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-actions.h"
46 #include "ofp-parse.h"
47 #include "ofp-print.h"
48 #include "ofproto-dpif-governor.h"
49 #include "ofproto-dpif-sflow.h"
50 #include "poll-loop.h"
54 #include "unaligned.h"
56 #include "vlan-bitmap.h"
59 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
61 COVERAGE_DEFINE(ofproto_dpif_expired);
62 COVERAGE_DEFINE(ofproto_dpif_xlate);
63 COVERAGE_DEFINE(facet_changed_rule);
64 COVERAGE_DEFINE(facet_revalidate);
65 COVERAGE_DEFINE(facet_unexpected);
66 COVERAGE_DEFINE(facet_suppress);
68 /* Maximum depth of flow table recursion (due to resubmit actions) in a
69 * flow translation. */
70 #define MAX_RESUBMIT_RECURSION 64
72 /* Number of implemented OpenFlow tables. */
73 enum { N_TABLES = 255 };
74 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
75 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
86 * - Do include packets and bytes from facets that have been deleted or
87 * whose own statistics have been folded into the rule.
89 * - Do include packets and bytes sent "by hand" that were accounted to
90 * the rule without any facet being involved (this is a rare corner
91 * case in rule_execute()).
93 * - Do not include packet or bytes that can be obtained from any facet's
94 * packet_count or byte_count member or that can be obtained from the
95 * datapath by, e.g., dpif_flow_get() for any subfacet.
97 uint64_t packet_count; /* Number of packets received. */
98 uint64_t byte_count; /* Number of bytes received. */
100 tag_type tag; /* Caches rule_calculate_tag() result. */
102 struct list facets; /* List of "struct facet"s. */
105 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
107 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
110 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
111 const struct flow *);
112 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
115 static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto,
116 const struct flow *flow);
118 static void rule_credit_stats(struct rule_dpif *,
119 const struct dpif_flow_stats *);
120 static void flow_push_stats(struct rule_dpif *, const struct flow *,
121 const struct dpif_flow_stats *);
122 static tag_type rule_calculate_tag(const struct flow *,
123 const struct minimask *, uint32_t basis);
124 static void rule_invalidate(const struct rule_dpif *);
126 #define MAX_MIRRORS 32
127 typedef uint32_t mirror_mask_t;
128 #define MIRROR_MASK_C(X) UINT32_C(X)
129 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
131 struct ofproto_dpif *ofproto; /* Owning ofproto. */
132 size_t idx; /* In ofproto's "mirrors" array. */
133 void *aux; /* Key supplied by ofproto's client. */
134 char *name; /* Identifier for log messages. */
136 /* Selection criteria. */
137 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
138 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
139 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
141 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
142 struct ofbundle *out; /* Output port or NULL. */
143 int out_vlan; /* Output VLAN or -1. */
144 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
147 int64_t packet_count; /* Number of packets sent. */
148 int64_t byte_count; /* Number of bytes sent. */
151 static void mirror_destroy(struct ofmirror *);
152 static void update_mirror_stats(struct ofproto_dpif *ofproto,
153 mirror_mask_t mirrors,
154 uint64_t packets, uint64_t bytes);
157 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
158 struct ofproto_dpif *ofproto; /* Owning ofproto. */
159 void *aux; /* Key supplied by ofproto's client. */
160 char *name; /* Identifier for log messages. */
163 struct list ports; /* Contains "struct ofport"s. */
164 enum port_vlan_mode vlan_mode; /* VLAN mode */
165 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
166 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
167 * NULL if all VLANs are trunked. */
168 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
169 struct bond *bond; /* Nonnull iff more than one port. */
170 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
173 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
175 /* Port mirroring info. */
176 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
177 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
178 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
181 static void bundle_remove(struct ofport *);
182 static void bundle_update(struct ofbundle *);
183 static void bundle_destroy(struct ofbundle *);
184 static void bundle_del_port(struct ofport_dpif *);
185 static void bundle_run(struct ofbundle *);
186 static void bundle_wait(struct ofbundle *);
187 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
188 uint16_t in_port, bool warn,
189 struct ofport_dpif **in_ofportp);
191 /* A controller may use OFPP_NONE as the ingress port to indicate that
192 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
193 * when an input bundle is needed for validation (e.g., mirroring or
194 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
195 * any 'port' structs, so care must be taken when dealing with it. */
196 static struct ofbundle ofpp_none_bundle = {
198 .vlan_mode = PORT_VLAN_TRUNK
201 static void stp_run(struct ofproto_dpif *ofproto);
202 static void stp_wait(struct ofproto_dpif *ofproto);
203 static int set_stp_port(struct ofport *,
204 const struct ofproto_port_stp_settings *);
206 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
208 struct action_xlate_ctx {
209 /* action_xlate_ctx_init() initializes these members. */
212 struct ofproto_dpif *ofproto;
214 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
215 * this flow when actions change header fields. */
218 /* The packet corresponding to 'flow', or a null pointer if we are
219 * revalidating without a packet to refer to. */
220 const struct ofpbuf *packet;
222 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
223 * actions update the flow table?
225 * We want to update these tables if we are actually processing a packet,
226 * or if we are accounting for packets that the datapath has processed, but
227 * not if we are just revalidating. */
230 /* The rule that we are currently translating, or NULL. */
231 struct rule_dpif *rule;
233 /* Union of the set of TCP flags seen so far in this flow. (Used only by
234 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
238 /* If nonnull, flow translation calls this function just before executing a
239 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
240 * when the recursion depth is exceeded.
242 * 'rule' is the rule being submitted into. It will be null if the
243 * resubmit or OFPP_TABLE action didn't find a matching rule.
245 * This is normally null so the client has to set it manually after
246 * calling action_xlate_ctx_init(). */
247 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule);
249 /* If nonnull, flow translation calls this function to report some
250 * significant decision, e.g. to explain why OFPP_NORMAL translation
251 * dropped a packet. */
252 void (*report_hook)(struct action_xlate_ctx *, const char *s);
254 /* If nonnull, flow translation credits the specified statistics to each
255 * rule reached through a resubmit or OFPP_TABLE action.
257 * This is normally null so the client has to set it manually after
258 * calling action_xlate_ctx_init(). */
259 const struct dpif_flow_stats *resubmit_stats;
261 /* xlate_actions() initializes and uses these members. The client might want
262 * to look at them after it returns. */
264 struct ofpbuf *odp_actions; /* Datapath actions. */
265 tag_type tags; /* Tags associated with actions. */
266 enum slow_path_reason slow; /* 0 if fast path may be used. */
267 bool has_learn; /* Actions include NXAST_LEARN? */
268 bool has_normal; /* Actions output to OFPP_NORMAL? */
269 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
270 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
271 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
273 /* xlate_actions() initializes and uses these members, but the client has no
274 * reason to look at them. */
276 int recurse; /* Recursion level, via xlate_table_action. */
277 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
278 struct flow base_flow; /* Flow at the last commit. */
279 uint32_t orig_skb_priority; /* Priority when packet arrived. */
280 uint8_t table_id; /* OpenFlow table ID where flow was found. */
281 uint32_t sflow_n_outputs; /* Number of output ports. */
282 uint32_t sflow_odp_port; /* Output port for composing sFlow action. */
283 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
284 bool exit; /* No further actions should be processed. */
285 struct flow orig_flow; /* Copy of original flow. */
288 static void action_xlate_ctx_init(struct action_xlate_ctx *,
289 struct ofproto_dpif *, const struct flow *,
290 ovs_be16 initial_tci, struct rule_dpif *,
291 uint8_t tcp_flags, const struct ofpbuf *);
292 static void xlate_actions(struct action_xlate_ctx *,
293 const struct ofpact *ofpacts, size_t ofpacts_len,
294 struct ofpbuf *odp_actions);
295 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
296 const struct ofpact *ofpacts,
299 static size_t put_userspace_action(const struct ofproto_dpif *,
300 struct ofpbuf *odp_actions,
302 const union user_action_cookie *);
304 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
305 enum slow_path_reason,
306 uint64_t *stub, size_t stub_size,
307 const struct nlattr **actionsp,
308 size_t *actions_lenp);
310 static void xlate_report(struct action_xlate_ctx *ctx, const char *s);
312 /* A subfacet (see "struct subfacet" below) has three possible installation
315 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
316 * case just after the subfacet is created, just before the subfacet is
317 * destroyed, or if the datapath returns an error when we try to install a
320 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
322 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
323 * ofproto_dpif is installed in the datapath.
326 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
327 SF_FAST_PATH, /* Full actions are installed. */
328 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
331 static const char *subfacet_path_to_string(enum subfacet_path);
333 /* A dpif flow and actions associated with a facet.
335 * See also the large comment on struct facet. */
338 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
339 struct list list_node; /* In struct facet's 'facets' list. */
340 struct facet *facet; /* Owning facet. */
344 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
345 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
346 * regenerate the ODP flow key from ->facet->flow. */
347 enum odp_key_fitness key_fitness;
351 long long int used; /* Time last used; time created if not used. */
353 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
354 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
358 * These should be essentially identical for every subfacet in a facet, but
359 * may differ in trivial ways due to VLAN splinters. */
360 size_t actions_len; /* Number of bytes in actions[]. */
361 struct nlattr *actions; /* Datapath actions. */
363 enum slow_path_reason slow; /* 0 if fast path may be used. */
364 enum subfacet_path path; /* Installed in datapath? */
366 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
367 * splinters can cause it to differ. This value should be removed when
368 * the VLAN splinters feature is no longer needed. */
369 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
371 /* Datapath port the packet arrived on. This is needed to remove
372 * flows for ports that are no longer part of the bridge. Since the
373 * flow definition only has the OpenFlow port number and the port is
374 * no longer part of the bridge, we can't determine the datapath port
375 * number needed to delete the flow from the datapath. */
376 uint32_t odp_in_port;
379 #define SUBFACET_DESTROY_MAX_BATCH 50
381 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
383 static struct subfacet *subfacet_find(struct ofproto_dpif *,
384 const struct nlattr *key, size_t key_len,
386 const struct flow *flow);
387 static void subfacet_destroy(struct subfacet *);
388 static void subfacet_destroy__(struct subfacet *);
389 static void subfacet_destroy_batch(struct ofproto_dpif *,
390 struct subfacet **, int n);
391 static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *,
393 static void subfacet_reset_dp_stats(struct subfacet *,
394 struct dpif_flow_stats *);
395 static void subfacet_update_time(struct subfacet *, long long int used);
396 static void subfacet_update_stats(struct subfacet *,
397 const struct dpif_flow_stats *);
398 static void subfacet_make_actions(struct subfacet *,
399 const struct ofpbuf *packet,
400 struct ofpbuf *odp_actions);
401 static int subfacet_install(struct subfacet *,
402 const struct nlattr *actions, size_t actions_len,
403 struct dpif_flow_stats *, enum slow_path_reason);
404 static void subfacet_uninstall(struct subfacet *);
406 static enum subfacet_path subfacet_want_path(enum slow_path_reason);
408 /* An exact-match instantiation of an OpenFlow flow.
410 * A facet associates a "struct flow", which represents the Open vSwitch
411 * userspace idea of an exact-match flow, with one or more subfacets. Each
412 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
413 * the facet. When the kernel module (or other dpif implementation) and Open
414 * vSwitch userspace agree on the definition of a flow key, there is exactly
415 * one subfacet per facet. If the dpif implementation supports more-specific
416 * flow matching than userspace, however, a facet can have more than one
417 * subfacet, each of which corresponds to some distinction in flow that
418 * userspace simply doesn't understand.
420 * Flow expiration works in terms of subfacets, so a facet must have at least
421 * one subfacet or it will never expire, leaking memory. */
424 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
425 struct list list_node; /* In owning rule's 'facets' list. */
426 struct rule_dpif *rule; /* Owning rule. */
429 struct list subfacets;
430 long long int used; /* Time last used; time created if not used. */
437 * - Do include packets and bytes sent "by hand", e.g. with
440 * - Do include packets and bytes that were obtained from the datapath
441 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
442 * DPIF_FP_ZERO_STATS).
444 * - Do not include packets or bytes that can be obtained from the
445 * datapath for any existing subfacet.
447 uint64_t packet_count; /* Number of packets received. */
448 uint64_t byte_count; /* Number of bytes received. */
450 /* Resubmit statistics. */
451 uint64_t prev_packet_count; /* Number of packets from last stats push. */
452 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
453 long long int prev_used; /* Used time from last stats push. */
456 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
457 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
458 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
460 /* Properties of datapath actions.
462 * Every subfacet has its own actions because actions can differ slightly
463 * between splintered and non-splintered subfacets due to the VLAN tag
464 * being initially different (present vs. absent). All of them have these
465 * properties in common so we just store one copy of them here. */
466 bool has_learn; /* Actions include NXAST_LEARN? */
467 bool has_normal; /* Actions output to OFPP_NORMAL? */
468 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
469 tag_type tags; /* Tags that would require revalidation. */
470 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
472 /* Storage for a single subfacet, to reduce malloc() time and space
473 * overhead. (A facet always has at least one subfacet and in the common
474 * case has exactly one subfacet.) */
475 struct subfacet one_subfacet;
478 static struct facet *facet_create(struct rule_dpif *,
479 const struct flow *, uint32_t hash);
480 static void facet_remove(struct facet *);
481 static void facet_free(struct facet *);
483 static struct facet *facet_find(struct ofproto_dpif *,
484 const struct flow *, uint32_t hash);
485 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
486 const struct flow *, uint32_t hash);
487 static void facet_revalidate(struct facet *);
488 static bool facet_check_consistency(struct facet *);
490 static void facet_flush_stats(struct facet *);
492 static void facet_update_time(struct facet *, long long int used);
493 static void facet_reset_counters(struct facet *);
494 static void facet_push_stats(struct facet *);
495 static void facet_learn(struct facet *);
496 static void facet_account(struct facet *);
498 static bool facet_is_controller_flow(struct facet *);
501 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
505 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
506 struct list bundle_node; /* In struct ofbundle's "ports" list. */
507 struct cfm *cfm; /* Connectivity Fault Management, if any. */
508 tag_type tag; /* Tag associated with this port. */
509 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
510 bool may_enable; /* May be enabled in bonds. */
511 long long int carrier_seq; /* Carrier status changes. */
514 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
515 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
516 long long int stp_state_entered;
518 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
520 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
522 * This is deprecated. It is only for compatibility with broken device
523 * drivers in old versions of Linux that do not properly support VLANs when
524 * VLAN devices are not used. When broken device drivers are no longer in
525 * widespread use, we will delete these interfaces. */
526 uint16_t realdev_ofp_port;
530 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
531 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
532 * traffic egressing the 'ofport' with that priority should be marked with. */
533 struct priority_to_dscp {
534 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
535 uint32_t priority; /* Priority of this queue (see struct flow). */
537 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
540 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
542 * This is deprecated. It is only for compatibility with broken device drivers
543 * in old versions of Linux that do not properly support VLANs when VLAN
544 * devices are not used. When broken device drivers are no longer in
545 * widespread use, we will delete these interfaces. */
546 struct vlan_splinter {
547 struct hmap_node realdev_vid_node;
548 struct hmap_node vlandev_node;
549 uint16_t realdev_ofp_port;
550 uint16_t vlandev_ofp_port;
554 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
555 uint32_t realdev, ovs_be16 vlan_tci);
556 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
557 static void vsp_remove(struct ofport_dpif *);
558 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
560 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
562 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
565 static struct ofport_dpif *
566 ofport_dpif_cast(const struct ofport *ofport)
568 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
569 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
572 static void port_run(struct ofport_dpif *);
573 static void port_run_fast(struct ofport_dpif *);
574 static void port_wait(struct ofport_dpif *);
575 static int set_cfm(struct ofport *, const struct cfm_settings *);
576 static void ofport_clear_priorities(struct ofport_dpif *);
578 struct dpif_completion {
579 struct list list_node;
580 struct ofoperation *op;
583 /* Extra information about a classifier table.
584 * Currently used just for optimized flow revalidation. */
586 /* If either of these is nonnull, then this table has a form that allows
587 * flows to be tagged to avoid revalidating most flows for the most common
588 * kinds of flow table changes. */
589 struct cls_table *catchall_table; /* Table that wildcards all fields. */
590 struct cls_table *other_table; /* Table with any other wildcard set. */
591 uint32_t basis; /* Keeps each table's tags separate. */
594 /* Reasons that we might need to revalidate every facet, and corresponding
597 * A value of 0 means that there is no need to revalidate.
