw_frag & FLOW_NW_FRAG_ANY
4484 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4485 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
4486 * are unavailable. */
4487 struct flow ofpc_normal_flow = *flow;
4488 ofpc_normal_flow.tp_src = htons(0);
4489 ofpc_normal_flow.tp_dst = htons(0);
4490 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
4492 cls_rule = classifier_lookup(cls, flow);
4494 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
4498 complete_operation(struct rule_dpif *rule)
4500 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4502 rule_invalidate(rule);
4504 struct dpif_completion *c = xmalloc(sizeof *c);
4505 c->op = rule->up.pending;
4506 list_push_back(&ofproto->completions, &c->list_node);
4508 ofoperation_complete(rule->up.pending, 0);
4512 static struct rule *
4515 struct rule_dpif *rule = xmalloc(sizeof *rule);
4520 rule_dealloc(struct rule *rule_)
4522 struct rule_dpif *rule = rule_dpif_cast(rule_);
4527 rule_construct(struct rule *rule_)
4529 struct rule_dpif *rule = rule_dpif_cast(rule_);
4530 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4531 struct rule_dpif *victim;
4535 error = validate_actions(rule->up.actions, rule->up.n_actions,
4536 &rule->up.cr.flow, ofproto->max_ports);
4541 rule->packet_count = 0;
4542 rule->byte_count = 0;
4544 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
4545 if (victim && !list_is_empty(&victim->facets)) {
4546 struct facet *facet;
4548 rule->facets = victim->facets;
4549 list_moved(&rule->facets);
4550 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4551 /* XXX: We're only clearing our local counters here. It's possible
4552 * that quite a few packets are unaccounted for in the datapath
4553 * statistics. These will be accounted to the new rule instead of
4554 * cleared as required. This could be fixed by clearing out the
4555 * datapath statistics for this facet, but currently it doesn't
4557 facet_reset_counters(facet);
4561 /* Must avoid list_moved() in this case. */
4562 list_init(&rule->facets);
4565 table_id = rule->up.table_id;
4566 rule->tag = (victim ? victim->tag
4568 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
4569 ofproto->tables[table_id].basis));
4571 complete_operation(rule);
4576 rule_destruct(struct rule *rule_)
4578 struct rule_dpif *rule = rule_dpif_cast(rule_);
4579 struct facet *facet, *next_facet;
4581 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4582 facet_revalidate(facet);
4585 complete_operation(rule);
4589 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4591 struct rule_dpif *rule = rule_dpif_cast(rule_);
4592 struct facet *facet;
4594 /* Start from historical data for 'rule' itself that are no longer tracked
4595 * in facets. This counts, for example, facets that have expired. */
4596 *packets = rule->packet_count;
4597 *bytes = rule->byte_count;
4599 /* Add any statistics that are tracked by facets. This includes
4600 * statistical data recently updated by ofproto_update_stats() as well as
4601 * stats for packets that were executed "by hand" via dpif_execute(). */
4602 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4603 *packets += facet->packet_count;
4604 *bytes += facet->byte_count;
4609 rule_execute(struct rule *rule_, const struct flow *flow,
4610 struct ofpbuf *packet)
4612 struct rule_dpif *rule = rule_dpif_cast(rule_);
4613 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4615 struct dpif_flow_stats stats;
4617 struct action_xlate_ctx ctx;
4618 uint64_t odp_actions_stub[1024 / 8];
4619 struct ofpbuf odp_actions;
4621 dpif_flow_stats_extract(flow, packet, &stats);
4622 rule_credit_stats(rule, &stats);
4624 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4625 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
4626 rule, stats.tcp_flags, packet);
4627 ctx.resubmit_stats = &stats;
4628 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions, &odp_actions);
4630 execute_odp_actions(ofproto, flow, odp_actions.data,
4631 odp_actions.size, packet);
4633 ofpbuf_uninit(&odp_actions);
4639 rule_modify_actions(struct rule *rule_)
4641 struct rule_dpif *rule = rule_dpif_cast(rule_);
4642 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4645 error = validate_actions(rule->up.actions, rule->up.n_actions,
4646 &rule->up.cr.flow, ofproto->max_ports);
4648 ofoperation_complete(rule->up.pending, error);
4652 complete_operation(rule);
4655 /* Sends 'packet' out 'ofport'.
4656 * May modify 'packet'.
4657 * Returns 0 if successful, otherwise a positive errno value. */
4659 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4661 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4662 struct ofpbuf key, odp_actions;
4663 struct odputil_keybuf keybuf;
4668 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
4669 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4671 if (odp_port != ofport->odp_port) {
4672 eth_pop_vlan(packet);
4673 flow.vlan_tci = htons(0);
4676 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4677 odp_flow_key_from_flow(&key, &flow);
4679 ofpbuf_init(&odp_actions, 32);
4680 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4682 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4683 error = dpif_execute(ofproto->dpif,
4685 odp_actions.data, odp_actions.size,
4687 ofpbuf_uninit(&odp_actions);
4690 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4691 ofproto->up.name, odp_port, strerror(error));
4693 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
4697 /* OpenFlow to datapath action translation. */
4699 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
4700 struct action_xlate_ctx *ctx);
4701 static void xlate_normal(struct action_xlate_ctx *);
4703 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
4704 * The action will state 'slow' as the reason that the action is in the slow
4705 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
4706 * dump-flows" output to see why a flow is in the slow path.)
4708 * The 'stub_size' bytes in 'stub' will be used to store the action.
4709 * 'stub_size' must be large enough for the action.
4711 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
4714 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
4715 enum slow_path_reason slow,
4716 uint64_t *stub, size_t stub_size,
4717 const struct nlattr **actionsp, size_t *actions_lenp)
4719 union user_action_cookie cookie;
4722 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
4723 cookie.slow_path.unused = 0;
4724 cookie.slow_path.reason = slow;
4726 ofpbuf_use_stack(&buf, stub, stub_size);
4727 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
4728 uint32_t pid = dpif_port_get_pid(ofproto->dpif, UINT16_MAX);
4729 odp_put_userspace_action(pid, &cookie, &buf);
4731 put_userspace_action(ofproto, &buf, flow, &cookie);
4733 *actionsp = buf.data;
4734 *actions_lenp = buf.size;
4738 put_userspace_action(const struct ofproto_dpif *ofproto,
4739 struct ofpbuf *odp_actions,
4740 const struct flow *flow,
4741 const union user_action_cookie *cookie)
4745 pid = dpif_port_get_pid(ofproto->dpif,
4746 ofp_port_to_odp_port(flow->in_port));
4748 return odp_put_userspace_action(pid, cookie, odp_actions);
4752 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
4753 ovs_be16 vlan_tci, uint32_t odp_port,
4754 unsigned int n_outputs, union user_action_cookie *cookie)
4758 cookie->type = USER_ACTION_COOKIE_SFLOW;
4759 cookie->sflow.vlan_tci = vlan_tci;
4761 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
4762 * port information") for the interpretation of cookie->output. */
4763 switch (n_outputs) {
4765 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
4766 cookie->sflow.output = 0x40000000 | 256;
4770 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4772 cookie->sflow.output = ifindex;
4777 /* 0x80000000 means "multiple output ports. */
4778 cookie->sflow.output = 0x80000000 | n_outputs;
4783 /* Compose SAMPLE action for sFlow. */
4785 compose_sflow_action(const struct ofproto_dpif *ofproto,
4786 struct ofpbuf *odp_actions,
4787 const struct flow *flow,
4790 uint32_t probability;
4791 union user_action_cookie cookie;
4792 size_t sample_offset, actions_offset;
4795 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4799 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4801 /* Number of packets out of UINT_MAX to sample. */
4802 probability = dpif_sflow_get_probability(ofproto->sflow);
4803 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4805 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4806 compose_sflow_cookie(ofproto, htons(0), odp_port,
4807 odp_port == OVSP_NONE ? 0 : 1, &cookie);
4808 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4810 nl_msg_end_nested(odp_actions, actions_offset);
4811 nl_msg_end_nested(odp_actions, sample_offset);
4812 return cookie_offset;
4815 /* SAMPLE action must be first action in any given list of actions.
