2 * Copyright (c) 2009, 2010, 2011, 2012 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-provider.h"
26 #include "byte-order.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
36 #include "mac-learning.h"
37 #include "meta-flow.h"
38 #include "multipath.h"
45 #include "ofp-parse.h"
46 #include "ofp-print.h"
47 #include "ofproto-dpif-governor.h"
48 #include "ofproto-dpif-sflow.h"
49 #include "poll-loop.h"
52 #include "unaligned.h"
54 #include "vlan-bitmap.h"
57 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
59 COVERAGE_DEFINE(ofproto_dpif_expired);
60 COVERAGE_DEFINE(ofproto_dpif_xlate);
61 COVERAGE_DEFINE(facet_changed_rule);
62 COVERAGE_DEFINE(facet_revalidate);
63 COVERAGE_DEFINE(facet_unexpected);
64 COVERAGE_DEFINE(facet_suppress);
66 /* Maximum depth of flow table recursion (due to resubmit actions) in a
67 * flow translation. */
68 #define MAX_RESUBMIT_RECURSION 32
70 /* Number of implemented OpenFlow tables. */
71 enum { N_TABLES = 255 };
72 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
73 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
83 * - Do include packets and bytes from facets that have been deleted or
84 * whose own statistics have been folded into the rule.
86 * - Do include packets and bytes sent "by hand" that were accounted to
87 * the rule without any facet being involved (this is a rare corner
88 * case in rule_execute()).
90 * - Do not include packet or bytes that can be obtained from any facet's
91 * packet_count or byte_count member or that can be obtained from the
92 * datapath by, e.g., dpif_flow_get() for any subfacet.
94 uint64_t packet_count; /* Number of packets received. */
95 uint64_t byte_count; /* Number of bytes received. */
97 tag_type tag; /* Caches rule_calculate_tag() result. */
99 struct list facets; /* List of "struct facet"s. */
102 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
104 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
107 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
108 const struct flow *);
109 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
113 static void rule_credit_stats(struct rule_dpif *,
114 const struct dpif_flow_stats *);
115 static void flow_push_stats(struct rule_dpif *, const struct flow *,
116 const struct dpif_flow_stats *);
117 static tag_type rule_calculate_tag(const struct flow *,
118 const struct flow_wildcards *,
120 static void rule_invalidate(const struct rule_dpif *);
122 #define MAX_MIRRORS 32
123 typedef uint32_t mirror_mask_t;
124 #define MIRROR_MASK_C(X) UINT32_C(X)
125 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
127 struct ofproto_dpif *ofproto; /* Owning ofproto. */
128 size_t idx; /* In ofproto's "mirrors" array. */
129 void *aux; /* Key supplied by ofproto's client. */
130 char *name; /* Identifier for log messages. */
132 /* Selection criteria. */
133 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
134 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
135 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
137 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
138 struct ofbundle *out; /* Output port or NULL. */
139 int out_vlan; /* Output VLAN or -1. */
140 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
143 int64_t packet_count; /* Number of packets sent. */
144 int64_t byte_count; /* Number of bytes sent. */
147 static void mirror_destroy(struct ofmirror *);
148 static void update_mirror_stats(struct ofproto_dpif *ofproto,
149 mirror_mask_t mirrors,
150 uint64_t packets, uint64_t bytes);
153 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
154 struct ofproto_dpif *ofproto; /* Owning ofproto. */
155 void *aux; /* Key supplied by ofproto's client. */
156 char *name; /* Identifier for log messages. */
159 struct list ports; /* Contains "struct ofport"s. */
160 enum port_vlan_mode vlan_mode; /* VLAN mode */
161 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
162 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
163 * NULL if all VLANs are trunked. */
164 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
165 struct bond *bond; /* Nonnull iff more than one port. */
166 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
169 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
171 /* Port mirroring info. */
172 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
173 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
174 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
177 static void bundle_remove(struct ofport *);
178 static void bundle_update(struct ofbundle *);
179 static void bundle_destroy(struct ofbundle *);
180 static void bundle_del_port(struct ofport_dpif *);
181 static void bundle_run(struct ofbundle *);
182 static void bundle_wait(struct ofbundle *);
183 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
184 uint16_t in_port, bool warn,
185 struct ofport_dpif **in_ofportp);
187 /* A controller may use OFPP_NONE as the ingress port to indicate that
188 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
189 * when an input bundle is needed for validation (e.g., mirroring or
190 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
191 * any 'port' structs, so care must be taken when dealing with it. */
192 static struct ofbundle ofpp_none_bundle = {
194 .vlan_mode = PORT_VLAN_TRUNK
197 static void stp_run(struct ofproto_dpif *ofproto);
198 static void stp_wait(struct ofproto_dpif *ofproto);
199 static int set_stp_port(struct ofport *,
200 const struct ofproto_port_stp_settings *);
202 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
204 struct action_xlate_ctx {
205 /* action_xlate_ctx_init() initializes these members. */
208 struct ofproto_dpif *ofproto;
210 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
211 * this flow when actions change header fields. */
214 /* The packet corresponding to 'flow', or a null pointer if we are
215 * revalidating without a packet to refer to. */
216 const struct ofpbuf *packet;
218 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
219 * actions update the flow table?
221 * We want to update these tables if we are actually processing a packet,
222 * or if we are accounting for packets that the datapath has processed, but
223 * not if we are just revalidating. */
226 /* The rule that we are currently translating, or NULL. */
227 struct rule_dpif *rule;
229 /* Union of the set of TCP flags seen so far in this flow. (Used only by
230 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
234 /* If nonnull, flow translation calls this function just before executing a
235 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
236 * when the recursion depth is exceeded.
238 * 'rule' is the rule being submitted into. It will be null if the
239 * resubmit or OFPP_TABLE action didn't find a matching rule.
241 * This is normally null so the client has to set it manually after
242 * calling action_xlate_ctx_init(). */
243 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule);
245 /* If nonnull, flow translation credits the specified statistics to each
246 * rule reached through a resubmit or OFPP_TABLE action.
248 * This is normally null so the client has to set it manually after
249 * calling action_xlate_ctx_init(). */
250 const struct dpif_flow_stats *resubmit_stats;
252 /* xlate_actions() initializes and uses these members. The client might want
253 * to look at them after it returns. */
255 struct ofpbuf *odp_actions; /* Datapath actions. */
256 tag_type tags; /* Tags associated with actions. */
257 enum slow_path_reason slow; /* 0 if fast path may be used. */
258 bool has_learn; /* Actions include NXAST_LEARN? */
259 bool has_normal; /* Actions output to OFPP_NORMAL? */
260 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
261 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
262 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
264 /* xlate_actions() initializes and uses these members, but the client has no
265 * reason to look at them. */
267 int recurse; /* Recursion level, via xlate_table_action. */
268 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
269 struct flow base_flow; /* Flow at the last commit. */
270 uint32_t orig_skb_priority; /* Priority when packet arrived. */
271 uint8_t table_id; /* OpenFlow table ID where flow was found. */
272 uint32_t sflow_n_outputs; /* Number of output ports. */
273 uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
274 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
275 bool exit; /* No further actions should be processed. */
276 struct flow orig_flow; /* Copy of original flow. */
279 static void action_xlate_ctx_init(struct action_xlate_ctx *,
280 struct ofproto_dpif *, const struct flow *,
281 ovs_be16 initial_tci, struct rule_dpif *,
282 uint8_t tcp_flags, const struct ofpbuf *);
283 static void xlate_actions(struct action_xlate_ctx *,
284 const union ofp_action *in, size_t n_in,
285 struct ofpbuf *odp_actions);
286 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
287 const union ofp_action *in,
290 static size_t put_userspace_action(const struct ofproto_dpif *,
291 struct ofpbuf *odp_actions,
293 const union user_action_cookie *);
295 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
296 enum slow_path_reason,
297 uint64_t *stub, size_t stub_size,
298 const struct nlattr **actionsp,
299 size_t *actions_lenp);
301 /* A subfacet (see "struct subfacet" below) has three possible installation
304 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
305 * case just after the subfacet is created, just before the subfacet is
306 * destroyed, or if the datapath returns an error when we try to install a
309 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
311 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
312 * ofproto_dpif is installed in the datapath.
315 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
316 SF_FAST_PATH, /* Full actions are installed. */
317 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
320 static const char *subfacet_path_to_string(enum subfacet_path);
322 /* A dpif flow and actions associated with a facet.
324 * See also the large comment on struct facet. */
327 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
328 struct list list_node; /* In struct facet's 'facets' list. */
329 struct facet *facet; /* Owning facet. */
333 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
334 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
335 * regenerate the ODP flow key from ->facet->flow. */
336 enum odp_key_fitness key_fitness;
340 long long int used; /* Time last used; time created if not used. */
342 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
343 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
347 * These should be essentially identical for every subfacet in a facet, but
348 * may differ in trivial ways due to VLAN splinters. */
349 size_t actions_len; /* Number of bytes in actions[]. */
350 struct nlattr *actions; /* Datapath actions. */
352 enum slow_path_reason slow; /* 0 if fast path may be used. */
353 enum subfacet_path path; /* Installed in datapath? */
355 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
356 * splinters can cause it to differ. This value should be removed when
357 * the VLAN splinters feature is no longer needed. */
358 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
361 static struct subfacet *subfacet_create(struct facet *, enum odp_key_fitness,
362 const struct nlattr *key,
363 size_t key_len, ovs_be16 initial_tci);
364 static struct subfacet *subfacet_find(struct ofproto_dpif *,
365 const struct nlattr *key, size_t key_len);
366 static void subfacet_destroy(struct subfacet *);
367 static void subfacet_destroy__(struct subfacet *);
368 static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *,
370 static void subfacet_reset_dp_stats(struct subfacet *,
371 struct dpif_flow_stats *);
372 static void subfacet_update_time(struct subfacet *, long long int used);
373 static void subfacet_update_stats(struct subfacet *,
374 const struct dpif_flow_stats *);
375 static void subfacet_make_actions(struct subfacet *,
376 const struct ofpbuf *packet,
377 struct ofpbuf *odp_actions);
378 static int subfacet_install(struct subfacet *,
379 const struct nlattr *actions, size_t actions_len,
380 struct dpif_flow_stats *, enum slow_path_reason);
381 static void subfacet_uninstall(struct subfacet *);
383 static enum subfacet_path subfacet_want_path(enum slow_path_reason);
385 /* An exact-match instantiation of an OpenFlow flow.
387 * A facet associates a "struct flow", which represents the Open vSwitch
388 * userspace idea of an exact-match flow, with one or more subfacets. Each
389 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
390 * the facet. When the kernel module (or other dpif implementation) and Open
391 * vSwitch userspace agree on the definition of a flow key, there is exactly
392 * one subfacet per facet. If the dpif implementation supports more-specific
393 * flow matching than userspace, however, a facet can have more than one
394 * subfacet, each of which corresponds to some distinction in flow that
395 * userspace simply doesn't understand.
397 * Flow expiration works in terms of subfacets, so a facet must have at least
398 * one subfacet or it will never expire, leaking memory. */
401 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
402 struct list list_node; /* In owning rule's 'facets' list. */
403 struct rule_dpif *rule; /* Owning rule. */
406 struct list subfacets;
407 long long int used; /* Time last used; time created if not used. */
414 * - Do include packets and bytes sent "by hand", e.g. with
417 * - Do include packets and bytes that were obtained from the datapath
418 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
419 * DPIF_FP_ZERO_STATS).
421 * - Do not include packets or bytes that can be obtained from the
422 * datapath for any existing subfacet.
424 uint64_t packet_count; /* Number of packets received. */
425 uint64_t byte_count; /* Number of bytes received. */
427 /* Resubmit statistics. */
428 uint64_t prev_packet_count; /* Number of packets from last stats push. */
429 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
430 long long int prev_used; /* Used time from last stats push. */
433 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
434 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
435 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
437 /* Properties of datapath actions.
439 * Every subfacet has its own actions because actions can differ slightly
440 * between splintered and non-splintered subfacets due to the VLAN tag
441 * being initially different (present vs. absent). All of them have these
442 * properties in common so we just store one copy of them here. */
443 bool has_learn; /* Actions include NXAST_LEARN? */
444 bool has_normal; /* Actions output to OFPP_NORMAL? */
445 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
446 tag_type tags; /* Tags that would require revalidation. */
447 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
449 /* Storage for a single subfacet, to reduce malloc() time and space
450 * overhead. (A facet always has at least one subfacet and in the common
451 * case has exactly one subfacet.) */
452 struct subfacet one_subfacet;
455 static struct facet *facet_create(struct rule_dpif *,
456 const struct flow *, uint32_t hash);
457 static void facet_remove(struct facet *);
458 static void facet_free(struct facet *);
460 static struct facet *facet_find(struct ofproto_dpif *,
461 const struct flow *, uint32_t hash);
462 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
463 const struct flow *, uint32_t hash);
464 static void facet_revalidate(struct facet *);
465 static bool facet_check_consistency(struct facet *);
467 static void facet_flush_stats(struct facet *);
469 static void facet_update_time(struct facet *, long long int used);
470 static void facet_reset_counters(struct facet *);
471 static void facet_push_stats(struct facet *);
472 static void facet_learn(struct facet *);
473 static void facet_account(struct facet *);
475 static bool facet_is_controller_flow(struct facet *);
481 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
482 struct list bundle_node; /* In struct ofbundle's "ports" list. */
483 struct cfm *cfm; /* Connectivity Fault Management, if any. */
484 tag_type tag; /* Tag associated with this port. */
485 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
486 bool may_enable; /* May be enabled in bonds. */
487 long long int carrier_seq; /* Carrier status changes. */
490 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
491 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
492 long long int stp_state_entered;
494 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
496 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
498 * This is deprecated. It is only for compatibility with broken device
499 * drivers in old versions of Linux that do not properly support VLANs when
500 * VLAN devices are not used. When broken device drivers are no longer in
501 * widespread use, we will delete these interfaces. */
502 uint16_t realdev_ofp_port;
506 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
507 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
508 * traffic egressing the 'ofport' with that priority should be marked with. */
509 struct priority_to_dscp {
510 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
511 uint32_t priority; /* Priority of this queue (see struct flow). */
513 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
516 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
518 * This is deprecated. It is only for compatibility with broken device drivers
519 * in old versions of Linux that do not properly support VLANs when VLAN
520 * devices are not used. When broken device drivers are no longer in
521 * widespread use, we will delete these interfaces. */
522 struct vlan_splinter {
523 struct hmap_node realdev_vid_node;
524 struct hmap_node vlandev_node;
525 uint16_t realdev_ofp_port;
526 uint16_t vlandev_ofp_port;
530 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
531 uint32_t realdev, ovs_be16 vlan_tci);
532 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
533 static void vsp_remove(struct ofport_dpif *);
534 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
536 static struct ofport_dpif *
537 ofport_dpif_cast(const struct ofport *ofport)
539 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
540 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
543 static void port_run(struct ofport_dpif *);
544 static void port_wait(struct ofport_dpif *);
545 static int set_cfm(struct ofport *, const struct cfm_settings *);
546 static void ofport_clear_priorities(struct ofport_dpif *);
548 struct dpif_completion {
549 struct list list_node;
550 struct ofoperation *op;
553 /* Extra information about a classifier table.
554 * Currently used just for optimized flow revalidation. */
556 /* If either of these is nonnull, then this table has a form that allows
557 * flows to be tagged to avoid revalidating most flows for the most common
558 * kinds of flow table changes. */
559 struct cls_table *catchall_table; /* Table that wildcards all fields. */
560 struct cls_table *other_table; /* Table with any other wildcard set. */
561 uint32_t basis; /* Keeps each table's tags separate. */
564 struct ofproto_dpif {
565 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
570 /* Special OpenFlow rules. */
571 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
572 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
578 struct netflow *netflow;
579 struct dpif_sflow *sflow;
580 struct hmap bundles; /* Contains "struct ofbundle"s. */
581 struct mac_learning *ml;
582 struct ofmirror *mirrors[MAX_MIRRORS];
584 bool has_bonded_bundles;
587 struct timer next_expiration;
591 struct hmap subfacets;
592 struct governor *governor;
595 struct table_dpif tables[N_TABLES];
596 bool need_revalidate;
597 struct tag_set revalidate_set;
599 /* Support for debugging async flow mods. */
600 struct list completions;
602 bool has_bundle_action; /* True when the first bundle action appears. */
603 struct netdev_stats stats; /* To account packets generated and consumed in
608 long long int stp_last_tick;
610 /* VLAN splinters. */
611 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
612 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
615 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
616 * for debugging the asynchronous flow_mod implementation.) */
619 /* All existing ofproto_dpif instances, indexed by ->up.name. */
620 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
622 static void ofproto_dpif_unixctl_init(void);
624 static struct ofproto_dpif *
625 ofproto_dpif_cast(const struct ofproto *ofproto)
627 assert(ofproto->ofproto_class == &ofproto_dpif_class);
628 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
631 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
633 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
635 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
636 const struct ofpbuf *, ovs_be16 initial_tci,
639 /* Packet processing. */
640 static void update_learning_table(struct ofproto_dpif *,
641 const struct flow *, int vlan,
644 #define FLOW_MISS_MAX_BATCH 50
645 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
647 /* Flow expiration. */
648 static int expire(struct ofproto_dpif *);
651 static void send_netflow_active_timeouts(struct ofproto_dpif *);
654 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
655 static size_t compose_sflow_action(const struct ofproto_dpif *,
656 struct ofpbuf *odp_actions,
657 const struct flow *, uint32_t odp_port);
658 static void add_mirror_actions(struct action_xlate_ctx *ctx,
659 const struct flow *flow);
660 /* Global variables. */
661 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
663 /* Factory functions. */
666 enumerate_types(struct sset *types)
668 dp_enumerate_types(types);
672 enumerate_names(const char *type, struct sset *names)
674 return dp_enumerate_names(type, names);
678 del(const char *type, const char *name)
683 error = dpif_open(name, type, &dpif);
685 error = dpif_delete(dpif);
691 /* Basic life-cycle. */
693 static int add_internal_flows(struct ofproto_dpif *);
695 static struct ofproto *
698 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
703 dealloc(struct ofproto *ofproto_)
705 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
710 construct(struct ofproto *ofproto_)
712 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
713 const char *name = ofproto->up.name;
717 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
719 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
723 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
724 ofproto->n_matches = 0;
726 dpif_flow_flush(ofproto->dpif);
727 dpif_recv_purge(ofproto->dpif);
729 error = dpif_recv_set(ofproto->dpif, true);
731 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
732 dpif_close(ofproto->dpif);
736 ofproto->netflow = NULL;
737 ofproto->sflow = NULL;
739 hmap_init(&ofproto->bundles);
740 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
741 for (i = 0; i < MAX_MIRRORS; i++) {
742 ofproto->mirrors[i] = NULL;
744 ofproto->has_bonded_bundles = false;
746 timer_set_duration(&ofproto->next_expiration, 1000);
748 hmap_init(&ofproto->facets);
749 hmap_init(&ofproto->subfacets);
750 ofproto->governor = NULL;
752 for (i = 0; i < N_TABLES; i++) {
753 struct table_dpif *table = &ofproto->tables[i];
755 table->catchall_table = NULL;
756 table->other_table = NULL;
757 table->basis = random_uint32();
759 ofproto->need_revalidate = false;
760 tag_set_init(&ofproto->revalidate_set);
762 list_init(&ofproto->completions);
764 ofproto_dpif_unixctl_init();
766 ofproto->has_mirrors = false;
767 ofproto->has_bundle_action = false;
769 hmap_init(&ofproto->vlandev_map);
770 hmap_init(&ofproto->realdev_vid_map);
772 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
773 hash_string(ofproto->up.name, 0));
774 memset(&ofproto->stats, 0, sizeof ofproto->stats);
776 ofproto_init_tables(ofproto_, N_TABLES);
777 error = add_internal_flows(ofproto);
778 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
784 add_internal_flow(struct ofproto_dpif *ofproto, int id,
785 const struct ofpbuf *actions, struct rule_dpif **rulep)
787 struct ofputil_flow_mod fm;
790 cls_rule_init_catchall(&fm.cr, 0);
791 cls_rule_set_reg(&fm.cr, 0, id);
792 fm.new_cookie = htonll(0);
793 fm.cookie = htonll(0);
794 fm.cookie_mask = htonll(0);
795 fm.table_id = TBL_INTERNAL;
796 fm.command = OFPFC_ADD;
802 fm.actions = actions->data;
803 fm.n_actions = actions->size / sizeof(union ofp_action);
805 error = ofproto_flow_mod(&ofproto->up, &fm);
807 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
808 id, ofperr_to_string(error));
812 *rulep = rule_dpif_lookup__(ofproto, &fm.cr.flow, TBL_INTERNAL);
813 assert(*rulep != NULL);
819 add_internal_flows(struct ofproto_dpif *ofproto)
821 struct nx_action_controller *nac;
822 uint64_t actions_stub[128 / 8];
823 struct ofpbuf actions;
827 ofpbuf_use_stack(&actions, actions_stub, sizeof actions_stub);
830 nac = ofputil_put_NXAST_CONTROLLER(&actions);
831 nac->max_len = htons(UINT16_MAX);
832 nac->controller_id = htons(0);
833 nac->reason = OFPR_NO_MATCH;
834 error = add_internal_flow(ofproto, id++, &actions, &ofproto->miss_rule);
839 ofpbuf_clear(&actions);
840 error = add_internal_flow(ofproto, id++, &actions,
841 &ofproto->no_packet_in_rule);
846 complete_operations(struct ofproto_dpif *ofproto)
848 struct dpif_completion *c, *next;
850 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
851 ofoperation_complete(c->op, 0);
852 list_remove(&c->list_node);
858 destruct(struct ofproto *ofproto_)
860 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
861 struct rule_dpif *rule, *next_rule;
862 struct oftable *table;
865 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
866 complete_operations(ofproto);
868 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
869 struct cls_cursor cursor;
871 cls_cursor_init(&cursor, &table->cls, NULL);
872 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
873 ofproto_rule_destroy(&rule->up);
877 for (i = 0; i < MAX_MIRRORS; i++) {
878 mirror_destroy(ofproto->mirrors[i]);
881 netflow_destroy(ofproto->netflow);
882 dpif_sflow_destroy(ofproto->sflow);
883 hmap_destroy(&ofproto->bundles);
884 mac_learning_destroy(ofproto->ml);
886 hmap_destroy(&ofproto->facets);
887 hmap_destroy(&ofproto->subfacets);
888 governor_destroy(ofproto->governor);
890 hmap_destroy(&ofproto->vlandev_map);
891 hmap_destroy(&ofproto->realdev_vid_map);
893 dpif_close(ofproto->dpif);
897 run_fast(struct ofproto *ofproto_)
899 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
902 /* Handle one or more batches of upcalls, until there's nothing left to do
903 * or until we do a fixed total amount of work.