599 * It would be nice to have some cleaner way to integrate with coverage
600 * counters, but with only a few reasons I guess this is good enough for
602 enum revalidate_reason {
603 REV_RECONFIGURE = 1, /* Switch configuration changed. */
604 REV_STP, /* Spanning tree protocol port status change. */
605 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
606 REV_FLOW_TABLE, /* Flow table changed. */
607 REV_INCONSISTENCY /* Facet self-check failed. */
609 COVERAGE_DEFINE(rev_reconfigure);
610 COVERAGE_DEFINE(rev_stp);
611 COVERAGE_DEFINE(rev_port_toggled);
612 COVERAGE_DEFINE(rev_flow_table);
613 COVERAGE_DEFINE(rev_inconsistency);
615 /* All datapaths of a given type share a single dpif backer instance. */
620 struct timer next_expiration;
621 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
624 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
625 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
627 static struct ofport_dpif *
628 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
630 struct ofproto_dpif {
631 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
633 struct dpif_backer *backer;
635 /* Special OpenFlow rules. */
636 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
637 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
643 struct netflow *netflow;
644 struct dpif_sflow *sflow;
645 struct hmap bundles; /* Contains "struct ofbundle"s. */
646 struct mac_learning *ml;
647 struct ofmirror *mirrors[MAX_MIRRORS];
649 bool has_bonded_bundles;
653 struct hmap subfacets;
654 struct governor *governor;
657 struct table_dpif tables[N_TABLES];
658 enum revalidate_reason need_revalidate;
659 struct tag_set revalidate_set;
661 /* Support for debugging async flow mods. */
662 struct list completions;
664 bool has_bundle_action; /* True when the first bundle action appears. */
665 struct netdev_stats stats; /* To account packets generated and consumed in
670 long long int stp_last_tick;
672 /* VLAN splinters. */
673 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
674 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
677 struct sset ports; /* Set of port names. */
678 struct sset port_poll_set; /* Queued names for port_poll() reply. */
679 int port_poll_errno; /* Last errno for port_poll() reply. */
682 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
683 * for debugging the asynchronous flow_mod implementation.) */
686 /* All existing ofproto_dpif instances, indexed by ->up.name. */
687 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
689 static void ofproto_dpif_unixctl_init(void);
691 static struct ofproto_dpif *
692 ofproto_dpif_cast(const struct ofproto *ofproto)
694 assert(ofproto->ofproto_class == &ofproto_dpif_class);
695 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
698 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
700 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
702 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
703 const struct ofpbuf *, ovs_be16 initial_tci,
706 /* Packet processing. */
707 static void update_learning_table(struct ofproto_dpif *,
708 const struct flow *, int vlan,
711 #define FLOW_MISS_MAX_BATCH 50
712 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
714 /* Flow expiration. */
715 static int expire(struct dpif_backer *);
718 static void send_netflow_active_timeouts(struct ofproto_dpif *);
721 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
722 static size_t compose_sflow_action(const struct ofproto_dpif *,
723 struct ofpbuf *odp_actions,
724 const struct flow *, uint32_t odp_port);
725 static void add_mirror_actions(struct action_xlate_ctx *ctx,
726 const struct flow *flow);
727 /* Global variables. */
728 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
730 /* Initial mappings of port to bridge mappings. */
731 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
733 /* Factory functions. */
736 init(const struct shash *iface_hints)
738 struct shash_node *node;
740 /* Make a local copy, since we don't own 'iface_hints' elements. */
741 SHASH_FOR_EACH(node, iface_hints) {
742 const struct iface_hint *orig_hint = node->data;
743 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
745 new_hint->br_name = xstrdup(orig_hint->br_name);
746 new_hint->br_type = xstrdup(orig_hint->br_type);
747 new_hint->ofp_port = orig_hint->ofp_port;
749 shash_add(&init_ofp_ports, node->name, new_hint);
754 enumerate_types(struct sset *types)
756 dp_enumerate_types(types);
760 enumerate_names(const char *type, struct sset *names)
762 struct ofproto_dpif *ofproto;
765 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
766 if (strcmp(type, ofproto->up.type)) {
769 sset_add(names, ofproto->up.name);
776 del(const char *type, const char *name)
781 error = dpif_open(name, type, &dpif);
783 error = dpif_delete(dpif);
790 port_open_type(const char *datapath_type, const char *port_type)
792 return dpif_port_open_type(datapath_type, port_type);
795 /* Type functions. */
797 static struct ofproto_dpif *
798 lookup_ofproto_dpif_by_port_name(const char *name)
800 struct ofproto_dpif *ofproto;
802 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
803 if (sset_contains(&ofproto->ports, name)) {
812 type_run(const char *type)
814 struct dpif_backer *backer;
818 backer = shash_find_data(&all_dpif_backers, type);
820 /* This is not necessarily a problem, since backers are only
821 * created on demand. */
825 dpif_run(backer->dpif);
827 if (timer_expired(&backer->next_expiration)) {
828 int delay = expire(backer);
829 timer_set_duration(&backer->next_expiration, delay);
832 /* Check for port changes in the dpif. */
833 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
834 struct ofproto_dpif *ofproto;
835 struct dpif_port port;
837 /* Don't report on the datapath's device. */
838 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
842 ofproto = lookup_ofproto_dpif_by_port_name(devname);
843 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
844 /* The port was removed. If we know the datapath,
845 * report it through poll_set(). If we don't, it may be
846 * notifying us of a removal we initiated, so ignore it.
847 * If there's a pending ENOBUFS, let it stand, since
848 * everything will be reevaluated. */
849 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
850 sset_add(&ofproto->port_poll_set, devname);
851 ofproto->port_poll_errno = 0;
853 } else if (!ofproto) {
854 /* The port was added, but we don't know with which
855 * ofproto we should associate it. Delete it. */
856 dpif_port_del(backer->dpif, port.port_no);
858 dpif_port_destroy(&port);
864 if (error != EAGAIN) {
865 struct ofproto_dpif *ofproto;
867 /* There was some sort of error, so propagate it to all
868 * ofprotos that use this backer. */
869 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
870 &all_ofproto_dpifs) {
871 if (ofproto->backer == backer) {
872 sset_clear(&ofproto->port_poll_set);
873 ofproto->port_poll_errno = error;
882 type_run_fast(const char *type)
884 struct dpif_backer *backer;
887 backer = shash_find_data(&all_dpif_backers, type);
889 /* This is not necessarily a problem, since backers are only
890 * created on demand. */
894 /* Handle one or more batches of upcalls, until there's nothing left to do
895 * or until we do a fixed total amount of work.
897 * We do work in batches because it can be much cheaper to set up a number
898 * of flows and fire off their patches all at once. We do multiple batches
899 * because in some cases handling a packet can cause another packet to be
900 * queued almost immediately as part of the return flow. Both
901 * optimizations can make major improvements on some benchmarks and
902 * presumably for real traffic as well. */
904 while (work < FLOW_MISS_MAX_BATCH) {
905 int retval = handle_upcalls(backer, FLOW_MISS_MAX_BATCH - work);
916 type_wait(const char *type)
918 struct dpif_backer *backer;
920 backer = shash_find_data(&all_dpif_backers, type);
922 /* This is not necessarily a problem, since backers are only
923 * created on demand. */
927 timer_wait(&backer->next_expiration);
930 /* Basic life-cycle. */
932 static int add_internal_flows(struct ofproto_dpif *);
934 static struct ofproto *
937 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
942 dealloc(struct ofproto *ofproto_)
944 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
949 close_dpif_backer(struct dpif_backer *backer)
951 struct shash_node *node;
953 assert(backer->refcount > 0);
955 if (--backer->refcount) {
959 hmap_destroy(&backer->odp_to_ofport_map);
960 node = shash_find(&all_dpif_backers, backer->type);
962 shash_delete(&all_dpif_backers, node);
963 dpif_close(backer->dpif);
968 /* Datapath port slated for removal from datapath. */
970 struct list list_node;
975 open_dpif_backer(const char *type, struct dpif_backer **backerp)
977 struct dpif_backer *backer;
978 struct dpif_port_dump port_dump;
979 struct dpif_port port;
980 struct shash_node *node;
981 struct list garbage_list;
982 struct odp_garbage *garbage, *next;
988 backer = shash_find_data(&all_dpif_backers, type);
995 backer_name = xasprintf("ovs-%s", type);
997 /* Remove any existing datapaths, since we assume we're the only
998 * userspace controlling the datapath. */
1000 dp_enumerate_names(type, &names);
1001 SSET_FOR_EACH(name, &names) {
1002 struct dpif *old_dpif;
1004 /* Don't remove our backer if it exists. */
1005 if (!strcmp(name, backer_name)) {
1009 if (dpif_open(name, type, &old_dpif)) {
1010 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1012 dpif_delete(old_dpif);
1013 dpif_close(old_dpif);
1016 sset_destroy(&names);
1018 backer = xmalloc(sizeof *backer);
1020 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1023 VLOG_ERR("failed to open datapath of type %s: %s", type,
1028 backer->type = xstrdup(type);
1029 backer->refcount = 1;
1030 hmap_init(&backer->odp_to_ofport_map);
1031 timer_set_duration(&backer->next_expiration, 1000);
1034 dpif_flow_flush(backer->dpif);
1036 /* Loop through the ports already on the datapath and remove any
1037 * that we don't need anymore. */
1038 list_init(&garbage_list);
1039 dpif_port_dump_start(&port_dump, backer->dpif);
1040 while (dpif_port_dump_next(&port_dump, &port)) {
1041 node = shash_find(&init_ofp_ports, port.name);
1042 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1043 garbage = xmalloc(sizeof *garbage);
1044 garbage->odp_port = port.port_no;
1045 list_push_front(&garbage_list, &garbage->list_node);
1048 dpif_port_dump_done(&port_dump);
1050 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1051 dpif_port_del(backer->dpif, garbage->odp_port);
1052 list_remove(&garbage->list_node);
1056 shash_add(&all_dpif_backers, type, backer);
1058 error = dpif_recv_set(backer->dpif, true);
1060 VLOG_ERR("failed to listen on datapath of type %s: %s",
1061 type, strerror(error));
1062 close_dpif_backer(backer);
1070 construct(struct ofproto *ofproto_)
1072 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1073 struct shash_node *node, *next;
1078 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1083 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1084 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1086 ofproto->n_matches = 0;
1088 ofproto->netflow = NULL;
1089 ofproto->sflow = NULL;
1090 ofproto->stp = NULL;
1091 hmap_init(&ofproto->bundles);
1092 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1093 for (i = 0; i < MAX_MIRRORS; i++) {
1094 ofproto->mirrors[i] = NULL;
1096 ofproto->has_bonded_bundles = false;
1098 hmap_init(&ofproto->facets);
1099 hmap_init(&ofproto->subfacets);
1100 ofproto->governor = NULL;
1102 for (i = 0; i < N_TABLES; i++) {
1103 struct table_dpif *table = &ofproto->tables[i];
1105 table->catchall_table = NULL;
1106 table->other_table = NULL;
1107 table->basis = random_uint32();
1109 ofproto->need_revalidate = 0;
1110 tag_set_init(&ofproto->revalidate_set);
1112 list_init(&ofproto->completions);
1114 ofproto_dpif_unixctl_init();
1116 ofproto->has_mirrors = false;
1117 ofproto->has_bundle_action = false;
1119 hmap_init(&ofproto->vlandev_map);
1120 hmap_init(&ofproto->realdev_vid_map);
1122 sset_init(&ofproto->ports);
1123 sset_init(&ofproto->port_poll_set);
1124 ofproto->port_poll_errno = 0;
1126 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1127 const struct iface_hint *iface_hint = node->data;
1129 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1130 /* Check if the datapath already has this port. */
1131 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1132 sset_add(&ofproto->ports, node->name);
1135 free(iface_hint->br_name);
1136 free(iface_hint->br_type);
1137 shash_delete(&init_ofp_ports, node);
1141 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1142 hash_string(ofproto->up.name, 0));
1143 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1145 ofproto_init_tables(ofproto_, N_TABLES);
1146 error = add_internal_flows(ofproto);
1147 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1153 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1154 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1156 struct ofputil_flow_mod fm;
1159 match_init_catchall(&fm.match);
1161 match_set_reg(&fm.match, 0, id);
1162 fm.new_cookie = htonll(0);
1163 fm.cookie = htonll(0);
1164 fm.cookie_mask = htonll(0);
1165 fm.table_id = TBL_INTERNAL;
1166 fm.command = OFPFC_ADD;
1167 fm.idle_timeout = 0;
1168 fm.hard_timeout = 0;
1172 fm.ofpacts = ofpacts->data;
1173 fm.ofpacts_len = ofpacts->size;
1175 error = ofproto_flow_mod(&ofproto->up, &fm);
1177 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1178 id, ofperr_to_string(error));
1182 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1183 assert(*rulep != NULL);
1189 add_internal_flows(struct ofproto_dpif *ofproto)
1191 struct ofpact_controller *controller;
1192 uint64_t ofpacts_stub[128 / 8];
1193 struct ofpbuf ofpacts;
1197 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1200 controller = ofpact_put_CONTROLLER(&ofpacts);
1201 controller->max_len = UINT16_MAX;
1202 controller->controller_id = 0;
1203 controller->reason = OFPR_NO_MATCH;
1204 ofpact_pad(&ofpacts);
1206 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1211 ofpbuf_clear(&ofpacts);
1212 error = add_internal_flow(ofproto, id++, &ofpacts,
1213 &ofproto->no_packet_in_rule);
1218 complete_operations(struct ofproto_dpif *ofproto)
1220 struct dpif_completion *c, *next;
1222 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1223 ofoperation_complete(c->op, 0);
1224 list_remove(&c->list_node);
1230 destruct(struct ofproto *ofproto_)
1232 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1233 struct rule_dpif *rule, *next_rule;
1234 struct oftable *table;
1237 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1238 complete_operations(ofproto);
1240 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1241 struct cls_cursor cursor;
1243 cls_cursor_init(&cursor, &table->cls, NULL);
1244 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1245 ofproto_rule_destroy(&rule->up);
1249 for (i = 0; i < MAX_MIRRORS; i++) {
1250 mirror_destroy(ofproto->mirrors[i]);
1253 netflow_destroy(ofproto->netflow);
1254 dpif_sflow_destroy(ofproto->sflow);
1255 hmap_destroy(&ofproto->bundles);
1256 mac_learning_destroy(ofproto->ml);
1258 hmap_destroy(&ofproto->facets);
1259 hmap_destroy(&ofproto->subfacets);
1260 governor_destroy(ofproto->governor);
1262 hmap_destroy(&ofproto->vlandev_map);
1263 hmap_destroy(&ofproto->realdev_vid_map);
1265 sset_destroy(&ofproto->ports);
1266 sset_destroy(&ofproto->port_poll_set);
1268 close_dpif_backer(ofproto->backer);
1272 run_fast(struct ofproto *ofproto_)
1274 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1275 struct ofport_dpif *ofport;
1277 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1278 port_run_fast(ofport);
1285 run(struct ofproto *ofproto_)
1287 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1288 struct ofport_dpif *ofport;
1289 struct ofbundle *bundle;
1293 complete_operations(ofproto);
1296 error = run_fast(ofproto_);
1301 if (ofproto->netflow) {
1302 if (netflow_run(ofproto->netflow)) {
1303 send_netflow_active_timeouts(ofproto);
1306 if (ofproto->sflow) {
1307 dpif_sflow_run(ofproto->sflow);
1310 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1313 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1318 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
1320 /* Now revalidate if there's anything to do. */
1321 if (ofproto->need_revalidate
1322 || !tag_set_is_empty(&ofproto->revalidate_set)) {
1323 struct tag_set revalidate_set = ofproto->revalidate_set;
1324 bool revalidate_all = ofproto->need_revalidate;
1325 struct facet *facet;
1327 switch (ofproto->need_revalidate) {
1328 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
1329 case REV_STP: COVERAGE_INC(rev_stp); break;
1330 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
1331 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
1332 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
1335 /* Clear the revalidation flags. */
1336 tag_set_init(&ofproto->revalidate_set);
1337 ofproto->need_revalidate = 0;
1339 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
1341 || tag_set_intersects(&revalidate_set, facet->tags)) {
1342 facet_revalidate(facet);
1347 /* Check the consistency of a random facet, to aid debugging. */
1348 if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) {
1349 struct facet *facet;
1351 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1352 struct facet, hmap_node);
1353 if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) {
1354 if (!facet_check_consistency(facet)) {
1355 ofproto->need_revalidate = REV_INCONSISTENCY;
1360 if (ofproto->governor) {
1363 governor_run(ofproto->governor);
1365 /* If the governor has shrunk to its minimum size and the number of
1366 * subfacets has dwindled, then drop the governor entirely.