4816 * At this point we do not have all information required to build it. So try to
4817 * build sample action as complete as possible. */
4819 add_sflow_action(struct action_xlate_ctx *ctx)
4821 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4823 &ctx->flow, OVSP_NONE);
4824 ctx->sflow_odp_port = 0;
4825 ctx->sflow_n_outputs = 0;
4828 /* Fix SAMPLE action according to data collected while composing ODP actions.
4829 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4830 * USERSPACE action's user-cookie which is required for sflow. */
4832 fix_sflow_action(struct action_xlate_ctx *ctx)
4834 const struct flow *base = &ctx->base_flow;
4835 union user_action_cookie *cookie;
4837 if (!ctx->user_cookie_offset) {
4841 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4843 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4845 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
4846 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
4850 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4853 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4854 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4855 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4856 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4860 struct priority_to_dscp *pdscp;
4862 if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD
4863 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4867 pdscp = get_priority(ofport, ctx->flow.skb_priority);
4869 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4870 ctx->flow.nw_tos |= pdscp->dscp;
4873 /* We may not have an ofport record for this port, but it doesn't hurt
4874 * to allow forwarding to it anyhow. Maybe such a port will appear
4875 * later and we're pre-populating the flow table. */
4878 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4879 ctx->flow.vlan_tci);
4880 if (out_port != odp_port) {
4881 ctx->flow.vlan_tci = htons(0);
4883 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4884 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4886 ctx->sflow_odp_port = odp_port;
4887 ctx->sflow_n_outputs++;
4888 ctx->nf_output_iface = ofp_port;
4889 ctx->flow.vlan_tci = flow_vlan_tci;
4890 ctx->flow.nw_tos = flow_nw_tos;
4894 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4896 compose_output_action__(ctx, ofp_port, true);
4900 xlate_table_action(struct action_xlate_ctx *ctx,
4901 uint16_t in_port, uint8_t table_id)
4903 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4904 struct ofproto_dpif *ofproto = ctx->ofproto;
4905 struct rule_dpif *rule;
4906 uint16_t old_in_port;
4907 uint8_t old_table_id;
4909 old_table_id = ctx->table_id;
4910 ctx->table_id = table_id;
4912 /* Look up a flow with 'in_port' as the input port. */
4913 old_in_port = ctx->flow.in_port;
4914 ctx->flow.in_port = in_port;
4915 rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id);
4918 if (table_id > 0 && table_id < N_TABLES) {
4919 struct table_dpif *table = &ofproto->tables[table_id];
4920 if (table->other_table) {
4921 ctx->tags |= (rule && rule->tag
4923 : rule_calculate_tag(&ctx->flow,
4924 &table->other_table->wc,
4929 /* Restore the original input port. Otherwise OFPP_NORMAL and
4930 * OFPP_IN_PORT will have surprising behavior. */
4931 ctx->flow.in_port = old_in_port;
4933 if (ctx->resubmit_hook) {
4934 ctx->resubmit_hook(ctx, rule);
4938 struct rule_dpif *old_rule = ctx->rule;
4940 if (ctx->resubmit_stats) {
4941 rule_credit_stats(rule, ctx->resubmit_stats);
4946 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
4947 ctx->rule = old_rule;
4951 ctx->table_id = old_table_id;
4953 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4955 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
4956 MAX_RESUBMIT_RECURSION);
4957 ctx->max_resubmit_trigger = true;
4962 xlate_resubmit_table(struct action_xlate_ctx *ctx,
4963 const struct nx_action_resubmit *nar)
4968 in_port = (nar->in_port == htons(OFPP_IN_PORT)
4970 : ntohs(nar->in_port));
4971 table_id = nar->table == 255 ? ctx->table_id : nar->table;
4973 xlate_table_action(ctx, in_port, table_id);
4977 flood_packets(struct action_xlate_ctx *ctx, bool all)
4979 struct ofport_dpif *ofport;
4981 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
4982 uint16_t ofp_port = ofport->up.ofp_port;
4984 if (ofp_port == ctx->flow.in_port) {
4989 compose_output_action__(ctx, ofp_port, false);
4990 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
4991 compose_output_action(ctx, ofp_port);
4995 ctx->nf_output_iface = NF_OUT_FLOOD;
4999 execute_controller_action(struct action_xlate_ctx *ctx, int len,
5000 enum ofp_packet_in_reason reason,
5001 uint16_t controller_id)
5003 struct ofputil_packet_in pin;
5004 struct ofpbuf *packet;
5006 ctx->slow |= SLOW_CONTROLLER;
5011 packet = ofpbuf_clone(ctx->packet);
5013 if (packet->l2 && packet->l3) {
5014 struct eth_header *eh;
5016 eth_pop_vlan(packet);
5019 /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
5020 * LLC frame. Calculating the Ethernet type of these frames is more
5021 * trouble than seems appropriate for a simple assertion. */
5022 assert(ntohs(eh->eth_type) < ETH_TYPE_MIN
5023 || eh->eth_type == ctx->flow.dl_type);
5025 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
5026 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
5028 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
5029 eth_push_vlan(packet, ctx->flow.vlan_tci);
5033 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5034 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
5035 ctx->flow.nw_tos, ctx->flow.nw_ttl);
5039 if (ctx->flow.nw_proto == IPPROTO_TCP) {
5040 packet_set_tcp_port(packet, ctx->flow.tp_src,
5042 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
5043 packet_set_udp_port(packet, ctx->flow.tp_src,
5050 pin.packet = packet->data;
5051 pin.packet_len = packet->size;
5052 pin.reason = reason;
5053 pin.controller_id = controller_id;
5054 pin.table_id = ctx->table_id;
5055 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
5058 flow_get_metadata(&ctx->flow, &pin.fmd);
5060 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
5061 ofpbuf_delete(packet);
5065 compose_dec_ttl(struct action_xlate_ctx *ctx)
5067 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
5068 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
5072 if (ctx->flow.nw_ttl > 1) {
5076 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
5078 /* Stop processing for current table. */
5084 xlate_output_action__(struct action_xlate_ctx *ctx,
5085 uint16_t port, uint16_t max_len)
5087 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
5089 ctx->nf_output_iface = NF_OUT_DROP;
5093 compose_output_action(ctx, ctx->flow.in_port);
5096 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
5102 flood_packets(ctx, false);
5105 flood_packets(ctx, true);
5107 case OFPP_CONTROLLER:
5108 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
5114 if (port != ctx->flow.in_port) {
5115 compose_output_action(ctx, port);
5120 if (prev_nf_output_iface == NF_OUT_FLOOD) {
5121 ctx->nf_output_iface = NF_OUT_FLOOD;
5122 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
5123 ctx->nf_output_iface = prev_nf_output_iface;
5124 } else if (prev_nf_output_iface != NF_OUT_DROP &&
5125 ctx->nf_output_iface != NF_OUT_FLOOD) {
5126 ctx->nf_output_iface = NF_OUT_MULTI;
5131 xlate_output_reg_action(struct action_xlate_ctx *ctx,
5132 const struct nx_action_output_reg *naor)
5134 struct mf_subfield src;
5137 nxm_decode(&src, naor->src, naor->ofs_nbits);
5138 ofp_port = mf_get_subfield(&src, &ctx->flow);
5140 if (ofp_port <= UINT16_MAX) {