905 * We do work in batches because it can be much cheaper to set up a number
906 * of flows and fire off their patches all at once. We do multiple batches
907 * because in some cases handling a packet can cause another packet to be
908 * queued almost immediately as part of the return flow. Both
909 * optimizations can make major improvements on some benchmarks and
910 * presumably for real traffic as well. */
912 while (work < FLOW_MISS_MAX_BATCH) {
913 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
923 run(struct ofproto *ofproto_)
925 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
926 struct ofport_dpif *ofport;
927 struct ofbundle *bundle;
931 complete_operations(ofproto);
933 dpif_run(ofproto->dpif);
935 error = run_fast(ofproto_);
940 if (timer_expired(&ofproto->next_expiration)) {
941 int delay = expire(ofproto);
942 timer_set_duration(&ofproto->next_expiration, delay);
945 if (ofproto->netflow) {
946 if (netflow_run(ofproto->netflow)) {
947 send_netflow_active_timeouts(ofproto);
950 if (ofproto->sflow) {
951 dpif_sflow_run(ofproto->sflow);
954 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
957 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
962 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
964 /* Now revalidate if there's anything to do. */
965 if (ofproto->need_revalidate
966 || !tag_set_is_empty(&ofproto->revalidate_set)) {
967 struct tag_set revalidate_set = ofproto->revalidate_set;
968 bool revalidate_all = ofproto->need_revalidate;
971 /* Clear the revalidation flags. */
972 tag_set_init(&ofproto->revalidate_set);
973 ofproto->need_revalidate = false;
975 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
977 || tag_set_intersects(&revalidate_set, facet->tags)) {
978 facet_revalidate(facet);
983 /* Check the consistency of a random facet, to aid debugging. */
984 if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) {
987 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
988 struct facet, hmap_node);
989 if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) {
990 if (!facet_check_consistency(facet)) {
991 ofproto->need_revalidate = true;
996 if (ofproto->governor) {
999 governor_run(ofproto->governor);
1001 /* If the governor has shrunk to its minimum size and the number of
1002 * subfacets has dwindled, then drop the governor entirely.
1004 * For hysteresis, the number of subfacets to drop the governor is
1005 * smaller than the number needed to trigger its creation. */
1006 n_subfacets = hmap_count(&ofproto->subfacets);
1007 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1008 && governor_is_idle(ofproto->governor)) {
1009 governor_destroy(ofproto->governor);
1010 ofproto->governor = NULL;
1018 wait(struct ofproto *ofproto_)
1020 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1021 struct ofport_dpif *ofport;
1022 struct ofbundle *bundle;
1024 if (!clogged && !list_is_empty(&ofproto->completions)) {
1025 poll_immediate_wake();
1028 dpif_wait(ofproto->dpif);
1029 dpif_recv_wait(ofproto->dpif);
1030 if (ofproto->sflow) {
1031 dpif_sflow_wait(ofproto->sflow);
1033 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
1034 poll_immediate_wake();
1036 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1039 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1040 bundle_wait(bundle);
1042 if (ofproto->netflow) {
1043 netflow_wait(ofproto->netflow);
1045 mac_learning_wait(ofproto->ml);
1047 if (ofproto->need_revalidate) {
1048 /* Shouldn't happen, but if it does just go around again. */
1049 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1050 poll_immediate_wake();
1052 timer_wait(&ofproto->next_expiration);
1054 if (ofproto->governor) {
1055 governor_wait(ofproto->governor);
1060 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1062 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1064 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1065 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1069 flush(struct ofproto *ofproto_)
1071 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1072 struct facet *facet, *next_facet;
1074 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1075 /* Mark the facet as not installed so that facet_remove() doesn't
1076 * bother trying to uninstall it. There is no point in uninstalling it
1077 * individually since we are about to blow away all the facets with
1078 * dpif_flow_flush(). */
1079 struct subfacet *subfacet;
1081 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
1082 subfacet->path = SF_NOT_INSTALLED;
1083 subfacet->dp_packet_count = 0;
1084 subfacet->dp_byte_count = 0;
1086 facet_remove(facet);
1088 dpif_flow_flush(ofproto->dpif);
1092 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1093 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1095 *arp_match_ip = true;
1096 *actions = (OFPUTIL_A_OUTPUT |
1097 OFPUTIL_A_SET_VLAN_VID |
1098 OFPUTIL_A_SET_VLAN_PCP |
1099 OFPUTIL_A_STRIP_VLAN |
1100 OFPUTIL_A_SET_DL_SRC |
1101 OFPUTIL_A_SET_DL_DST |
1102 OFPUTIL_A_SET_NW_SRC |
1103 OFPUTIL_A_SET_NW_DST |
1104 OFPUTIL_A_SET_NW_TOS |
1105 OFPUTIL_A_SET_TP_SRC |
1106 OFPUTIL_A_SET_TP_DST |
1111 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
1113 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1114 struct dpif_dp_stats s;
1116 strcpy(ots->name, "classifier");
1118 dpif_get_dp_stats(ofproto->dpif, &s);
1119 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
1120 put_32aligned_be64(&ots->matched_count,
1121 htonll(s.n_hit + ofproto->n_matches));
1124 static struct ofport *
1127 struct ofport_dpif *port = xmalloc(sizeof *port);
1132 port_dealloc(struct ofport *port_)
1134 struct ofport_dpif *port = ofport_dpif_cast(port_);
1139 port_construct(struct ofport *port_)
1141 struct ofport_dpif *port = ofport_dpif_cast(port_);
1142 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1144 ofproto->need_revalidate = true;
1145 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
1146 port->bundle = NULL;
1148 port->tag = tag_create_random();
1149 port->may_enable = true;
1150 port->stp_port = NULL;
1151 port->stp_state = STP_DISABLED;
1152 hmap_init(&port->priorities);
1153 port->realdev_ofp_port = 0;
1154 port->vlandev_vid = 0;
1155 port->carrier_seq = netdev_get_carrier_resets(port->up.netdev);
1157 if (ofproto->sflow) {
1158 dpif_sflow_add_port(ofproto->sflow, port_);
1165 port_destruct(struct ofport *port_)
1167 struct ofport_dpif *port = ofport_dpif_cast(port_);
1168 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1170 ofproto->need_revalidate = true;
1171 bundle_remove(port_);
1172 set_cfm(port_, NULL);
1173 if (ofproto->sflow) {
1174 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1177 ofport_clear_priorities(port);
1178 hmap_destroy(&port->priorities);
1182 port_modified(struct ofport *port_)
1184 struct ofport_dpif *port = ofport_dpif_cast(port_);
1186 if (port->bundle && port->bundle->bond) {
1187 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1192 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1194 struct ofport_dpif *port = ofport_dpif_cast(port_);
1195 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1196 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1198 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1199 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1200 OFPUTIL_PC_NO_PACKET_IN)) {
1201 ofproto->need_revalidate = true;
1203 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1204 bundle_update(port->bundle);
1210 set_sflow(struct ofproto *ofproto_,
1211 const struct ofproto_sflow_options *sflow_options)
1213 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1214 struct dpif_sflow *ds = ofproto->sflow;
1216 if (sflow_options) {
1218 struct ofport_dpif *ofport;
1220 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1221 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1222 dpif_sflow_add_port(ds, &ofport->up);
1224 ofproto->need_revalidate = true;
1226 dpif_sflow_set_options(ds, sflow_options);
1229 dpif_sflow_destroy(ds);
1230 ofproto->need_revalidate = true;
1231 ofproto->sflow = NULL;
1238 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1240 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1247 struct ofproto_dpif *ofproto;
1249 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1250 ofproto->need_revalidate = true;
1251 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1254 if (cfm_configure(ofport->cfm, s)) {
1260 cfm_destroy(ofport->cfm);
1266 get_cfm_fault(const struct ofport *ofport_)
1268 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1270 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1274 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1277 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1280 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1288 get_cfm_health(const struct ofport *ofport_)
1290 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1292 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1295 /* Spanning Tree. */
1298 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1300 struct ofproto_dpif *ofproto = ofproto_;
1301 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1302 struct ofport_dpif *ofport;
1304 ofport = stp_port_get_aux(sp);
1306 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1307 ofproto->up.name, port_num);
1309 struct eth_header *eth = pkt->l2;
1311 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1312 if (eth_addr_is_zero(eth->eth_src)) {
1313 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1314 "with unknown MAC", ofproto->up.name, port_num);
1316 send_packet(ofport, pkt);
1322 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1324 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1326 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1328 /* Only revalidate flows if the configuration changed. */
1329 if (!s != !ofproto->stp) {
1330 ofproto->need_revalidate = true;
1334 if (!ofproto->stp) {
1335 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1336 send_bpdu_cb, ofproto);
1337 ofproto->stp_last_tick = time_msec();
1340 stp_set_bridge_id(ofproto->stp, s->system_id);
1341 stp_set_bridge_priority(ofproto->stp, s->priority);
1342 stp_set_hello_time(ofproto->stp, s->hello_time);
1343 stp_set_max_age(ofproto->stp, s->max_age);
1344 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1346 struct ofport *ofport;
1348 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1349 set_stp_port(ofport, NULL);
1352 stp_destroy(ofproto->stp);
1353 ofproto->stp = NULL;
1360 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1362 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1366 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1367 s->designated_root = stp_get_designated_root(ofproto->stp);
1368 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1377 update_stp_port_state(struct ofport_dpif *ofport)
1379 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1380 enum stp_state state;
1382 /* Figure out new state. */
1383 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1387 if (ofport->stp_state != state) {
1388 enum ofputil_port_state of_state;
1391 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1392 netdev_get_name(ofport->up.netdev),
1393 stp_state_name(ofport->stp_state),
1394 stp_state_name(state));
1395 if (stp_learn_in_state(ofport->stp_state)
1396 != stp_learn_in_state(state)) {
1397 /* xxx Learning action flows should also be flushed. */
1398 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1400 fwd_change = stp_forward_in_state(ofport->stp_state)
1401 != stp_forward_in_state(state);
1403 ofproto->need_revalidate = true;
1404 ofport->stp_state = state;
1405 ofport->stp_state_entered = time_msec();
1407 if (fwd_change && ofport->bundle) {
1408 bundle_update(ofport->bundle);
1411 /* Update the STP state bits in the OpenFlow port description. */
1412 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1413 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1414 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1415 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1416 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1418 ofproto_port_set_state(&ofport->up, of_state);
1422 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1423 * caller is responsible for assigning STP port numbers and ensuring
1424 * there are no duplicates. */
1426 set_stp_port(struct ofport *ofport_,
1427 const struct ofproto_port_stp_settings *s)
1429 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1430 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1431 struct stp_port *sp = ofport->stp_port;
1433 if (!s || !s->enable) {
1435 ofport->stp_port = NULL;
1436 stp_port_disable(sp);
1437 update_stp_port_state(ofport);
1440 } else if (sp && stp_port_no(sp) != s->port_num
1441 && ofport == stp_port_get_aux(sp)) {
1442 /* The port-id changed, so disable the old one if it's not
1443 * already in use by another port. */
1444 stp_port_disable(sp);
1447 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1448 stp_port_enable(sp);
1450 stp_port_set_aux(sp, ofport);
1451 stp_port_set_priority(sp, s->priority);
1452 stp_port_set_path_cost(sp, s->path_cost);
1454 update_stp_port_state(ofport);
1460 get_stp_port_status(struct ofport *ofport_,
1461 struct ofproto_port_stp_status *s)
1463 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1464 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1465 struct stp_port *sp = ofport->stp_port;
1467 if (!ofproto->stp || !sp) {
1473 s->port_id = stp_port_get_id(sp);
1474 s->state = stp_port_get_state(sp);
1475 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1476 s->role = stp_port_get_role(sp);
1477 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1483 stp_run(struct ofproto_dpif *ofproto)
1486 long long int now = time_msec();
1487 long long int elapsed = now - ofproto->stp_last_tick;
1488 struct stp_port *sp;
1491 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1492 ofproto->stp_last_tick = now;
1494 while (stp_get_changed_port(ofproto->stp, &sp)) {
1495 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1498 update_stp_port_state(ofport);
1502 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1503 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1509 stp_wait(struct ofproto_dpif *ofproto)
1512 poll_timer_wait(1000);
1516 /* Returns true if STP should process 'flow'. */
1518 stp_should_process_flow(const struct flow *flow)
1520 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1524 stp_process_packet(const struct ofport_dpif *ofport,
1525 const struct ofpbuf *packet)
1527 struct ofpbuf payload = *packet;
1528 struct eth_header *eth = payload.data;
1529 struct stp_port *sp = ofport->stp_port;
1531 /* Sink packets on ports that have STP disabled when the bridge has
1533 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1537 /* Trim off padding on payload. */
1538 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1539 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1542 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1543 stp_received_bpdu(sp, payload.data, payload.size);
1547 static struct priority_to_dscp *
1548 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1550 struct priority_to_dscp *pdscp;
1553 hash = hash_int(priority, 0);
1554 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1555 if (pdscp->priority == priority) {
1563 ofport_clear_priorities(struct ofport_dpif *ofport)
1565 struct priority_to_dscp *pdscp, *next;
1567 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1568 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1574 set_queues(struct ofport *ofport_,
1575 const struct ofproto_port_queue *qdscp_list,
1578 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1579 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1580 struct hmap new = HMAP_INITIALIZER(&new);
1583 for (i = 0; i < n_qdscp; i++) {
1584 struct priority_to_dscp *pdscp;
1588 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1589 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1594 pdscp = get_priority(ofport, priority);
1596 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1598 pdscp = xmalloc(sizeof *pdscp);
1599 pdscp->priority = priority;
1601 ofproto->need_revalidate = true;
1604 if (pdscp->dscp != dscp) {
1606 ofproto->need_revalidate = true;
1609 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1612 if (!hmap_is_empty(&ofport->priorities)) {
1613 ofport_clear_priorities(ofport);
1614 ofproto->need_revalidate = true;
1617 hmap_swap(&new, &ofport->priorities);
1625 /* Expires all MAC learning entries associated with 'bundle' and forces its
1626 * ofproto to revalidate every flow.