1368 * For hysteresis, the number of subfacets to drop the governor is
1369 * smaller than the number needed to trigger its creation. */
1370 n_subfacets = hmap_count(&ofproto->subfacets);
1371 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1372 && governor_is_idle(ofproto->governor)) {
1373 governor_destroy(ofproto->governor);
1374 ofproto->governor = NULL;
1382 wait(struct ofproto *ofproto_)
1384 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1385 struct ofport_dpif *ofport;
1386 struct ofbundle *bundle;
1388 if (!clogged && !list_is_empty(&ofproto->completions)) {
1389 poll_immediate_wake();
1392 dpif_wait(ofproto->backer->dpif);
1393 dpif_recv_wait(ofproto->backer->dpif);
1394 if (ofproto->sflow) {
1395 dpif_sflow_wait(ofproto->sflow);
1397 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
1398 poll_immediate_wake();
1400 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1403 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1404 bundle_wait(bundle);
1406 if (ofproto->netflow) {
1407 netflow_wait(ofproto->netflow);
1409 mac_learning_wait(ofproto->ml);
1411 if (ofproto->need_revalidate) {
1412 /* Shouldn't happen, but if it does just go around again. */
1413 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1414 poll_immediate_wake();
1416 if (ofproto->governor) {
1417 governor_wait(ofproto->governor);
1422 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1424 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1426 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1427 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1431 flush(struct ofproto *ofproto_)
1433 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1434 struct subfacet *subfacet, *next_subfacet;
1435 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1439 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1440 &ofproto->subfacets) {
1441 if (subfacet->path != SF_NOT_INSTALLED) {
1442 batch[n_batch++] = subfacet;
1443 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1444 subfacet_destroy_batch(ofproto, batch, n_batch);
1448 subfacet_destroy(subfacet);
1453 subfacet_destroy_batch(ofproto, batch, n_batch);
1458 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1459 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1461 *arp_match_ip = true;
1462 *actions = (OFPUTIL_A_OUTPUT |
1463 OFPUTIL_A_SET_VLAN_VID |
1464 OFPUTIL_A_SET_VLAN_PCP |
1465 OFPUTIL_A_STRIP_VLAN |
1466 OFPUTIL_A_SET_DL_SRC |
1467 OFPUTIL_A_SET_DL_DST |
1468 OFPUTIL_A_SET_NW_SRC |
1469 OFPUTIL_A_SET_NW_DST |
1470 OFPUTIL_A_SET_NW_TOS |
1471 OFPUTIL_A_SET_TP_SRC |
1472 OFPUTIL_A_SET_TP_DST |
1477 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1479 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1480 struct dpif_dp_stats s;
1482 strcpy(ots->name, "classifier");
1484 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1486 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1487 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1490 static struct ofport *
1493 struct ofport_dpif *port = xmalloc(sizeof *port);
1498 port_dealloc(struct ofport *port_)
1500 struct ofport_dpif *port = ofport_dpif_cast(port_);
1505 port_construct(struct ofport *port_)
1507 struct ofport_dpif *port = ofport_dpif_cast(port_);
1508 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1509 struct dpif_port dpif_port;
1512 ofproto->need_revalidate = REV_RECONFIGURE;
1513 port->bundle = NULL;
1515 port->tag = tag_create_random();
1516 port->may_enable = true;
1517 port->stp_port = NULL;
1518 port->stp_state = STP_DISABLED;
1519 hmap_init(&port->priorities);
1520 port->realdev_ofp_port = 0;
1521 port->vlandev_vid = 0;
1522 port->carrier_seq = netdev_get_carrier_resets(port->up.netdev);
1524 error = dpif_port_query_by_name(ofproto->backer->dpif,
1525 netdev_get_name(port->up.netdev),
1531 port->odp_port = dpif_port.port_no;
1533 /* Sanity-check that a mapping doesn't already exist. This
1534 * shouldn't happen. */
1535 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1536 VLOG_ERR("port %s already has an OpenFlow port number\n",
1541 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1542 hash_int(port->odp_port, 0));
1544 if (ofproto->sflow) {
1545 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1552 port_destruct(struct ofport *port_)
1554 struct ofport_dpif *port = ofport_dpif_cast(port_);
1555 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1556 const char *devname = netdev_get_name(port->up.netdev);
1558 if (dpif_port_exists(ofproto->backer->dpif, devname)) {
1559 /* The underlying device is still there, so delete it. This
1560 * happens when the ofproto is being destroyed, since the caller
1561 * assumes that removal of attached ports will happen as part of
1563 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1566 sset_find_and_delete(&ofproto->ports, devname);
1567 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1568 ofproto->need_revalidate = REV_RECONFIGURE;
1569 bundle_remove(port_);
1570 set_cfm(port_, NULL);
1571 if (ofproto->sflow) {
1572 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1575 ofport_clear_priorities(port);
1576 hmap_destroy(&port->priorities);
1580 port_modified(struct ofport *port_)
1582 struct ofport_dpif *port = ofport_dpif_cast(port_);
1584 if (port->bundle && port->bundle->bond) {
1585 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1590 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1592 struct ofport_dpif *port = ofport_dpif_cast(port_);
1593 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1594 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1596 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1597 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1598 OFPUTIL_PC_NO_PACKET_IN)) {
1599 ofproto->need_revalidate = REV_RECONFIGURE;
1601 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1602 bundle_update(port->bundle);
1608 set_sflow(struct ofproto *ofproto_,
1609 const struct ofproto_sflow_options *sflow_options)
1611 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1612 struct dpif_sflow *ds = ofproto->sflow;
1614 if (sflow_options) {
1616 struct ofport_dpif *ofport;
1618 ds = ofproto->sflow = dpif_sflow_create();
1619 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1620 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1622 ofproto->need_revalidate = REV_RECONFIGURE;
1624 dpif_sflow_set_options(ds, sflow_options);
1627 dpif_sflow_destroy(ds);
1628 ofproto->need_revalidate = REV_RECONFIGURE;
1629 ofproto->sflow = NULL;
1636 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1638 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1645 struct ofproto_dpif *ofproto;
1647 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1648 ofproto->need_revalidate = REV_RECONFIGURE;
1649 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1652 if (cfm_configure(ofport->cfm, s)) {
1658 cfm_destroy(ofport->cfm);
1664 get_cfm_fault(const struct ofport *ofport_)
1666 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1668 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1672 get_cfm_opup(const struct ofport *ofport_)
1674 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1676 return ofport->cfm ? cfm_get_opup(ofport->cfm) : -1;
1680 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1683 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1686 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1694 get_cfm_health(const struct ofport *ofport_)
1696 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1698 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1701 /* Spanning Tree. */
1704 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1706 struct ofproto_dpif *ofproto = ofproto_;
1707 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1708 struct ofport_dpif *ofport;
1710 ofport = stp_port_get_aux(sp);
1712 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1713 ofproto->up.name, port_num);
1715 struct eth_header *eth = pkt->l2;
1717 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1718 if (eth_addr_is_zero(eth->eth_src)) {
1719 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1720 "with unknown MAC", ofproto->up.name, port_num);
1722 send_packet(ofport, pkt);
1728 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1730 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1732 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1734 /* Only revalidate flows if the configuration changed. */
1735 if (!s != !ofproto->stp) {
1736 ofproto->need_revalidate = REV_RECONFIGURE;
1740 if (!ofproto->stp) {
1741 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1742 send_bpdu_cb, ofproto);
1743 ofproto->stp_last_tick = time_msec();
1746 stp_set_bridge_id(ofproto->stp, s->system_id);
1747 stp_set_bridge_priority(ofproto->stp, s->priority);
1748 stp_set_hello_time(ofproto->stp, s->hello_time);
1749 stp_set_max_age(ofproto->stp, s->max_age);
1750 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1752 struct ofport *ofport;
1754 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1755 set_stp_port(ofport, NULL);
1758 stp_destroy(ofproto->stp);
1759 ofproto->stp = NULL;
1766 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1768 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1772 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1773 s->designated_root = stp_get_designated_root(ofproto->stp);
1774 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1783 update_stp_port_state(struct ofport_dpif *ofport)
1785 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1786 enum stp_state state;
1788 /* Figure out new state. */
1789 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1793 if (ofport->stp_state != state) {
1794 enum ofputil_port_state of_state;
1797 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1798 netdev_get_name(ofport->up.netdev),
1799 stp_state_name(ofport->stp_state),
1800 stp_state_name(state));
1801 if (stp_learn_in_state(ofport->stp_state)
1802 != stp_learn_in_state(state)) {
1803 /* xxx Learning action flows should also be flushed. */
1804 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1806 fwd_change = stp_forward_in_state(ofport->stp_state)
1807 != stp_forward_in_state(state);
1809 ofproto->need_revalidate = REV_STP;
1810 ofport->stp_state = state;
1811 ofport->stp_state_entered = time_msec();
1813 if (fwd_change && ofport->bundle) {
1814 bundle_update(ofport->bundle);
1817 /* Update the STP state bits in the OpenFlow port description. */
1818 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1819 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1820 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1821 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1822 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1824 ofproto_port_set_state(&ofport->up, of_state);
1828 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1829 * caller is responsible for assigning STP port numbers and ensuring
1830 * there are no duplicates. */
1832 set_stp_port(struct ofport *ofport_,
1833 const struct ofproto_port_stp_settings *s)
1835 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1836 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1837 struct stp_port *sp = ofport->stp_port;
1839 if (!s || !s->enable) {
1841 ofport->stp_port = NULL;
1842 stp_port_disable(sp);
1843 update_stp_port_state(ofport);
1846 } else if (sp && stp_port_no(sp) != s->port_num
1847 && ofport == stp_port_get_aux(sp)) {
1848 /* The port-id changed, so disable the old one if it's not
1849 * already in use by another port. */
1850 stp_port_disable(sp);
1853 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1854 stp_port_enable(sp);
1856 stp_port_set_aux(sp, ofport);
1857 stp_port_set_priority(sp, s->priority);
1858 stp_port_set_path_cost(sp, s->path_cost);
1860 update_stp_port_state(ofport);
1866 get_stp_port_status(struct ofport *ofport_,
1867 struct ofproto_port_stp_status *s)
1869 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1870 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1871 struct stp_port *sp = ofport->stp_port;
1873 if (!ofproto->stp || !sp) {
1879 s->port_id = stp_port_get_id(sp);
1880 s->state = stp_port_get_state(sp);
1881 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1882 s->role = stp_port_get_role(sp);
1883 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1889 stp_run(struct ofproto_dpif *ofproto)
1892 long long int now = time_msec();
1893 long long int elapsed = now - ofproto->stp_last_tick;
1894 struct stp_port *sp;
1897 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1898 ofproto->stp_last_tick = now;
1900 while (stp_get_changed_port(ofproto->stp, &sp)) {
1901 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1904 update_stp_port_state(ofport);
1908 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1909 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1915 stp_wait(struct ofproto_dpif *ofproto)
1918 poll_timer_wait(1000);
1922 /* Returns true if STP should process 'flow'. */
1924 stp_should_process_flow(const struct flow *flow)
1926 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1930 stp_process_packet(const struct ofport_dpif *ofport,
1931 const struct ofpbuf *packet)
1933 struct ofpbuf payload = *packet;
1934 struct eth_header *eth = payload.data;
1935 struct stp_port *sp = ofport->stp_port;
1937 /* Sink packets on ports that have STP disabled when the bridge has
1939 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1943 /* Trim off padding on payload. */
1944 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1945 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1948 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1949 stp_received_bpdu(sp, payload.data, payload.size);
1953 static struct priority_to_dscp *
1954 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1956 struct priority_to_dscp *pdscp;
1959 hash = hash_int(priority, 0);
1960 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1961 if (pdscp->priority == priority) {
1969 ofport_clear_priorities(struct ofport_dpif *ofport)
1971 struct priority_to_dscp *pdscp, *next;
1973 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1974 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1980 set_queues(struct ofport *ofport_,
1981 const struct ofproto_port_queue *qdscp_list,
1984 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1985 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1986 struct hmap new = HMAP_INITIALIZER(&new);
1989 for (i = 0; i < n_qdscp; i++) {
1990 struct priority_to_dscp *pdscp;
1994 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1995 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2000 pdscp = get_priority(ofport, priority);
2002 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2004 pdscp = xmalloc(sizeof *pdscp);
2005 pdscp->priority = priority;
2007 ofproto->need_revalidate = REV_RECONFIGURE;
2010 if (pdscp->dscp != dscp) {
2012 ofproto->need_revalidate = REV_RECONFIGURE;
2015 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2018 if (!hmap_is_empty(&ofport->priorities)) {
2019 ofport_clear_priorities(ofport);
2020 ofproto->need_revalidate = REV_RECONFIGURE;
2023 hmap_swap(&new, &ofport->priorities);
2031 /* Expires all MAC learning entries associated with 'bundle' and forces its
2032 * ofproto to revalidate every flow.