5141 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
5146 xlate_output_action(struct action_xlate_ctx *ctx,
5147 const struct ofp_action_output *oao)
5149 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
5153 xlate_enqueue_action(struct action_xlate_ctx *ctx,
5154 const struct ofp_action_enqueue *oae)
5157 uint32_t flow_priority, priority;
5160 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
5163 /* Fall back to ordinary output action. */
5164 xlate_output_action__(ctx, ntohs(oae->port), 0);
5168 /* Figure out datapath output port. */
5169 ofp_port = ntohs(oae->port);
5170 if (ofp_port == OFPP_IN_PORT) {
5171 ofp_port = ctx->flow.in_port;
5172 } else if (ofp_port == ctx->flow.in_port) {
5176 /* Add datapath actions. */
5177 flow_priority = ctx->flow.skb_priority;
5178 ctx->flow.skb_priority = priority;
5179 compose_output_action(ctx, ofp_port);
5180 ctx->flow.skb_priority = flow_priority;
5182 /* Update NetFlow output port. */
5183 if (ctx->nf_output_iface == NF_OUT_DROP) {
5184 ctx->nf_output_iface = ofp_port;
5185 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
5186 ctx->nf_output_iface = NF_OUT_MULTI;
5191 xlate_set_queue_action(struct action_xlate_ctx *ctx,
5192 const struct nx_action_set_queue *nasq)
5197 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
5200 /* Couldn't translate queue to a priority, so ignore. A warning
5201 * has already been logged. */
5205 ctx->flow.skb_priority = priority;
5208 struct xlate_reg_state {
5214 xlate_autopath(struct action_xlate_ctx *ctx,
5215 const struct nx_action_autopath *naa)
5217 uint16_t ofp_port = ntohl(naa->id);
5218 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
5220 if (!port || !port->bundle) {
5221 ofp_port = OFPP_NONE;
5222 } else if (port->bundle->bond) {
5223 /* Autopath does not support VLAN hashing. */
5224 struct ofport_dpif *slave = bond_choose_output_slave(
5225 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
5227 ofp_port = slave->up.ofp_port;
5230 autopath_execute(naa, &ctx->flow, ofp_port);
5234 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
5236 struct ofproto_dpif *ofproto = ofproto_;
5237 struct ofport_dpif *port;
5247 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
5250 port = get_ofp_port(ofproto, ofp_port);
5251 return port ? port->may_enable : false;
5256 xlate_learn_action(struct action_xlate_ctx *ctx,
5257 const struct nx_action_learn *learn)
5259 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
5260 struct ofputil_flow_mod fm;
5263 learn_execute(learn, &ctx->flow, &fm);
5265 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
5266 if (error && !VLOG_DROP_WARN(&rl)) {
5267 VLOG_WARN("learning action failed to modify flow table (%s)",
5268 ofperr_get_name(error));
5274 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
5275 * means "infinite". */
5277 reduce_timeout(uint16_t max, uint16_t *timeout)
5279 if (max && (!*timeout || *timeout > max)) {
5285 xlate_fin_timeout(struct action_xlate_ctx *ctx,
5286 const struct nx_action_fin_timeout *naft)
5288 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
5289 struct rule_dpif *rule = ctx->rule;
5291 reduce_timeout(ntohs(naft->fin_idle_timeout), &rule->up.idle_timeout);
5292 reduce_timeout(ntohs(naft->fin_hard_timeout), &rule->up.hard_timeout);
5297 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
5299 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
5300 ? OFPUTIL_PC_NO_RECV_STP
5301 : OFPUTIL_PC_NO_RECV)) {
5305 /* Only drop packets here if both forwarding and learning are
5306 * disabled. If just learning is enabled, we need to have
5307 * OFPP_NORMAL and the learning action have a look at the packet
5308 * before we can drop it. */
5309 if (!stp_forward_in_state(port->stp_state)
5310 && !stp_learn_in_state(port->stp_state)) {
5318 do_xlate_actions(const union ofp_action *in, size_t n_in,
5319 struct action_xlate_ctx *ctx)
5321 const struct ofport_dpif *port;
5322 const union ofp_action *ia;
5323 bool was_evictable = true;
5326 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5327 if (port && !may_receive(port, ctx)) {
5328 /* Drop this flow. */
5333 /* Don't let the rule we're working on get evicted underneath us. */
5334 was_evictable = ctx->rule->up.evictable;
5335 ctx->rule->up.evictable = false;
5337 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
5338 const struct ofp_action_dl_addr *oada;
5339 const struct nx_action_resubmit *nar;
5340 const struct nx_action_set_tunnel *nast;
5341 const struct nx_action_set_queue *nasq;
5342 const struct nx_action_multipath *nam;
5343 const struct nx_action_autopath *naa;
5344 const struct nx_action_bundle *nab;
5345 const struct nx_action_output_reg *naor;
5346 const struct nx_action_controller *nac;
5347 enum ofputil_action_code code;
5354 code = ofputil_decode_action_unsafe(ia);
5356 case OFPUTIL_OFPAT10_OUTPUT:
5357 xlate_output_action(ctx, &ia->output);
5360 case OFPUTIL_OFPAT10_SET_VLAN_VID:
5361 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
5362 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
5365 case OFPUTIL_OFPAT10_SET_VLAN_PCP:
5366 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
5367 ctx->flow.vlan_tci |= htons(
5368 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
5371 case OFPUTIL_OFPAT10_STRIP_VLAN:
5372 ctx->flow.vlan_tci = htons(0);
5375 case OFPUTIL_OFPAT10_SET_DL_SRC:
5376 oada = ((struct ofp_action_dl_addr *) ia);
5377 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
5380 case OFPUTIL_OFPAT10_SET_DL_DST:
5381 oada = ((struct ofp_action_dl_addr *) ia);
5382 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
5385 case OFPUTIL_OFPAT10_SET_NW_SRC:
5386 ctx->flow.nw_src = ia->nw_addr.nw_addr;
5389 case OFPUTIL_OFPAT10_SET_NW_DST:
5390 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
5393 case OFPUTIL_OFPAT10_SET_NW_TOS:
5394 /* OpenFlow 1.0 only supports IPv4. */
5395 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5396 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5397 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
5401 case OFPUTIL_OFPAT10_SET_TP_SRC:
5402 ctx->flow.tp_src = ia->tp_port.tp_port;
5405 case OFPUTIL_OFPAT10_SET_TP_DST:
5406 ctx->flow.tp_dst = ia->tp_port.tp_port;
5409 case OFPUTIL_OFPAT10_ENQUEUE:
5410 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
5413 case OFPUTIL_NXAST_RESUBMIT:
5414 nar = (const struct nx_action_resubmit *) ia;
5415 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
5418 case OFPUTIL_NXAST_RESUBMIT_TABLE:
5419 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
5422 case OFPUTIL_NXAST_SET_TUNNEL:
5423 nast = (const struct nx_action_set_tunnel *) ia;
5424 tun_id = htonll(ntohl(nast->tun_id));
5425 ctx->flow.tun_id = tun_id;
5428 case OFPUTIL_NXAST_SET_QUEUE:
5429 nasq = (const struct nx_action_set_queue *) ia;
5430 xlate_set_queue_action(ctx, nasq);
5433 case OFPUTIL_NXAST_POP_QUEUE:
5434 ctx->flow.skb_priority = ctx->orig_skb_priority;
5437 case OFPUTIL_NXAST_REG_MOVE:
5438 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
5442 case OFPUTIL_NXAST_REG_LOAD:
5443 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
5447 case OFPUTIL_NXAST_NOTE:
5448 /* Nothing to do. */
5451 case OFPUTIL_NXAST_SET_TUNNEL64:
5452 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
5453 ctx->flow.tun_id = tun_id;
5456 case OFPUTIL_NXAST_MULTIPATH:
5457 nam = (const struct nx_action_multipath *) ia;
5458 multipath_execute(nam, &ctx->flow);
5461 case OFPUTIL_NXAST_AUTOPATH:
5462 naa = (const struct nx_action_autopath *) ia;
5463 xlate_autopath(ctx, naa);
5466 case OFPUTIL_NXAST_BUNDLE:
5467 ctx->ofproto->has_bundle_action = true;
5468 nab = (const struct nx_action_bundle *) ia;
5469 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
5474 case OFPUTIL_NXAST_BUNDLE_LOAD:
5475 ctx->ofproto->has_bundle_action = true;
5476 nab = (const struct nx_action_bundle *) ia;
5477 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
5481 case OFPUTIL_NXAST_OUTPUT_REG:
5482 naor = (const struct nx_action_output_reg *) ia;
5483 xlate_output_reg_action(ctx, naor);
5486 case OFPUTIL_NXAST_LEARN:
5487 ctx->has_learn = true;
5488 if (ctx->may_learn) {
5489 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
5493 case OFPUTIL_NXAST_DEC_TTL:
5494 if (compose_dec_ttl(ctx)) {
5499 case OFPUTIL_NXAST_EXIT:
5503 case OFPUTIL_NXAST_FIN_TIMEOUT:
5504 ctx->has_fin_timeout = true;
5505 xlate_fin_timeout(ctx, (const struct nx_action_fin_timeout *) ia);
5508 case OFPUTIL_NXAST_CONTROLLER:
5509 nac = (const struct nx_action_controller *) ia;
5510 execute_controller_action(ctx, ntohs(nac->max_len), nac->reason,
5511 ntohs(nac->controller_id));
5517 /* We've let OFPP_NORMAL and the learning action look at the packet,
5518 * so drop it now if forwarding is disabled. */
5519 if (port && !stp_forward_in_state(port->stp_state)) {
5520 ofpbuf_clear(ctx->odp_actions);
5521 add_sflow_action(ctx);
5524 ctx->rule->up.evictable = was_evictable;
5529 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
5530 struct ofproto_dpif *ofproto, const struct flow *flow,
5531 ovs_be16 initial_tci, struct rule_dpif *rule,
5532 uint8_t tcp_flags, const struct ofpbuf *packet)
5534 ctx->ofproto = ofproto;
5536 ctx->base_flow = ctx->flow;
5537 ctx->base_flow.tun_id = 0;
5538 ctx->base_flow.vlan_tci = initial_tci;
5540 ctx->packet = packet;
5541 ctx->may_learn = packet != NULL;
5542 ctx->tcp_flags = tcp_flags;
5543 ctx->resubmit_hook = NULL;
5544 ctx->resubmit_stats = NULL;
5547 /* Translates the 'n_in' "union ofp_action"s in 'in' into datapath actions in
5548 * 'odp_actions', using 'ctx'. */
5550 xlate_actions(struct action_xlate_ctx *ctx,
5551 const union ofp_action *in, size_t n_in,
5552 struct ofpbuf *odp_actions)
5554 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
5555 * that in the future we always keep a copy of the original flow for
5556 * tracing purposes. */
5557 static bool hit_resubmit_limit;
5559 enum slow_path_reason special;
5561 COVERAGE_INC(ofproto_dpif_xlate);
5563 ofpbuf_clear(odp_actions);
5564 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
5566 ctx->odp_actions = odp_actions;
5569 ctx->has_learn = false;
5570 ctx->has_normal = false;
5571 ctx->has_fin_timeout = false;
5572 ctx->nf_output_iface = NF_OUT_DROP;
5575 ctx->max_resubmit_trigger = false;
5576 ctx->orig_skb_priority = ctx->flow.skb_priority;
5580 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
5581 /* Do this conditionally because the copy is expensive enough that it
5582 * shows up in profiles.
5584 * We keep orig_flow in 'ctx' only because I couldn't make GCC 4.4
5585 * believe that I wasn't using it without initializing it if I kept it
5586 * in a local variable. */
5587 ctx->orig_flow = ctx->flow;
5590 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
5591 switch (ctx->ofproto->up.frag_handling) {
5592 case OFPC_FRAG_NORMAL:
5593 /* We must pretend that transport ports are unavailable. */
5594 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
5595 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
5598 case OFPC_FRAG_DROP:
5601 case OFPC_FRAG_REASM:
5604 case OFPC_FRAG_NX_MATCH:
5605 /* Nothing to do. */
5608 case OFPC_INVALID_TTL_TO_CONTROLLER:
5613 special = process_special(ctx->ofproto, &ctx->flow, ctx->packet);
5615 ctx->slow |= special;
5617 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5618 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
5620 add_sflow_action(ctx);
5621 do_xlate_actions(in, n_in, ctx);
5623 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
5624 if (!hit_resubmit_limit) {
5625 /* We didn't record the original flow. Make sure we do from
5627 hit_resubmit_limit = true;
5628 } else if (!VLOG_DROP_ERR(&trace_rl)) {
5629 struct ds ds = DS_EMPTY_INITIALIZER;
5631 ofproto_trace(ctx->ofproto, &ctx->orig_flow, ctx->packet,
5633 VLOG_ERR("Trace triggered by excessive resubmit "
5634 "recursion:\n%s", ds_cstr(&ds));
5639 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
5640 ctx->odp_actions->data,
5641 ctx->odp_actions->size)) {
5642 ctx->slow |= SLOW_IN_BAND;
5644 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
5646 compose_output_action(ctx, OFPP_LOCAL);
5649 if (ctx->ofproto->has_mirrors) {
5650 add_mirror_actions(ctx, &ctx->orig_flow);
5652 fix_sflow_action(ctx);
5656 /* Translates the 'n_in' "union ofp_action"s in 'in' into datapath actions,
5657 * using 'ctx', and discards the datapath actions. */
5659 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
5660 const union ofp_action *in, size_t n_in)
5662 uint64_t odp_actions_stub[1024 / 8];
5663 struct ofpbuf odp_actions;
5665 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5666 xlate_actions(ctx, in, n_in, &odp_actions);
5667 ofpbuf_uninit(&odp_actions);
5670 /* OFPP_NORMAL implementation. */
5672 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
5674 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
5675 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
5676 * the bundle on which the packet was received, returns the VLAN to which the
5679 * Both 'vid' and the return value are in the range 0...4095. */
5681 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
5683 switch (in_bundle->vlan_mode) {
5684 case PORT_VLAN_ACCESS:
5685 return in_bundle->vlan;
5688 case PORT_VLAN_TRUNK:
5691 case PORT_VLAN_NATIVE_UNTAGGED:
5692 case PORT_VLAN_NATIVE_TAGGED:
5693 return vid ? vid : in_bundle->vlan;
5700 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
5701 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
5704 * 'vid' should be the VID obtained from the 802.1Q header that was received as
5705 * part of a packet (specify 0 if there was no 802.1Q header), in the range
5708 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
5710 /* Allow any VID on the OFPP_NONE port. */
5711 if (in_bundle == &ofpp_none_bundle) {
5715 switch (in_bundle->vlan_mode) {
5716 case PORT_VLAN_ACCESS:
5719 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5720 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
5721 "packet received on port %s configured as VLAN "
5722 "%"PRIu16" access port",
5723 in_bundle->ofproto->up.name, vid,
5724 in_bundle->name, in_bundle->vlan);
5730 case PORT_VLAN_NATIVE_UNTAGGED:
5731 case PORT_VLAN_NATIVE_TAGGED:
5733 /* Port must always carry its native VLAN. */
5737 case PORT_VLAN_TRUNK:
5738 if (!ofbundle_includes_vlan(in_bundle, vid)) {
5740 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5741 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
5742 "received on port %s not configured for trunking "
5744 in_bundle->ofproto->up.name, vid,
5745 in_bundle->name, vid);
5757 /* Given 'vlan', the VLAN that a packet belongs to, and
5758 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
5759 * that should be included in the 802.1Q header. (If the return value is 0,
5760 * then the 802.1Q header should only be included in the packet if there is a
5763 * Both 'vlan' and the return value are in the range 0...4095. */
5765 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
5767 switch (out_bundle->vlan_mode) {
5768 case PORT_VLAN_ACCESS:
5771 case PORT_VLAN_TRUNK:
5772 case PORT_VLAN_NATIVE_TAGGED:
5775 case PORT_VLAN_NATIVE_UNTAGGED:
5776 return vlan == out_bundle->vlan ? 0 : vlan;
5784 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
5787 struct ofport_dpif *port;
5789 ovs_be16 tci, old_tci;
5791 vid = output_vlan_to_vid(out_bundle, vlan);
5792 if (!out_bundle->bond) {
5793 port = ofbundle_get_a_port(out_bundle);
5795 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
5798 /* No slaves enabled, so drop packet. */
5803 old_tci = ctx->flow.vlan_tci;
5805 if (tci || out_bundle->use_priority_tags) {
5806 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
5808 tci |= htons(VLAN_CFI);
5811 ctx->flow.vlan_tci = tci;
5813 compose_output_action(ctx, port->up.ofp_port);
5814 ctx->flow.vlan_tci = old_tci;
5818 mirror_mask_ffs(mirror_mask_t mask)
5820 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
5825 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
5827 return (bundle->vlan_mode != PORT_VLAN_ACCESS
5828 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
5832 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
5834 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
5837 /* Returns an arbitrary interface within 'bundle'. */
5838 static struct ofport_dpif *
5839 ofbundle_get_a_port(const struct ofbundle *bundle)
5841 return CONTAINER_OF(list_front(&bundle->ports),
5842 struct ofport_dpif, bundle_node);
5846 vlan_is_mirrored(const struct ofmirror *m, int vlan)
5848 return !m->vlans || bitmap_is_set(m->vlans, vlan);
5851 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
5852 * to a VLAN. In general most packets may be mirrored but we want to drop
5853 * protocols that may confuse switches. */
5855 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
5857 /* If you change this function's behavior, please update corresponding
5858 * documentation in vswitch.xml at the same time. */
5859 if (dst[0] != 0x01) {
5860 /* All the currently banned MACs happen to start with 01 currently, so
5861 * this is a quick way to eliminate most of the good ones. */
5863 if (eth_addr_is_reserved(dst)) {
5864 /* Drop STP, IEEE pause frames, and other reserved protocols
5865 * (01-80-c2-00-00-0x). */
5869 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
5871 if ((dst[3] & 0xfe) == 0xcc &&
5872 (dst[4] & 0xfe) == 0xcc &&
5873 (dst[5] & 0xfe) == 0xcc) {
5874 /* Drop the following protocols plus others following the same
5877 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
5878 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
5879 STP Uplink Fast (01-00-0c-cd-cd-cd) */
5883 if (!(dst[3] | dst[4] | dst[5])) {
5884 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
5893 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5895 struct ofproto_dpif *ofproto = ctx->ofproto;
5896 mirror_mask_t mirrors;
5897 struct ofbundle *in_bundle;
5900 const struct nlattr *a;
5903 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5904 ctx->packet != NULL, NULL);
5908 mirrors = in_bundle->src_mirrors;
5910 /* Drop frames on bundles reserved for mirroring. */
5911 if (in_bundle->mirror_out) {
5912 if (ctx->packet != NULL) {
5913 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5914 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5915 "%s, which is reserved exclusively for mirroring",
5916 ctx->ofproto->up.name, in_bundle->name);
5922 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5923 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5926 vlan = input_vid_to_vlan(in_bundle, vid);
5928 /* Look at the output ports to check for destination selections. */
5930 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5931 ctx->odp_actions->size) {
5932 enum ovs_action_attr type = nl_attr_type(a);
5933 struct ofport_dpif *ofport;
5935 if (type != OVS_ACTION_ATTR_OUTPUT) {
5939 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5940 if (ofport && ofport->bundle) {
5941 mirrors |= ofport->bundle->dst_mirrors;
5949 /* Restore the original packet before adding the mirror actions. */
5950 ctx->flow = *orig_flow;
5955 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5957 if (!vlan_is_mirrored(m, vlan)) {
5958 mirrors &= mirrors - 1;
5962 mirrors &= ~m->dup_mirrors;
5963 ctx->mirrors |= m->dup_mirrors;
5965 output_normal(ctx, m->out, vlan);
5966 } else if (eth_dst_may_rspan(orig_flow->dl_dst)
5967 && vlan != m->out_vlan) {
5968 struct ofbundle *bundle;
5970 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5971 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5972 && !bundle->mirror_out) {
5973 output_normal(ctx, bundle, m->out_vlan);
5981 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5982 uint64_t packets, uint64_t bytes)
5988 for (; mirrors; mirrors &= mirrors - 1) {
5991 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5994 /* In normal circumstances 'm' will not be NULL. However,
5995 * if mirrors are reconfigured, we can temporarily get out
5996 * of sync in facet_revalidate(). We could "correct" the
5997 * mirror list before reaching here, but doing that would
5998 * not properly account the traffic stats we've currently
5999 * accumulated for previous mirror configuration. */
6003 m->packet_count += packets;
6004 m->byte_count += bytes;
6008 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
6009 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
6010 * indicate this; newer upstream kernels use gratuitous ARP requests. */
6012 is_gratuitous_arp(const struct flow *flow)
6014 return (flow->dl_type == htons(ETH_TYPE_ARP)
6015 && eth_addr_is_broadcast(flow->dl_dst)
6016 && (flow->nw_proto == ARP_OP_REPLY
6017 || (flow->nw_proto == ARP_OP_REQUEST
6018 && flow->nw_src == flow->nw_dst)));
6022 update_learning_table(struct ofproto_dpif *ofproto,
6023 const struct flow *flow, int vlan,
6024 struct ofbundle *in_bundle)
6026 struct mac_entry *mac;
6028 /* Don't learn the OFPP_NONE port. */
6029 if (in_bundle == &ofpp_none_bundle) {
6033 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
6037 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
6038 if (is_gratuitous_arp(flow)) {
6039 /* We don't want to learn from gratuitous ARP packets that are
6040 * reflected back over bond slaves so we lock the learning table. */
6041 if (!in_bundle->bond) {
6042 mac_entry_set_grat_arp_lock(mac);
6043 } else if (mac_entry_is_grat_arp_locked(mac)) {
6048 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
6049 /* The log messages here could actually be useful in debugging,
6050 * so keep the rate limit relatively high. */
6051 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
6052 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
6053 "on port %s in VLAN %d",
6054 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
6055 in_bundle->name, vlan);
6057 mac->port.p = in_bundle;
6058 tag_set_add(&ofproto->revalidate_set,
6059 mac_learning_changed(ofproto->ml, mac));
6063 static struct ofbundle *
6064 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
6065 bool warn, struct ofport_dpif **in_ofportp)
6067 struct ofport_dpif *ofport;
6069 /* Find the port and bundle for the received packet. */
6070 ofport = get_ofp_port(ofproto, in_port);
6072 *in_ofportp = ofport;
6074 if (ofport && ofport->bundle) {
6075 return ofport->bundle;
6078 /* Special-case OFPP_NONE, which a controller may use as the ingress
6079 * port for traffic that it is sourcing. */
6080 if (in_port == OFPP_NONE) {
6081 return &ofpp_none_bundle;
6084 /* Odd. A few possible reasons here:
6086 * - We deleted a port but there are still a few packets queued up
6089 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
6090 * we don't know about.