1628 * Normally MAC learning entries are removed only from the ofproto associated
1629 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1630 * are removed from every ofproto. When patch ports and SLB bonds are in use
1631 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1632 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1633 * with the host from which it migrated. */
1635 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1637 struct ofproto_dpif *ofproto = bundle->ofproto;
1638 struct mac_learning *ml = ofproto->ml;
1639 struct mac_entry *mac, *next_mac;
1641 ofproto->need_revalidate = true;
1642 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1643 if (mac->port.p == bundle) {
1645 struct ofproto_dpif *o;
1647 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1649 struct mac_entry *e;
1651 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1654 tag_set_add(&o->revalidate_set, e->tag);
1655 mac_learning_expire(o->ml, e);
1661 mac_learning_expire(ml, mac);
1666 static struct ofbundle *
1667 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1669 struct ofbundle *bundle;
1671 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1672 &ofproto->bundles) {
1673 if (bundle->aux == aux) {
1680 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1681 * ones that are found to 'bundles'. */
1683 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1684 void **auxes, size_t n_auxes,
1685 struct hmapx *bundles)
1689 hmapx_init(bundles);
1690 for (i = 0; i < n_auxes; i++) {
1691 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1693 hmapx_add(bundles, bundle);
1699 bundle_update(struct ofbundle *bundle)
1701 struct ofport_dpif *port;
1703 bundle->floodable = true;
1704 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1705 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1706 || !stp_forward_in_state(port->stp_state)) {
1707 bundle->floodable = false;
1714 bundle_del_port(struct ofport_dpif *port)
1716 struct ofbundle *bundle = port->bundle;
1718 bundle->ofproto->need_revalidate = true;
1720 list_remove(&port->bundle_node);
1721 port->bundle = NULL;
1724 lacp_slave_unregister(bundle->lacp, port);
1727 bond_slave_unregister(bundle->bond, port);
1730 bundle_update(bundle);
1734 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1735 struct lacp_slave_settings *lacp,
1736 uint32_t bond_stable_id)
1738 struct ofport_dpif *port;
1740 port = get_ofp_port(bundle->ofproto, ofp_port);
1745 if (port->bundle != bundle) {
1746 bundle->ofproto->need_revalidate = true;
1748 bundle_del_port(port);
1751 port->bundle = bundle;
1752 list_push_back(&bundle->ports, &port->bundle_node);
1753 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1754 || !stp_forward_in_state(port->stp_state)) {
1755 bundle->floodable = false;
1759 port->bundle->ofproto->need_revalidate = true;
1760 lacp_slave_register(bundle->lacp, port, lacp);
1763 port->bond_stable_id = bond_stable_id;
1769 bundle_destroy(struct ofbundle *bundle)
1771 struct ofproto_dpif *ofproto;
1772 struct ofport_dpif *port, *next_port;
1779 ofproto = bundle->ofproto;
1780 for (i = 0; i < MAX_MIRRORS; i++) {
1781 struct ofmirror *m = ofproto->mirrors[i];
1783 if (m->out == bundle) {
1785 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1786 || hmapx_find_and_delete(&m->dsts, bundle)) {
1787 ofproto->need_revalidate = true;
1792 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1793 bundle_del_port(port);
1796 bundle_flush_macs(bundle, true);
1797 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1799 free(bundle->trunks);
1800 lacp_destroy(bundle->lacp);
1801 bond_destroy(bundle->bond);
1806 bundle_set(struct ofproto *ofproto_, void *aux,
1807 const struct ofproto_bundle_settings *s)
1809 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1810 bool need_flush = false;
1811 struct ofport_dpif *port;
1812 struct ofbundle *bundle;
1813 unsigned long *trunks;
1819 bundle_destroy(bundle_lookup(ofproto, aux));
1823 assert(s->n_slaves == 1 || s->bond != NULL);
1824 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1826 bundle = bundle_lookup(ofproto, aux);
1828 bundle = xmalloc(sizeof *bundle);
1830 bundle->ofproto = ofproto;
1831 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1832 hash_pointer(aux, 0));
1834 bundle->name = NULL;
1836 list_init(&bundle->ports);
1837 bundle->vlan_mode = PORT_VLAN_TRUNK;
1839 bundle->trunks = NULL;
1840 bundle->use_priority_tags = s->use_priority_tags;
1841 bundle->lacp = NULL;
1842 bundle->bond = NULL;
1844 bundle->floodable = true;
1846 bundle->src_mirrors = 0;
1847 bundle->dst_mirrors = 0;
1848 bundle->mirror_out = 0;
1851 if (!bundle->name || strcmp(s->name, bundle->name)) {
1853 bundle->name = xstrdup(s->name);
1858 if (!bundle->lacp) {
1859 ofproto->need_revalidate = true;
1860 bundle->lacp = lacp_create();
1862 lacp_configure(bundle->lacp, s->lacp);
1864 lacp_destroy(bundle->lacp);
1865 bundle->lacp = NULL;
1868 /* Update set of ports. */
1870 for (i = 0; i < s->n_slaves; i++) {
1871 if (!bundle_add_port(bundle, s->slaves[i],
1872 s->lacp ? &s->lacp_slaves[i] : NULL,
1873 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1877 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1878 struct ofport_dpif *next_port;
1880 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1881 for (i = 0; i < s->n_slaves; i++) {
1882 if (s->slaves[i] == port->up.ofp_port) {
1887 bundle_del_port(port);
1891 assert(list_size(&bundle->ports) <= s->n_slaves);
1893 if (list_is_empty(&bundle->ports)) {
1894 bundle_destroy(bundle);
1898 /* Set VLAN tagging mode */
1899 if (s->vlan_mode != bundle->vlan_mode
1900 || s->use_priority_tags != bundle->use_priority_tags) {
1901 bundle->vlan_mode = s->vlan_mode;
1902 bundle->use_priority_tags = s->use_priority_tags;
1907 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1908 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1910 if (vlan != bundle->vlan) {
1911 bundle->vlan = vlan;
1915 /* Get trunked VLANs. */
1916 switch (s->vlan_mode) {
1917 case PORT_VLAN_ACCESS:
1921 case PORT_VLAN_TRUNK:
1922 trunks = (unsigned long *) s->trunks;
1925 case PORT_VLAN_NATIVE_UNTAGGED:
1926 case PORT_VLAN_NATIVE_TAGGED:
1927 if (vlan != 0 && (!s->trunks
1928 || !bitmap_is_set(s->trunks, vlan)
1929 || bitmap_is_set(s->trunks, 0))) {
1930 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1932 trunks = bitmap_clone(s->trunks, 4096);
1934 trunks = bitmap_allocate1(4096);
1936 bitmap_set1(trunks, vlan);
1937 bitmap_set0(trunks, 0);
1939 trunks = (unsigned long *) s->trunks;
1946 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1947 free(bundle->trunks);
1948 if (trunks == s->trunks) {
1949 bundle->trunks = vlan_bitmap_clone(trunks);
1951 bundle->trunks = trunks;
1956 if (trunks != s->trunks) {
1961 if (!list_is_short(&bundle->ports)) {
1962 bundle->ofproto->has_bonded_bundles = true;
1964 if (bond_reconfigure(bundle->bond, s->bond)) {
1965 ofproto->need_revalidate = true;
1968 bundle->bond = bond_create(s->bond);
1969 ofproto->need_revalidate = true;
1972 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1973 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1977 bond_destroy(bundle->bond);
1978 bundle->bond = NULL;
1981 /* If we changed something that would affect MAC learning, un-learn
1982 * everything on this port and force flow revalidation. */
1984 bundle_flush_macs(bundle, false);
1991 bundle_remove(struct ofport *port_)
1993 struct ofport_dpif *port = ofport_dpif_cast(port_);
1994 struct ofbundle *bundle = port->bundle;
1997 bundle_del_port(port);
1998 if (list_is_empty(&bundle->ports)) {
1999 bundle_destroy(bundle);
2000 } else if (list_is_short(&bundle->ports)) {
2001 bond_destroy(bundle->bond);
2002 bundle->bond = NULL;
2008 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2010 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2011 struct ofport_dpif *port = port_;
2012 uint8_t ea[ETH_ADDR_LEN];
2015 error = netdev_get_etheraddr(port->up.netdev, ea);
2017 struct ofpbuf packet;
2020 ofpbuf_init(&packet, 0);
2021 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2023 memcpy(packet_pdu, pdu, pdu_size);
2025 send_packet(port, &packet);
2026 ofpbuf_uninit(&packet);
2028 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2029 "%s (%s)", port->bundle->name,
2030 netdev_get_name(port->up.netdev), strerror(error));
2035 bundle_send_learning_packets(struct ofbundle *bundle)
2037 struct ofproto_dpif *ofproto = bundle->ofproto;
2038 int error, n_packets, n_errors;
2039 struct mac_entry *e;
2041 error = n_packets = n_errors = 0;
2042 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2043 if (e->port.p != bundle) {
2044 struct ofpbuf *learning_packet;
2045 struct ofport_dpif *port;
2049 /* The assignment to "port" is unnecessary but makes "grep"ing for
2050 * struct ofport_dpif more effective. */
2051 learning_packet = bond_compose_learning_packet(bundle->bond,
2055 ret = send_packet(port, learning_packet);
2056 ofpbuf_delete(learning_packet);
2066 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2067 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2068 "packets, last error was: %s",
2069 bundle->name, n_errors, n_packets, strerror(error));
2071 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2072 bundle->name, n_packets);
2077 bundle_run(struct ofbundle *bundle)
2080 lacp_run(bundle->lacp, send_pdu_cb);
2083 struct ofport_dpif *port;
2085 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2086 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2089 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
2090 lacp_status(bundle->lacp));
2091 if (bond_should_send_learning_packets(bundle->bond)) {
2092 bundle_send_learning_packets(bundle);
2098 bundle_wait(struct ofbundle *bundle)
2101 lacp_wait(bundle->lacp);
2104 bond_wait(bundle->bond);
2111 mirror_scan(struct ofproto_dpif *ofproto)
2115 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2116 if (!ofproto->mirrors[idx]) {
2123 static struct ofmirror *
2124 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2128 for (i = 0; i < MAX_MIRRORS; i++) {
2129 struct ofmirror *mirror = ofproto->mirrors[i];
2130 if (mirror && mirror->aux == aux) {
2138 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2140 mirror_update_dups(struct ofproto_dpif *ofproto)
2144 for (i = 0; i < MAX_MIRRORS; i++) {
2145 struct ofmirror *m = ofproto->mirrors[i];
2148 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2152 for (i = 0; i < MAX_MIRRORS; i++) {
2153 struct ofmirror *m1 = ofproto->mirrors[i];
2160 for (j = i + 1; j < MAX_MIRRORS; j++) {
2161 struct ofmirror *m2 = ofproto->mirrors[j];
2163 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2164 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2165 m2->dup_mirrors |= m1->dup_mirrors;
2172 mirror_set(struct ofproto *ofproto_, void *aux,
2173 const struct ofproto_mirror_settings *s)
2175 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2176 mirror_mask_t mirror_bit;
2177 struct ofbundle *bundle;
2178 struct ofmirror *mirror;
2179 struct ofbundle *out;
2180 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2181 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2184 mirror = mirror_lookup(ofproto, aux);
2186 mirror_destroy(mirror);
2192 idx = mirror_scan(ofproto);
2194 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2196 ofproto->up.name, MAX_MIRRORS, s->name);
2200 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2201 mirror->ofproto = ofproto;
2204 mirror->out_vlan = -1;
2205 mirror->name = NULL;
2208 if (!mirror->name || strcmp(s->name, mirror->name)) {
2210 mirror->name = xstrdup(s->name);
2213 /* Get the new configuration. */
2214 if (s->out_bundle) {
2215 out = bundle_lookup(ofproto, s->out_bundle);
2217 mirror_destroy(mirror);
2223 out_vlan = s->out_vlan;
2225 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2226 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2228 /* If the configuration has not changed, do nothing. */
2229 if (hmapx_equals(&srcs, &mirror->srcs)
2230 && hmapx_equals(&dsts, &mirror->dsts)
2231 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2232 && mirror->out == out
2233 && mirror->out_vlan == out_vlan)
2235 hmapx_destroy(&srcs);
2236 hmapx_destroy(&dsts);
2240 hmapx_swap(&srcs, &mirror->srcs);
2241 hmapx_destroy(&srcs);
2243 hmapx_swap(&dsts, &mirror->dsts);
2244 hmapx_destroy(&dsts);
2246 free(mirror->vlans);
2247 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2250 mirror->out_vlan = out_vlan;
2252 /* Update bundles. */
2253 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2254 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2255 if (hmapx_contains(&mirror->srcs, bundle)) {
2256 bundle->src_mirrors |= mirror_bit;
2258 bundle->src_mirrors &= ~mirror_bit;
2261 if (hmapx_contains(&mirror->dsts, bundle)) {
2262 bundle->dst_mirrors |= mirror_bit;
2264 bundle->dst_mirrors &= ~mirror_bit;
2267 if (mirror->out == bundle) {
2268 bundle->mirror_out |= mirror_bit;
2270 bundle->mirror_out &= ~mirror_bit;
2274 ofproto->need_revalidate = true;
2275 ofproto->has_mirrors = true;
2276 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2277 mirror_update_dups(ofproto);
2283 mirror_destroy(struct ofmirror *mirror)
2285 struct ofproto_dpif *ofproto;
2286 mirror_mask_t mirror_bit;
2287 struct ofbundle *bundle;
2294 ofproto = mirror->ofproto;
2295 ofproto->need_revalidate = true;
2296 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2298 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2299 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2300 bundle->src_mirrors &= ~mirror_bit;
2301 bundle->dst_mirrors &= ~mirror_bit;
2302 bundle->mirror_out &= ~mirror_bit;
2305 hmapx_destroy(&mirror->srcs);
2306 hmapx_destroy(&mirror->dsts);
2307 free(mirror->vlans);
2309 ofproto->mirrors[mirror->idx] = NULL;
2313 mirror_update_dups(ofproto);
2315 ofproto->has_mirrors = false;
2316 for (i = 0; i < MAX_MIRRORS; i++) {
2317 if (ofproto->mirrors[i]) {
2318 ofproto->has_mirrors = true;
2325 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2326 uint64_t *packets, uint64_t *bytes)
2328 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2329 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2332 *packets = *bytes = UINT64_MAX;
2336 *packets = mirror->packet_count;
2337 *bytes = mirror->byte_count;
2343 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2345 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2346 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2347 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2353 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2355 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2356 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2357 return bundle && bundle->mirror_out != 0;
2361 forward_bpdu_changed(struct ofproto *ofproto_)
2363 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2364 /* Revalidate cached flows whenever forward_bpdu option changes. */
2365 ofproto->need_revalidate = true;
2369 set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time)
2371 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2372 mac_learning_set_idle_time(ofproto->ml, idle_time);
2377 static struct ofport_dpif *
2378 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2380 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2381 return ofport ? ofport_dpif_cast(ofport) : NULL;
2384 static struct ofport_dpif *
2385 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2387 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2391 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2392 struct dpif_port *dpif_port)
2394 ofproto_port->name = dpif_port->name;
2395 ofproto_port->type = dpif_port->type;
2396 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2400 port_run(struct ofport_dpif *ofport)
2402 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2403 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2404 bool enable = netdev_get_carrier(ofport->up.netdev);
2406 ofport->carrier_seq = carrier_seq;
2409 cfm_run(ofport->cfm);
2411 if (cfm_should_send_ccm(ofport->cfm)) {
2412 struct ofpbuf packet;
2414 ofpbuf_init(&packet, 0);
2415 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2416 send_packet(ofport, &packet);
2417 ofpbuf_uninit(&packet);
2420 enable = enable && !cfm_get_fault(ofport->cfm)
2421 && cfm_get_opup(ofport->cfm);
2424 if (ofport->bundle) {
2425 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2426 if (carrier_changed) {
2427 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2431 if (ofport->may_enable != enable) {
2432 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2434 if (ofproto->has_bundle_action) {
2435 ofproto->need_revalidate = true;
2439 ofport->may_enable = enable;
2443 port_wait(struct ofport_dpif *ofport)
2446 cfm_wait(ofport->cfm);
2451 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2452 struct ofproto_port *ofproto_port)
2454 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2455 struct dpif_port dpif_port;
2458 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2460 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2466 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2468 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2472 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2474 *ofp_portp = odp_port_to_ofp_port(odp_port);
2480 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2482 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2485 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2487 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2489 /* The caller is going to close ofport->up.netdev. If this is a
2490 * bonded port, then the bond is using that netdev, so remove it
2491 * from the bond. The client will need to reconfigure everything
2492 * after deleting ports, so then the slave will get re-added. */
2493 bundle_remove(&ofport->up);
2500 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2502 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2505 error = netdev_get_stats(ofport->up.netdev, stats);
2507 if (!error && ofport->odp_port == OVSP_LOCAL) {
2508 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2510 /* ofproto->stats.tx_packets represents packets that we created
2511 * internally and sent to some port (e.g. packets sent with
2512 * send_packet()). Account for them as if they had come from
2513 * OFPP_LOCAL and got forwarded. */
2515 if (stats->rx_packets != UINT64_MAX) {
2516 stats->rx_packets += ofproto->stats.tx_packets;
2519 if (stats->rx_bytes != UINT64_MAX) {
2520 stats->rx_bytes += ofproto->stats.tx_bytes;
2523 /* ofproto->stats.rx_packets represents packets that were received on
2524 * some port and we processed internally and dropped (e.g. STP).
2525 * Account fro them as if they had been forwarded to OFPP_LOCAL. */
2527 if (stats->tx_packets != UINT64_MAX) {
2528 stats->tx_packets += ofproto->stats.rx_packets;
2531 if (stats->tx_bytes != UINT64_MAX) {
2532 stats->tx_bytes += ofproto->stats.rx_bytes;
2539 /* Account packets for LOCAL port. */
2541 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2542 size_t tx_size, size_t rx_size)
2544 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2547 ofproto->stats.rx_packets++;
2548 ofproto->stats.rx_bytes += rx_size;
2551 ofproto->stats.tx_packets++;
2552 ofproto->stats.tx_bytes += tx_size;
2556 struct port_dump_state {
2557 struct dpif_port_dump dump;
2562 port_dump_start(const struct ofproto *ofproto_, void **statep)
2564 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2565 struct port_dump_state *state;
2567 *statep = state = xmalloc(sizeof *state);
2568 dpif_port_dump_start(&state->dump, ofproto->dpif);
2569 state->done = false;
2574 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2575 struct ofproto_port *port)
2577 struct port_dump_state *state = state_;
2578 struct dpif_port dpif_port;
2580 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2581 ofproto_port_from_dpif_port(port, &dpif_port);
2584 int error = dpif_port_dump_done(&state->dump);
2586 return error ? error : EOF;
2591 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2593 struct port_dump_state *state = state_;
2596 dpif_port_dump_done(&state->dump);
2603 port_poll(const struct ofproto *ofproto_, char **devnamep)
2605 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2606 return dpif_port_poll(ofproto->dpif, devnamep);
2610 port_poll_wait(const struct ofproto *ofproto_)
2612 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2613 dpif_port_poll_wait(ofproto->dpif);
2617 port_is_lacp_current(const struct ofport *ofport_)
2619 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2620 return (ofport->bundle && ofport->bundle->lacp
2621 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2625 /* Upcall handling. */
2627 /* Flow miss batching.
2629 * Some dpifs implement operations faster when you hand them off in a batch.
2630 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2631 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2632 * more packets, plus possibly installing the flow in the dpif.
2634 * So far we only batch the operations that affect flow setup time the most.
2635 * It's possible to batch more than that, but the benefit might be minimal. */
2637 struct hmap_node hmap_node;
2639 enum odp_key_fitness key_fitness;
2640 const struct nlattr *key;
2642 ovs_be16 initial_tci;
2643 struct list packets;
2644 enum dpif_upcall_type upcall_type;
2647 struct flow_miss_op {
2648 struct dpif_op dpif_op;
2649 struct subfacet *subfacet; /* Subfacet */
2650 void *garbage; /* Pointer to pass to free(), NULL if none. */
2651 uint64_t stub[1024 / 8]; /* Temporary buffer. */
2654 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2655 * OpenFlow controller as necessary according to their individual
2656 * configurations. */
2658 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
2659 const struct flow *flow)
2661 struct ofputil_packet_in pin;
2663 pin.packet = packet->data;
2664 pin.packet_len = packet->size;
2665 pin.reason = OFPR_NO_MATCH;
2666 pin.controller_id = 0;
2671 pin.send_len = 0; /* not used for flow table misses */
2673 flow_get_metadata(flow, &pin.fmd);
2675 /* Registers aren't meaningful on a miss. */
2676 memset(pin.fmd.reg_masks, 0, sizeof pin.fmd.reg_masks);
2678 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
2681 static enum slow_path_reason
2682 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2683 const struct ofpbuf *packet)
2685 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2691 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2693 cfm_process_heartbeat(ofport->cfm, packet);
2696 } else if (ofport->bundle && ofport->bundle->lacp
2697 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2699 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2702 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2704 stp_process_packet(ofport, packet);
2711 static struct flow_miss *
2712 flow_miss_find(struct hmap *todo, const struct flow *flow, uint32_t hash)
2714 struct flow_miss *miss;
2716 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2717 if (flow_equal(&miss->flow, flow)) {
2725 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
2726 * 'packet'. The caller must initialize op->actions and op->actions_len. If
2727 * 'miss' is associated with a subfacet the caller must also initialize the
2728 * returned op->subfacet, and if anything needs to be freed after processing
2729 * the op, the caller must initialize op->garbage also. */
2731 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
2732 struct flow_miss_op *op)
2734 if (miss->flow.vlan_tci != miss->initial_tci) {
2735 /* This packet was received on a VLAN splinter port. We
2736 * added a VLAN to the packet to make the packet resemble
2737 * the flow, but the actions were composed assuming that
2738 * the packet contained no VLAN. So, we must remove the
2739 * VLAN header from the packet before trying to execute the
2741 eth_pop_vlan(packet);
2744 op->subfacet = NULL;
2746 op->dpif_op.type = DPIF_OP_EXECUTE;
2747 op->dpif_op.u.execute.key = miss->key;
2748 op->dpif_op.u.execute.key_len = miss->key_len;
2749 op->dpif_op.u.execute.packet = packet;
2752 /* Helper for handle_flow_miss_without_facet() and
2753 * handle_flow_miss_with_facet(). */
2755 handle_flow_miss_common(struct rule_dpif *rule,
2756 struct ofpbuf *packet, const struct flow *flow)
2758 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2760 ofproto->n_matches++;
2762 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2764 * Extra-special case for fail-open mode.
2766 * We are in fail-open mode and the packet matched the fail-open
2767 * rule, but we are connected to a controller too. We should send
2768 * the packet up to the controller in the hope that it will try to
2769 * set up a flow and thereby allow us to exit fail-open.
2771 * See the top-level comment in fail-open.c for more information.