2034 * Normally MAC learning entries are removed only from the ofproto associated
2035 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2036 * are removed from every ofproto. When patch ports and SLB bonds are in use
2037 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2038 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2039 * with the host from which it migrated. */
2041 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2043 struct ofproto_dpif *ofproto = bundle->ofproto;
2044 struct mac_learning *ml = ofproto->ml;
2045 struct mac_entry *mac, *next_mac;
2047 ofproto->need_revalidate = REV_RECONFIGURE;
2048 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2049 if (mac->port.p == bundle) {
2051 struct ofproto_dpif *o;
2053 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2055 struct mac_entry *e;
2057 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2060 tag_set_add(&o->revalidate_set, e->tag);
2061 mac_learning_expire(o->ml, e);
2067 mac_learning_expire(ml, mac);
2072 static struct ofbundle *
2073 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2075 struct ofbundle *bundle;
2077 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2078 &ofproto->bundles) {
2079 if (bundle->aux == aux) {
2086 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2087 * ones that are found to 'bundles'. */
2089 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2090 void **auxes, size_t n_auxes,
2091 struct hmapx *bundles)
2095 hmapx_init(bundles);
2096 for (i = 0; i < n_auxes; i++) {
2097 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2099 hmapx_add(bundles, bundle);
2105 bundle_update(struct ofbundle *bundle)
2107 struct ofport_dpif *port;
2109 bundle->floodable = true;
2110 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2111 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2112 || !stp_forward_in_state(port->stp_state)) {
2113 bundle->floodable = false;
2120 bundle_del_port(struct ofport_dpif *port)
2122 struct ofbundle *bundle = port->bundle;
2124 bundle->ofproto->need_revalidate = REV_RECONFIGURE;
2126 list_remove(&port->bundle_node);
2127 port->bundle = NULL;
2130 lacp_slave_unregister(bundle->lacp, port);
2133 bond_slave_unregister(bundle->bond, port);
2136 bundle_update(bundle);
2140 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2141 struct lacp_slave_settings *lacp,
2142 uint32_t bond_stable_id)
2144 struct ofport_dpif *port;
2146 port = get_ofp_port(bundle->ofproto, ofp_port);
2151 if (port->bundle != bundle) {
2152 bundle->ofproto->need_revalidate = REV_RECONFIGURE;
2154 bundle_del_port(port);
2157 port->bundle = bundle;
2158 list_push_back(&bundle->ports, &port->bundle_node);
2159 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2160 || !stp_forward_in_state(port->stp_state)) {
2161 bundle->floodable = false;
2165 port->bundle->ofproto->need_revalidate = REV_RECONFIGURE;
2166 lacp_slave_register(bundle->lacp, port, lacp);
2169 port->bond_stable_id = bond_stable_id;
2175 bundle_destroy(struct ofbundle *bundle)
2177 struct ofproto_dpif *ofproto;
2178 struct ofport_dpif *port, *next_port;
2185 ofproto = bundle->ofproto;
2186 for (i = 0; i < MAX_MIRRORS; i++) {
2187 struct ofmirror *m = ofproto->mirrors[i];
2189 if (m->out == bundle) {
2191 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2192 || hmapx_find_and_delete(&m->dsts, bundle)) {
2193 ofproto->need_revalidate = REV_RECONFIGURE;
2198 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2199 bundle_del_port(port);
2202 bundle_flush_macs(bundle, true);
2203 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2205 free(bundle->trunks);
2206 lacp_destroy(bundle->lacp);
2207 bond_destroy(bundle->bond);
2212 bundle_set(struct ofproto *ofproto_, void *aux,
2213 const struct ofproto_bundle_settings *s)
2215 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2216 bool need_flush = false;
2217 struct ofport_dpif *port;
2218 struct ofbundle *bundle;
2219 unsigned long *trunks;
2225 bundle_destroy(bundle_lookup(ofproto, aux));
2229 assert(s->n_slaves == 1 || s->bond != NULL);
2230 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2232 bundle = bundle_lookup(ofproto, aux);
2234 bundle = xmalloc(sizeof *bundle);
2236 bundle->ofproto = ofproto;
2237 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2238 hash_pointer(aux, 0));
2240 bundle->name = NULL;
2242 list_init(&bundle->ports);
2243 bundle->vlan_mode = PORT_VLAN_TRUNK;
2245 bundle->trunks = NULL;
2246 bundle->use_priority_tags = s->use_priority_tags;
2247 bundle->lacp = NULL;
2248 bundle->bond = NULL;
2250 bundle->floodable = true;
2252 bundle->src_mirrors = 0;
2253 bundle->dst_mirrors = 0;
2254 bundle->mirror_out = 0;
2257 if (!bundle->name || strcmp(s->name, bundle->name)) {
2259 bundle->name = xstrdup(s->name);
2264 if (!bundle->lacp) {
2265 ofproto->need_revalidate = REV_RECONFIGURE;
2266 bundle->lacp = lacp_create();
2268 lacp_configure(bundle->lacp, s->lacp);
2270 lacp_destroy(bundle->lacp);
2271 bundle->lacp = NULL;
2274 /* Update set of ports. */
2276 for (i = 0; i < s->n_slaves; i++) {
2277 if (!bundle_add_port(bundle, s->slaves[i],
2278 s->lacp ? &s->lacp_slaves[i] : NULL,
2279 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
2283 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2284 struct ofport_dpif *next_port;
2286 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2287 for (i = 0; i < s->n_slaves; i++) {
2288 if (s->slaves[i] == port->up.ofp_port) {
2293 bundle_del_port(port);
2297 assert(list_size(&bundle->ports) <= s->n_slaves);
2299 if (list_is_empty(&bundle->ports)) {
2300 bundle_destroy(bundle);
2304 /* Set VLAN tagging mode */
2305 if (s->vlan_mode != bundle->vlan_mode
2306 || s->use_priority_tags != bundle->use_priority_tags) {
2307 bundle->vlan_mode = s->vlan_mode;
2308 bundle->use_priority_tags = s->use_priority_tags;
2313 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2314 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2316 if (vlan != bundle->vlan) {
2317 bundle->vlan = vlan;
2321 /* Get trunked VLANs. */
2322 switch (s->vlan_mode) {
2323 case PORT_VLAN_ACCESS:
2327 case PORT_VLAN_TRUNK:
2328 trunks = CONST_CAST(unsigned long *, s->trunks);
2331 case PORT_VLAN_NATIVE_UNTAGGED:
2332 case PORT_VLAN_NATIVE_TAGGED:
2333 if (vlan != 0 && (!s->trunks
2334 || !bitmap_is_set(s->trunks, vlan)
2335 || bitmap_is_set(s->trunks, 0))) {
2336 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2338 trunks = bitmap_clone(s->trunks, 4096);
2340 trunks = bitmap_allocate1(4096);
2342 bitmap_set1(trunks, vlan);
2343 bitmap_set0(trunks, 0);
2345 trunks = CONST_CAST(unsigned long *, s->trunks);
2352 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2353 free(bundle->trunks);
2354 if (trunks == s->trunks) {
2355 bundle->trunks = vlan_bitmap_clone(trunks);
2357 bundle->trunks = trunks;
2362 if (trunks != s->trunks) {
2367 if (!list_is_short(&bundle->ports)) {
2368 bundle->ofproto->has_bonded_bundles = true;
2370 if (bond_reconfigure(bundle->bond, s->bond)) {
2371 ofproto->need_revalidate = REV_RECONFIGURE;
2374 bundle->bond = bond_create(s->bond);
2375 ofproto->need_revalidate = REV_RECONFIGURE;
2378 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2379 bond_slave_register(bundle->bond, port, port->bond_stable_id,
2383 bond_destroy(bundle->bond);
2384 bundle->bond = NULL;
2387 /* If we changed something that would affect MAC learning, un-learn
2388 * everything on this port and force flow revalidation. */
2390 bundle_flush_macs(bundle, false);
2397 bundle_remove(struct ofport *port_)
2399 struct ofport_dpif *port = ofport_dpif_cast(port_);
2400 struct ofbundle *bundle = port->bundle;
2403 bundle_del_port(port);
2404 if (list_is_empty(&bundle->ports)) {
2405 bundle_destroy(bundle);
2406 } else if (list_is_short(&bundle->ports)) {
2407 bond_destroy(bundle->bond);
2408 bundle->bond = NULL;
2414 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2416 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2417 struct ofport_dpif *port = port_;
2418 uint8_t ea[ETH_ADDR_LEN];
2421 error = netdev_get_etheraddr(port->up.netdev, ea);
2423 struct ofpbuf packet;
2426 ofpbuf_init(&packet, 0);
2427 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2429 memcpy(packet_pdu, pdu, pdu_size);
2431 send_packet(port, &packet);
2432 ofpbuf_uninit(&packet);
2434 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2435 "%s (%s)", port->bundle->name,
2436 netdev_get_name(port->up.netdev), strerror(error));
2441 bundle_send_learning_packets(struct ofbundle *bundle)
2443 struct ofproto_dpif *ofproto = bundle->ofproto;
2444 int error, n_packets, n_errors;
2445 struct mac_entry *e;
2447 error = n_packets = n_errors = 0;
2448 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2449 if (e->port.p != bundle) {
2450 struct ofpbuf *learning_packet;
2451 struct ofport_dpif *port;
2455 /* The assignment to "port" is unnecessary but makes "grep"ing for
2456 * struct ofport_dpif more effective. */
2457 learning_packet = bond_compose_learning_packet(bundle->bond,
2461 ret = send_packet(port, learning_packet);
2462 ofpbuf_delete(learning_packet);
2472 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2473 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2474 "packets, last error was: %s",
2475 bundle->name, n_errors, n_packets, strerror(error));
2477 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2478 bundle->name, n_packets);
2483 bundle_run(struct ofbundle *bundle)
2486 lacp_run(bundle->lacp, send_pdu_cb);
2489 struct ofport_dpif *port;
2491 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2492 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2495 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
2496 lacp_status(bundle->lacp));
2497 if (bond_should_send_learning_packets(bundle->bond)) {
2498 bundle_send_learning_packets(bundle);
2504 bundle_wait(struct ofbundle *bundle)
2507 lacp_wait(bundle->lacp);
2510 bond_wait(bundle->bond);
2517 mirror_scan(struct ofproto_dpif *ofproto)
2521 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2522 if (!ofproto->mirrors[idx]) {
2529 static struct ofmirror *
2530 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2534 for (i = 0; i < MAX_MIRRORS; i++) {
2535 struct ofmirror *mirror = ofproto->mirrors[i];
2536 if (mirror && mirror->aux == aux) {
2544 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2546 mirror_update_dups(struct ofproto_dpif *ofproto)
2550 for (i = 0; i < MAX_MIRRORS; i++) {
2551 struct ofmirror *m = ofproto->mirrors[i];
2554 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2558 for (i = 0; i < MAX_MIRRORS; i++) {
2559 struct ofmirror *m1 = ofproto->mirrors[i];
2566 for (j = i + 1; j < MAX_MIRRORS; j++) {
2567 struct ofmirror *m2 = ofproto->mirrors[j];
2569 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2570 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2571 m2->dup_mirrors |= m1->dup_mirrors;
2578 mirror_set(struct ofproto *ofproto_, void *aux,
2579 const struct ofproto_mirror_settings *s)
2581 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2582 mirror_mask_t mirror_bit;
2583 struct ofbundle *bundle;
2584 struct ofmirror *mirror;
2585 struct ofbundle *out;
2586 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2587 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2590 mirror = mirror_lookup(ofproto, aux);
2592 mirror_destroy(mirror);
2598 idx = mirror_scan(ofproto);
2600 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2602 ofproto->up.name, MAX_MIRRORS, s->name);
2606 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2607 mirror->ofproto = ofproto;
2610 mirror->out_vlan = -1;
2611 mirror->name = NULL;
2614 if (!mirror->name || strcmp(s->name, mirror->name)) {
2616 mirror->name = xstrdup(s->name);
2619 /* Get the new configuration. */
2620 if (s->out_bundle) {
2621 out = bundle_lookup(ofproto, s->out_bundle);
2623 mirror_destroy(mirror);
2629 out_vlan = s->out_vlan;
2631 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2632 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2634 /* If the configuration has not changed, do nothing. */
2635 if (hmapx_equals(&srcs, &mirror->srcs)
2636 && hmapx_equals(&dsts, &mirror->dsts)
2637 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2638 && mirror->out == out
2639 && mirror->out_vlan == out_vlan)
2641 hmapx_destroy(&srcs);
2642 hmapx_destroy(&dsts);
2646 hmapx_swap(&srcs, &mirror->srcs);
2647 hmapx_destroy(&srcs);
2649 hmapx_swap(&dsts, &mirror->dsts);
2650 hmapx_destroy(&dsts);
2652 free(mirror->vlans);
2653 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2656 mirror->out_vlan = out_vlan;
2658 /* Update bundles. */
2659 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2660 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2661 if (hmapx_contains(&mirror->srcs, bundle)) {
2662 bundle->src_mirrors |= mirror_bit;
2664 bundle->src_mirrors &= ~mirror_bit;
2667 if (hmapx_contains(&mirror->dsts, bundle)) {
2668 bundle->dst_mirrors |= mirror_bit;
2670 bundle->dst_mirrors &= ~mirror_bit;
2673 if (mirror->out == bundle) {
2674 bundle->mirror_out |= mirror_bit;
2676 bundle->mirror_out &= ~mirror_bit;
2680 ofproto->need_revalidate = REV_RECONFIGURE;
2681 ofproto->has_mirrors = true;
2682 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2683 mirror_update_dups(ofproto);
2689 mirror_destroy(struct ofmirror *mirror)
2691 struct ofproto_dpif *ofproto;
2692 mirror_mask_t mirror_bit;
2693 struct ofbundle *bundle;
2700 ofproto = mirror->ofproto;
2701 ofproto->need_revalidate = REV_RECONFIGURE;
2702 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2704 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2705 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2706 bundle->src_mirrors &= ~mirror_bit;
2707 bundle->dst_mirrors &= ~mirror_bit;
2708 bundle->mirror_out &= ~mirror_bit;
2711 hmapx_destroy(&mirror->srcs);
2712 hmapx_destroy(&mirror->dsts);
2713 free(mirror->vlans);
2715 ofproto->mirrors[mirror->idx] = NULL;
2719 mirror_update_dups(ofproto);
2721 ofproto->has_mirrors = false;
2722 for (i = 0; i < MAX_MIRRORS; i++) {
2723 if (ofproto->mirrors[i]) {
2724 ofproto->has_mirrors = true;
2731 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2732 uint64_t *packets, uint64_t *bytes)
2734 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2735 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2738 *packets = *bytes = UINT64_MAX;
2742 *packets = mirror->packet_count;
2743 *bytes = mirror->byte_count;
2749 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2751 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2752 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2753 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2759 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2761 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2762 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2763 return bundle && bundle->mirror_out != 0;
2767 forward_bpdu_changed(struct ofproto *ofproto_)
2769 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2770 ofproto->need_revalidate = REV_RECONFIGURE;
2774 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2777 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2778 mac_learning_set_idle_time(ofproto->ml, idle_time);
2779 mac_learning_set_max_entries(ofproto->ml, max_entries);
2784 static struct ofport_dpif *
2785 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2787 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2788 return ofport ? ofport_dpif_cast(ofport) : NULL;
2791 static struct ofport_dpif *
2792 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2794 return get_ofp_port(ofproto, odp_port_to_ofp_port(ofproto, odp_port));
2798 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2799 struct ofproto_port *ofproto_port,
2800 struct dpif_port *dpif_port)
2802 ofproto_port->name = dpif_port->name;
2803 ofproto_port->type = dpif_port->type;
2804 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2808 port_run_fast(struct ofport_dpif *ofport)
2810 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2811 struct ofpbuf packet;
2813 ofpbuf_init(&packet, 0);
2814 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2815 send_packet(ofport, &packet);
2816 ofpbuf_uninit(&packet);
2821 port_run(struct ofport_dpif *ofport)
2823 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2824 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2825 bool enable = netdev_get_carrier(ofport->up.netdev);
2827 ofport->carrier_seq = carrier_seq;
2829 port_run_fast(ofport);
2831 int cfm_opup = cfm_get_opup(ofport->cfm);
2833 cfm_run(ofport->cfm);
2834 enable = enable && !cfm_get_fault(ofport->cfm);
2836 if (cfm_opup >= 0) {
2837 enable = enable && cfm_opup;
2841 if (ofport->bundle) {
2842 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2843 if (carrier_changed) {
2844 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2848 if (ofport->may_enable != enable) {
2849 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2851 if (ofproto->has_bundle_action) {
2852 ofproto->need_revalidate = REV_PORT_TOGGLED;
2856 ofport->may_enable = enable;
2860 port_wait(struct ofport_dpif *ofport)
2863 cfm_wait(ofport->cfm);
2868 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2869 struct ofproto_port *ofproto_port)
2871 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2872 struct dpif_port dpif_port;
2875 if (!sset_contains(&ofproto->ports, devname)) {
2878 error = dpif_port_query_by_name(ofproto->backer->dpif,
2879 devname, &dpif_port);
2881 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2887 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2889 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2890 uint32_t odp_port = UINT32_MAX;
2893 error = dpif_port_add(ofproto->backer->dpif, netdev, &odp_port);
2895 sset_add(&ofproto->ports, netdev_get_name(netdev));
2901 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2903 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2904 uint32_t odp_port = ofp_port_to_odp_port(ofproto, ofp_port);
2907 if (odp_port != OFPP_NONE) {
2908 error = dpif_port_del(ofproto->backer->dpif, odp_port);
2911 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2913 /* The caller is going to close ofport->up.netdev. If this is a
2914 * bonded port, then the bond is using that netdev, so remove it
2915 * from the bond. The client will need to reconfigure everything
2916 * after deleting ports, so then the slave will get re-added. */
2917 bundle_remove(&ofport->up);
2924 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2926 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2929 error = netdev_get_stats(ofport->up.netdev, stats);
2931 if (!error && ofport->odp_port == OVSP_LOCAL) {
2932 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2934 /* ofproto->stats.tx_packets represents packets that we created
2935 * internally and sent to some port (e.g. packets sent with
2936 * send_packet()). Account for them as if they had come from
2937 * OFPP_LOCAL and got forwarded. */
2939 if (stats->rx_packets != UINT64_MAX) {
2940 stats->rx_packets += ofproto->stats.tx_packets;
2943 if (stats->rx_bytes != UINT64_MAX) {
2944 stats->rx_bytes += ofproto->stats.tx_bytes;
2947 /* ofproto->stats.rx_packets represents packets that were received on
2948 * some port and we processed internally and dropped (e.g. STP).
2949 * Account for them as if they had been forwarded to OFPP_LOCAL. */
2951 if (stats->tx_packets != UINT64_MAX) {
2952 stats->tx_packets += ofproto->stats.rx_packets;
2955 if (stats->tx_bytes != UINT64_MAX) {
2956 stats->tx_bytes += ofproto->stats.rx_bytes;
2963 /* Account packets for LOCAL port. */
2965 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2966 size_t tx_size, size_t rx_size)
2968 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2971 ofproto->stats.rx_packets++;
2972 ofproto->stats.rx_bytes += rx_size;
2975 ofproto->stats.tx_packets++;
2976 ofproto->stats.tx_bytes += tx_size;
2980 struct port_dump_state {
2986 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
2988 struct port_dump_state *state;
2990 *statep = state = xmalloc(sizeof *state);
2997 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2998 struct ofproto_port *port)
3000 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3001 struct port_dump_state *state = state_;
3002 struct sset_node *node;
3004 while ((node = sset_at_position(&ofproto->ports, &state->bucket,
3008 error = port_query_by_name(ofproto_, node->name, port);
3009 if (error != ENODEV) {
3018 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3020 struct port_dump_state *state = state_;
3027 port_poll(const struct ofproto *ofproto_, char **devnamep)
3029 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3031 if (ofproto->port_poll_errno) {
3032 int error = ofproto->port_poll_errno;
3033 ofproto->port_poll_errno = 0;
3037 if (sset_is_empty(&ofproto->port_poll_set)) {
3041 *devnamep = sset_pop(&ofproto->port_poll_set);
3046 port_poll_wait(const struct ofproto *ofproto_)
3048 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3049 dpif_port_poll_wait(ofproto->backer->dpif);
3053 port_is_lacp_current(const struct ofport *ofport_)
3055 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3056 return (ofport->bundle && ofport->bundle->lacp
3057 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3061 /* Upcall handling. */
3063 /* Flow miss batching.
3065 * Some dpifs implement operations faster when you hand them off in a batch.
3066 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3067 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3068 * more packets, plus possibly installing the flow in the dpif.
3070 * So far we only batch the operations that affect flow setup time the most.