6092 * - The ofproto client didn't configure the port as part of a bundle.
6095 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6097 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
6098 "port %"PRIu16, ofproto->up.name, in_port);
6103 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
6104 * dropped. Returns true if they may be forwarded, false if they should be
6107 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
6108 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
6110 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
6111 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
6112 * checked by input_vid_is_valid().
6114 * May also add tags to '*tags', although the current implementation only does
6115 * so in one special case.
6118 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
6119 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
6121 struct ofbundle *in_bundle = in_port->bundle;
6123 /* Drop frames for reserved multicast addresses
6124 * only if forward_bpdu option is absent. */
6125 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
6129 if (in_bundle->bond) {
6130 struct mac_entry *mac;
6132 switch (bond_check_admissibility(in_bundle->bond, in_port,
6133 flow->dl_dst, tags)) {
6140 case BV_DROP_IF_MOVED:
6141 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
6142 if (mac && mac->port.p != in_bundle &&
6143 (!is_gratuitous_arp(flow)
6144 || mac_entry_is_grat_arp_locked(mac))) {
6155 xlate_normal(struct action_xlate_ctx *ctx)
6157 struct ofport_dpif *in_port;
6158 struct ofbundle *in_bundle;
6159 struct mac_entry *mac;
6163 ctx->has_normal = true;
6165 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
6166 ctx->packet != NULL, &in_port);
6171 /* Drop malformed frames. */
6172 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
6173 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
6174 if (ctx->packet != NULL) {
6175 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6176 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
6177 "VLAN tag received on port %s",
6178 ctx->ofproto->up.name, in_bundle->name);
6183 /* Drop frames on bundles reserved for mirroring. */
6184 if (in_bundle->mirror_out) {
6185 if (ctx->packet != NULL) {
6186 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6187 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6188 "%s, which is reserved exclusively for mirroring",
6189 ctx->ofproto->up.name, in_bundle->name);
6195 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
6196 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6199 vlan = input_vid_to_vlan(in_bundle, vid);
6201 /* Check other admissibility requirements. */
6203 !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
6207 /* Learn source MAC. */
6208 if (ctx->may_learn) {
6209 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
6212 /* Determine output bundle. */
6213 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
6216 if (mac->port.p != in_bundle) {
6217 output_normal(ctx, mac->port.p, vlan);
6220 struct ofbundle *bundle;
6222 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
6223 if (bundle != in_bundle
6224 && ofbundle_includes_vlan(bundle, vlan)
6225 && bundle->floodable
6226 && !bundle->mirror_out) {
6227 output_normal(ctx, bundle, vlan);
6230 ctx->nf_output_iface = NF_OUT_FLOOD;
6234 /* Optimized flow revalidation.
6236 * It's a difficult problem, in general, to tell which facets need to have
6237 * their actions recalculated whenever the OpenFlow flow table changes. We
6238 * don't try to solve that general problem: for most kinds of OpenFlow flow
6239 * table changes, we recalculate the actions for every facet. This is
6240 * relatively expensive, but it's good enough if the OpenFlow flow table
6241 * doesn't change very often.
6243 * However, we can expect one particular kind of OpenFlow flow table change to
6244 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
6245 * of CPU on revalidating every facet whenever MAC learning modifies the flow
6246 * table, we add a special case that applies to flow tables in which every rule
6247 * has the same form (that is, the same wildcards), except that the table is
6248 * also allowed to have a single "catch-all" flow that matches all packets. We
6249 * optimize this case by tagging all of the facets that resubmit into the table
6250 * and invalidating the same tag whenever a flow changes in that table. The
6251 * end result is that we revalidate just the facets that need it (and sometimes
6252 * a few more, but not all of the facets or even all of the facets that
6253 * resubmit to the table modified by MAC learning). */
6255 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
6256 * into an OpenFlow table with the given 'basis'. */
6258 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
6261 if (flow_wildcards_is_catchall(wc)) {
6264 struct flow tag_flow = *flow;
6265 flow_zero_wildcards(&tag_flow, wc);
6266 return tag_create_deterministic(flow_hash(&tag_flow, secret));
6270 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
6271 * taggability of that table.
6273 * This function must be called after *each* change to a flow table. If you
6274 * skip calling it on some changes then the pointer comparisons at the end can
6275 * be invalid if you get unlucky. For example, if a flow removal causes a
6276 * cls_table to be destroyed and then a flow insertion causes a cls_table with
6277 * different wildcards to be created with the same address, then this function
6278 * will incorrectly skip revalidation. */
6280 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
6282 struct table_dpif *table = &ofproto->tables[table_id];
6283 const struct oftable *oftable = &ofproto->up.tables[table_id];
6284 struct cls_table *catchall, *other;
6285 struct cls_table *t;
6287 catchall = other = NULL;
6289 switch (hmap_count(&oftable->cls.tables)) {
6291 /* We could tag this OpenFlow table but it would make the logic a
6292 * little harder and it's a corner case that doesn't seem worth it
6298 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
6299 if (cls_table_is_catchall(t)) {
6301 } else if (!other) {
6304 /* Indicate that we can't tag this by setting both tables to
6305 * NULL. (We know that 'catchall' is already NULL.) */
6312 /* Can't tag this table. */
6316 if (table->catchall_table != catchall || table->other_table != other) {
6317 table->catchall_table = catchall;
6318 table->other_table = other;
6319 ofproto->need_revalidate = true;
6323 /* Given 'rule' that has changed in some way (either it is a rule being
6324 * inserted, a rule being deleted, or a rule whose actions are being
6325 * modified), marks facets for revalidation to ensure that packets will be
6326 * forwarded correctly according to the new state of the flow table.