2773 send_packet_in_miss(ofproto, packet, flow);
2777 /* Figures out whether a flow that missed in 'ofproto', whose details are in
2778 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
2779 * installing a datapath flow. The answer is usually "yes" (a return value of
2780 * true). However, for short flows the cost of bookkeeping is much higher than
2781 * the benefits, so when the datapath holds a large number of flows we impose
2782 * some heuristics to decide which flows are likely to be worth tracking. */
2784 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
2785 struct flow_miss *miss, uint32_t hash)
2787 if (!ofproto->governor) {
2790 n_subfacets = hmap_count(&ofproto->subfacets);
2791 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
2795 ofproto->governor = governor_create(ofproto->up.name);
2798 return governor_should_install_flow(ofproto->governor, hash,
2799 list_size(&miss->packets));
2802 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
2803 * or creating any datapath flow. May add an "execute" operation to 'ops' and
2804 * increment '*n_ops'. */
2806 handle_flow_miss_without_facet(struct flow_miss *miss,
2807 struct rule_dpif *rule,
2808 struct flow_miss_op *ops, size_t *n_ops)
2810 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2811 struct action_xlate_ctx ctx;
2812 struct ofpbuf *packet;
2814 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2815 struct flow_miss_op *op = &ops[*n_ops];
2816 struct dpif_flow_stats stats;
2817 struct ofpbuf odp_actions;
2819 COVERAGE_INC(facet_suppress);
2821 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
2823 dpif_flow_stats_extract(&miss->flow, packet, &stats);
2824 rule_credit_stats(rule, &stats);
2826 action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci,
2828 ctx.resubmit_stats = &stats;
2829 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions,
2832 if (odp_actions.size) {
2833 struct dpif_execute *execute = &op->dpif_op.u.execute;
2835 init_flow_miss_execute_op(miss, packet, op);
2836 execute->actions = odp_actions.data;
2837 execute->actions_len = odp_actions.size;
2838 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
2842 ofpbuf_uninit(&odp_actions);
2847 /* Handles 'miss', which matches 'facet'. May add any required datapath
2848 * operations to 'ops', incrementing '*n_ops' for each new op. */
2850 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
2851 struct flow_miss_op *ops, size_t *n_ops)
2853 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
2854 enum subfacet_path want_path;
2855 struct subfacet *subfacet;
2856 struct ofpbuf *packet;
2858 subfacet = subfacet_create(facet,
2859 miss->key_fitness, miss->key, miss->key_len,
2862 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2863 struct flow_miss_op *op = &ops[*n_ops];
2864 struct dpif_flow_stats stats;
2865 struct ofpbuf odp_actions;
2867 handle_flow_miss_common(facet->rule, packet, &miss->flow);
2869 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
2870 if (!subfacet->actions || subfacet->slow) {
2871 subfacet_make_actions(subfacet, packet, &odp_actions);
2874 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2875 subfacet_update_stats(subfacet, &stats);
2877 if (subfacet->actions_len) {
2878 struct dpif_execute *execute = &op->dpif_op.u.execute;
2880 init_flow_miss_execute_op(miss, packet, op);
2881 op->subfacet = subfacet;
2882 if (!subfacet->slow) {
2883 execute->actions = subfacet->actions;
2884 execute->actions_len = subfacet->actions_len;
2885 ofpbuf_uninit(&odp_actions);
2887 execute->actions = odp_actions.data;
2888 execute->actions_len = odp_actions.size;
2889 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
2894 ofpbuf_uninit(&odp_actions);
2898 want_path = subfacet_want_path(subfacet->slow);
2899 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
2900 struct flow_miss_op *op = &ops[(*n_ops)++];
2901 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
2903 op->subfacet = subfacet;
2905 op->dpif_op.type = DPIF_OP_FLOW_PUT;
2906 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2907 put->key = miss->key;
2908 put->key_len = miss->key_len;
2909 if (want_path == SF_FAST_PATH) {
2910 put->actions = subfacet->actions;
2911 put->actions_len = subfacet->actions_len;
2913 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
2914 op->stub, sizeof op->stub,
2915 &put->actions, &put->actions_len);
2921 /* Handles flow miss 'miss' on 'ofproto'. May add any required datapath
2922 * operations to 'ops', incrementing '*n_ops' for each new op. */
2924 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2925 struct flow_miss_op *ops, size_t *n_ops)
2927 struct facet *facet;
2930 /* The caller must ensure that miss->hmap_node.hash contains
2931 * flow_hash(miss->flow, 0). */
2932 hash = miss->hmap_node.hash;
2934 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
2936 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
2938 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
2939 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
2943 facet = facet_create(rule, &miss->flow, hash);
2945 handle_flow_miss_with_facet(miss, facet, ops, n_ops);
2948 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
2949 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
2950 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
2951 * what a flow key should contain.
2953 * This function also includes some logic to help make VLAN splinters
2954 * transparent to the rest of the upcall processing logic. In particular, if
2955 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
2956 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
2957 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
2959 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
2960 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
2961 * (This differs from the value returned in flow->vlan_tci only for packets
2962 * received on VLAN splinters.)
2964 static enum odp_key_fitness
2965 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
2966 const struct nlattr *key, size_t key_len,
2967 struct flow *flow, ovs_be16 *initial_tci,
2968 struct ofpbuf *packet)
2970 enum odp_key_fitness fitness;
2972 fitness = odp_flow_key_to_flow(key, key_len, flow);
2973 if (fitness == ODP_FIT_ERROR) {
2976 *initial_tci = flow->vlan_tci;
2978 if (vsp_adjust_flow(ofproto, flow)) {
2980 /* Make the packet resemble the flow, so that it gets sent to an
2981 * OpenFlow controller properly, so that it looks correct for
2982 * sFlow, and so that flow_extract() will get the correct vlan_tci
2983 * if it is called on 'packet'.
2985 * The allocated space inside 'packet' probably also contains
2986 * 'key', that is, both 'packet' and 'key' are probably part of a
2987 * struct dpif_upcall (see the large comment on that structure
2988 * definition), so pushing data on 'packet' is in general not a
2989 * good idea since it could overwrite 'key' or free it as a side
2990 * effect. However, it's OK in this special case because we know
2991 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
2992 * will just overwrite the 4-byte "struct nlattr", which is fine
2993 * since we don't need that header anymore. */
2994 eth_push_vlan(packet, flow->vlan_tci);
2997 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
2998 if (fitness == ODP_FIT_PERFECT) {
2999 fitness = ODP_FIT_TOO_MUCH;
3007 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
3010 struct dpif_upcall *upcall;
3011 struct flow_miss *miss;
3012 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3013 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3014 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3024 /* Construct the to-do list.
3026 * This just amounts to extracting the flow from each packet and sticking
3027 * the packets that have the same flow in the same "flow_miss" structure so
3028 * that we can process them together. */
3031 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3032 struct flow_miss *miss = &misses[n_misses];
3033 struct flow_miss *existing_miss;
3036 /* Obtain metadata and check userspace/kernel agreement on flow match,
3037 * then set 'flow''s header pointers. */
3038 miss->key_fitness = ofproto_dpif_extract_flow_key(
3039 ofproto, upcall->key, upcall->key_len,
3040 &miss->flow, &miss->initial_tci, upcall->packet);
3041 if (miss->key_fitness == ODP_FIT_ERROR) {
3044 flow_extract(upcall->packet, miss->flow.skb_priority,
3045 miss->flow.tun_id, miss->flow.in_port, &miss->flow);
3047 /* Add other packets to a to-do list. */
3048 hash = flow_hash(&miss->flow, 0);
3049 existing_miss = flow_miss_find(&todo, &miss->flow, hash);
3050 if (!existing_miss) {
3051 hmap_insert(&todo, &miss->hmap_node, hash);
3052 miss->key = upcall->key;
3053 miss->key_len = upcall->key_len;
3054 miss->upcall_type = upcall->type;
3055 list_init(&miss->packets);
3059 miss = existing_miss;
3061 list_push_back(&miss->packets, &upcall->packet->list_node);
3064 /* Process each element in the to-do list, constructing the set of
3065 * operations to batch. */
3067 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3068 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
3070 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3072 /* Execute batch. */
3073 for (i = 0; i < n_ops; i++) {
3074 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3076 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
3078 /* Free memory and update facets. */
3079 for (i = 0; i < n_ops; i++) {
3080 struct flow_miss_op *op = &flow_miss_ops[i];
3082 switch (op->dpif_op.type) {
3083 case DPIF_OP_EXECUTE:
3086 case DPIF_OP_FLOW_PUT:
3087 if (!op->dpif_op.error) {
3088 op->subfacet->path = subfacet_want_path(op->subfacet->slow);
3092 case DPIF_OP_FLOW_DEL:
3098 hmap_destroy(&todo);
3101 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
3102 classify_upcall(const struct dpif_upcall *upcall)
3104 union user_action_cookie cookie;
3106 /* First look at the upcall type. */
3107 switch (upcall->type) {
3108 case DPIF_UC_ACTION:
3114 case DPIF_N_UC_TYPES:
3116 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3120 /* "action" upcalls need a closer look. */
3121 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3122 switch (cookie.type) {
3123 case USER_ACTION_COOKIE_SFLOW:
3124 return SFLOW_UPCALL;
3126 case USER_ACTION_COOKIE_SLOW_PATH:
3129 case USER_ACTION_COOKIE_UNSPEC:
3131 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
3137 handle_sflow_upcall(struct ofproto_dpif *ofproto,
3138 const struct dpif_upcall *upcall)
3140 union user_action_cookie cookie;
3141 enum odp_key_fitness fitness;
3142 ovs_be16 initial_tci;
3145 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
3146 upcall->key_len, &flow,
3147 &initial_tci, upcall->packet);
3148 if (fitness == ODP_FIT_ERROR) {
3152 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3153 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, &cookie);
3157 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
3159 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3160 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3161 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3166 assert(max_batch <= FLOW_MISS_MAX_BATCH);
3169 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3170 struct dpif_upcall *upcall = &misses[n_misses];
3171 struct ofpbuf *buf = &miss_bufs[n_misses];
3174 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3175 sizeof miss_buf_stubs[n_misses]);
3176 error = dpif_recv(ofproto->dpif, upcall, buf);
3182 switch (classify_upcall(upcall)) {
3184 /* Handle it later. */
3189 if (ofproto->sflow) {
3190 handle_sflow_upcall(ofproto, upcall);
3201 /* Handle deferred MISS_UPCALL processing. */
3202 handle_miss_upcalls(ofproto, misses, n_misses);
3203 for (i = 0; i < n_misses; i++) {
3204 ofpbuf_uninit(&miss_bufs[i]);
3210 /* Flow expiration. */
3212 static int subfacet_max_idle(const struct ofproto_dpif *);
3213 static void update_stats(struct ofproto_dpif *);
3214 static void rule_expire(struct rule_dpif *);
3215 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
3217 /* This function is called periodically by run(). Its job is to collect
3218 * updates for the flows that have been installed into the datapath, most
3219 * importantly when they last were used, and then use that information to
3220 * expire flows that have not been used recently.
3222 * Returns the number of milliseconds after which it should be called again. */
3224 expire(struct ofproto_dpif *ofproto)
3226 struct rule_dpif *rule, *next_rule;
3227 struct oftable *table;
3230 /* Update stats for each flow in the datapath. */
3231 update_stats(ofproto);
3233 /* Expire subfacets that have been idle too long. */
3234 dp_max_idle = subfacet_max_idle(ofproto);
3235 expire_subfacets(ofproto, dp_max_idle);
3237 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
3238 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
3239 struct cls_cursor cursor;
3241 cls_cursor_init(&cursor, &table->cls, NULL);
3242 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
3247 /* All outstanding data in existing flows has been accounted, so it's a
3248 * good time to do bond rebalancing. */
3249 if (ofproto->has_bonded_bundles) {
3250 struct ofbundle *bundle;
3252 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3254 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
3259 return MIN(dp_max_idle, 1000);
3262 /* Updates flow table statistics given that the datapath just reported 'stats'
3263 * as 'subfacet''s statistics. */
3265 update_subfacet_stats(struct subfacet *subfacet,
3266 const struct dpif_flow_stats *stats)
3268 struct facet *facet = subfacet->facet;
3270 if (stats->n_packets >= subfacet->dp_packet_count) {
3271 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
3272 facet->packet_count += extra;
3274 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3277 if (stats->n_bytes >= subfacet->dp_byte_count) {
3278 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
3280 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3283 subfacet->dp_packet_count = stats->n_packets;
3284 subfacet->dp_byte_count = stats->n_bytes;
3286 facet->tcp_flags |= stats->tcp_flags;
3288 subfacet_update_time(subfacet, stats->used);
3289 if (facet->accounted_bytes < facet->byte_count) {
3291 facet_account(facet);
3292 facet->accounted_bytes = facet->byte_count;
3294 facet_push_stats(facet);
3297 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3298 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3300 delete_unexpected_flow(struct dpif *dpif,
3301 const struct nlattr *key, size_t key_len)
3303 if (!VLOG_DROP_WARN(&rl)) {
3307 odp_flow_key_format(key, key_len, &s);
3308 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
3312 COVERAGE_INC(facet_unexpected);
3313 dpif_flow_del(dpif, key, key_len, NULL);
3316 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3318 * This function also pushes statistics updates to rules which each facet
3319 * resubmits into. Generally these statistics will be accurate. However, if a
3320 * facet changes the rule it resubmits into at some time in between
3321 * update_stats() runs, it is possible that statistics accrued to the
3322 * old rule will be incorrectly attributed to the new rule. This could be
3323 * avoided by calling update_stats() whenever rules are created or
3324 * deleted. However, the performance impact of making so many calls to the
3325 * datapath do not justify the benefit of having perfectly accurate statistics.
3328 update_stats(struct ofproto_dpif *p)
3330 const struct dpif_flow_stats *stats;
3331 struct dpif_flow_dump dump;
3332 const struct nlattr *key;
3335 dpif_flow_dump_start(&dump, p->dpif);
3336 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
3337 struct subfacet *subfacet;
3339 subfacet = subfacet_find(p, key, key_len);
3340 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3342 update_subfacet_stats(subfacet, stats);
3346 /* Stats are updated per-packet. */
3349 case SF_NOT_INSTALLED:
3351 delete_unexpected_flow(p->dpif, key, key_len);
3355 dpif_flow_dump_done(&dump);
3358 /* Calculates and returns the number of milliseconds of idle time after which
3359 * subfacets should expire from the datapath. When a subfacet expires, we fold
3360 * its statistics into its facet, and when a facet's last subfacet expires, we
3361 * fold its statistic into its rule. */
3363 subfacet_max_idle(const struct ofproto_dpif *ofproto)
3366 * Idle time histogram.
3368 * Most of the time a switch has a relatively small number of subfacets.
3369 * When this is the case we might as well keep statistics for all of them
3370 * in userspace and to cache them in the kernel datapath for performance as
3373 * As the number of subfacets increases, the memory required to maintain
3374 * statistics about them in userspace and in the kernel becomes
3375 * significant. However, with a large number of subfacets it is likely
3376 * that only a few of them are "heavy hitters" that consume a large amount
3377 * of bandwidth. At this point, only heavy hitters are worth caching in
3378 * the kernel and maintaining in userspaces; other subfacets we can
3381 * The technique used to compute the idle time is to build a histogram with
3382 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3383 * that is installed in the kernel gets dropped in the appropriate bucket.
3384 * After the histogram has been built, we compute the cutoff so that only
3385 * the most-recently-used 1% of subfacets (but at least
3386 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3387 * the most-recently-used bucket of subfacets is kept, so actually an
3388 * arbitrary number of subfacets can be kept in any given expiration run
3389 * (though the next run will delete most of those unless they receive
3392 * This requires a second pass through the subfacets, in addition to the
3393 * pass made by update_stats(), because the former function never looks at
3394 * uninstallable subfacets.
3396 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3397 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3398 int buckets[N_BUCKETS] = { 0 };
3399 int total, subtotal, bucket;
3400 struct subfacet *subfacet;
3404 total = hmap_count(&ofproto->subfacets);
3405 if (total <= ofproto->up.flow_eviction_threshold) {
3406 return N_BUCKETS * BUCKET_WIDTH;
3409 /* Build histogram. */
3411 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3412 long long int idle = now - subfacet->used;
3413 int bucket = (idle <= 0 ? 0
3414 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3415 : (unsigned int) idle / BUCKET_WIDTH);
3419 /* Find the first bucket whose flows should be expired. */
3420 subtotal = bucket = 0;
3422 subtotal += buckets[bucket++];
3423 } while (bucket < N_BUCKETS &&
3424 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3426 if (VLOG_IS_DBG_ENABLED()) {
3430 ds_put_cstr(&s, "keep");
3431 for (i = 0; i < N_BUCKETS; i++) {
3433 ds_put_cstr(&s, ", drop");
3436 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3439 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3443 return bucket * BUCKET_WIDTH;
3446 enum { EXPIRE_MAX_BATCH = 50 };
3449 expire_batch(struct ofproto_dpif *ofproto, struct subfacet **subfacets, int n)
3451 struct odputil_keybuf keybufs[EXPIRE_MAX_BATCH];
3452 struct dpif_op ops[EXPIRE_MAX_BATCH];
3453 struct dpif_op *opsp[EXPIRE_MAX_BATCH];
3454 struct ofpbuf keys[EXPIRE_MAX_BATCH];
3455 struct dpif_flow_stats stats[EXPIRE_MAX_BATCH];
3458 for (i = 0; i < n; i++) {
3459 ops[i].type = DPIF_OP_FLOW_DEL;
3460 subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]);
3461 ops[i].u.flow_del.key = keys[i].data;
3462 ops[i].u.flow_del.key_len = keys[i].size;
3463 ops[i].u.flow_del.stats = &stats[i];
3467 dpif_operate(ofproto->dpif, opsp, n);
3468 for (i = 0; i < n; i++) {
3469 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
3470 subfacets[i]->path = SF_NOT_INSTALLED;
3471 subfacet_destroy(subfacets[i]);
3476 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3478 /* Cutoff time for most flows. */
3479 long long int normal_cutoff = time_msec() - dp_max_idle;
3481 /* We really want to keep flows for special protocols around, so use a more
3482 * conservative cutoff. */
3483 long long int special_cutoff = time_msec() - 10000;
3485 struct subfacet *subfacet, *next_subfacet;
3486 struct subfacet *batch[EXPIRE_MAX_BATCH];
3490 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3491 &ofproto->subfacets) {
3492 long long int cutoff;
3494 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
3497 if (subfacet->used < cutoff) {
3498 if (subfacet->path != SF_NOT_INSTALLED) {
3499 batch[n_batch++] = subfacet;
3500 if (n_batch >= EXPIRE_MAX_BATCH) {
3501 expire_batch(ofproto, batch, n_batch);
3505 subfacet_destroy(subfacet);
3511 expire_batch(ofproto, batch, n_batch);
3515 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3516 * then delete it entirely. */
3518 rule_expire(struct rule_dpif *rule)
3520 struct facet *facet, *next_facet;
3524 /* Has 'rule' expired? */
3526 if (rule->up.hard_timeout
3527 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3528 reason = OFPRR_HARD_TIMEOUT;
3529 } else if (rule->up.idle_timeout
3530 && now > rule->up.used + rule->up.idle_timeout * 1000) {
3531 reason = OFPRR_IDLE_TIMEOUT;
3536 COVERAGE_INC(ofproto_dpif_expired);
3538 /* Update stats. (This is a no-op if the rule expired due to an idle
3539 * timeout, because that only happens when the rule has no facets left.) */
3540 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3541 facet_remove(facet);
3544 /* Get rid of the rule. */
3545 ofproto_rule_expire(&rule->up, reason);
3550 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3552 * The caller must already have determined that no facet with an identical
3553 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3554 * the ofproto's classifier table.
3556 * 'hash' must be the return value of flow_hash(flow, 0).
3558 * The facet will initially have no subfacets. The caller should create (at
3559 * least) one subfacet with subfacet_create(). */
3560 static struct facet *
3561 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
3563 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3564 struct facet *facet;
3566 facet = xzalloc(sizeof *facet);
3567 facet->used = time_msec();
3568 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
3569 list_push_back(&rule->facets, &facet->list_node);
3571 facet->flow = *flow;
3572 list_init(&facet->subfacets);
3573 netflow_flow_init(&facet->nf_flow);
3574 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3580 facet_free(struct facet *facet)
3585 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3586 * 'packet', which arrived on 'in_port'.
3588 * Takes ownership of 'packet'. */
3590 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3591 const struct nlattr *odp_actions, size_t actions_len,
3592 struct ofpbuf *packet)
3594 struct odputil_keybuf keybuf;
3598 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3599 odp_flow_key_from_flow(&key, flow);
3601 error = dpif_execute(ofproto->dpif, key.data, key.size,
3602 odp_actions, actions_len, packet);
3604 ofpbuf_delete(packet);
3608 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3610 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3611 * rule's statistics, via subfacet_uninstall().
3613 * - Removes 'facet' from its rule and from ofproto->facets.
3616 facet_remove(struct facet *facet)
3618 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3619 struct subfacet *subfacet, *next_subfacet;
3621 assert(!list_is_empty(&facet->subfacets));
3623 /* First uninstall all of the subfacets to get final statistics. */
3624 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3625 subfacet_uninstall(subfacet);
3628 /* Flush the final stats to the rule.