3071 * It's possible to batch more than that, but the benefit might be minimal. */
3073 struct hmap_node hmap_node;
3074 struct ofproto_dpif *ofproto;
3076 enum odp_key_fitness key_fitness;
3077 const struct nlattr *key;
3079 ovs_be16 initial_tci;
3080 struct list packets;
3081 enum dpif_upcall_type upcall_type;
3082 uint32_t odp_in_port;
3085 struct flow_miss_op {
3086 struct dpif_op dpif_op;
3087 struct subfacet *subfacet; /* Subfacet */
3088 void *garbage; /* Pointer to pass to free(), NULL if none. */
3089 uint64_t stub[1024 / 8]; /* Temporary buffer. */
3092 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3093 * OpenFlow controller as necessary according to their individual
3094 * configurations. */
3096 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3097 const struct flow *flow)
3099 struct ofputil_packet_in pin;
3101 pin.packet = packet->data;
3102 pin.packet_len = packet->size;
3103 pin.reason = OFPR_NO_MATCH;
3104 pin.controller_id = 0;
3109 pin.send_len = 0; /* not used for flow table misses */
3111 flow_get_metadata(flow, &pin.fmd);
3113 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3116 static enum slow_path_reason
3117 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3118 const struct ofpbuf *packet)
3120 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
3126 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3128 cfm_process_heartbeat(ofport->cfm, packet);
3131 } else if (ofport->bundle && ofport->bundle->lacp
3132 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3134 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3137 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3139 stp_process_packet(ofport, packet);
3146 static struct flow_miss *
3147 flow_miss_find(struct hmap *todo, const struct flow *flow, uint32_t hash)
3149 struct flow_miss *miss;
3151 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3152 if (flow_equal(&miss->flow, flow)) {
3160 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3161 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3162 * 'miss' is associated with a subfacet the caller must also initialize the
3163 * returned op->subfacet, and if anything needs to be freed after processing
3164 * the op, the caller must initialize op->garbage also. */
3166 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3167 struct flow_miss_op *op)
3169 if (miss->flow.vlan_tci != miss->initial_tci) {
3170 /* This packet was received on a VLAN splinter port. We
3171 * added a VLAN to the packet to make the packet resemble
3172 * the flow, but the actions were composed assuming that
3173 * the packet contained no VLAN. So, we must remove the
3174 * VLAN header from the packet before trying to execute the
3176 eth_pop_vlan(packet);
3179 op->subfacet = NULL;
3181 op->dpif_op.type = DPIF_OP_EXECUTE;
3182 op->dpif_op.u.execute.key = miss->key;
3183 op->dpif_op.u.execute.key_len = miss->key_len;
3184 op->dpif_op.u.execute.packet = packet;
3187 /* Helper for handle_flow_miss_without_facet() and
3188 * handle_flow_miss_with_facet(). */
3190 handle_flow_miss_common(struct rule_dpif *rule,
3191 struct ofpbuf *packet, const struct flow *flow)
3193 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3195 ofproto->n_matches++;
3197 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3199 * Extra-special case for fail-open mode.
3201 * We are in fail-open mode and the packet matched the fail-open
3202 * rule, but we are connected to a controller too. We should send
3203 * the packet up to the controller in the hope that it will try to
3204 * set up a flow and thereby allow us to exit fail-open.
3206 * See the top-level comment in fail-open.c for more information.
3208 send_packet_in_miss(ofproto, packet, flow);
3212 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3213 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3214 * installing a datapath flow. The answer is usually "yes" (a return value of
3215 * true). However, for short flows the cost of bookkeeping is much higher than
3216 * the benefits, so when the datapath holds a large number of flows we impose
3217 * some heuristics to decide which flows are likely to be worth tracking. */
3219 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3220 struct flow_miss *miss, uint32_t hash)
3222 if (!ofproto->governor) {
3225 n_subfacets = hmap_count(&ofproto->subfacets);
3226 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3230 ofproto->governor = governor_create(ofproto->up.name);
3233 return governor_should_install_flow(ofproto->governor, hash,
3234 list_size(&miss->packets));
3237 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3238 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3239 * increment '*n_ops'. */
3241 handle_flow_miss_without_facet(struct flow_miss *miss,
3242 struct rule_dpif *rule,
3243 struct flow_miss_op *ops, size_t *n_ops)
3245 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3246 long long int now = time_msec();
3247 struct action_xlate_ctx ctx;
3248 struct ofpbuf *packet;
3250 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3251 struct flow_miss_op *op = &ops[*n_ops];
3252 struct dpif_flow_stats stats;
3253 struct ofpbuf odp_actions;
3255 COVERAGE_INC(facet_suppress);
3257 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3259 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3260 rule_credit_stats(rule, &stats);
3262 action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci,
3264 ctx.resubmit_stats = &stats;
3265 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3268 if (odp_actions.size) {
3269 struct dpif_execute *execute = &op->dpif_op.u.execute;
3271 init_flow_miss_execute_op(miss, packet, op);
3272 execute->actions = odp_actions.data;
3273 execute->actions_len = odp_actions.size;
3274 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3278 ofpbuf_uninit(&odp_actions);
3283 /* Handles 'miss', which matches 'facet'. May add any required datapath
3284 * operations to 'ops', incrementing '*n_ops' for each new op.
3286 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3287 * This is really important only for new facets: if we just called time_msec()
3288 * here, then the new subfacet or its packets could look (occasionally) as
3289 * though it was used some time after the facet was used. That can make a
3290 * one-packet flow look like it has a nonzero duration, which looks odd in
3291 * e.g. NetFlow statistics. */
3293 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3295 struct flow_miss_op *ops, size_t *n_ops)
3297 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3298 enum subfacet_path want_path;
3299 struct subfacet *subfacet;
3300 struct ofpbuf *packet;
3302 subfacet = subfacet_create(facet, miss, now);
3304 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3305 struct flow_miss_op *op = &ops[*n_ops];
3306 struct dpif_flow_stats stats;
3307 struct ofpbuf odp_actions;
3309 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3311 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3312 if (!subfacet->actions || subfacet->slow) {
3313 subfacet_make_actions(subfacet, packet, &odp_actions);
3316 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3317 subfacet_update_stats(subfacet, &stats);
3319 if (subfacet->actions_len) {
3320 struct dpif_execute *execute = &op->dpif_op.u.execute;
3322 init_flow_miss_execute_op(miss, packet, op);
3323 op->subfacet = subfacet;
3324 if (!subfacet->slow) {
3325 execute->actions = subfacet->actions;
3326 execute->actions_len = subfacet->actions_len;
3327 ofpbuf_uninit(&odp_actions);
3329 execute->actions = odp_actions.data;
3330 execute->actions_len = odp_actions.size;
3331 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3336 ofpbuf_uninit(&odp_actions);
3340 want_path = subfacet_want_path(subfacet->slow);
3341 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3342 struct flow_miss_op *op = &ops[(*n_ops)++];
3343 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3345 op->subfacet = subfacet;
3347 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3348 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3349 put->key = miss->key;
3350 put->key_len = miss->key_len;
3351 if (want_path == SF_FAST_PATH) {
3352 put->actions = subfacet->actions;
3353 put->actions_len = subfacet->actions_len;
3355 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
3356 op->stub, sizeof op->stub,
3357 &put->actions, &put->actions_len);
3363 /* Handles flow miss 'miss'. May add any required datapath operations
3364 * to 'ops', incrementing '*n_ops' for each new op. */
3366 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3369 struct ofproto_dpif *ofproto = miss->ofproto;
3370 struct facet *facet;
3374 /* The caller must ensure that miss->hmap_node.hash contains
3375 * flow_hash(miss->flow, 0). */
3376 hash = miss->hmap_node.hash;
3378 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3380 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
3382 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
3383 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
3387 facet = facet_create(rule, &miss->flow, hash);
3392 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3395 /* This function does post-processing on data returned from
3396 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the
3397 * rest of the upcall processing logic. In particular, if the extracted
3398 * in_port is a VLAN splinter port, it replaces flow->in_port by the "real"
3399 * port, sets flow->vlan_tci correctly for the VLAN of the VLAN splinter
3400 * port, and pushes a VLAN header onto 'packet' (if it is nonnull). The
3401 * caller must have called odp_flow_key_to_flow() and supply 'fitness' and
3402 * 'flow' from its output. The 'flow' argument must have had the "in_port"
3403 * member converted to the OpenFlow number.
3405 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
3406 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
3407 * (This differs from the value returned in flow->vlan_tci only for packets
3408 * received on VLAN splinters.) */
3409 static enum odp_key_fitness
3410 ofproto_dpif_vsp_adjust(const struct ofproto_dpif *ofproto,
3411 enum odp_key_fitness fitness,
3412 struct flow *flow, ovs_be16 *initial_tci,
3413 struct ofpbuf *packet)
3415 if (fitness == ODP_FIT_ERROR) {
3418 *initial_tci = flow->vlan_tci;
3420 if (vsp_adjust_flow(ofproto, flow)) {
3422 /* Make the packet resemble the flow, so that it gets sent to an
3423 * OpenFlow controller properly, so that it looks correct for
3424 * sFlow, and so that flow_extract() will get the correct vlan_tci
3425 * if it is called on 'packet'.
3427 * The allocated space inside 'packet' probably also contains
3428 * 'key', that is, both 'packet' and 'key' are probably part of a
3429 * struct dpif_upcall (see the large comment on that structure
3430 * definition), so pushing data on 'packet' is in general not a
3431 * good idea since it could overwrite 'key' or free it as a side
3432 * effect. However, it's OK in this special case because we know
3433 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
3434 * will just overwrite the 4-byte "struct nlattr", which is fine
3435 * since we don't need that header anymore. */
3436 eth_push_vlan(packet, flow->vlan_tci);
3439 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
3440 if (fitness == ODP_FIT_PERFECT) {
3441 fitness = ODP_FIT_TOO_MUCH;
3449 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3452 struct dpif_upcall *upcall;
3453 struct flow_miss *miss;
3454 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3455 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3456 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3466 /* Construct the to-do list.
3468 * This just amounts to extracting the flow from each packet and sticking
3469 * the packets that have the same flow in the same "flow_miss" structure so
3470 * that we can process them together. */
3473 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3474 struct flow_miss *miss = &misses[n_misses];
3475 struct flow_miss *existing_miss;
3476 enum odp_key_fitness fitness;
3477 struct ofproto_dpif *ofproto;
3478 struct ofport_dpif *port;
3479 uint32_t odp_in_port;
3483 fitness = odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
3484 port = odp_port_to_ofport(backer, flow.in_port);
3486 /* Received packet on port for which we couldn't associate
3487 * an ofproto. This can happen if a port is removed while
3488 * traffic is being received. Print a rate-limited message
3489 * in case it happens frequently. */
3490 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
3494 ofproto = ofproto_dpif_cast(port->up.ofproto);
3495 odp_in_port = flow.in_port;
3496 flow.in_port = port->up.ofp_port;
3498 /* Obtain metadata and check userspace/kernel agreement on flow match,
3499 * then set 'flow''s header pointers. */
3500 miss->key_fitness = ofproto_dpif_vsp_adjust(ofproto, fitness,
3501 &flow, &miss->initial_tci, upcall->packet);
3502 if (miss->key_fitness == ODP_FIT_ERROR) {
3505 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3506 &flow.tunnel, flow.in_port, &miss->flow);
3508 /* Add other packets to a to-do list. */
3509 hash = flow_hash(&miss->flow, 0);
3510 existing_miss = flow_miss_find(&todo, &miss->flow, hash);
3511 if (!existing_miss) {
3512 hmap_insert(&todo, &miss->hmap_node, hash);
3513 miss->ofproto = ofproto;
3514 miss->key = upcall->key;
3515 miss->key_len = upcall->key_len;
3516 miss->upcall_type = upcall->type;
3517 miss->odp_in_port = odp_in_port;
3518 list_init(&miss->packets);
3522 miss = existing_miss;
3524 list_push_back(&miss->packets, &upcall->packet->list_node);
3527 /* Process each element in the to-do list, constructing the set of
3528 * operations to batch. */
3530 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3531 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3533 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3535 /* Execute batch. */
3536 for (i = 0; i < n_ops; i++) {
3537 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3539 dpif_operate(backer->dpif, dpif_ops, n_ops);
3541 /* Free memory and update facets. */
3542 for (i = 0; i < n_ops; i++) {
3543 struct flow_miss_op *op = &flow_miss_ops[i];
3545 switch (op->dpif_op.type) {
3546 case DPIF_OP_EXECUTE:
3549 case DPIF_OP_FLOW_PUT:
3550 if (!op->dpif_op.error) {
3551 op->subfacet->path = subfacet_want_path(op->subfacet->slow);
3555 case DPIF_OP_FLOW_DEL:
3561 hmap_destroy(&todo);
3564 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
3565 classify_upcall(const struct dpif_upcall *upcall)
3567 union user_action_cookie cookie;
3569 /* First look at the upcall type. */
3570 switch (upcall->type) {
3571 case DPIF_UC_ACTION:
3577 case DPIF_N_UC_TYPES:
3579 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3583 /* "action" upcalls need a closer look. */
3584 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3585 switch (cookie.type) {
3586 case USER_ACTION_COOKIE_SFLOW:
3587 return SFLOW_UPCALL;
3589 case USER_ACTION_COOKIE_SLOW_PATH:
3592 case USER_ACTION_COOKIE_UNSPEC:
3594 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
3600 handle_sflow_upcall(struct dpif_backer *backer,
3601 const struct dpif_upcall *upcall)
3603 struct ofproto_dpif *ofproto;
3604 union user_action_cookie cookie;
3605 enum odp_key_fitness fitness;
3606 struct ofport_dpif *port;
3607 ovs_be16 initial_tci;
3609 uint32_t odp_in_port;
3611 fitness = odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
3613 port = odp_port_to_ofport(backer, flow.in_port);
3618 ofproto = ofproto_dpif_cast(port->up.ofproto);
3619 if (!ofproto->sflow) {
3623 odp_in_port = flow.in_port;
3624 flow.in_port = port->up.ofp_port;
3625 fitness = ofproto_dpif_vsp_adjust(ofproto, fitness, &flow,
3626 &initial_tci, upcall->packet);
3627 if (fitness == ODP_FIT_ERROR) {
3631 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3632 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3633 odp_in_port, &cookie);
3637 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3639 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3640 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3641 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3646 assert(max_batch <= FLOW_MISS_MAX_BATCH);
3649 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3650 struct dpif_upcall *upcall = &misses[n_misses];
3651 struct ofpbuf *buf = &miss_bufs[n_misses];
3654 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3655 sizeof miss_buf_stubs[n_misses]);
3656 error = dpif_recv(backer->dpif, upcall, buf);
3662 switch (classify_upcall(upcall)) {
3664 /* Handle it later. */
3669 handle_sflow_upcall(backer, upcall);
3679 /* Handle deferred MISS_UPCALL processing. */
3680 handle_miss_upcalls(backer, misses, n_misses);
3681 for (i = 0; i < n_misses; i++) {
3682 ofpbuf_uninit(&miss_bufs[i]);
3688 /* Flow expiration. */
3690 static int subfacet_max_idle(const struct ofproto_dpif *);
3691 static void update_stats(struct dpif_backer *);
3692 static void rule_expire(struct rule_dpif *);
3693 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
3695 /* This function is called periodically by run(). Its job is to collect
3696 * updates for the flows that have been installed into the datapath, most
3697 * importantly when they last were used, and then use that information to
3698 * expire flows that have not been used recently.
3700 * Returns the number of milliseconds after which it should be called again. */
3702 expire(struct dpif_backer *backer)
3704 struct ofproto_dpif *ofproto;
3705 int max_idle = INT32_MAX;
3707 /* Update stats for each flow in the backer. */
3708 update_stats(backer);
3710 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3711 struct rule_dpif *rule, *next_rule;
3712 struct oftable *table;
3715 if (ofproto->backer != backer) {
3719 /* Expire subfacets that have been idle too long. */
3720 dp_max_idle = subfacet_max_idle(ofproto);
3721 expire_subfacets(ofproto, dp_max_idle);
3723 max_idle = MIN(max_idle, dp_max_idle);
3725 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3727 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
3728 struct cls_cursor cursor;
3730 cls_cursor_init(&cursor, &table->cls, NULL);
3731 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
3736 /* All outstanding data in existing flows has been accounted, so it's a
3737 * good time to do bond rebalancing. */
3738 if (ofproto->has_bonded_bundles) {
3739 struct ofbundle *bundle;
3741 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3743 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
3749 return MIN(max_idle, 1000);
3752 /* Updates flow table statistics given that the datapath just reported 'stats'
3753 * as 'subfacet''s statistics. */
3755 update_subfacet_stats(struct subfacet *subfacet,
3756 const struct dpif_flow_stats *stats)
3758 struct facet *facet = subfacet->facet;
3760 if (stats->n_packets >= subfacet->dp_packet_count) {
3761 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
3762 facet->packet_count += extra;
3764 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3767 if (stats->n_bytes >= subfacet->dp_byte_count) {
3768 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
3770 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3773 subfacet->dp_packet_count = stats->n_packets;
3774 subfacet->dp_byte_count = stats->n_bytes;
3776 facet->tcp_flags |= stats->tcp_flags;
3778 subfacet_update_time(subfacet, stats->used);
3779 if (facet->accounted_bytes < facet->byte_count) {
3781 facet_account(facet);
3782 facet->accounted_bytes = facet->byte_count;
3784 facet_push_stats(facet);
3787 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3788 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3790 delete_unexpected_flow(struct ofproto_dpif *ofproto,
3791 const struct nlattr *key, size_t key_len)
3793 if (!VLOG_DROP_WARN(&rl)) {
3797 odp_flow_key_format(key, key_len, &s);
3798 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
3802 COVERAGE_INC(facet_unexpected);
3803 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
3806 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3808 * This function also pushes statistics updates to rules which each facet
3809 * resubmits into. Generally these statistics will be accurate. However, if a
3810 * facet changes the rule it resubmits into at some time in between
3811 * update_stats() runs, it is possible that statistics accrued to the
3812 * old rule will be incorrectly attributed to the new rule. This could be
3813 * avoided by calling update_stats() whenever rules are created or
3814 * deleted. However, the performance impact of making so many calls to the
3815 * datapath do not justify the benefit of having perfectly accurate statistics.