6328 * This function must be called after *each* change to a flow table. See
6329 * the comment on table_update_taggable() for more information. */
6331 rule_invalidate(const struct rule_dpif *rule)
6333 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
6335 table_update_taggable(ofproto, rule->up.table_id);
6337 if (!ofproto->need_revalidate) {
6338 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
6340 if (table->other_table && rule->tag) {
6341 tag_set_add(&ofproto->revalidate_set, rule->tag);
6343 ofproto->need_revalidate = true;
6349 set_frag_handling(struct ofproto *ofproto_,
6350 enum ofp_config_flags frag_handling)
6352 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6354 if (frag_handling != OFPC_FRAG_REASM) {
6355 ofproto->need_revalidate = true;
6363 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
6364 const struct flow *flow,
6365 const union ofp_action *ofp_actions, size_t n_ofp_actions)
6367 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6370 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
6371 return OFPERR_NXBRC_BAD_IN_PORT;
6374 error = validate_actions(ofp_actions, n_ofp_actions, flow,
6375 ofproto->max_ports);
6377 struct odputil_keybuf keybuf;
6378 struct dpif_flow_stats stats;
6382 struct action_xlate_ctx ctx;
6383 uint64_t odp_actions_stub[1024 / 8];
6384 struct ofpbuf odp_actions;
6386 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
6387 odp_flow_key_from_flow(&key, flow);
6389 dpif_flow_stats_extract(flow, packet, &stats);
6391 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL,
6392 packet_get_tcp_flags(packet, flow), packet);
6393 ctx.resubmit_stats = &stats;
6395 ofpbuf_use_stub(&odp_actions,
6396 odp_actions_stub, sizeof odp_actions_stub);
6397 xlate_actions(&ctx, ofp_actions, n_ofp_actions, &odp_actions);
6398 dpif_execute(ofproto->dpif, key.data, key.size,
6399 odp_actions.data, odp_actions.size, packet);
6400 ofpbuf_uninit(&odp_actions);
6408 set_netflow(struct ofproto *ofproto_,
6409 const struct netflow_options *netflow_options)
6411 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6413 if (netflow_options) {
6414 if (!ofproto->netflow) {
6415 ofproto->netflow = netflow_create();
6417 return netflow_set_options(ofproto->netflow, netflow_options);
6419 netflow_destroy(ofproto->netflow);
6420 ofproto->netflow = NULL;
6426 get_netflow_ids(const struct ofproto *ofproto_,
6427 uint8_t *engine_type, uint8_t *engine_id)
6429 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6431 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
6435 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
6437 if (!facet_is_controller_flow(facet) &&
6438 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
6439 struct subfacet *subfacet;
6440 struct ofexpired expired;
6442 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
6443 if (subfacet->path == SF_FAST_PATH) {
6444 struct dpif_flow_stats stats;
6446 subfacet_reinstall(subfacet, &stats);
6447 subfacet_update_stats(subfacet, &stats);
6451 expired.flow = facet->flow;
6452 expired.packet_count = facet->packet_count;
6453 expired.byte_count = facet->byte_count;
6454 expired.used = facet->used;
6455 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
6460 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
6462 struct facet *facet;
6464 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6465 send_active_timeout(ofproto, facet);
6469 static struct ofproto_dpif *
6470 ofproto_dpif_lookup(const char *name)
6472 struct ofproto_dpif *ofproto;
6474 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
6475 hash_string(name, 0), &all_ofproto_dpifs) {
6476 if (!strcmp(ofproto->up.name, name)) {
6484 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
6485 const char *argv[], void *aux OVS_UNUSED)
6487 struct ofproto_dpif *ofproto;
6490 ofproto = ofproto_dpif_lookup(argv[1]);
6492 unixctl_command_reply_error(conn, "no such bridge");
6495 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6497 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6498 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6502 unixctl_command_reply(conn, "table successfully flushed");
6506 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6507 const char *argv[], void *aux OVS_UNUSED)
6509 struct ds ds = DS_EMPTY_INITIALIZER;
6510 const struct ofproto_dpif *ofproto;
6511 const struct mac_entry *e;
6513 ofproto = ofproto_dpif_lookup(argv[1]);
6515 unixctl_command_reply_error(conn, "no such bridge");
6519 ds_put_cstr(&ds, " port VLAN MAC Age\n");
6520 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
6521 struct ofbundle *bundle = e->port.p;
6522 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
6523 ofbundle_get_a_port(bundle)->odp_port,
6524 e->vlan, ETH_ADDR_ARGS(e->mac),
6525 mac_entry_age(ofproto->ml, e));
6527 unixctl_command_reply(conn, ds_cstr(&ds));
6532 struct action_xlate_ctx ctx;
6538 trace_format_rule(struct ds *result, uint8_t table_id, int level,
6539 const struct rule_dpif *rule)
6541 ds_put_char_multiple(result, '\t', level);
6543 ds_put_cstr(result, "No match\n");
6547 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
6548 table_id, ntohll(rule->up.flow_cookie));
6549 cls_rule_format(&rule->up.cr, result);
6550 ds_put_char(result, '\n');
6552 ds_put_char_multiple(result, '\t', level);
6553 ds_put_cstr(result, "OpenFlow ");
6554 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
6555 ds_put_char(result, '\n');
6559 trace_format_flow(struct ds *result, int level, const char *title,
6560 struct trace_ctx *trace)
6562 ds_put_char_multiple(result, '\t', level);
6563 ds_put_format(result, "%s: ", title);
6564 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
6565 ds_put_cstr(result, "unchanged");
6567 flow_format(result, &trace->ctx.flow);
6568 trace->flow = trace->ctx.flow;
6570 ds_put_char(result, '\n');
6574 trace_format_regs(struct ds *result, int level, const char *title,
6575 struct trace_ctx *trace)
6579 ds_put_char_multiple(result, '\t', level);
6580 ds_put_format(result, "%s:", title);
6581 for (i = 0; i < FLOW_N_REGS; i++) {
6582 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
6584 ds_put_char(result, '\n');
6588 trace_format_odp(struct ds *result, int level, const char *title,
6589 struct trace_ctx *trace)
6591 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
6593 ds_put_char_multiple(result, '\t', level);
6594 ds_put_format(result, "%s: ", title);
6595 format_odp_actions(result, odp_actions->data, odp_actions->size);
6596 ds_put_char(result, '\n');
6600 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
6602 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
6603 struct ds *result = trace->result;
6605 ds_put_char(result, '\n');
6606 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
6607 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
6608 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
6609 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
6613 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
6614 void *aux OVS_UNUSED)
6616 const char *dpname = argv[1];
6617 struct ofproto_dpif *ofproto;
6618 struct ofpbuf odp_key;
6619 struct ofpbuf *packet;
6620 ovs_be16 initial_tci;
6626 ofpbuf_init(&odp_key, 0);
6629 ofproto = ofproto_dpif_lookup(dpname);
6631 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
6635 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
6636 /* ofproto/trace dpname flow [-generate] */
6637 const char *flow_s = argv[2];
6638 const char *generate_s = argv[3];
6640 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
6641 * flow. We guess which type it is based on whether 'flow_s' contains
6642 * an '(', since a datapath flow always contains '(') but an
6643 * OpenFlow-like flow should not (in fact it's allowed but I believe
6644 * that's not documented anywhere).
6646 * An alternative would be to try to parse 'flow_s' both ways, but then
6647 * it would be tricky giving a sensible error message. After all, do
6648 * you just say "syntax error" or do you present both error messages?
6649 * Both choices seem lousy. */
6650 if (strchr(flow_s, '(')) {
6653 /* Convert string to datapath key. */
6654 ofpbuf_init(&odp_key, 0);
6655 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
6657 unixctl_command_reply_error(conn, "Bad flow syntax");
6661 /* Convert odp_key to flow. */
6662 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
6663 odp_key.size, &flow,
6664 &initial_tci, NULL);
6665 if (error == ODP_FIT_ERROR) {
6666 unixctl_command_reply_error(conn, "Invalid flow");
6672 error_s = parse_ofp_exact_flow(&flow, argv[2]);
6674 unixctl_command_reply_error(conn, error_s);
6679 initial_tci = flow.vlan_tci;
6680 vsp_adjust_flow(ofproto, &flow);
6683 /* Generate a packet, if requested. */
6685 packet = ofpbuf_new(0);
6686 flow_compose(packet, &flow);
6688 } else if (argc == 6) {
6689 /* ofproto/trace dpname priority tun_id in_port packet */
6690 const char *priority_s = argv[2];
6691 const char *tun_id_s = argv[3];
6692 const char *in_port_s = argv[4];
6693 const char *packet_s = argv[5];
6694 uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s));
6695 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
6696 uint32_t priority = atoi(priority_s);
6699 msg = eth_from_hex(packet_s, &packet);
6701 unixctl_command_reply_error(conn, msg);
6705 ds_put_cstr(&result, "Packet: ");
6706 s = ofp_packet_to_string(packet->data, packet->size);
6707 ds_put_cstr(&result, s);
6710 flow_extract(packet, priority, tun_id, in_port, &flow);
6711 initial_tci = flow.vlan_tci;
6713 unixctl_command_reply_error(conn, "Bad command syntax");