3630 * This might require us to have at least one subfacet around so that we
3631 * can use its actions for accounting in facet_account(), which is why we
3632 * have uninstalled but not yet destroyed the subfacets. */
3633 facet_flush_stats(facet);
3635 /* Now we're really all done so destroy everything. */
3636 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3637 &facet->subfacets) {
3638 subfacet_destroy__(subfacet);
3640 hmap_remove(&ofproto->facets, &facet->hmap_node);
3641 list_remove(&facet->list_node);
3645 /* Feed information from 'facet' back into the learning table to keep it in
3646 * sync with what is actually flowing through the datapath. */
3648 facet_learn(struct facet *facet)
3650 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3651 struct action_xlate_ctx ctx;
3653 if (!facet->has_learn
3654 && !facet->has_normal
3655 && (!facet->has_fin_timeout
3656 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
3660 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3661 facet->flow.vlan_tci,
3662 facet->rule, facet->tcp_flags, NULL);
3663 ctx.may_learn = true;
3664 xlate_actions_for_side_effects(&ctx, facet->rule->up.actions,
3665 facet->rule->up.n_actions);
3669 facet_account(struct facet *facet)
3671 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3672 struct subfacet *subfacet;
3673 const struct nlattr *a;
3678 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3681 n_bytes = facet->byte_count - facet->accounted_bytes;
3683 /* This loop feeds byte counters to bond_account() for rebalancing to use
3684 * as a basis. We also need to track the actual VLAN on which the packet
3685 * is going to be sent to ensure that it matches the one passed to
3686 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3689 * We use the actions from an arbitrary subfacet because they should all
3690 * be equally valid for our purpose. */
3691 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3692 struct subfacet, list_node);
3693 vlan_tci = facet->flow.vlan_tci;
3694 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3695 subfacet->actions, subfacet->actions_len) {
3696 const struct ovs_action_push_vlan *vlan;
3697 struct ofport_dpif *port;
3699 switch (nl_attr_type(a)) {
3700 case OVS_ACTION_ATTR_OUTPUT:
3701 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3702 if (port && port->bundle && port->bundle->bond) {
3703 bond_account(port->bundle->bond, &facet->flow,
3704 vlan_tci_to_vid(vlan_tci), n_bytes);
3708 case OVS_ACTION_ATTR_POP_VLAN:
3709 vlan_tci = htons(0);
3712 case OVS_ACTION_ATTR_PUSH_VLAN:
3713 vlan = nl_attr_get(a);
3714 vlan_tci = vlan->vlan_tci;
3720 /* Returns true if the only action for 'facet' is to send to the controller.
3721 * (We don't report NetFlow expiration messages for such facets because they
3722 * are just part of the control logic for the network, not real traffic). */
3724 facet_is_controller_flow(struct facet *facet)
3727 && facet->rule->up.n_actions == 1
3728 && action_outputs_to_port(&facet->rule->up.actions[0],
3729 htons(OFPP_CONTROLLER)));
3732 /* Folds all of 'facet''s statistics into its rule. Also updates the
3733 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3734 * 'facet''s statistics in the datapath should have been zeroed and folded into
3735 * its packet and byte counts before this function is called. */
3737 facet_flush_stats(struct facet *facet)
3739 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3740 struct subfacet *subfacet;
3742 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3743 assert(!subfacet->dp_byte_count);
3744 assert(!subfacet->dp_packet_count);
3747 facet_push_stats(facet);
3748 if (facet->accounted_bytes < facet->byte_count) {
3749 facet_account(facet);
3750 facet->accounted_bytes = facet->byte_count;
3753 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3754 struct ofexpired expired;
3755 expired.flow = facet->flow;
3756 expired.packet_count = facet->packet_count;
3757 expired.byte_count = facet->byte_count;
3758 expired.used = facet->used;
3759 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3762 facet->rule->packet_count += facet->packet_count;
3763 facet->rule->byte_count += facet->byte_count;
3765 /* Reset counters to prevent double counting if 'facet' ever gets
3767 facet_reset_counters(facet);
3769 netflow_flow_clear(&facet->nf_flow);
3770 facet->tcp_flags = 0;
3773 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3774 * Returns it if found, otherwise a null pointer.
3776 * 'hash' must be the return value of flow_hash(flow, 0).
3778 * The returned facet might need revalidation; use facet_lookup_valid()
3779 * instead if that is important. */
3780 static struct facet *
3781 facet_find(struct ofproto_dpif *ofproto,
3782 const struct flow *flow, uint32_t hash)
3784 struct facet *facet;
3786 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
3787 if (flow_equal(flow, &facet->flow)) {
3795 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3796 * Returns it if found, otherwise a null pointer.
3798 * 'hash' must be the return value of flow_hash(flow, 0).
3800 * The returned facet is guaranteed to be valid. */
3801 static struct facet *
3802 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
3805 struct facet *facet;
3807 facet = facet_find(ofproto, flow, hash);
3809 && (ofproto->need_revalidate
3810 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))) {
3811 facet_revalidate(facet);
3818 subfacet_path_to_string(enum subfacet_path path)
3821 case SF_NOT_INSTALLED:
3822 return "not installed";
3824 return "in fast path";
3826 return "in slow path";
3832 /* Returns the path in which a subfacet should be installed if its 'slow'
3833 * member has the specified value. */
3834 static enum subfacet_path
3835 subfacet_want_path(enum slow_path_reason slow)
3837 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
3840 /* Returns true if 'subfacet' needs to have its datapath flow updated,
3841 * supposing that its actions have been recalculated as 'want_actions' and that
3842 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
3844 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
3845 const struct ofpbuf *want_actions)
3847 enum subfacet_path want_path = subfacet_want_path(slow);
3848 return (want_path != subfacet->path
3849 || (want_path == SF_FAST_PATH
3850 && (subfacet->actions_len != want_actions->size
3851 || memcmp(subfacet->actions, want_actions->data,
3852 subfacet->actions_len))));
3856 facet_check_consistency(struct facet *facet)
3858 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3860 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3862 uint64_t odp_actions_stub[1024 / 8];
3863 struct ofpbuf odp_actions;
3865 struct rule_dpif *rule;
3866 struct subfacet *subfacet;
3867 bool may_log = false;
3870 /* Check the rule for consistency. */
3871 rule = rule_dpif_lookup(ofproto, &facet->flow);
3872 ok = rule == facet->rule;
3874 may_log = !VLOG_DROP_WARN(&rl);
3879 flow_format(&s, &facet->flow);
3880 ds_put_format(&s, ": facet associated with wrong rule (was "
3881 "table=%"PRIu8",", facet->rule->up.table_id);
3882 cls_rule_format(&facet->rule->up.cr, &s);
3883 ds_put_format(&s, ") (should have been table=%"PRIu8",",
3885 cls_rule_format(&rule->up.cr, &s);
3886 ds_put_char(&s, ')');
3888 VLOG_WARN("%s", ds_cstr(&s));
3893 /* Check the datapath actions for consistency. */
3894 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
3895 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3896 enum subfacet_path want_path;
3897 struct odputil_keybuf keybuf;
3898 struct action_xlate_ctx ctx;
3902 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3903 subfacet->initial_tci, rule, 0, NULL);
3904 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions,
3907 if (subfacet->path == SF_NOT_INSTALLED) {
3908 /* This only happens if the datapath reported an error when we
3909 * tried to install the flow. Don't flag another error here. */
3913 want_path = subfacet_want_path(subfacet->slow);
3914 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
3915 /* The actions for slow-path flows may legitimately vary from one
3916 * packet to the next. We're done. */
3920 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
3924 /* Inconsistency! */
3926 may_log = !VLOG_DROP_WARN(&rl);
3930 /* Rate-limited, skip reporting. */
3935 subfacet_get_key(subfacet, &keybuf, &key);
3936 odp_flow_key_format(key.data, key.size, &s);
3938 ds_put_cstr(&s, ": inconsistency in subfacet");
3939 if (want_path != subfacet->path) {
3940 enum odp_key_fitness fitness = subfacet->key_fitness;
3942 ds_put_format(&s, " (%s, fitness=%s)",
3943 subfacet_path_to_string(subfacet->path),
3944 odp_key_fitness_to_string(fitness));
3945 ds_put_format(&s, " (should have been %s)",
3946 subfacet_path_to_string(want_path));
3947 } else if (want_path == SF_FAST_PATH) {
3948 ds_put_cstr(&s, " (actions were: ");
3949 format_odp_actions(&s, subfacet->actions,
3950 subfacet->actions_len);
3951 ds_put_cstr(&s, ") (correct actions: ");
3952 format_odp_actions(&s, odp_actions.data, odp_actions.size);
3953 ds_put_char(&s, ')');
3955 ds_put_cstr(&s, " (actions: ");
3956 format_odp_actions(&s, subfacet->actions,
3957 subfacet->actions_len);
3958 ds_put_char(&s, ')');
3960 VLOG_WARN("%s", ds_cstr(&s));
3963 ofpbuf_uninit(&odp_actions);
3968 /* Re-searches the classifier for 'facet':
3970 * - If the rule found is different from 'facet''s current rule, moves
3971 * 'facet' to the new rule and recompiles its actions.
3973 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3974 * where it is and recompiles its actions anyway. */
3976 facet_revalidate(struct facet *facet)
3978 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3980 struct nlattr *odp_actions;
3983 struct actions *new_actions;
3985 struct action_xlate_ctx ctx;
3986 uint64_t odp_actions_stub[1024 / 8];
3987 struct ofpbuf odp_actions;
3989 struct rule_dpif *new_rule;
3990 struct subfacet *subfacet;
3993 COVERAGE_INC(facet_revalidate);
3995 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
3997 /* Calculate new datapath actions.
3999 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4000 * emit a NetFlow expiration and, if so, we need to have the old state
4001 * around to properly compose it. */
4003 /* If the datapath actions changed or the installability changed,
4004 * then we need to talk to the datapath. */
4007 memset(&ctx, 0, sizeof ctx);
4008 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4009 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4010 enum slow_path_reason slow;
4012 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4013 subfacet->initial_tci, new_rule, 0, NULL);
4014 xlate_actions(&ctx, new_rule->up.actions, new_rule->up.n_actions,
4017 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4018 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
4019 struct dpif_flow_stats stats;
4021 subfacet_install(subfacet,
4022 odp_actions.data, odp_actions.size, &stats, slow);
4023 subfacet_update_stats(subfacet, &stats);
4026 new_actions = xcalloc(list_size(&facet->subfacets),
4027 sizeof *new_actions);
4029 new_actions[i].odp_actions = xmemdup(odp_actions.data,
4031 new_actions[i].actions_len = odp_actions.size;
4036 ofpbuf_uninit(&odp_actions);
4039 facet_flush_stats(facet);
4042 /* Update 'facet' now that we've taken care of all the old state. */
4043 facet->tags = ctx.tags;
4044 facet->nf_flow.output_iface = ctx.nf_output_iface;
4045 facet->has_learn = ctx.has_learn;
4046 facet->has_normal = ctx.has_normal;
4047 facet->has_fin_timeout = ctx.has_fin_timeout;
4048 facet->mirrors = ctx.mirrors;
4051 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4052 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4054 if (new_actions && new_actions[i].odp_actions) {
4055 free(subfacet->actions);
4056 subfacet->actions = new_actions[i].odp_actions;
4057 subfacet->actions_len = new_actions[i].actions_len;
4063 if (facet->rule != new_rule) {
4064 COVERAGE_INC(facet_changed_rule);
4065 list_remove(&facet->list_node);
4066 list_push_back(&new_rule->facets, &facet->list_node);
4067 facet->rule = new_rule;
4068 facet->used = new_rule->up.created;
4069 facet->prev_used = facet->used;
4073 /* Updates 'facet''s used time. Caller is responsible for calling
4074 * facet_push_stats() to update the flows which 'facet' resubmits into. */
4076 facet_update_time(struct facet *facet, long long int used)
4078 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4079 if (used > facet->used) {
4081 ofproto_rule_update_used(&facet->rule->up, used);
4082 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
4087 facet_reset_counters(struct facet *facet)
4089 facet->packet_count = 0;
4090 facet->byte_count = 0;
4091 facet->prev_packet_count = 0;
4092 facet->prev_byte_count = 0;
4093 facet->accounted_bytes = 0;
4097 facet_push_stats(struct facet *facet)
4099 struct dpif_flow_stats stats;
4101 assert(facet->packet_count >= facet->prev_packet_count);
4102 assert(facet->byte_count >= facet->prev_byte_count);
4103 assert(facet->used >= facet->prev_used);
4105 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4106 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4107 stats.used = facet->used;
4108 stats.tcp_flags = 0;
4110 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
4111 facet->prev_packet_count = facet->packet_count;
4112 facet->prev_byte_count = facet->byte_count;
4113 facet->prev_used = facet->used;
4115 flow_push_stats(facet->rule, &facet->flow, &stats);
4117 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
4118 facet->mirrors, stats.n_packets, stats.n_bytes);
4123 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4125 rule->packet_count += stats->n_packets;
4126 rule->byte_count += stats->n_bytes;
4127 ofproto_rule_update_used(&rule->up, stats->used);
4130 /* Pushes flow statistics to the rules which 'flow' resubmits into given
4131 * 'rule''s actions and mirrors. */
4133 flow_push_stats(struct rule_dpif *rule,
4134 const struct flow *flow, const struct dpif_flow_stats *stats)
4136 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4137 struct action_xlate_ctx ctx;
4139 ofproto_rule_update_used(&rule->up, stats->used);
4141 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule,
4143 ctx.resubmit_stats = stats;
4144 xlate_actions_for_side_effects(&ctx, rule->up.actions, rule->up.n_actions);
4149 static struct subfacet *
4150 subfacet_find__(struct ofproto_dpif *ofproto,
4151 const struct nlattr *key, size_t key_len, uint32_t key_hash,
4152 const struct flow *flow)
4154 struct subfacet *subfacet;
4156 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4157 &ofproto->subfacets) {
4159 ? (subfacet->key_len == key_len
4160 && !memcmp(key, subfacet->key, key_len))
4161 : flow_equal(flow, &subfacet->facet->flow)) {
4169 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4170 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
4171 * there is one, otherwise creates and returns a new subfacet.
4173 * If the returned subfacet is new, then subfacet->actions will be NULL, in
4174 * which case the caller must populate the actions with
4175 * subfacet_make_actions(). */
4176 static struct subfacet *
4177 subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness,
4178 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
4180 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4181 uint32_t key_hash = odp_flow_key_hash(key, key_len);
4182 struct subfacet *subfacet;
4184 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
4186 if (subfacet->facet == facet) {
4190 /* This shouldn't happen. */
4191 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4192 subfacet_destroy(subfacet);
4195 subfacet = (list_is_empty(&facet->subfacets)
4196 ? &facet->one_subfacet
4197 : xmalloc(sizeof *subfacet));
4198 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
4199 list_push_back(&facet->subfacets, &subfacet->list_node);
4200 subfacet->facet = facet;
4201 subfacet->key_fitness = key_fitness;
4202 if (key_fitness != ODP_FIT_PERFECT) {
4203 subfacet->key = xmemdup(key, key_len);
4204 subfacet->key_len = key_len;
4206 subfacet->key = NULL;
4207 subfacet->key_len = 0;
4209 subfacet->used = time_msec();
4210 subfacet->dp_packet_count = 0;
4211 subfacet->dp_byte_count = 0;
4212 subfacet->actions_len = 0;
4213 subfacet->actions = NULL;
4214 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
4217 subfacet->path = SF_NOT_INSTALLED;
4218 subfacet->initial_tci = initial_tci;
4223 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
4224 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
4225 static struct subfacet *
4226 subfacet_find(struct ofproto_dpif *ofproto,
4227 const struct nlattr *key, size_t key_len)
4229 uint32_t key_hash = odp_flow_key_hash(key, key_len);
4230 enum odp_key_fitness fitness;
4233 fitness = odp_flow_key_to_flow(key, key_len, &flow);
4234 if (fitness == ODP_FIT_ERROR) {
4238 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
4241 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4242 * its facet within 'ofproto', and frees it. */
4244 subfacet_destroy__(struct subfacet *subfacet)
4246 struct facet *facet = subfacet->facet;
4247 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4249 subfacet_uninstall(subfacet);
4250 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
4251 list_remove(&subfacet->list_node);
4252 free(subfacet->key);
4253 free(subfacet->actions);
4254 if (subfacet != &facet->one_subfacet) {
4259 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4260 * last remaining subfacet in its facet destroys the facet too. */
4262 subfacet_destroy(struct subfacet *subfacet)
4264 struct facet *facet = subfacet->facet;
4266 if (list_is_singleton(&facet->subfacets)) {
4267 /* facet_remove() needs at least one subfacet (it will remove it). */
4268 facet_remove(facet);
4270 subfacet_destroy__(subfacet);
4274 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
4275 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
4276 * for use as temporary storage. */
4278 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
4281 if (!subfacet->key) {
4282 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
4283 odp_flow_key_from_flow(key, &subfacet->facet->flow);
4285 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
4289 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
4290 * Translates the actions into 'odp_actions', which the caller must have
4291 * initialized and is responsible for uninitializing. */
4293 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
4294 struct ofpbuf *odp_actions)
4296 struct facet *facet = subfacet->facet;
4297 struct rule_dpif *rule = facet->rule;
4298 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4300 struct action_xlate_ctx ctx;
4302 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
4304 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions, odp_actions);
4305 facet->tags = ctx.tags;
4306 facet->has_learn = ctx.has_learn;
4307 facet->has_normal = ctx.has_normal;
4308 facet->has_fin_timeout = ctx.has_fin_timeout;
4309 facet->nf_flow.output_iface = ctx.nf_output_iface;
4310 facet->mirrors = ctx.mirrors;
4312 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4313 if (subfacet->actions_len != odp_actions->size
4314 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
4315 free(subfacet->actions);
4316 subfacet->actions_len = odp_actions->size;
4317 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
4321 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4322 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4323 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4324 * since 'subfacet' was last updated.