3818 update_stats(struct dpif_backer *backer)
3820 const struct dpif_flow_stats *stats;
3821 struct dpif_flow_dump dump;
3822 const struct nlattr *key;
3825 dpif_flow_dump_start(&dump, backer->dpif);
3826 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
3828 struct subfacet *subfacet;
3829 enum odp_key_fitness fitness;
3830 struct ofproto_dpif *ofproto;
3831 struct ofport_dpif *port;
3834 fitness = odp_flow_key_to_flow(key, key_len, &flow);
3835 if (fitness == ODP_FIT_ERROR) {
3839 port = odp_port_to_ofport(backer, flow.in_port);
3841 /* This flow is for a port for which we couldn't associate an
3842 * ofproto. This can happen if a port is removed while
3843 * traffic is being received. Ignore this flow, since it
3844 * will get timed out. */
3848 ofproto = ofproto_dpif_cast(port->up.ofproto);
3849 flow.in_port = port->up.ofp_port;
3850 key_hash = odp_flow_key_hash(key, key_len);
3852 subfacet = subfacet_find(ofproto, key, key_len, key_hash, &flow);
3853 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3855 update_subfacet_stats(subfacet, stats);
3859 /* Stats are updated per-packet. */
3862 case SF_NOT_INSTALLED:
3864 delete_unexpected_flow(ofproto, key, key_len);
3868 dpif_flow_dump_done(&dump);
3871 /* Calculates and returns the number of milliseconds of idle time after which
3872 * subfacets should expire from the datapath. When a subfacet expires, we fold
3873 * its statistics into its facet, and when a facet's last subfacet expires, we
3874 * fold its statistic into its rule. */
3876 subfacet_max_idle(const struct ofproto_dpif *ofproto)
3879 * Idle time histogram.
3881 * Most of the time a switch has a relatively small number of subfacets.
3882 * When this is the case we might as well keep statistics for all of them
3883 * in userspace and to cache them in the kernel datapath for performance as
3886 * As the number of subfacets increases, the memory required to maintain
3887 * statistics about them in userspace and in the kernel becomes
3888 * significant. However, with a large number of subfacets it is likely
3889 * that only a few of them are "heavy hitters" that consume a large amount
3890 * of bandwidth. At this point, only heavy hitters are worth caching in
3891 * the kernel and maintaining in userspaces; other subfacets we can
3894 * The technique used to compute the idle time is to build a histogram with
3895 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3896 * that is installed in the kernel gets dropped in the appropriate bucket.
3897 * After the histogram has been built, we compute the cutoff so that only
3898 * the most-recently-used 1% of subfacets (but at least
3899 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3900 * the most-recently-used bucket of subfacets is kept, so actually an
3901 * arbitrary number of subfacets can be kept in any given expiration run
3902 * (though the next run will delete most of those unless they receive
3905 * This requires a second pass through the subfacets, in addition to the
3906 * pass made by update_stats(), because the former function never looks at
3907 * uninstallable subfacets.
3909 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3910 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3911 int buckets[N_BUCKETS] = { 0 };
3912 int total, subtotal, bucket;
3913 struct subfacet *subfacet;
3917 total = hmap_count(&ofproto->subfacets);
3918 if (total <= ofproto->up.flow_eviction_threshold) {
3919 return N_BUCKETS * BUCKET_WIDTH;
3922 /* Build histogram. */
3924 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3925 long long int idle = now - subfacet->used;
3926 int bucket = (idle <= 0 ? 0
3927 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3928 : (unsigned int) idle / BUCKET_WIDTH);
3932 /* Find the first bucket whose flows should be expired. */
3933 subtotal = bucket = 0;
3935 subtotal += buckets[bucket++];
3936 } while (bucket < N_BUCKETS &&
3937 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3939 if (VLOG_IS_DBG_ENABLED()) {
3943 ds_put_cstr(&s, "keep");
3944 for (i = 0; i < N_BUCKETS; i++) {
3946 ds_put_cstr(&s, ", drop");
3949 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3952 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3956 return bucket * BUCKET_WIDTH;
3960 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3962 /* Cutoff time for most flows. */
3963 long long int normal_cutoff = time_msec() - dp_max_idle;
3965 /* We really want to keep flows for special protocols around, so use a more
3966 * conservative cutoff. */
3967 long long int special_cutoff = time_msec() - 10000;
3969 struct subfacet *subfacet, *next_subfacet;
3970 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3974 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3975 &ofproto->subfacets) {
3976 long long int cutoff;
3978 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
3981 if (subfacet->used < cutoff) {
3982 if (subfacet->path != SF_NOT_INSTALLED) {
3983 batch[n_batch++] = subfacet;
3984 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
3985 subfacet_destroy_batch(ofproto, batch, n_batch);
3989 subfacet_destroy(subfacet);
3995 subfacet_destroy_batch(ofproto, batch, n_batch);
3999 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4000 * then delete it entirely. */
4002 rule_expire(struct rule_dpif *rule)
4004 struct facet *facet, *next_facet;
4008 if (rule->up.pending) {
4009 /* We'll have to expire it later. */
4013 /* Has 'rule' expired? */
4015 if (rule->up.hard_timeout
4016 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4017 reason = OFPRR_HARD_TIMEOUT;
4018 } else if (rule->up.idle_timeout
4019 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4020 reason = OFPRR_IDLE_TIMEOUT;
4025 COVERAGE_INC(ofproto_dpif_expired);
4027 /* Update stats. (This is a no-op if the rule expired due to an idle
4028 * timeout, because that only happens when the rule has no facets left.) */
4029 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4030 facet_remove(facet);
4033 /* Get rid of the rule. */
4034 ofproto_rule_expire(&rule->up, reason);
4039 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
4041 * The caller must already have determined that no facet with an identical
4042 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
4043 * the ofproto's classifier table.
4045 * 'hash' must be the return value of flow_hash(flow, 0).
4047 * The facet will initially have no subfacets. The caller should create (at
4048 * least) one subfacet with subfacet_create(). */
4049 static struct facet *
4050 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
4052 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4053 struct facet *facet;
4055 facet = xzalloc(sizeof *facet);
4056 facet->used = time_msec();
4057 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4058 list_push_back(&rule->facets, &facet->list_node);
4060 facet->flow = *flow;
4061 list_init(&facet->subfacets);
4062 netflow_flow_init(&facet->nf_flow);
4063 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4069 facet_free(struct facet *facet)
4074 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4075 * 'packet', which arrived on 'in_port'.
4077 * Takes ownership of 'packet'. */
4079 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4080 const struct nlattr *odp_actions, size_t actions_len,
4081 struct ofpbuf *packet)
4083 struct odputil_keybuf keybuf;
4087 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4088 odp_flow_key_from_flow(&key, flow,
4089 ofp_port_to_odp_port(ofproto, flow->in_port));
4091 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4092 odp_actions, actions_len, packet);
4094 ofpbuf_delete(packet);
4098 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4100 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4101 * rule's statistics, via subfacet_uninstall().
4103 * - Removes 'facet' from its rule and from ofproto->facets.
4106 facet_remove(struct facet *facet)
4108 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4109 struct subfacet *subfacet, *next_subfacet;
4111 assert(!list_is_empty(&facet->subfacets));
4113 /* First uninstall all of the subfacets to get final statistics. */
4114 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4115 subfacet_uninstall(subfacet);
4118 /* Flush the final stats to the rule.
4120 * This might require us to have at least one subfacet around so that we
4121 * can use its actions for accounting in facet_account(), which is why we
4122 * have uninstalled but not yet destroyed the subfacets. */
4123 facet_flush_stats(facet);
4125 /* Now we're really all done so destroy everything. */
4126 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4127 &facet->subfacets) {
4128 subfacet_destroy__(subfacet);
4130 hmap_remove(&ofproto->facets, &facet->hmap_node);
4131 list_remove(&facet->list_node);
4135 /* Feed information from 'facet' back into the learning table to keep it in
4136 * sync with what is actually flowing through the datapath. */
4138 facet_learn(struct facet *facet)
4140 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4141 struct action_xlate_ctx ctx;
4143 if (!facet->has_learn
4144 && !facet->has_normal
4145 && (!facet->has_fin_timeout
4146 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4150 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4151 facet->flow.vlan_tci,
4152 facet->rule, facet->tcp_flags, NULL);
4153 ctx.may_learn = true;
4154 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
4155 facet->rule->up.ofpacts_len);
4159 facet_account(struct facet *facet)
4161 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4162 struct subfacet *subfacet;
4163 const struct nlattr *a;
4168 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
4171 n_bytes = facet->byte_count - facet->accounted_bytes;
4173 /* This loop feeds byte counters to bond_account() for rebalancing to use
4174 * as a basis. We also need to track the actual VLAN on which the packet
4175 * is going to be sent to ensure that it matches the one passed to
4176 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4179 * We use the actions from an arbitrary subfacet because they should all
4180 * be equally valid for our purpose. */
4181 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
4182 struct subfacet, list_node);
4183 vlan_tci = facet->flow.vlan_tci;
4184 NL_ATTR_FOR_EACH_UNSAFE (a, left,
4185 subfacet->actions, subfacet->actions_len) {
4186 const struct ovs_action_push_vlan *vlan;
4187 struct ofport_dpif *port;
4189 switch (nl_attr_type(a)) {
4190 case OVS_ACTION_ATTR_OUTPUT:
4191 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4192 if (port && port->bundle && port->bundle->bond) {
4193 bond_account(port->bundle->bond, &facet->flow,
4194 vlan_tci_to_vid(vlan_tci), n_bytes);
4198 case OVS_ACTION_ATTR_POP_VLAN:
4199 vlan_tci = htons(0);
4202 case OVS_ACTION_ATTR_PUSH_VLAN:
4203 vlan = nl_attr_get(a);
4204 vlan_tci = vlan->vlan_tci;
4210 /* Returns true if the only action for 'facet' is to send to the controller.
4211 * (We don't report NetFlow expiration messages for such facets because they
4212 * are just part of the control logic for the network, not real traffic). */
4214 facet_is_controller_flow(struct facet *facet)
4217 const struct rule *rule = &facet->rule->up;
4218 const struct ofpact *ofpacts = rule->ofpacts;
4219 size_t ofpacts_len = rule->ofpacts_len;
4221 if (ofpacts_len > 0 &&
4222 ofpacts->type == OFPACT_CONTROLLER &&
4223 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4230 /* Folds all of 'facet''s statistics into its rule. Also updates the
4231 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4232 * 'facet''s statistics in the datapath should have been zeroed and folded into
4233 * its packet and byte counts before this function is called. */
4235 facet_flush_stats(struct facet *facet)
4237 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4238 struct subfacet *subfacet;
4240 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4241 assert(!subfacet->dp_byte_count);
4242 assert(!subfacet->dp_packet_count);
4245 facet_push_stats(facet);
4246 if (facet->accounted_bytes < facet->byte_count) {
4247 facet_account(facet);
4248 facet->accounted_bytes = facet->byte_count;
4251 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4252 struct ofexpired expired;
4253 expired.flow = facet->flow;
4254 expired.packet_count = facet->packet_count;
4255 expired.byte_count = facet->byte_count;
4256 expired.used = facet->used;
4257 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4260 facet->rule->packet_count += facet->packet_count;
4261 facet->rule->byte_count += facet->byte_count;
4263 /* Reset counters to prevent double counting if 'facet' ever gets
4265 facet_reset_counters(facet);
4267 netflow_flow_clear(&facet->nf_flow);
4268 facet->tcp_flags = 0;
4271 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4272 * Returns it if found, otherwise a null pointer.
4274 * 'hash' must be the return value of flow_hash(flow, 0).
4276 * The returned facet might need revalidation; use facet_lookup_valid()
4277 * instead if that is important. */
4278 static struct facet *
4279 facet_find(struct ofproto_dpif *ofproto,
4280 const struct flow *flow, uint32_t hash)
4282 struct facet *facet;
4284 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4285 if (flow_equal(flow, &facet->flow)) {
4293 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4294 * Returns it if found, otherwise a null pointer.
4296 * 'hash' must be the return value of flow_hash(flow, 0).
4298 * The returned facet is guaranteed to be valid. */
4299 static struct facet *
4300 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4303 struct facet *facet;
4305 facet = facet_find(ofproto, flow, hash);
4307 && (ofproto->need_revalidate
4308 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))) {
4309 facet_revalidate(facet);
4316 subfacet_path_to_string(enum subfacet_path path)
4319 case SF_NOT_INSTALLED:
4320 return "not installed";
4322 return "in fast path";
4324 return "in slow path";
4330 /* Returns the path in which a subfacet should be installed if its 'slow'
4331 * member has the specified value. */
4332 static enum subfacet_path
4333 subfacet_want_path(enum slow_path_reason slow)
4335 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
4338 /* Returns true if 'subfacet' needs to have its datapath flow updated,
4339 * supposing that its actions have been recalculated as 'want_actions' and that
4340 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
4342 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
4343 const struct ofpbuf *want_actions)
4345 enum subfacet_path want_path = subfacet_want_path(slow);
4346 return (want_path != subfacet->path
4347 || (want_path == SF_FAST_PATH
4348 && (subfacet->actions_len != want_actions->size
4349 || memcmp(subfacet->actions, want_actions->data,
4350 subfacet->actions_len))));
4354 facet_check_consistency(struct facet *facet)
4356 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4358 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4360 uint64_t odp_actions_stub[1024 / 8];
4361 struct ofpbuf odp_actions;
4363 struct rule_dpif *rule;
4364 struct subfacet *subfacet;
4365 bool may_log = false;
4368 /* Check the rule for consistency. */
4369 rule = rule_dpif_lookup(ofproto, &facet->flow);
4370 ok = rule == facet->rule;
4372 may_log = !VLOG_DROP_WARN(&rl);
4377 flow_format(&s, &facet->flow);
4378 ds_put_format(&s, ": facet associated with wrong rule (was "
4379 "table=%"PRIu8",", facet->rule->up.table_id);
4380 cls_rule_format(&facet->rule->up.cr, &s);
4381 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4383 cls_rule_format(&rule->up.cr, &s);
4384 ds_put_char(&s, ')');
4386 VLOG_WARN("%s", ds_cstr(&s));
4391 /* Check the datapath actions for consistency. */
4392 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4393 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4394 enum subfacet_path want_path;
4395 struct odputil_keybuf keybuf;
4396 struct action_xlate_ctx ctx;
4400 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4401 subfacet->initial_tci, rule, 0, NULL);
4402 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
4405 if (subfacet->path == SF_NOT_INSTALLED) {
4406 /* This only happens if the datapath reported an error when we
4407 * tried to install the flow. Don't flag another error here. */
4411 want_path = subfacet_want_path(subfacet->slow);
4412 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
4413 /* The actions for slow-path flows may legitimately vary from one
4414 * packet to the next. We're done. */
4418 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
4422 /* Inconsistency! */
4424 may_log = !VLOG_DROP_WARN(&rl);
4428 /* Rate-limited, skip reporting. */
4433 subfacet_get_key(subfacet, &keybuf, &key);
4434 odp_flow_key_format(key.data, key.size, &s);
4436 ds_put_cstr(&s, ": inconsistency in subfacet");
4437 if (want_path != subfacet->path) {
4438 enum odp_key_fitness fitness = subfacet->key_fitness;
4440 ds_put_format(&s, " (%s, fitness=%s)",
4441 subfacet_path_to_string(subfacet->path),
4442 odp_key_fitness_to_string(fitness));
4443 ds_put_format(&s, " (should have been %s)",
4444 subfacet_path_to_string(want_path));
4445 } else if (want_path == SF_FAST_PATH) {
4446 ds_put_cstr(&s, " (actions were: ");
4447 format_odp_actions(&s, subfacet->actions,
4448 subfacet->actions_len);
4449 ds_put_cstr(&s, ") (correct actions: ");
4450 format_odp_actions(&s, odp_actions.data, odp_actions.size);
4451 ds_put_char(&s, ')');
4453 ds_put_cstr(&s, " (actions: ");
4454 format_odp_actions(&s, subfacet->actions,
4455 subfacet->actions_len);
4456 ds_put_char(&s, ')');
4458 VLOG_WARN("%s", ds_cstr(&s));
4461 ofpbuf_uninit(&odp_actions);
4466 /* Re-searches the classifier for 'facet':
4468 * - If the rule found is different from 'facet''s current rule, moves
4469 * 'facet' to the new rule and recompiles its actions.