6717 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
6718 unixctl_command_reply(conn, ds_cstr(&result));
6721 ds_destroy(&result);
6722 ofpbuf_delete(packet);
6723 ofpbuf_uninit(&odp_key);
6727 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
6728 const struct ofpbuf *packet, ovs_be16 initial_tci,
6731 struct rule_dpif *rule;
6733 ds_put_cstr(ds, "Flow: ");
6734 flow_format(ds, flow);
6735 ds_put_char(ds, '\n');
6737 rule = rule_dpif_lookup(ofproto, flow);
6739 trace_format_rule(ds, 0, 0, rule);
6740 if (rule == ofproto->miss_rule) {
6741 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
6742 } else if (rule == ofproto->no_packet_in_rule) {
6743 ds_put_cstr(ds, "\nNo match, packets dropped because "
6744 "OFPPC_NO_PACKET_IN is set on in_port.\n");
6748 uint64_t odp_actions_stub[1024 / 8];
6749 struct ofpbuf odp_actions;
6751 struct trace_ctx trace;
6754 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
6757 ofpbuf_use_stub(&odp_actions,
6758 odp_actions_stub, sizeof odp_actions_stub);
6759 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
6760 rule, tcp_flags, packet);
6761 trace.ctx.resubmit_hook = trace_resubmit;
6762 xlate_actions(&trace.ctx, rule->up.actions, rule->up.n_actions,
6765 ds_put_char(ds, '\n');
6766 trace_format_flow(ds, 0, "Final flow", &trace);
6767 ds_put_cstr(ds, "Datapath actions: ");
6768 format_odp_actions(ds, odp_actions.data, odp_actions.size);
6769 ofpbuf_uninit(&odp_actions);
6771 if (trace.ctx.slow) {
6772 enum slow_path_reason slow;
6774 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
6775 "slow path because it:");
6776 for (slow = trace.ctx.slow; slow; ) {
6777 enum slow_path_reason bit = rightmost_1bit(slow);
6781 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
6784 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
6787 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
6790 ds_put_cstr(ds, "\n\t- Needs in-band special case "
6793 ds_put_cstr(ds, "\n\t (The datapath actions are "
6794 "incomplete--for complete actions, "
6795 "please supply a packet.)");
6798 case SLOW_CONTROLLER:
6799 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
6800 "to the OpenFlow controller.");
6803 ds_put_cstr(ds, "\n\t- Needs more specific matching "
6804 "than the datapath supports.");
6811 if (slow & ~SLOW_MATCH) {
6812 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
6813 "the special slow-path processing.");
6820 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6821 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6824 unixctl_command_reply(conn, NULL);
6828 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6829 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6832 unixctl_command_reply(conn, NULL);
6835 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
6836 * 'reply' describing the results. */
6838 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
6840 struct facet *facet;
6844 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6845 if (!facet_check_consistency(facet)) {
6850 ofproto->need_revalidate = true;
6854 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
6855 ofproto->up.name, errors);
6857 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
6862 ofproto_dpif_self_check(struct unixctl_conn *conn,
6863 int argc, const char *argv[], void *aux OVS_UNUSED)
6865 struct ds reply = DS_EMPTY_INITIALIZER;
6866 struct ofproto_dpif *ofproto;
6869 ofproto = ofproto_dpif_lookup(argv[1]);
6871 unixctl_command_reply_error(conn, "Unknown ofproto (use "
6872 "ofproto/list for help)");
6875 ofproto_dpif_self_check__(ofproto, &reply);
6877 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6878 ofproto_dpif_self_check__(ofproto, &reply);
6882 unixctl_command_reply(conn, ds_cstr(&reply));
6887 ofproto_dpif_unixctl_init(void)
6889 static bool registered;
6895 unixctl_command_register(
6897 "bridge {tun_id in_port packet | odp_flow [-generate]}",
6898 2, 5, ofproto_unixctl_trace, NULL);
6899 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6900 ofproto_unixctl_fdb_flush, NULL);
6901 unixctl_command_register("fdb/show", "bridge", 1, 1,
6902 ofproto_unixctl_fdb_show, NULL);
6903 unixctl_command_register("ofproto/clog", "", 0, 0,
6904 ofproto_dpif_clog, NULL);
6905 unixctl_command_register("ofproto/unclog", "", 0, 0,
6906 ofproto_dpif_unclog, NULL);
6907 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6908 ofproto_dpif_self_check, NULL);
6911 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6913 * This is deprecated. It is only for compatibility with broken device drivers
6914 * in old versions of Linux that do not properly support VLANs when VLAN
6915 * devices are not used. When broken device drivers are no longer in
6916 * widespread use, we will delete these interfaces. */
6919 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
6921 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6922 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6924 if (realdev_ofp_port == ofport->realdev_ofp_port
6925 && vid == ofport->vlandev_vid) {
6929 ofproto->need_revalidate = true;
6931 if (ofport->realdev_ofp_port) {
6934 if (realdev_ofp_port && ofport->bundle) {
6935 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6936 * themselves be part of a bundle. */
6937 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6940 ofport->realdev_ofp_port = realdev_ofp_port;
6941 ofport->vlandev_vid = vid;
6943 if (realdev_ofp_port) {
6944 vsp_add(ofport, realdev_ofp_port, vid);
6951 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
6953 return hash_2words(realdev_ofp_port, vid);
6956 /* Returns the ODP port number of the Linux VLAN device that corresponds to
6957 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
6958 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
6959 * it would return the port number of eth0.9.
6961 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
6962 * function just returns its 'realdev_odp_port' argument. */
6964 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6965 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
6967 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6968 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
6969 int vid = vlan_tci_to_vid(vlan_tci);
6970 const struct vlan_splinter *vsp;
6972 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6973 hash_realdev_vid(realdev_ofp_port, vid),
6974 &ofproto->realdev_vid_map) {
6975 if (vsp->realdev_ofp_port == realdev_ofp_port
6976 && vsp->vid == vid) {
6977 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
6981 return realdev_odp_port;
6984 static struct vlan_splinter *
6985 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
6987 struct vlan_splinter *vsp;
6989 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
6990 &ofproto->vlandev_map) {
6991 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6999 /* Returns the OpenFlow port number of the "real" device underlying the Linux
7000 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
7001 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
7002 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
7003 * eth0 and store 9 in '*vid'.
7005 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
7006 * VLAN device. Unless VLAN splinters are enabled, this is what this function
7009 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
7010 uint16_t vlandev_ofp_port, int *vid)
7012 if (!hmap_is_empty(&ofproto->vlandev_map)) {
7013 const struct vlan_splinter *vsp;
7015 vsp = vlandev_find(ofproto, vlandev_ofp_port);
7020 return vsp->realdev_ofp_port;
7026 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
7027 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
7028 * 'flow->in_port' to the "real" device backing the VLAN device, sets
7029 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
7030 * always the case unless VLAN splinters are enabled), returns false without
7031 * making any changes. */
7033 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
7038 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
7043 /* Cause the flow to be processed as if it came in on the real device with
7044 * the VLAN device's VLAN ID. */
7045 flow->in_port = realdev;
7046 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
7051 vsp_remove(struct ofport_dpif *port)
7053 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7054 struct vlan_splinter *vsp;
7056 vsp = vlandev_find(ofproto, port->up.ofp_port);
7058 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
7059 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
7062 port->realdev_ofp_port = 0;
7064 VLOG_ERR("missing vlan device record");
7069 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
7071 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7073 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
7074 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
7075 == realdev_ofp_port)) {
7076 struct vlan_splinter *vsp;
7078 vsp = xmalloc(sizeof *vsp);
7079 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
7080 hash_int(port->up.ofp_port, 0));
7081 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
7082 hash_realdev_vid(realdev_ofp_port, vid));
7083 vsp->realdev_ofp_port = realdev_ofp_port;
7084 vsp->vlandev_ofp_port = port->up.ofp_port;
7087 port->realdev_ofp_port = realdev_ofp_port;
7089 VLOG_ERR("duplicate vlan device record");
7093 const struct ofproto_class ofproto_dpif_class = {
7122 port_is_lacp_current,
7123 NULL, /* rule_choose_table */
7130 rule_modify_actions,
7138 get_cfm_remote_mpids,
7143 get_stp_port_status,
7150 is_mirror_output_bundle,
7151 forward_bpdu_changed,