4326 * Returns 0 if successful, otherwise a positive errno value. */
4328 subfacet_install(struct subfacet *subfacet,
4329 const struct nlattr *actions, size_t actions_len,
4330 struct dpif_flow_stats *stats,
4331 enum slow_path_reason slow)
4333 struct facet *facet = subfacet->facet;
4334 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4335 enum subfacet_path path = subfacet_want_path(slow);
4336 uint64_t slow_path_stub[128 / 8];
4337 struct odputil_keybuf keybuf;
4338 enum dpif_flow_put_flags flags;
4342 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
4344 flags |= DPIF_FP_ZERO_STATS;
4347 if (path == SF_SLOW_PATH) {
4348 compose_slow_path(ofproto, &facet->flow, slow,
4349 slow_path_stub, sizeof slow_path_stub,
4350 &actions, &actions_len);
4353 subfacet_get_key(subfacet, &keybuf, &key);
4354 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
4355 actions, actions_len, stats);
4358 subfacet_reset_dp_stats(subfacet, stats);
4362 subfacet->path = path;
4368 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
4370 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
4371 stats, subfacet->slow);
4374 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4376 subfacet_uninstall(struct subfacet *subfacet)
4378 if (subfacet->path != SF_NOT_INSTALLED) {
4379 struct rule_dpif *rule = subfacet->facet->rule;
4380 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4381 struct odputil_keybuf keybuf;
4382 struct dpif_flow_stats stats;
4386 subfacet_get_key(subfacet, &keybuf, &key);
4387 error = dpif_flow_del(ofproto->dpif, key.data, key.size, &stats);
4388 subfacet_reset_dp_stats(subfacet, &stats);
4390 subfacet_update_stats(subfacet, &stats);
4392 subfacet->path = SF_NOT_INSTALLED;
4394 assert(subfacet->dp_packet_count == 0);
4395 assert(subfacet->dp_byte_count == 0);
4399 /* Resets 'subfacet''s datapath statistics counters. This should be called
4400 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4401 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4402 * was reset in the datapath. 'stats' will be modified to include only
4403 * statistics new since 'subfacet' was last updated. */
4405 subfacet_reset_dp_stats(struct subfacet *subfacet,
4406 struct dpif_flow_stats *stats)
4409 && subfacet->dp_packet_count <= stats->n_packets
4410 && subfacet->dp_byte_count <= stats->n_bytes) {
4411 stats->n_packets -= subfacet->dp_packet_count;
4412 stats->n_bytes -= subfacet->dp_byte_count;
4415 subfacet->dp_packet_count = 0;
4416 subfacet->dp_byte_count = 0;
4419 /* Updates 'subfacet''s used time. The caller is responsible for calling
4420 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
4422 subfacet_update_time(struct subfacet *subfacet, long long int used)
4424 if (used > subfacet->used) {
4425 subfacet->used = used;
4426 facet_update_time(subfacet->facet, used);
4430 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4432 * Because of the meaning of a subfacet's counters, it only makes sense to do
4433 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4434 * represents a packet that was sent by hand or if it represents statistics
4435 * that have been cleared out of the datapath. */
4437 subfacet_update_stats(struct subfacet *subfacet,
4438 const struct dpif_flow_stats *stats)
4440 if (stats->n_packets || stats->used > subfacet->used) {
4441 struct facet *facet = subfacet->facet;
4443 subfacet_update_time(subfacet, stats->used);
4444 facet->packet_count += stats->n_packets;
4445 facet->byte_count += stats->n_bytes;
4446 facet->tcp_flags |= stats->tcp_flags;
4447 facet_push_stats(facet);
4448 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
4454 static struct rule_dpif *
4455 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
4457 struct ofport_dpif *port;
4458 struct rule_dpif *rule;
4460 rule = rule_dpif_lookup__(ofproto, flow, 0);
4465 port = get_ofp_port(ofproto, flow->in_port);
4467 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
4468 return ofproto->miss_rule;
4471 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
4472 return ofproto->no_packet_in_rule;
4474 return ofproto->miss_rule;
4477 static struct rule_dpif *
4478 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
4481 struct cls_rule *cls_rule;
4482 struct classifier *cls;
4484 if (table_id >= N_TABLES) {
4488 cls = &ofproto->up.tables[table_id].cls;
4489 if (flow->nw_frag & FLOW_NW_FRAG_ANY
4490 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4491 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
4492 * are unavailable. */
4493 struct flow ofpc_normal_flow = *flow;
4494 ofpc_normal_flow.tp_src = htons(0);
4495 ofpc_normal_flow.tp_dst = htons(0);
4496 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
4498 cls_rule = classifier_lookup(cls, flow);
4500 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
4504 complete_operation(struct rule_dpif *rule)
4506 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4508 rule_invalidate(rule);
4510 struct dpif_completion *c = xmalloc(sizeof *c);
4511 c->op = rule->up.pending;
4512 list_push_back(&ofproto->completions, &c->list_node);
4514 ofoperation_complete(rule->up.pending, 0);
4518 static struct rule *
4521 struct rule_dpif *rule = xmalloc(sizeof *rule);
4526 rule_dealloc(struct rule *rule_)
4528 struct rule_dpif *rule = rule_dpif_cast(rule_);
4533 rule_construct(struct rule *rule_)
4535 struct rule_dpif *rule = rule_dpif_cast(rule_);
4536 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4537 struct rule_dpif *victim;
4541 error = validate_actions(rule->up.actions, rule->up.n_actions,
4542 &rule->up.cr.flow, ofproto->max_ports);
4547 rule->packet_count = 0;
4548 rule->byte_count = 0;
4550 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
4551 if (victim && !list_is_empty(&victim->facets)) {
4552 struct facet *facet;
4554 rule->facets = victim->facets;
4555 list_moved(&rule->facets);
4556 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4557 /* XXX: We're only clearing our local counters here. It's possible
4558 * that quite a few packets are unaccounted for in the datapath
4559 * statistics. These will be accounted to the new rule instead of
4560 * cleared as required. This could be fixed by clearing out the
4561 * datapath statistics for this facet, but currently it doesn't
4563 facet_reset_counters(facet);
4567 /* Must avoid list_moved() in this case. */
4568 list_init(&rule->facets);
4571 table_id = rule->up.table_id;
4572 rule->tag = (victim ? victim->tag
4574 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
4575 ofproto->tables[table_id].basis));
4577 complete_operation(rule);
4582 rule_destruct(struct rule *rule_)
4584 struct rule_dpif *rule = rule_dpif_cast(rule_);
4585 struct facet *facet, *next_facet;
4587 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4588 facet_revalidate(facet);
4591 complete_operation(rule);
4595 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4597 struct rule_dpif *rule = rule_dpif_cast(rule_);
4598 struct facet *facet;
4600 /* Start from historical data for 'rule' itself that are no longer tracked
4601 * in facets. This counts, for example, facets that have expired. */
4602 *packets = rule->packet_count;
4603 *bytes = rule->byte_count;
4605 /* Add any statistics that are tracked by facets. This includes
4606 * statistical data recently updated by ofproto_update_stats() as well as
4607 * stats for packets that were executed "by hand" via dpif_execute(). */
4608 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4609 *packets += facet->packet_count;
4610 *bytes += facet->byte_count;
4615 rule_execute(struct rule *rule_, const struct flow *flow,
4616 struct ofpbuf *packet)
4618 struct rule_dpif *rule = rule_dpif_cast(rule_);
4619 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4621 struct dpif_flow_stats stats;
4623 struct action_xlate_ctx ctx;
4624 uint64_t odp_actions_stub[1024 / 8];
4625 struct ofpbuf odp_actions;
4627 dpif_flow_stats_extract(flow, packet, &stats);
4628 rule_credit_stats(rule, &stats);
4630 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4631 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
4632 rule, stats.tcp_flags, packet);
4633 ctx.resubmit_stats = &stats;
4634 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions, &odp_actions);
4636 execute_odp_actions(ofproto, flow, odp_actions.data,
4637 odp_actions.size, packet);
4639 ofpbuf_uninit(&odp_actions);
4645 rule_modify_actions(struct rule *rule_)
4647 struct rule_dpif *rule = rule_dpif_cast(rule_);
4648 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4651 error = validate_actions(rule->up.actions, rule->up.n_actions,
4652 &rule->up.cr.flow, ofproto->max_ports);
4654 ofoperation_complete(rule->up.pending, error);
4658 complete_operation(rule);
4661 /* Sends 'packet' out 'ofport'.
4662 * May modify 'packet'.
4663 * Returns 0 if successful, otherwise a positive errno value. */
4665 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4667 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4668 struct ofpbuf key, odp_actions;
4669 struct odputil_keybuf keybuf;
4674 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
4675 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4677 if (odp_port != ofport->odp_port) {
4678 eth_pop_vlan(packet);
4679 flow.vlan_tci = htons(0);
4682 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4683 odp_flow_key_from_flow(&key, &flow);
4685 ofpbuf_init(&odp_actions, 32);
4686 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4688 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4689 error = dpif_execute(ofproto->dpif,
4691 odp_actions.data, odp_actions.size,
4693 ofpbuf_uninit(&odp_actions);
4696 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4697 ofproto->up.name, odp_port, strerror(error));
4699 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
4703 /* OpenFlow to datapath action translation. */
4705 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
4706 struct action_xlate_ctx *ctx);
4707 static void xlate_normal(struct action_xlate_ctx *);
4709 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
4710 * The action will state 'slow' as the reason that the action is in the slow
4711 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
4712 * dump-flows" output to see why a flow is in the slow path.)
4714 * The 'stub_size' bytes in 'stub' will be used to store the action.
4715 * 'stub_size' must be large enough for the action.
4717 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
4720 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
4721 enum slow_path_reason slow,
4722 uint64_t *stub, size_t stub_size,
4723 const struct nlattr **actionsp, size_t *actions_lenp)
4725 union user_action_cookie cookie;
4728 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
4729 cookie.slow_path.unused = 0;
4730 cookie.slow_path.reason = slow;
4732 ofpbuf_use_stack(&buf, stub, stub_size);
4733 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
4734 uint32_t pid = dpif_port_get_pid(ofproto->dpif, UINT16_MAX);
4735 odp_put_userspace_action(pid, &cookie, &buf);
4737 put_userspace_action(ofproto, &buf, flow, &cookie);
4739 *actionsp = buf.data;
4740 *actions_lenp = buf.size;
4744 put_userspace_action(const struct ofproto_dpif *ofproto,
4745 struct ofpbuf *odp_actions,
4746 const struct flow *flow,
4747 const union user_action_cookie *cookie)
4751 pid = dpif_port_get_pid(ofproto->dpif,
4752 ofp_port_to_odp_port(flow->in_port));
4754 return odp_put_userspace_action(pid, cookie, odp_actions);
4758 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
4759 ovs_be16 vlan_tci, uint32_t odp_port,
4760 unsigned int n_outputs, union user_action_cookie *cookie)
4764 cookie->type = USER_ACTION_COOKIE_SFLOW;
4765 cookie->sflow.vlan_tci = vlan_tci;
4767 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
4768 * port information") for the interpretation of cookie->output. */
4769 switch (n_outputs) {
4771 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
4772 cookie->sflow.output = 0x40000000 | 256;
4776 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4778 cookie->sflow.output = ifindex;
4783 /* 0x80000000 means "multiple output ports. */
4784 cookie->sflow.output = 0x80000000 | n_outputs;
4789 /* Compose SAMPLE action for sFlow. */
4791 compose_sflow_action(const struct ofproto_dpif *ofproto,
4792 struct ofpbuf *odp_actions,
4793 const struct flow *flow,
4796 uint32_t probability;
4797 union user_action_cookie cookie;
4798 size_t sample_offset, actions_offset;
4801 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4805 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4807 /* Number of packets out of UINT_MAX to sample. */
4808 probability = dpif_sflow_get_probability(ofproto->sflow);
4809 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4811 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4812 compose_sflow_cookie(ofproto, htons(0), odp_port,
4813 odp_port == OVSP_NONE ? 0 : 1, &cookie);
4814 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4816 nl_msg_end_nested(odp_actions, actions_offset);
4817 nl_msg_end_nested(odp_actions, sample_offset);
4818 return cookie_offset;
4821 /* SAMPLE action must be first action in any given list of actions.
4822 * At this point we do not have all information required to build it. So try to
4823 * build sample action as complete as possible. */
4825 add_sflow_action(struct action_xlate_ctx *ctx)
4827 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4829 &ctx->flow, OVSP_NONE);
4830 ctx->sflow_odp_port = 0;
4831 ctx->sflow_n_outputs = 0;
4834 /* Fix SAMPLE action according to data collected while composing ODP actions.
4835 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4836 * USERSPACE action's user-cookie which is required for sflow. */
4838 fix_sflow_action(struct action_xlate_ctx *ctx)
4840 const struct flow *base = &ctx->base_flow;
4841 union user_action_cookie *cookie;
4843 if (!ctx->user_cookie_offset) {
4847 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4849 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4851 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
4852 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
4856 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4859 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4860 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4861 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4862 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4866 struct priority_to_dscp *pdscp;
4868 if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD
4869 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4873 pdscp = get_priority(ofport, ctx->flow.skb_priority);
4875 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4876 ctx->flow.nw_tos |= pdscp->dscp;
4879 /* We may not have an ofport record for this port, but it doesn't hurt
4880 * to allow forwarding to it anyhow. Maybe such a port will appear
4881 * later and we're pre-populating the flow table. */
4884 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4885 ctx->flow.vlan_tci);
4886 if (out_port != odp_port) {
4887 ctx->flow.vlan_tci = htons(0);
4889 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4890 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4892 ctx->sflow_odp_port = odp_port;
4893 ctx->sflow_n_outputs++;
4894 ctx->nf_output_iface = ofp_port;
4895 ctx->flow.vlan_tci = flow_vlan_tci;
4896 ctx->flow.nw_tos = flow_nw_tos;
4900 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4902 compose_output_action__(ctx, ofp_port, true);
4906 xlate_table_action(struct action_xlate_ctx *ctx,
4907 uint16_t in_port, uint8_t table_id)
4909 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4910 struct ofproto_dpif *ofproto = ctx->ofproto;
4911 struct rule_dpif *rule;
4912 uint16_t old_in_port;
4913 uint8_t old_table_id;
4915 old_table_id = ctx->table_id;
4916 ctx->table_id = table_id;
4918 /* Look up a flow with 'in_port' as the input port. */
4919 old_in_port = ctx->flow.in_port;
4920 ctx->flow.in_port = in_port;
4921 rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id);
4924 if (table_id > 0 && table_id < N_TABLES) {
4925 struct table_dpif *table = &ofproto->tables[table_id];
4926 if (table->other_table) {
4927 ctx->tags |= (rule && rule->tag
4929 : rule_calculate_tag(&ctx->flow,
4930 &table->other_table->wc,
4935 /* Restore the original input port. Otherwise OFPP_NORMAL and
4936 * OFPP_IN_PORT will have surprising behavior. */
4937 ctx->flow.in_port = old_in_port;
4939 if (ctx->resubmit_hook) {
4940 ctx->resubmit_hook(ctx, rule);
4944 struct rule_dpif *old_rule = ctx->rule;
4946 if (ctx->resubmit_stats) {
4947 rule_credit_stats(rule, ctx->resubmit_stats);
4952 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
4953 ctx->rule = old_rule;
4957 ctx->table_id = old_table_id;
4959 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4961 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
4962 MAX_RESUBMIT_RECURSION);
4963 ctx->max_resubmit_trigger = true;
4968 xlate_resubmit_table(struct action_xlate_ctx *ctx,
4969 const struct nx_action_resubmit *nar)
4974 in_port = (nar->in_port == htons(OFPP_IN_PORT)
4976 : ntohs(nar->in_port));
4977 table_id = nar->table == 255 ? ctx->table_id : nar->table;
4979 xlate_table_action(ctx, in_port, table_id);
4983 flood_packets(struct action_xlate_ctx *ctx, bool all)
4985 struct ofport_dpif *ofport;
4987 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
4988 uint16_t ofp_port = ofport->up.ofp_port;
4990 if (ofp_port == ctx->flow.in_port) {
4995 compose_output_action__(ctx, ofp_port, false);
4996 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
4997 compose_output_action(ctx, ofp_port);
5001 ctx->nf_output_iface = NF_OUT_FLOOD;
5005 execute_controller_action(struct action_xlate_ctx *ctx, int len,
5006 enum ofp_packet_in_reason reason,
5007 uint16_t controller_id)
5009 struct ofputil_packet_in pin;
5010 struct ofpbuf *packet;
5012 ctx->slow |= SLOW_CONTROLLER;
5017 packet = ofpbuf_clone(ctx->packet);
5019 if (packet->l2 && packet->l3) {
5020 struct eth_header *eh;
5022 eth_pop_vlan(packet);
5025 /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
5026 * LLC frame. Calculating the Ethernet type of these frames is more
5027 * trouble than seems appropriate for a simple assertion. */
5028 assert(ntohs(eh->eth_type) < ETH_TYPE_MIN
5029 || eh->eth_type == ctx->flow.dl_type);
5031 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
5032 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
5034 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
5035 eth_push_vlan(packet, ctx->flow.vlan_tci);
5039 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5040 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
5041 ctx->flow.nw_tos, ctx->flow.nw_ttl);
5045 if (ctx->flow.nw_proto == IPPROTO_TCP) {
5046 packet_set_tcp_port(packet, ctx->flow.tp_src,
5048 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
5049 packet_set_udp_port(packet, ctx->flow.tp_src,
5056 pin.packet = packet->data;
5057 pin.packet_len = packet->size;
5058 pin.reason = reason;
5059 pin.controller_id = controller_id;
5060 pin.table_id = ctx->table_id;
5061 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
5064 flow_get_metadata(&ctx->flow, &pin.fmd);
5066 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
5067 ofpbuf_delete(packet);
5071 compose_dec_ttl(struct action_xlate_ctx *ctx)
5073 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
5074 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
5078 if (ctx->flow.nw_ttl > 1) {
5082 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
5084 /* Stop processing for current table. */
5090 xlate_output_action__(struct action_xlate_ctx *ctx,
5091 uint16_t port, uint16_t max_len)
5093 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
5095 ctx->nf_output_iface = NF_OUT_DROP;
5099 compose_output_action(ctx, ctx->flow.in_port);
5102 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
5108 flood_packets(ctx, false);
5111 flood_packets(ctx, true);
5113 case OFPP_CONTROLLER:
5114 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
5120 if (port != ctx->flow.in_port) {
5121 compose_output_action(ctx, port);
5126 if (prev_nf_output_iface == NF_OUT_FLOOD) {
5127 ctx->nf_output_iface = NF_OUT_FLOOD;
5128 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
5129 ctx->nf_output_iface = prev_nf_output_iface;
5130 } else if (prev_nf_output_iface != NF_OUT_DROP &&
5131 ctx->nf_output_iface != NF_OUT_FLOOD) {
5132 ctx->nf_output_iface = NF_OUT_MULTI;
5137 xlate_output_reg_action(struct action_xlate_ctx *ctx,
5138 const struct nx_action_output_reg *naor)
5140 struct mf_subfield src;
5143 nxm_decode(&src, naor->src, naor->ofs_nbits);
5144 ofp_port = mf_get_subfield(&src, &ctx->flow);
5146 if (ofp_port <= UINT16_MAX) {
5147 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
5152 xlate_output_action(struct action_xlate_ctx *ctx,
5153 const struct ofp_action_output *oao)
5155 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
5159 xlate_enqueue_action(struct action_xlate_ctx *ctx,
5160 const struct ofp_action_enqueue *oae)
5163 uint32_t flow_priority, priority;
5166 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
5169 /* Fall back to ordinary output action. */
5170 xlate_output_action__(ctx, ntohs(oae->port), 0);
5174 /* Figure out datapath output port. */
5175 ofp_port = ntohs(oae->port);
5176 if (ofp_port == OFPP_IN_PORT) {
5177 ofp_port = ctx->flow.in_port;
5178 } else if (ofp_port == ctx->flow.in_port) {
5182 /* Add datapath actions. */
5183 flow_priority = ctx->flow.skb_priority;
5184 ctx->flow.skb_priority = priority;
5185 compose_output_action(ctx, ofp_port);
5186 ctx->flow.skb_priority = flow_priority;
5188 /* Update NetFlow output port. */
5189 if (ctx->nf_output_iface == NF_OUT_DROP) {
5190 ctx->nf_output_iface = ofp_port;
5191 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
5192 ctx->nf_output_iface = NF_OUT_MULTI;
5197 xlate_set_queue_action(struct action_xlate_ctx *ctx,
5198 const struct nx_action_set_queue *nasq)
5203 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
5206 /* Couldn't translate queue to a priority, so ignore. A warning
5207 * has already been logged. */
5211 ctx->flow.skb_priority = priority;
5214 struct xlate_reg_state {
5220 xlate_autopath(struct action_xlate_ctx *ctx,
5221 const struct nx_action_autopath *naa)
5223 uint16_t ofp_port = ntohl(naa->id);
5224 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
5226 if (!port || !port->bundle) {
5227 ofp_port = OFPP_NONE;
5228 } else if (port->bundle->bond) {
5229 /* Autopath does not support VLAN hashing. */
5230 struct ofport_dpif *slave = bond_choose_output_slave(
5231 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
5233 ofp_port = slave->up.ofp_port;
5236 autopath_execute(naa, &ctx->flow, ofp_port);
5240 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