4471 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4472 * where it is and recompiles its actions anyway. */
4474 facet_revalidate(struct facet *facet)
4476 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4478 struct nlattr *odp_actions;
4481 struct actions *new_actions;
4483 struct action_xlate_ctx ctx;
4484 uint64_t odp_actions_stub[1024 / 8];
4485 struct ofpbuf odp_actions;
4487 struct rule_dpif *new_rule;
4488 struct subfacet *subfacet;
4491 COVERAGE_INC(facet_revalidate);
4493 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
4495 /* Calculate new datapath actions.
4497 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4498 * emit a NetFlow expiration and, if so, we need to have the old state
4499 * around to properly compose it. */
4501 /* If the datapath actions changed or the installability changed,
4502 * then we need to talk to the datapath. */
4505 memset(&ctx, 0, sizeof ctx);
4506 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4507 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4508 enum slow_path_reason slow;
4510 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4511 subfacet->initial_tci, new_rule, 0, NULL);
4512 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
4515 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4516 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
4517 struct dpif_flow_stats stats;
4519 subfacet_install(subfacet,
4520 odp_actions.data, odp_actions.size, &stats, slow);
4521 subfacet_update_stats(subfacet, &stats);
4524 new_actions = xcalloc(list_size(&facet->subfacets),
4525 sizeof *new_actions);
4527 new_actions[i].odp_actions = xmemdup(odp_actions.data,
4529 new_actions[i].actions_len = odp_actions.size;
4534 ofpbuf_uninit(&odp_actions);
4537 facet_flush_stats(facet);
4540 /* Update 'facet' now that we've taken care of all the old state. */
4541 facet->tags = ctx.tags;
4542 facet->nf_flow.output_iface = ctx.nf_output_iface;
4543 facet->has_learn = ctx.has_learn;
4544 facet->has_normal = ctx.has_normal;
4545 facet->has_fin_timeout = ctx.has_fin_timeout;
4546 facet->mirrors = ctx.mirrors;
4549 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4550 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4552 if (new_actions && new_actions[i].odp_actions) {
4553 free(subfacet->actions);
4554 subfacet->actions = new_actions[i].odp_actions;
4555 subfacet->actions_len = new_actions[i].actions_len;
4561 if (facet->rule != new_rule) {
4562 COVERAGE_INC(facet_changed_rule);
4563 list_remove(&facet->list_node);
4564 list_push_back(&new_rule->facets, &facet->list_node);
4565 facet->rule = new_rule;
4566 facet->used = new_rule->up.created;
4567 facet->prev_used = facet->used;
4571 /* Updates 'facet''s used time. Caller is responsible for calling
4572 * facet_push_stats() to update the flows which 'facet' resubmits into. */
4574 facet_update_time(struct facet *facet, long long int used)
4576 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4577 if (used > facet->used) {
4579 ofproto_rule_update_used(&facet->rule->up, used);
4580 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
4585 facet_reset_counters(struct facet *facet)
4587 facet->packet_count = 0;
4588 facet->byte_count = 0;
4589 facet->prev_packet_count = 0;
4590 facet->prev_byte_count = 0;
4591 facet->accounted_bytes = 0;
4595 facet_push_stats(struct facet *facet)
4597 struct dpif_flow_stats stats;
4599 assert(facet->packet_count >= facet->prev_packet_count);
4600 assert(facet->byte_count >= facet->prev_byte_count);
4601 assert(facet->used >= facet->prev_used);
4603 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4604 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4605 stats.used = facet->used;
4606 stats.tcp_flags = 0;
4608 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
4609 facet->prev_packet_count = facet->packet_count;
4610 facet->prev_byte_count = facet->byte_count;
4611 facet->prev_used = facet->used;
4613 flow_push_stats(facet->rule, &facet->flow, &stats);
4615 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
4616 facet->mirrors, stats.n_packets, stats.n_bytes);
4621 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4623 rule->packet_count += stats->n_packets;
4624 rule->byte_count += stats->n_bytes;
4625 ofproto_rule_update_used(&rule->up, stats->used);
4628 /* Pushes flow statistics to the rules which 'flow' resubmits into given
4629 * 'rule''s actions and mirrors. */
4631 flow_push_stats(struct rule_dpif *rule,
4632 const struct flow *flow, const struct dpif_flow_stats *stats)
4634 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4635 struct action_xlate_ctx ctx;
4637 ofproto_rule_update_used(&rule->up, stats->used);
4639 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule,
4641 ctx.resubmit_stats = stats;
4642 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
4643 rule->up.ofpacts_len);
4648 static struct subfacet *
4649 subfacet_find(struct ofproto_dpif *ofproto,
4650 const struct nlattr *key, size_t key_len, uint32_t key_hash,
4651 const struct flow *flow)
4653 struct subfacet *subfacet;
4655 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4656 &ofproto->subfacets) {
4658 ? (subfacet->key_len == key_len
4659 && !memcmp(key, subfacet->key, key_len))
4660 : flow_equal(flow, &subfacet->facet->flow)) {
4668 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4669 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4670 * existing subfacet if there is one, otherwise creates and returns a
4673 * If the returned subfacet is new, then subfacet->actions will be NULL, in
4674 * which case the caller must populate the actions with
4675 * subfacet_make_actions(). */
4676 static struct subfacet *
4677 subfacet_create(struct facet *facet, struct flow_miss *miss,
4680 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4681 enum odp_key_fitness key_fitness = miss->key_fitness;
4682 const struct nlattr *key = miss->key;
4683 size_t key_len = miss->key_len;
4685 struct subfacet *subfacet;
4687 key_hash = odp_flow_key_hash(key, key_len);
4689 if (list_is_empty(&facet->subfacets)) {
4690 subfacet = &facet->one_subfacet;
4692 subfacet = subfacet_find(ofproto, key, key_len, key_hash,
4695 if (subfacet->facet == facet) {
4699 /* This shouldn't happen. */
4700 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4701 subfacet_destroy(subfacet);
4704 subfacet = xmalloc(sizeof *subfacet);
4707 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
4708 list_push_back(&facet->subfacets, &subfacet->list_node);
4709 subfacet->facet = facet;
4710 subfacet->key_fitness = key_fitness;
4711 if (key_fitness != ODP_FIT_PERFECT) {
4712 subfacet->key = xmemdup(key, key_len);
4713 subfacet->key_len = key_len;
4715 subfacet->key = NULL;
4716 subfacet->key_len = 0;
4718 subfacet->used = now;
4719 subfacet->dp_packet_count = 0;
4720 subfacet->dp_byte_count = 0;
4721 subfacet->actions_len = 0;
4722 subfacet->actions = NULL;
4723 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
4726 subfacet->path = SF_NOT_INSTALLED;
4727 subfacet->initial_tci = miss->initial_tci;
4728 subfacet->odp_in_port = miss->odp_in_port;
4733 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4734 * its facet within 'ofproto', and frees it. */
4736 subfacet_destroy__(struct subfacet *subfacet)
4738 struct facet *facet = subfacet->facet;
4739 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4741 subfacet_uninstall(subfacet);
4742 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
4743 list_remove(&subfacet->list_node);
4744 free(subfacet->key);
4745 free(subfacet->actions);
4746 if (subfacet != &facet->one_subfacet) {
4751 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4752 * last remaining subfacet in its facet destroys the facet too. */
4754 subfacet_destroy(struct subfacet *subfacet)
4756 struct facet *facet = subfacet->facet;
4758 if (list_is_singleton(&facet->subfacets)) {
4759 /* facet_remove() needs at least one subfacet (it will remove it). */
4760 facet_remove(facet);
4762 subfacet_destroy__(subfacet);
4767 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
4768 struct subfacet **subfacets, int n)
4770 struct odputil_keybuf keybufs[SUBFACET_DESTROY_MAX_BATCH];
4771 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4772 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4773 struct ofpbuf keys[SUBFACET_DESTROY_MAX_BATCH];
4774 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4777 for (i = 0; i < n; i++) {
4778 ops[i].type = DPIF_OP_FLOW_DEL;
4779 subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]);
4780 ops[i].u.flow_del.key = keys[i].data;
4781 ops[i].u.flow_del.key_len = keys[i].size;
4782 ops[i].u.flow_del.stats = &stats[i];
4786 dpif_operate(ofproto->backer->dpif, opsp, n);
4787 for (i = 0; i < n; i++) {
4788 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4789 subfacets[i]->path = SF_NOT_INSTALLED;
4790 subfacet_destroy(subfacets[i]);
4794 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
4795 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
4796 * for use as temporary storage. */
4798 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
4802 if (!subfacet->key) {
4803 struct flow *flow = &subfacet->facet->flow;
4805 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
4806 odp_flow_key_from_flow(key, flow, subfacet->odp_in_port);
4808 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
4812 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
4813 * Translates the actions into 'odp_actions', which the caller must have
4814 * initialized and is responsible for uninitializing. */
4816 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
4817 struct ofpbuf *odp_actions)
4819 struct facet *facet = subfacet->facet;
4820 struct rule_dpif *rule = facet->rule;
4821 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4823 struct action_xlate_ctx ctx;
4825 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
4827 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
4828 facet->tags = ctx.tags;
4829 facet->has_learn = ctx.has_learn;
4830 facet->has_normal = ctx.has_normal;
4831 facet->has_fin_timeout = ctx.has_fin_timeout;
4832 facet->nf_flow.output_iface = ctx.nf_output_iface;
4833 facet->mirrors = ctx.mirrors;
4835 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4836 if (subfacet->actions_len != odp_actions->size
4837 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
4838 free(subfacet->actions);
4839 subfacet->actions_len = odp_actions->size;
4840 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
4844 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4845 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4846 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4847 * since 'subfacet' was last updated.
4849 * Returns 0 if successful, otherwise a positive errno value. */
4851 subfacet_install(struct subfacet *subfacet,
4852 const struct nlattr *actions, size_t actions_len,
4853 struct dpif_flow_stats *stats,
4854 enum slow_path_reason slow)
4856 struct facet *facet = subfacet->facet;
4857 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4858 enum subfacet_path path = subfacet_want_path(slow);
4859 uint64_t slow_path_stub[128 / 8];
4860 struct odputil_keybuf keybuf;
4861 enum dpif_flow_put_flags flags;
4865 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
4867 flags |= DPIF_FP_ZERO_STATS;
4870 if (path == SF_SLOW_PATH) {
4871 compose_slow_path(ofproto, &facet->flow, slow,
4872 slow_path_stub, sizeof slow_path_stub,
4873 &actions, &actions_len);
4876 subfacet_get_key(subfacet, &keybuf, &key);
4877 ret = dpif_flow_put(ofproto->backer->dpif, flags, key.data, key.size,
4878 actions, actions_len, stats);
4881 subfacet_reset_dp_stats(subfacet, stats);
4885 subfacet->path = path;
4891 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
4893 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
4894 stats, subfacet->slow);
4897 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4899 subfacet_uninstall(struct subfacet *subfacet)
4901 if (subfacet->path != SF_NOT_INSTALLED) {
4902 struct rule_dpif *rule = subfacet->facet->rule;
4903 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4904 struct odputil_keybuf keybuf;
4905 struct dpif_flow_stats stats;
4909 subfacet_get_key(subfacet, &keybuf, &key);
4910 error = dpif_flow_del(ofproto->backer->dpif,
4911 key.data, key.size, &stats);
4912 subfacet_reset_dp_stats(subfacet, &stats);
4914 subfacet_update_stats(subfacet, &stats);
4916 subfacet->path = SF_NOT_INSTALLED;
4918 assert(subfacet->dp_packet_count == 0);
4919 assert(subfacet->dp_byte_count == 0);
4923 /* Resets 'subfacet''s datapath statistics counters. This should be called
4924 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4925 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4926 * was reset in the datapath. 'stats' will be modified to include only
4927 * statistics new since 'subfacet' was last updated. */
4929 subfacet_reset_dp_stats(struct subfacet *subfacet,
4930 struct dpif_flow_stats *stats)
4933 && subfacet->dp_packet_count <= stats->n_packets
4934 && subfacet->dp_byte_count <= stats->n_bytes) {
4935 stats->n_packets -= subfacet->dp_packet_count;
4936 stats->n_bytes -= subfacet->dp_byte_count;
4939 subfacet->dp_packet_count = 0;
4940 subfacet->dp_byte_count = 0;
4943 /* Updates 'subfacet''s used time. The caller is responsible for calling
4944 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
4946 subfacet_update_time(struct subfacet *subfacet, long long int used)
4948 if (used > subfacet->used) {
4949 subfacet->used = used;
4950 facet_update_time(subfacet->facet, used);
4954 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4956 * Because of the meaning of a subfacet's counters, it only makes sense to do
4957 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4958 * represents a packet that was sent by hand or if it represents statistics
4959 * that have been cleared out of the datapath. */
4961 subfacet_update_stats(struct subfacet *subfacet,
4962 const struct dpif_flow_stats *stats)
4964 if (stats->n_packets || stats->used > subfacet->used) {
4965 struct facet *facet = subfacet->facet;
4967 subfacet_update_time(subfacet, stats->used);
4968 facet->packet_count += stats->n_packets;
4969 facet->byte_count += stats->n_bytes;
4970 facet->tcp_flags |= stats->tcp_flags;
4971 facet_push_stats(facet);
4972 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
4978 static struct rule_dpif *
4979 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
4981 struct rule_dpif *rule;
4983 rule = rule_dpif_lookup__(ofproto, flow, 0);
4988 return rule_dpif_miss_rule(ofproto, flow);
4991 static struct rule_dpif *
4992 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
4995 struct cls_rule *cls_rule;
4996 struct classifier *cls;
4998 if (table_id >= N_TABLES) {
5002 cls = &ofproto->up.tables[table_id].cls;
5003 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5004 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5005 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5006 * are unavailable. */
5007 struct flow ofpc_normal_flow = *flow;
5008 ofpc_normal_flow.tp_src = htons(0);
5009 ofpc_normal_flow.tp_dst = htons(0);
5010 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5012 cls_rule = classifier_lookup(cls, flow);
5014 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5017 static struct rule_dpif *
5018 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5020 struct ofport_dpif *port;
5022 port = get_ofp_port(ofproto, flow->in_port);
5024 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5025 return ofproto->miss_rule;
5028 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5029 return ofproto->no_packet_in_rule;
5031 return ofproto->miss_rule;
5035 complete_operation(struct rule_dpif *rule)
5037 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5039 rule_invalidate(rule);
5041 struct dpif_completion *c = xmalloc(sizeof *c);
5042 c->op = rule->up.pending;
5043 list_push_back(&ofproto->completions, &c->list_node);
5045 ofoperation_complete(rule->up.pending, 0);
5049 static struct rule *
5052 struct rule_dpif *rule = xmalloc(sizeof *rule);
5057 rule_dealloc(struct rule *rule_)
5059 struct rule_dpif *rule = rule_dpif_cast(rule_);
5064 rule_construct(struct rule *rule_)
5066 struct rule_dpif *rule = rule_dpif_cast(rule_);
5067 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5068 struct rule_dpif *victim;
5071 rule->packet_count = 0;
5072 rule->byte_count = 0;
5074 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5075 if (victim && !list_is_empty(&victim->facets)) {
5076 struct facet *facet;
5078 rule->facets = victim->facets;
5079 list_moved(&rule->facets);
5080 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5081 /* XXX: We're only clearing our local counters here. It's possible
5082 * that quite a few packets are unaccounted for in the datapath
5083 * statistics. These will be accounted to the new rule instead of
5084 * cleared as required. This could be fixed by clearing out the
5085 * datapath statistics for this facet, but currently it doesn't
5087 facet_reset_counters(facet);
5091 /* Must avoid list_moved() in this case. */
5092 list_init(&rule->facets);
5095 table_id = rule->up.table_id;
5097 rule->tag = victim->tag;
5098 } else if (table_id == 0) {
5103 miniflow_expand(&rule->up.cr.match.flow, &flow);
5104 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5105 ofproto->tables[table_id].basis);
5108 complete_operation(rule);
5113 rule_destruct(struct rule *rule_)
5115 struct rule_dpif *rule = rule_dpif_cast(rule_);
5116 struct facet *facet, *next_facet;
5118 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5119 facet_revalidate(facet);
5122 complete_operation(rule);
5126 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5128 struct rule_dpif *rule = rule_dpif_cast(rule_);
5129 struct facet *facet;
5131 /* Start from historical data for 'rule' itself that are no longer tracked
5132 * in facets. This counts, for example, facets that have expired. */
5133 *packets = rule->packet_count;
5134 *bytes = rule->byte_count;
5136 /* Add any statistics that are tracked by facets. This includes
5137 * statistical data recently updated by ofproto_update_stats() as well as
5138 * stats for packets that were executed "by hand" via dpif_execute(). */
5139 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5140 *packets += facet->packet_count;
5141 *bytes += facet->byte_count;
5146 rule_execute(struct rule *rule_, const struct flow *flow,
5147 struct ofpbuf *packet)
5149 struct rule_dpif *rule = rule_dpif_cast(rule_);
5150 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5152 struct dpif_flow_stats stats;
5154 struct action_xlate_ctx ctx;
5155 uint64_t odp_actions_stub[1024 / 8];
5156 struct ofpbuf odp_actions;
5158 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5159 rule_credit_stats(rule, &stats);
5161 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5162 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
5163 rule, stats.tcp_flags, packet);
5164 ctx.resubmit_stats = &stats;
5165 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
5167 execute_odp_actions(ofproto, flow, odp_actions.data,
5168 odp_actions.size, packet);
5170 ofpbuf_uninit(&odp_actions);
5176 rule_modify_actions(struct rule *rule_)
5178 struct rule_dpif *rule = rule_dpif_cast(rule_);
5180 complete_operation(rule);
5183 /* Sends 'packet' out 'ofport'.