5242 struct ofproto_dpif *ofproto = ofproto_;
5243 struct ofport_dpif *port;
5253 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
5256 port = get_ofp_port(ofproto, ofp_port);
5257 return port ? port->may_enable : false;
5262 xlate_learn_action(struct action_xlate_ctx *ctx,
5263 const struct nx_action_learn *learn)
5265 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
5266 struct ofputil_flow_mod fm;
5269 learn_execute(learn, &ctx->flow, &fm);
5271 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
5272 if (error && !VLOG_DROP_WARN(&rl)) {
5273 VLOG_WARN("learning action failed to modify flow table (%s)",
5274 ofperr_get_name(error));
5280 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
5281 * means "infinite". */
5283 reduce_timeout(uint16_t max, uint16_t *timeout)
5285 if (max && (!*timeout || *timeout > max)) {
5291 xlate_fin_timeout(struct action_xlate_ctx *ctx,
5292 const struct nx_action_fin_timeout *naft)
5294 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
5295 struct rule_dpif *rule = ctx->rule;
5297 reduce_timeout(ntohs(naft->fin_idle_timeout), &rule->up.idle_timeout);
5298 reduce_timeout(ntohs(naft->fin_hard_timeout), &rule->up.hard_timeout);
5303 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
5305 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
5306 ? OFPUTIL_PC_NO_RECV_STP
5307 : OFPUTIL_PC_NO_RECV)) {
5311 /* Only drop packets here if both forwarding and learning are
5312 * disabled. If just learning is enabled, we need to have
5313 * OFPP_NORMAL and the learning action have a look at the packet
5314 * before we can drop it. */
5315 if (!stp_forward_in_state(port->stp_state)
5316 && !stp_learn_in_state(port->stp_state)) {
5324 do_xlate_actions(const union ofp_action *in, size_t n_in,
5325 struct action_xlate_ctx *ctx)
5327 const struct ofport_dpif *port;
5328 const union ofp_action *ia;
5329 bool was_evictable = true;
5332 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5333 if (port && !may_receive(port, ctx)) {
5334 /* Drop this flow. */
5339 /* Don't let the rule we're working on get evicted underneath us. */
5340 was_evictable = ctx->rule->up.evictable;
5341 ctx->rule->up.evictable = false;
5343 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
5344 const struct ofp_action_dl_addr *oada;
5345 const struct nx_action_resubmit *nar;
5346 const struct nx_action_set_tunnel *nast;
5347 const struct nx_action_set_queue *nasq;
5348 const struct nx_action_multipath *nam;
5349 const struct nx_action_autopath *naa;
5350 const struct nx_action_bundle *nab;
5351 const struct nx_action_output_reg *naor;
5352 const struct nx_action_controller *nac;
5353 enum ofputil_action_code code;
5360 code = ofputil_decode_action_unsafe(ia);
5362 case OFPUTIL_OFPAT10_OUTPUT:
5363 xlate_output_action(ctx, &ia->output);
5366 case OFPUTIL_OFPAT10_SET_VLAN_VID:
5367 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
5368 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
5371 case OFPUTIL_OFPAT10_SET_VLAN_PCP:
5372 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
5373 ctx->flow.vlan_tci |= htons(
5374 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
5377 case OFPUTIL_OFPAT10_STRIP_VLAN:
5378 ctx->flow.vlan_tci = htons(0);
5381 case OFPUTIL_OFPAT10_SET_DL_SRC:
5382 oada = ((struct ofp_action_dl_addr *) ia);
5383 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
5386 case OFPUTIL_OFPAT10_SET_DL_DST:
5387 oada = ((struct ofp_action_dl_addr *) ia);
5388 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
5391 case OFPUTIL_OFPAT10_SET_NW_SRC:
5392 ctx->flow.nw_src = ia->nw_addr.nw_addr;
5395 case OFPUTIL_OFPAT10_SET_NW_DST:
5396 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
5399 case OFPUTIL_OFPAT10_SET_NW_TOS:
5400 /* OpenFlow 1.0 only supports IPv4. */
5401 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5402 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5403 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
5407 case OFPUTIL_OFPAT10_SET_TP_SRC:
5408 ctx->flow.tp_src = ia->tp_port.tp_port;
5411 case OFPUTIL_OFPAT10_SET_TP_DST:
5412 ctx->flow.tp_dst = ia->tp_port.tp_port;
5415 case OFPUTIL_OFPAT10_ENQUEUE:
5416 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
5419 case OFPUTIL_NXAST_RESUBMIT:
5420 nar = (const struct nx_action_resubmit *) ia;
5421 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
5424 case OFPUTIL_NXAST_RESUBMIT_TABLE:
5425 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
5428 case OFPUTIL_NXAST_SET_TUNNEL:
5429 nast = (const struct nx_action_set_tunnel *) ia;
5430 tun_id = htonll(ntohl(nast->tun_id));
5431 ctx->flow.tun_id = tun_id;
5434 case OFPUTIL_NXAST_SET_QUEUE:
5435 nasq = (const struct nx_action_set_queue *) ia;
5436 xlate_set_queue_action(ctx, nasq);
5439 case OFPUTIL_NXAST_POP_QUEUE:
5440 ctx->flow.skb_priority = ctx->orig_skb_priority;
5443 case OFPUTIL_NXAST_REG_MOVE:
5444 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
5448 case OFPUTIL_NXAST_REG_LOAD:
5449 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
5453 case OFPUTIL_NXAST_NOTE:
5454 /* Nothing to do. */
5457 case OFPUTIL_NXAST_SET_TUNNEL64:
5458 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
5459 ctx->flow.tun_id = tun_id;
5462 case OFPUTIL_NXAST_MULTIPATH:
5463 nam = (const struct nx_action_multipath *) ia;
5464 multipath_execute(nam, &ctx->flow);
5467 case OFPUTIL_NXAST_AUTOPATH:
5468 naa = (const struct nx_action_autopath *) ia;
5469 xlate_autopath(ctx, naa);
5472 case OFPUTIL_NXAST_BUNDLE:
5473 ctx->ofproto->has_bundle_action = true;
5474 nab = (const struct nx_action_bundle *) ia;
5475 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
5480 case OFPUTIL_NXAST_BUNDLE_LOAD:
5481 ctx->ofproto->has_bundle_action = true;
5482 nab = (const struct nx_action_bundle *) ia;
5483 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
5487 case OFPUTIL_NXAST_OUTPUT_REG:
5488 naor = (const struct nx_action_output_reg *) ia;
5489 xlate_output_reg_action(ctx, naor);
5492 case OFPUTIL_NXAST_LEARN:
5493 ctx->has_learn = true;
5494 if (ctx->may_learn) {
5495 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
5499 case OFPUTIL_NXAST_DEC_TTL:
5500 if (compose_dec_ttl(ctx)) {
5505 case OFPUTIL_NXAST_EXIT:
5509 case OFPUTIL_NXAST_FIN_TIMEOUT:
5510 ctx->has_fin_timeout = true;
5511 xlate_fin_timeout(ctx, (const struct nx_action_fin_timeout *) ia);
5514 case OFPUTIL_NXAST_CONTROLLER:
5515 nac = (const struct nx_action_controller *) ia;
5516 execute_controller_action(ctx, ntohs(nac->max_len), nac->reason,
5517 ntohs(nac->controller_id));
5523 /* We've let OFPP_NORMAL and the learning action look at the packet,
5524 * so drop it now if forwarding is disabled. */
5525 if (port && !stp_forward_in_state(port->stp_state)) {
5526 ofpbuf_clear(ctx->odp_actions);
5527 add_sflow_action(ctx);
5530 ctx->rule->up.evictable = was_evictable;
5535 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
5536 struct ofproto_dpif *ofproto, const struct flow *flow,
5537 ovs_be16 initial_tci, struct rule_dpif *rule,
5538 uint8_t tcp_flags, const struct ofpbuf *packet)
5540 ctx->ofproto = ofproto;
5542 ctx->base_flow = ctx->flow;
5543 ctx->base_flow.tun_id = 0;
5544 ctx->base_flow.vlan_tci = initial_tci;
5546 ctx->packet = packet;
5547 ctx->may_learn = packet != NULL;
5548 ctx->tcp_flags = tcp_flags;
5549 ctx->resubmit_hook = NULL;
5550 ctx->resubmit_stats = NULL;
5553 /* Translates the 'n_in' "union ofp_action"s in 'in' into datapath actions in
5554 * 'odp_actions', using 'ctx'. */
5556 xlate_actions(struct action_xlate_ctx *ctx,
5557 const union ofp_action *in, size_t n_in,
5558 struct ofpbuf *odp_actions)
5560 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
5561 * that in the future we always keep a copy of the original flow for
5562 * tracing purposes. */
5563 static bool hit_resubmit_limit;
5565 enum slow_path_reason special;
5567 COVERAGE_INC(ofproto_dpif_xlate);
5569 ofpbuf_clear(odp_actions);
5570 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
5572 ctx->odp_actions = odp_actions;
5575 ctx->has_learn = false;
5576 ctx->has_normal = false;
5577 ctx->has_fin_timeout = false;
5578 ctx->nf_output_iface = NF_OUT_DROP;
5581 ctx->max_resubmit_trigger = false;
5582 ctx->orig_skb_priority = ctx->flow.skb_priority;
5586 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
5587 /* Do this conditionally because the copy is expensive enough that it
5588 * shows up in profiles.
5590 * We keep orig_flow in 'ctx' only because I couldn't make GCC 4.4
5591 * believe that I wasn't using it without initializing it if I kept it
5592 * in a local variable. */
5593 ctx->orig_flow = ctx->flow;
5596 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
5597 switch (ctx->ofproto->up.frag_handling) {
5598 case OFPC_FRAG_NORMAL:
5599 /* We must pretend that transport ports are unavailable. */
5600 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
5601 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
5604 case OFPC_FRAG_DROP:
5607 case OFPC_FRAG_REASM:
5610 case OFPC_FRAG_NX_MATCH:
5611 /* Nothing to do. */
5614 case OFPC_INVALID_TTL_TO_CONTROLLER:
5619 special = process_special(ctx->ofproto, &ctx->flow, ctx->packet);
5621 ctx->slow |= special;
5623 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5624 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
5626 add_sflow_action(ctx);
5627 do_xlate_actions(in, n_in, ctx);
5629 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
5630 if (!hit_resubmit_limit) {
5631 /* We didn't record the original flow. Make sure we do from
5633 hit_resubmit_limit = true;
5634 } else if (!VLOG_DROP_ERR(&trace_rl)) {
5635 struct ds ds = DS_EMPTY_INITIALIZER;
5637 ofproto_trace(ctx->ofproto, &ctx->orig_flow, ctx->packet,
5639 VLOG_ERR("Trace triggered by excessive resubmit "
5640 "recursion:\n%s", ds_cstr(&ds));
5645 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
5646 ctx->odp_actions->data,
5647 ctx->odp_actions->size)) {
5648 ctx->slow |= SLOW_IN_BAND;
5650 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
5652 compose_output_action(ctx, OFPP_LOCAL);
5655 if (ctx->ofproto->has_mirrors) {
5656 add_mirror_actions(ctx, &ctx->orig_flow);
5658 fix_sflow_action(ctx);
5662 /* Translates the 'n_in' "union ofp_action"s in 'in' into datapath actions,
5663 * using 'ctx', and discards the datapath actions. */
5665 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
5666 const union ofp_action *in, size_t n_in)
5668 uint64_t odp_actions_stub[1024 / 8];
5669 struct ofpbuf odp_actions;
5671 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5672 xlate_actions(ctx, in, n_in, &odp_actions);
5673 ofpbuf_uninit(&odp_actions);
5676 /* OFPP_NORMAL implementation. */
5678 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
5680 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
5681 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
5682 * the bundle on which the packet was received, returns the VLAN to which the
5685 * Both 'vid' and the return value are in the range 0...4095. */
5687 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
5689 switch (in_bundle->vlan_mode) {
5690 case PORT_VLAN_ACCESS:
5691 return in_bundle->vlan;
5694 case PORT_VLAN_TRUNK:
5697 case PORT_VLAN_NATIVE_UNTAGGED:
5698 case PORT_VLAN_NATIVE_TAGGED:
5699 return vid ? vid : in_bundle->vlan;
5706 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
5707 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
5710 * 'vid' should be the VID obtained from the 802.1Q header that was received as
5711 * part of a packet (specify 0 if there was no 802.1Q header), in the range
5714 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
5716 /* Allow any VID on the OFPP_NONE port. */
5717 if (in_bundle == &ofpp_none_bundle) {
5721 switch (in_bundle->vlan_mode) {
5722 case PORT_VLAN_ACCESS:
5725 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5726 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
5727 "packet received on port %s configured as VLAN "
5728 "%"PRIu16" access port",
5729 in_bundle->ofproto->up.name, vid,
5730 in_bundle->name, in_bundle->vlan);
5736 case PORT_VLAN_NATIVE_UNTAGGED:
5737 case PORT_VLAN_NATIVE_TAGGED:
5739 /* Port must always carry its native VLAN. */
5743 case PORT_VLAN_TRUNK:
5744 if (!ofbundle_includes_vlan(in_bundle, vid)) {
5746 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5747 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
5748 "received on port %s not configured for trunking "
5750 in_bundle->ofproto->up.name, vid,
5751 in_bundle->name, vid);
5763 /* Given 'vlan', the VLAN that a packet belongs to, and
5764 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
5765 * that should be included in the 802.1Q header. (If the return value is 0,
5766 * then the 802.1Q header should only be included in the packet if there is a
5769 * Both 'vlan' and the return value are in the range 0...4095. */
5771 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
5773 switch (out_bundle->vlan_mode) {
5774 case PORT_VLAN_ACCESS:
5777 case PORT_VLAN_TRUNK:
5778 case PORT_VLAN_NATIVE_TAGGED:
5781 case PORT_VLAN_NATIVE_UNTAGGED:
5782 return vlan == out_bundle->vlan ? 0 : vlan;
5790 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
5793 struct ofport_dpif *port;
5795 ovs_be16 tci, old_tci;
5797 vid = output_vlan_to_vid(out_bundle, vlan);
5798 if (!out_bundle->bond) {
5799 port = ofbundle_get_a_port(out_bundle);
5801 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
5804 /* No slaves enabled, so drop packet. */
5809 old_tci = ctx->flow.vlan_tci;
5811 if (tci || out_bundle->use_priority_tags) {
5812 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
5814 tci |= htons(VLAN_CFI);
5817 ctx->flow.vlan_tci = tci;
5819 compose_output_action(ctx, port->up.ofp_port);
5820 ctx->flow.vlan_tci = old_tci;
5824 mirror_mask_ffs(mirror_mask_t mask)
5826 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
5831 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
5833 return (bundle->vlan_mode != PORT_VLAN_ACCESS
5834 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
5838 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
5840 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
5843 /* Returns an arbitrary interface within 'bundle'. */
5844 static struct ofport_dpif *
5845 ofbundle_get_a_port(const struct ofbundle *bundle)
5847 return CONTAINER_OF(list_front(&bundle->ports),
5848 struct ofport_dpif, bundle_node);
5852 vlan_is_mirrored(const struct ofmirror *m, int vlan)
5854 return !m->vlans || bitmap_is_set(m->vlans, vlan);
5858 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5860 struct ofproto_dpif *ofproto = ctx->ofproto;
5861 mirror_mask_t mirrors;
5862 struct ofbundle *in_bundle;
5865 const struct nlattr *a;
5868 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5869 ctx->packet != NULL, NULL);
5873 mirrors = in_bundle->src_mirrors;
5875 /* Drop frames on bundles reserved for mirroring. */
5876 if (in_bundle->mirror_out) {
5877 if (ctx->packet != NULL) {
5878 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5879 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5880 "%s, which is reserved exclusively for mirroring",
5881 ctx->ofproto->up.name, in_bundle->name);
5887 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5888 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5891 vlan = input_vid_to_vlan(in_bundle, vid);
5893 /* Look at the output ports to check for destination selections. */
5895 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5896 ctx->odp_actions->size) {
5897 enum ovs_action_attr type = nl_attr_type(a);
5898 struct ofport_dpif *ofport;
5900 if (type != OVS_ACTION_ATTR_OUTPUT) {
5904 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5905 if (ofport && ofport->bundle) {
5906 mirrors |= ofport->bundle->dst_mirrors;
5914 /* Restore the original packet before adding the mirror actions. */
5915 ctx->flow = *orig_flow;
5920 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5922 if (!vlan_is_mirrored(m, vlan)) {
5923 mirrors &= mirrors - 1;
5927 mirrors &= ~m->dup_mirrors;
5928 ctx->mirrors |= m->dup_mirrors;
5930 output_normal(ctx, m->out, vlan);
5931 } else if (vlan != m->out_vlan
5932 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
5933 struct ofbundle *bundle;
5935 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5936 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5937 && !bundle->mirror_out) {
5938 output_normal(ctx, bundle, m->out_vlan);
5946 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5947 uint64_t packets, uint64_t bytes)
5953 for (; mirrors; mirrors &= mirrors - 1) {
5956 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5959 /* In normal circumstances 'm' will not be NULL. However,
5960 * if mirrors are reconfigured, we can temporarily get out
5961 * of sync in facet_revalidate(). We could "correct" the
5962 * mirror list before reaching here, but doing that would
5963 * not properly account the traffic stats we've currently
5964 * accumulated for previous mirror configuration. */
5968 m->packet_count += packets;
5969 m->byte_count += bytes;
5973 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
5974 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
5975 * indicate this; newer upstream kernels use gratuitous ARP requests. */
5977 is_gratuitous_arp(const struct flow *flow)
5979 return (flow->dl_type == htons(ETH_TYPE_ARP)
5980 && eth_addr_is_broadcast(flow->dl_dst)
5981 && (flow->nw_proto == ARP_OP_REPLY
5982 || (flow->nw_proto == ARP_OP_REQUEST
5983 && flow->nw_src == flow->nw_dst)));
5987 update_learning_table(struct ofproto_dpif *ofproto,
5988 const struct flow *flow, int vlan,
5989 struct ofbundle *in_bundle)
5991 struct mac_entry *mac;
5993 /* Don't learn the OFPP_NONE port. */
5994 if (in_bundle == &ofpp_none_bundle) {
5998 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
6002 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
6003 if (is_gratuitous_arp(flow)) {
6004 /* We don't want to learn from gratuitous ARP packets that are
6005 * reflected back over bond slaves so we lock the learning table. */
6006 if (!in_bundle->bond) {
6007 mac_entry_set_grat_arp_lock(mac);
6008 } else if (mac_entry_is_grat_arp_locked(mac)) {
6013 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
6014 /* The log messages here could actually be useful in debugging,
6015 * so keep the rate limit relatively high. */
6016 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
6017 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
6018 "on port %s in VLAN %d",
6019 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
6020 in_bundle->name, vlan);
6022 mac->port.p = in_bundle;
6023 tag_set_add(&ofproto->revalidate_set,
6024 mac_learning_changed(ofproto->ml, mac));
6028 static struct ofbundle *
6029 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
6030 bool warn, struct ofport_dpif **in_ofportp)
6032 struct ofport_dpif *ofport;
6034 /* Find the port and bundle for the received packet. */
6035 ofport = get_ofp_port(ofproto, in_port);
6037 *in_ofportp = ofport;
6039 if (ofport && ofport->bundle) {
6040 return ofport->bundle;
6043 /* Special-case OFPP_NONE, which a controller may use as the ingress
6044 * port for traffic that it is sourcing. */
6045 if (in_port == OFPP_NONE) {
6046 return &ofpp_none_bundle;
6049 /* Odd. A few possible reasons here:
6051 * - We deleted a port but there are still a few packets queued up
6054 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
6055 * we don't know about.
6057 * - The ofproto client didn't configure the port as part of a bundle.
6060 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6062 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
6063 "port %"PRIu16, ofproto->up.name, in_port);
6068 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
6069 * dropped. Returns true if they may be forwarded, false if they should be
6072 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
6073 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
6075 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
6076 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
6077 * checked by input_vid_is_valid().
6079 * May also add tags to '*tags', although the current implementation only does
6080 * so in one special case.
6083 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
6084 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
6086 struct ofbundle *in_bundle = in_port->bundle;
6088 /* Drop frames for reserved multicast addresses
6089 * only if forward_bpdu option is absent. */
6090 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
6094 if (in_bundle->bond) {
6095 struct mac_entry *mac;
6097 switch (bond_check_admissibility(in_bundle->bond, in_port,
6098 flow->dl_dst, tags)) {
6105 case BV_DROP_IF_MOVED:
6106 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
6107 if (mac && mac->port.p != in_bundle &&
6108 (!is_gratuitous_arp(flow)
6109 || mac_entry_is_grat_arp_locked(mac))) {
6120 xlate_normal(struct action_xlate_ctx *ctx)
6122 struct ofport_dpif *in_port;
6123 struct ofbundle *in_bundle;
6124 struct mac_entry *mac;
6128 ctx->has_normal = true;
6130 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
6131 ctx->packet != NULL, &in_port);
6136 /* Drop malformed frames. */
6137 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
6138 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
6139 if (ctx->packet != NULL) {
6140 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6141 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
6142 "VLAN tag received on port %s",
6143 ctx->ofproto->up.name, in_bundle->name);
6148 /* Drop frames on bundles reserved for mirroring. */
6149 if (in_bundle->mirror_out) {
6150 if (ctx->packet != NULL) {
6151 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6152 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6153 "%s, which is reserved exclusively for mirroring",
6154 ctx->ofproto->up.name, in_bundle->name);
6160 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
6161 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6164 vlan = input_vid_to_vlan(in_bundle, vid);
6166 /* Check other admissibility requirements. */
6168 !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
6172 /* Learn source MAC. */
6173 if (ctx->may_learn) {
6174 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
6177 /* Determine output bundle. */
6178 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
6181 if (mac->port.p != in_bundle) {
6182 output_normal(ctx, mac->port.p, vlan);
6185 struct ofbundle *bundle;
6187 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
6188 if (bundle != in_bundle
6189 && ofbundle_includes_vlan(bundle, vlan)
6190 && bundle->floodable
6191 && !bundle->mirror_out) {
6192 output_normal(ctx, bundle, vlan);
6195 ctx->nf_output_iface = NF_OUT_FLOOD;
6199 /* Optimized flow revalidation.
6201 * It's a difficult problem, in general, to tell which facets need to have
6202 * their actions recalculated whenever the OpenFlow flow table changes. We
6203 * don't try to solve that general problem: for most kinds of OpenFlow flow
6204 * table changes, we recalculate the actions for every facet. This is
6205 * relatively expensive, but it's good enough if the OpenFlow flow table
6206 * doesn't change very often.
6208 * However, we can expect one particular kind of OpenFlow flow table change to
6209 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
6210 * of CPU on revalidating every facet whenever MAC learning modifies the flow
6211 * table, we add a special case that applies to flow tables in which every rule
6212 * has the same form (that is, the same wildcards), except that the table is
6213 * also allowed to have a single "catch-all" flow that matches all packets. We
6214 * optimize this case by tagging all of the facets that resubmit into the table
6215 * and invalidating the same tag whenever a flow changes in that table. The
6216 * end result is that we revalidate just the facets that need it (and sometimes
6217 * a few more, but not all of the facets or even all of the facets that
6218 * resubmit to the table modified by MAC learning). */
6220 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
6221 * into an OpenFlow table with the given 'basis'. */
6223 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
6226 if (flow_wildcards_is_catchall(wc)) {
6229 struct flow tag_flow = *flow;
6230 flow_zero_wildcards(&tag_flow, wc);
6231 return tag_create_deterministic(flow_hash(&tag_flow, secret));
6235 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
6236 * taggability of that table.
6238 * This function must be called after *each* change to a flow table. If you
6239 * skip calling it on some changes then the pointer comparisons at the end can
6240 * be invalid if you get unlucky. For example, if a flow removal causes a
6241 * cls_table to be destroyed and then a flow insertion causes a cls_table with
6242 * different wildcards to be created with the same address, then this function
6243 * will incorrectly skip revalidation. */
6245 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
6247 struct table_dpif *table = &ofproto->tables[table_id];
6248 const struct oftable *oftable = &ofproto->up.tables[table_id];
6249 struct cls_table *catchall, *other;
6250 struct cls_table *t;
6252 catchall = other = NULL;
6254 switch (hmap_count(&oftable->cls.tables)) {
6256 /* We could tag this OpenFlow table but it would make the logic a
6257 * little harder and it's a corner case that doesn't seem worth it
6263 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
6264 if (cls_table_is_catchall(t)) {
6266 } else if (!other) {
6269 /* Indicate that we can't tag this by setting both tables to
6270 * NULL. (We know that 'catchall' is already NULL.) */
6277 /* Can't tag this table. */
6281 if (table->catchall_table != catchall || table->other_table != other) {
6282 table->catchall_table = catchall;
6283 table->other_table = other;
6284 ofproto->need_revalidate = true;
6288 /* Given 'rule' that has changed in some way (either it is a rule being
6289 * inserted, a rule being deleted, or a rule whose actions are being
6290 * modified), marks facets for revalidation to ensure that packets will be
6291 * forwarded correctly according to the new state of the flow table.
6293 * This function must be called after *each* change to a flow table. See
6294 * the comment on table_update_taggable() for more information. */
6296 rule_invalidate(const struct rule_dpif *rule)
6298 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
6300 table_update_taggable(ofproto, rule->up.table_id);
6302 if (!ofproto->need_revalidate) {
6303 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
6305 if (table->other_table && rule->tag) {
6306 tag_set_add(&ofproto->revalidate_set, rule->tag);
6308 ofproto->need_revalidate = true;
6314 set_frag_handling(struct ofproto *ofproto_,
6315 enum ofp_config_flags frag_handling)
6317 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6319 if (frag_handling != OFPC_FRAG_REASM) {
6320 ofproto->need_revalidate = true;
6328 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
6329 const struct flow *flow,
6330 const union ofp_action *ofp_actions, size_t n_ofp_actions)
6332 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6335 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
6336 return OFPERR_NXBRC_BAD_IN_PORT;
6339 error = validate_actions(ofp_actions, n_ofp_actions, flow,
6340 ofproto->max_ports);
6342 struct odputil_keybuf keybuf;
6343 struct dpif_flow_stats stats;
6347 struct action_xlate_ctx ctx;
6348 uint64_t odp_actions_stub[1024 / 8];
6349 struct ofpbuf odp_actions;
6351 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
6352 odp_flow_key_from_flow(&key, flow);
6354 dpif_flow_stats_extract(flow, packet, &stats);
6356 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL,
6357 packet_get_tcp_flags(packet, flow), packet);
6358 ctx.resubmit_stats = &stats;
6360 ofpbuf_use_stub(&odp_actions,
6361 odp_actions_stub, sizeof odp_actions_stub);
6362 xlate_actions(&ctx, ofp_actions, n_ofp_actions, &odp_actions);
6363 dpif_execute(ofproto->dpif, key.data, key.size,
6364 odp_actions.data, odp_actions.size, packet);
6365 ofpbuf_uninit(&odp_actions);
6373 set_netflow(struct ofproto *ofproto_,
6374 const struct netflow_options *netflow_options)
6376 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6378 if (netflow_options) {
6379 if (!ofproto->netflow) {
6380 ofproto->netflow = netflow_create();
6382 return netflow_set_options(ofproto->netflow, netflow_options);
6384 netflow_destroy(ofproto->netflow);
6385 ofproto->netflow = NULL;
6391 get_netflow_ids(const struct ofproto *ofproto_,
6392 uint8_t *engine_type, uint8_t *engine_id)
6394 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6396 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
6400 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
6402 if (!facet_is_controller_flow(facet) &&
6403 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
6404 struct subfacet *subfacet;
6405 struct ofexpired expired;
6407 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
6408 if (subfacet->path == SF_FAST_PATH) {
6409 struct dpif_flow_stats stats;
6411 subfacet_reinstall(subfacet, &stats);
6412 subfacet_update_stats(subfacet, &stats);
6416 expired.flow = facet->flow;
6417 expired.packet_count = facet->packet_count;
6418 expired.byte_count = facet->byte_count;
6419 expired.used = facet->used;
6420 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
6425 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
6427 struct facet *facet;
6429 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6430 send_active_timeout(ofproto, facet);
6434 static struct ofproto_dpif *
6435 ofproto_dpif_lookup(const char *name)
6437 struct ofproto_dpif *ofproto;
6439 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
6440 hash_string(name, 0), &all_ofproto_dpifs) {
6441 if (!strcmp(ofproto->up.name, name)) {
6449 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
6450 const char *argv[], void *aux OVS_UNUSED)
6452 struct ofproto_dpif *ofproto;
6455 ofproto = ofproto_dpif_lookup(argv[1]);
6457 unixctl_command_reply_error(conn, "no such bridge");
6460 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6462 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6463 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6467 unixctl_command_reply(conn, "table successfully flushed");
6471 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6472 const char *argv[], void *aux OVS_UNUSED)
6474 struct ds ds = DS_EMPTY_INITIALIZER;
6475 const struct ofproto_dpif *ofproto;
6476 const struct mac_entry *e;
6478 ofproto = ofproto_dpif_lookup(argv[1]);
6480 unixctl_command_reply_error(conn, "no such bridge");
6484 ds_put_cstr(&ds, " port VLAN MAC Age\n");
6485 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
6486 struct ofbundle *bundle = e->port.p;
6487 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
6488 ofbundle_get_a_port(bundle)->odp_port,
6489 e->vlan, ETH_ADDR_ARGS(e->mac),
6490 mac_entry_age(ofproto->ml, e));
6492 unixctl_command_reply(conn, ds_cstr(&ds));
6497 struct action_xlate_ctx ctx;
6503 trace_format_rule(struct ds *result, uint8_t table_id, int level,
6504 const struct rule_dpif *rule)
6506 ds_put_char_multiple(result, '\t', level);
6508 ds_put_cstr(result, "No match\n");
6512 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
6513 table_id, ntohll(rule->up.flow_cookie));
6514 cls_rule_format(&rule->up.cr, result);
6515 ds_put_char(result, '\n');
6517 ds_put_char_multiple(result, '\t', level);
6518 ds_put_cstr(result, "OpenFlow ");
6519 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
6520 ds_put_char(result, '\n');
6524 trace_format_flow(struct ds *result, int level, const char *title,
6525 struct trace_ctx *trace)
6527 ds_put_char_multiple(result, '\t', level);
6528 ds_put_format(result, "%s: ", title);
6529 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
6530 ds_put_cstr(result, "unchanged");
6532 flow_format(result, &trace->ctx.flow);
6533 trace->flow = trace->ctx.flow;
6535 ds_put_char(result, '\n');
6539 trace_format_regs(struct ds *result, int level, const char *title,
6540 struct trace_ctx *trace)
6544 ds_put_char_multiple(result, '\t', level);
6545 ds_put_format(result, "%s:", title);
6546 for (i = 0; i < FLOW_N_REGS; i++) {
6547 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
6549 ds_put_char(result, '\n');
6553 trace_format_odp(struct ds *result, int level, const char *title,
6554 struct trace_ctx *trace)
6556 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
6558 ds_put_char_multiple(result, '\t', level);
6559 ds_put_format(result, "%s: ", title);
6560 format_odp_actions(result, odp_actions->data, odp_actions->size);
6561 ds_put_char(result, '\n');
6565 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
6567 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
6568 struct ds *result = trace->result;
6570 ds_put_char(result, '\n');
6571 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
6572 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
6573 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
6574 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
6578 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
6579 void *aux OVS_UNUSED)
6581 const char *dpname = argv[1];
6582 struct ofproto_dpif *ofproto;
6583 struct ofpbuf odp_key;
6584 struct ofpbuf *packet;
6585 ovs_be16 initial_tci;
6591 ofpbuf_init(&odp_key, 0);
6594 ofproto = ofproto_dpif_lookup(dpname);
6596 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
6600 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
6601 /* ofproto/trace dpname flow [-generate] */
6602 const char *flow_s = argv[2];
6603 const char *generate_s = argv[3];
6605 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
6606 * flow. We guess which type it is based on whether 'flow_s' contains
6607 * an '(', since a datapath flow always contains '(') but an
6608 * OpenFlow-like flow should not (in fact it's allowed but I believe
6609 * that's not documented anywhere).
6611 * An alternative would be to try to parse 'flow_s' both ways, but then
6612 * it would be tricky giving a sensible error message. After all, do
6613 * you just say "syntax error" or do you present both error messages?
6614 * Both choices seem lousy. */
6615 if (strchr(flow_s, '(')) {
6618 /* Convert string to datapath key. */
6619 ofpbuf_init(&odp_key, 0);
6620 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
6622 unixctl_command_reply_error(conn, "Bad flow syntax");
6626 /* Convert odp_key to flow. */
6627 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
6628 odp_key.size, &flow,
6629 &initial_tci, NULL);
6630 if (error == ODP_FIT_ERROR) {
6631 unixctl_command_reply_error(conn, "Invalid flow");
6637 error_s = parse_ofp_exact_flow(&flow, argv[2]);
6639 unixctl_command_reply_error(conn, error_s);
6644 initial_tci = flow.vlan_tci;
6645 vsp_adjust_flow(ofproto, &flow);
6648 /* Generate a packet, if requested. */
6650 packet = ofpbuf_new(0);
6651 flow_compose(packet, &flow);
6653 } else if (argc == 6) {
6654 /* ofproto/trace dpname priority tun_id in_port packet */
6655 const char *priority_s = argv[2];
6656 const char *tun_id_s = argv[3];
6657 const char *in_port_s = argv[4];
6658 const char *packet_s = argv[5];
6659 uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s));
6660 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
6661 uint32_t priority = atoi(priority_s);
6664 msg = eth_from_hex(packet_s, &packet);
6666 unixctl_command_reply_error(conn, msg);
6670 ds_put_cstr(&result, "Packet: ");
6671 s = ofp_packet_to_string(packet->data, packet->size);
6672 ds_put_cstr(&result, s);
6675 flow_extract(packet, priority, tun_id, in_port, &flow);
6676 initial_tci = flow.vlan_tci;
6678 unixctl_command_reply_error(conn, "Bad command syntax");
6682 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
6683 unixctl_command_reply(conn, ds_cstr(&result));
6686 ds_destroy(&result);
6687 ofpbuf_delete(packet);
6688 ofpbuf_uninit(&odp_key);
6692 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
6693 const struct ofpbuf *packet, ovs_be16 initial_tci,
6696 struct rule_dpif *rule;
6698 ds_put_cstr(ds, "Flow: ");
6699 flow_format(ds, flow);
6700 ds_put_char(ds, '\n');
6702 rule = rule_dpif_lookup(ofproto, flow);
6704 trace_format_rule(ds, 0, 0, rule);
6705 if (rule == ofproto->miss_rule) {
6706 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
6707 } else if (rule == ofproto->no_packet_in_rule) {
6708 ds_put_cstr(ds, "\nNo match, packets dropped because "
6709 "OFPPC_NO_PACKET_IN is set on in_port.\n");
6713 uint64_t odp_actions_stub[1024 / 8];
6714 struct ofpbuf odp_actions;
6716 struct trace_ctx trace;
6719 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
6722 ofpbuf_use_stub(&odp_actions,
6723 odp_actions_stub, sizeof odp_actions_stub);
6724 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
6725 rule, tcp_flags, packet);
6726 trace.ctx.resubmit_hook = trace_resubmit;
6727 xlate_actions(&trace.ctx, rule->up.actions, rule->up.n_actions,
6730 ds_put_char(ds, '\n');
6731 trace_format_flow(ds, 0, "Final flow", &trace);
6732 ds_put_cstr(ds, "Datapath actions: ");
6733 format_odp_actions(ds, odp_actions.data, odp_actions.size);
6734 ofpbuf_uninit(&odp_actions);
6736 if (trace.ctx.slow) {
6737 enum slow_path_reason slow;
6739 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
6740 "slow path because it:");
6741 for (slow = trace.ctx.slow; slow; ) {
6742 enum slow_path_reason bit = rightmost_1bit(slow);
6746 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
6749 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
6752 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
6755 ds_put_cstr(ds, "\n\t- Needs in-band special case "
6758 ds_put_cstr(ds, "\n\t (The datapath actions are "
6759 "incomplete--for complete actions, "
6760 "please supply a packet.)");
6763 case SLOW_CONTROLLER:
6764 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
6765 "to the OpenFlow controller.");
6768 ds_put_cstr(ds, "\n\t- Needs more specific matching "
6769 "than the datapath supports.");
6776 if (slow & ~SLOW_MATCH) {
6777 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
6778 "the special slow-path processing.");
6785 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6786 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6789 unixctl_command_reply(conn, NULL);
6793 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6794 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6797 unixctl_command_reply(conn, NULL);
6800 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
6801 * 'reply' describing the results. */
6803 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
6805 struct facet *facet;
6809 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6810 if (!facet_check_consistency(facet)) {
6815 ofproto->need_revalidate = true;
6819 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
6820 ofproto->up.name, errors);
6822 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
6827 ofproto_dpif_self_check(struct unixctl_conn *conn,
6828 int argc, const char *argv[], void *aux OVS_UNUSED)
6830 struct ds reply = DS_EMPTY_INITIALIZER;
6831 struct ofproto_dpif *ofproto;
6834 ofproto = ofproto_dpif_lookup(argv[1]);
6836 unixctl_command_reply_error(conn, "Unknown ofproto (use "
6837 "ofproto/list for help)");
6840 ofproto_dpif_self_check__(ofproto, &reply);
6842 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6843 ofproto_dpif_self_check__(ofproto, &reply);
6847 unixctl_command_reply(conn, ds_cstr(&reply));
6852 ofproto_dpif_unixctl_init(void)
6854 static bool registered;
6860 unixctl_command_register(
6862 "bridge {tun_id in_port packet | odp_flow [-generate]}",
6863 2, 5, ofproto_unixctl_trace, NULL);
6864 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6865 ofproto_unixctl_fdb_flush, NULL);
6866 unixctl_command_register("fdb/show", "bridge", 1, 1,
6867 ofproto_unixctl_fdb_show, NULL);
6868 unixctl_command_register("ofproto/clog", "", 0, 0,
6869 ofproto_dpif_clog, NULL);
6870 unixctl_command_register("ofproto/unclog", "", 0, 0,
6871 ofproto_dpif_unclog, NULL);
6872 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6873 ofproto_dpif_self_check, NULL);
6876 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6878 * This is deprecated. It is only for compatibility with broken device drivers
6879 * in old versions of Linux that do not properly support VLANs when VLAN
6880 * devices are not used. When broken device drivers are no longer in
6881 * widespread use, we will delete these interfaces. */
6884 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
6886 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6887 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6889 if (realdev_ofp_port == ofport->realdev_ofp_port
6890 && vid == ofport->vlandev_vid) {
6894 ofproto->need_revalidate = true;
6896 if (ofport->realdev_ofp_port) {
6899 if (realdev_ofp_port && ofport->bundle) {
6900 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6901 * themselves be part of a bundle. */
6902 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6905 ofport->realdev_ofp_port = realdev_ofp_port;
6906 ofport->vlandev_vid = vid;
6908 if (realdev_ofp_port) {
6909 vsp_add(ofport, realdev_ofp_port, vid);
6916 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
6918 return hash_2words(realdev_ofp_port, vid);
6921 /* Returns the ODP port number of the Linux VLAN device that corresponds to
6922 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
6923 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
6924 * it would return the port number of eth0.9.
6926 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
6927 * function just returns its 'realdev_odp_port' argument. */
6929 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6930 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
6932 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6933 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
6934 int vid = vlan_tci_to_vid(vlan_tci);
6935 const struct vlan_splinter *vsp;
6937 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6938 hash_realdev_vid(realdev_ofp_port, vid),
6939 &ofproto->realdev_vid_map) {
6940 if (vsp->realdev_ofp_port == realdev_ofp_port
6941 && vsp->vid == vid) {
6942 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
6946 return realdev_odp_port;
6949 static struct vlan_splinter *
6950 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
6952 struct vlan_splinter *vsp;
6954 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
6955 &ofproto->vlandev_map) {
6956 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6964 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6965 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6966 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6967 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6968 * eth0 and store 9 in '*vid'.
6970 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6971 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6974 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6975 uint16_t vlandev_ofp_port, int *vid)
6977 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6978 const struct vlan_splinter *vsp;
6980 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6985 return vsp->realdev_ofp_port;
6991 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6992 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
6993 * 'flow->in_port' to the "real" device backing the VLAN device, sets
6994 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
6995 * always the case unless VLAN splinters are enabled), returns false without
6996 * making any changes. */
6998 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
7003 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
7008 /* Cause the flow to be processed as if it came in on the real device with
7009 * the VLAN device's VLAN ID. */
7010 flow->in_port = realdev;
7011 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
7016 vsp_remove(struct ofport_dpif *port)
7018 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7019 struct vlan_splinter *vsp;
7021 vsp = vlandev_find(ofproto, port->up.ofp_port);
7023 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
7024 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
7027 port->realdev_ofp_port = 0;
7029 VLOG_ERR("missing vlan device record");
7034 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
7036 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7038 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
7039 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
7040 == realdev_ofp_port)) {
7041 struct vlan_splinter *vsp;
7043 vsp = xmalloc(sizeof *vsp);
7044 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
7045 hash_int(port->up.ofp_port, 0));
7046 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
7047 hash_realdev_vid(realdev_ofp_port, vid));
7048 vsp->realdev_ofp_port = realdev_ofp_port;
7049 vsp->vlandev_ofp_port = port->up.ofp_port;
7052 port->realdev_ofp_port = realdev_ofp_port;
7054 VLOG_ERR("duplicate vlan device record");
7058 const struct ofproto_class ofproto_dpif_class = {
7088 port_is_lacp_current,
7089 NULL, /* rule_choose_table */
7096 rule_modify_actions,
7104 get_cfm_remote_mpids,
7109 get_stp_port_status,
7116 is_mirror_output_bundle,
7117 forward_bpdu_changed,