5184 * May modify 'packet'.
5185 * Returns 0 if successful, otherwise a positive errno value. */
5187 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5189 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5190 struct ofpbuf key, odp_actions;
5191 struct odputil_keybuf keybuf;
5196 flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow);
5197 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5199 if (odp_port != ofport->odp_port) {
5200 eth_pop_vlan(packet);
5201 flow.vlan_tci = htons(0);
5204 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5205 odp_flow_key_from_flow(&key, &flow,
5206 ofp_port_to_odp_port(ofproto, flow.in_port));
5208 ofpbuf_init(&odp_actions, 32);
5209 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5211 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5212 error = dpif_execute(ofproto->backer->dpif,
5214 odp_actions.data, odp_actions.size,
5216 ofpbuf_uninit(&odp_actions);
5219 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5220 ofproto->up.name, odp_port, strerror(error));
5222 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5226 /* OpenFlow to datapath action translation. */
5228 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5229 struct action_xlate_ctx *);
5230 static void xlate_normal(struct action_xlate_ctx *);
5232 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5233 * The action will state 'slow' as the reason that the action is in the slow
5234 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5235 * dump-flows" output to see why a flow is in the slow path.)
5237 * The 'stub_size' bytes in 'stub' will be used to store the action.
5238 * 'stub_size' must be large enough for the action.
5240 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5243 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5244 enum slow_path_reason slow,
5245 uint64_t *stub, size_t stub_size,
5246 const struct nlattr **actionsp, size_t *actions_lenp)
5248 union user_action_cookie cookie;
5251 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5252 cookie.slow_path.unused = 0;
5253 cookie.slow_path.reason = slow;
5255 ofpbuf_use_stack(&buf, stub, stub_size);
5256 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
5257 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT16_MAX);
5258 odp_put_userspace_action(pid, &cookie, &buf);
5260 put_userspace_action(ofproto, &buf, flow, &cookie);
5262 *actionsp = buf.data;
5263 *actions_lenp = buf.size;
5267 put_userspace_action(const struct ofproto_dpif *ofproto,
5268 struct ofpbuf *odp_actions,
5269 const struct flow *flow,
5270 const union user_action_cookie *cookie)
5274 pid = dpif_port_get_pid(ofproto->backer->dpif,
5275 ofp_port_to_odp_port(ofproto, flow->in_port));
5277 return odp_put_userspace_action(pid, cookie, odp_actions);
5281 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5282 ovs_be16 vlan_tci, uint32_t odp_port,
5283 unsigned int n_outputs, union user_action_cookie *cookie)
5287 cookie->type = USER_ACTION_COOKIE_SFLOW;
5288 cookie->sflow.vlan_tci = vlan_tci;
5290 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5291 * port information") for the interpretation of cookie->output. */
5292 switch (n_outputs) {
5294 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5295 cookie->sflow.output = 0x40000000 | 256;
5299 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5301 cookie->sflow.output = ifindex;
5306 /* 0x80000000 means "multiple output ports. */
5307 cookie->sflow.output = 0x80000000 | n_outputs;
5312 /* Compose SAMPLE action for sFlow. */
5314 compose_sflow_action(const struct ofproto_dpif *ofproto,
5315 struct ofpbuf *odp_actions,
5316 const struct flow *flow,
5319 uint32_t probability;
5320 union user_action_cookie cookie;
5321 size_t sample_offset, actions_offset;
5324 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5328 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5330 /* Number of packets out of UINT_MAX to sample. */
5331 probability = dpif_sflow_get_probability(ofproto->sflow);
5332 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5334 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5335 compose_sflow_cookie(ofproto, htons(0), odp_port,
5336 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5337 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
5339 nl_msg_end_nested(odp_actions, actions_offset);
5340 nl_msg_end_nested(odp_actions, sample_offset);
5341 return cookie_offset;
5344 /* SAMPLE action must be first action in any given list of actions.
5345 * At this point we do not have all information required to build it. So try to
5346 * build sample action as complete as possible. */
5348 add_sflow_action(struct action_xlate_ctx *ctx)
5350 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5352 &ctx->flow, OVSP_NONE);
5353 ctx->sflow_odp_port = 0;
5354 ctx->sflow_n_outputs = 0;
5357 /* Fix SAMPLE action according to data collected while composing ODP actions.
5358 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5359 * USERSPACE action's user-cookie which is required for sflow. */
5361 fix_sflow_action(struct action_xlate_ctx *ctx)
5363 const struct flow *base = &ctx->base_flow;
5364 union user_action_cookie *cookie;
5366 if (!ctx->user_cookie_offset) {
5370 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
5372 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5374 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5375 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5379 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
5382 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5383 uint32_t odp_port = ofp_port_to_odp_port(ctx->ofproto, ofp_port);
5384 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
5385 uint8_t flow_nw_tos = ctx->flow.nw_tos;
5389 struct priority_to_dscp *pdscp;
5391 if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5392 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5394 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5395 xlate_report(ctx, "STP not in forwarding state, skipping output");
5399 pdscp = get_priority(ofport, ctx->flow.skb_priority);
5401 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5402 ctx->flow.nw_tos |= pdscp->dscp;
5405 /* We may not have an ofport record for this port, but it doesn't hurt
5406 * to allow forwarding to it anyhow. Maybe such a port will appear
5407 * later and we're pre-populating the flow table. */
5410 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
5411 ctx->flow.vlan_tci);
5412 if (out_port != odp_port) {
5413 ctx->flow.vlan_tci = htons(0);
5415 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
5416 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
5418 ctx->sflow_odp_port = odp_port;
5419 ctx->sflow_n_outputs++;
5420 ctx->nf_output_iface = ofp_port;
5421 ctx->flow.vlan_tci = flow_vlan_tci;
5422 ctx->flow.nw_tos = flow_nw_tos;
5426 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
5428 compose_output_action__(ctx, ofp_port, true);
5432 xlate_table_action(struct action_xlate_ctx *ctx,
5433 uint16_t in_port, uint8_t table_id, bool may_packet_in)
5435 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
5436 struct ofproto_dpif *ofproto = ctx->ofproto;
5437 struct rule_dpif *rule;
5438 uint16_t old_in_port;
5439 uint8_t old_table_id;
5441 old_table_id = ctx->table_id;
5442 ctx->table_id = table_id;
5444 /* Look up a flow with 'in_port' as the input port. */
5445 old_in_port = ctx->flow.in_port;
5446 ctx->flow.in_port = in_port;
5447 rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id);
5450 if (table_id > 0 && table_id < N_TABLES) {
5451 struct table_dpif *table = &ofproto->tables[table_id];
5452 if (table->other_table) {
5453 ctx->tags |= (rule && rule->tag
5455 : rule_calculate_tag(&ctx->flow,
5456 &table->other_table->mask,
5461 /* Restore the original input port. Otherwise OFPP_NORMAL and
5462 * OFPP_IN_PORT will have surprising behavior. */
5463 ctx->flow.in_port = old_in_port;
5465 if (ctx->resubmit_hook) {
5466 ctx->resubmit_hook(ctx, rule);
5469 if (rule == NULL && may_packet_in) {
5471 * check if table configuration flags
5472 * OFPTC_TABLE_MISS_CONTROLLER, default.
5473 * OFPTC_TABLE_MISS_CONTINUE,
5474 * OFPTC_TABLE_MISS_DROP
5475 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
5477 rule = rule_dpif_miss_rule(ofproto, &ctx->flow);
5481 struct rule_dpif *old_rule = ctx->rule;
5483 if (ctx->resubmit_stats) {
5484 rule_credit_stats(rule, ctx->resubmit_stats);
5489 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
5490 ctx->rule = old_rule;
5494 ctx->table_id = old_table_id;
5496 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5498 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
5499 MAX_RESUBMIT_RECURSION);
5500 ctx->max_resubmit_trigger = true;
5505 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
5506 const struct ofpact_resubmit *resubmit)
5511 in_port = resubmit->in_port;
5512 if (in_port == OFPP_IN_PORT) {
5513 in_port = ctx->flow.in_port;
5516 table_id = resubmit->table_id;
5517 if (table_id == 255) {
5518 table_id = ctx->table_id;
5521 xlate_table_action(ctx, in_port, table_id, false);
5525 flood_packets(struct action_xlate_ctx *ctx, bool all)
5527 struct ofport_dpif *ofport;
5529 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
5530 uint16_t ofp_port = ofport->up.ofp_port;
5532 if (ofp_port == ctx->flow.in_port) {
5537 compose_output_action__(ctx, ofp_port, false);
5538 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
5539 compose_output_action(ctx, ofp_port);
5543 ctx->nf_output_iface = NF_OUT_FLOOD;
5547 execute_controller_action(struct action_xlate_ctx *ctx, int len,
5548 enum ofp_packet_in_reason reason,
5549 uint16_t controller_id)
5551 struct ofputil_packet_in pin;
5552 struct ofpbuf *packet;
5554 ctx->slow |= SLOW_CONTROLLER;
5559 packet = ofpbuf_clone(ctx->packet);
5561 if (packet->l2 && packet->l3) {
5562 struct eth_header *eh;
5564 eth_pop_vlan(packet);
5567 /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
5568 * LLC frame. Calculating the Ethernet type of these frames is more
5569 * trouble than seems appropriate for a simple assertion. */
5570 assert(ntohs(eh->eth_type) < ETH_TYPE_MIN
5571 || eh->eth_type == ctx->flow.dl_type);
5573 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
5574 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
5576 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
5577 eth_push_vlan(packet, ctx->flow.vlan_tci);
5581 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5582 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
5583 ctx->flow.nw_tos, ctx->flow.nw_ttl);
5587 if (ctx->flow.nw_proto == IPPROTO_TCP) {
5588 packet_set_tcp_port(packet, ctx->flow.tp_src,
5590 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
5591 packet_set_udp_port(packet, ctx->flow.tp_src,
5598 pin.packet = packet->data;
5599 pin.packet_len = packet->size;
5600 pin.reason = reason;
5601 pin.controller_id = controller_id;
5602 pin.table_id = ctx->table_id;
5603 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
5606 flow_get_metadata(&ctx->flow, &pin.fmd);
5608 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
5609 ofpbuf_delete(packet);
5613 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
5615 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
5616 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
5620 if (ctx->flow.nw_ttl > 1) {
5626 for (i = 0; i < ids->n_controllers; i++) {
5627 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
5631 /* Stop processing for current table. */
5637 xlate_output_action(struct action_xlate_ctx *ctx,
5638 uint16_t port, uint16_t max_len, bool may_packet_in)
5640 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
5642 ctx->nf_output_iface = NF_OUT_DROP;
5646 compose_output_action(ctx, ctx->flow.in_port);
5649 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
5655 flood_packets(ctx, false);
5658 flood_packets(ctx, true);
5660 case OFPP_CONTROLLER:
5661 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
5667 if (port != ctx->flow.in_port) {
5668 compose_output_action(ctx, port);
5670 xlate_report(ctx, "skipping output to input port");
5675 if (prev_nf_output_iface == NF_OUT_FLOOD) {
5676 ctx->nf_output_iface = NF_OUT_FLOOD;
5677 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
5678 ctx->nf_output_iface = prev_nf_output_iface;
5679 } else if (prev_nf_output_iface != NF_OUT_DROP &&
5680 ctx->nf_output_iface != NF_OUT_FLOOD) {
5681 ctx->nf_output_iface = NF_OUT_MULTI;
5686 xlate_output_reg_action(struct action_xlate_ctx *ctx,
5687 const struct ofpact_output_reg *or)
5689 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
5690 if (port <= UINT16_MAX) {
5691 xlate_output_action(ctx, port, or->max_len, false);
5696 xlate_enqueue_action(struct action_xlate_ctx *ctx,
5697 const struct ofpact_enqueue *enqueue)
5699 uint16_t ofp_port = enqueue->port;
5700 uint32_t queue_id = enqueue->queue;
5701 uint32_t flow_priority, priority;
5704 /* Translate queue to priority. */
5705 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
5706 queue_id, &priority);
5708 /* Fall back to ordinary output action. */
5709 xlate_output_action(ctx, enqueue->port, 0, false);
5713 /* Check output port. */
5714 if (ofp_port == OFPP_IN_PORT) {
5715 ofp_port = ctx->flow.in_port;
5716 } else if (ofp_port == ctx->flow.in_port) {
5720 /* Add datapath actions. */
5721 flow_priority = ctx->flow.skb_priority;
5722 ctx->flow.skb_priority = priority;
5723 compose_output_action(ctx, ofp_port);
5724 ctx->flow.skb_priority = flow_priority;
5726 /* Update NetFlow output port. */
5727 if (ctx->nf_output_iface == NF_OUT_DROP) {
5728 ctx->nf_output_iface = ofp_port;
5729 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
5730 ctx->nf_output_iface = NF_OUT_MULTI;
5735 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
5737 uint32_t skb_priority;
5739 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
5740 queue_id, &skb_priority)) {
5741 ctx->flow.skb_priority = skb_priority;
5743 /* Couldn't translate queue to a priority. Nothing to do. A warning
5744 * has already been logged. */
5748 struct xlate_reg_state {
5754 xlate_autopath(struct action_xlate_ctx *ctx,
5755 const struct ofpact_autopath *ap)
5757 uint16_t ofp_port = ap->port;
5758 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
5760 if (!port || !port->bundle) {
5761 ofp_port = OFPP_NONE;
5762 } else if (port->bundle->bond) {
5763 /* Autopath does not support VLAN hashing. */
5764 struct ofport_dpif *slave = bond_choose_output_slave(
5765 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
5767 ofp_port = slave->up.ofp_port;
5770 nxm_reg_load(&ap->dst, ofp_port, &ctx->flow);
5774 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
5776 struct ofproto_dpif *ofproto = ofproto_;
5777 struct ofport_dpif *port;
5787 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
5790 port = get_ofp_port(ofproto, ofp_port);
5791 return port ? port->may_enable : false;
5796 xlate_bundle_action(struct action_xlate_ctx *ctx,
5797 const struct ofpact_bundle *bundle)
5801 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
5802 if (bundle->dst.field) {
5803 nxm_reg_load(&bundle->dst, port, &ctx->flow);
5805 xlate_output_action(ctx, port, 0, false);
5810 xlate_learn_action(struct action_xlate_ctx *ctx,
5811 const struct ofpact_learn *learn)
5813 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
5814 struct ofputil_flow_mod fm;
5815 uint64_t ofpacts_stub[1024 / 8];
5816 struct ofpbuf ofpacts;
5819 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
5820 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
5822 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
5823 if (error && !VLOG_DROP_WARN(&rl)) {
5824 VLOG_WARN("learning action failed to modify flow table (%s)",
5825 ofperr_get_name(error));
5828 ofpbuf_uninit(&ofpacts);
5831 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
5832 * means "infinite". */
5834 reduce_timeout(uint16_t max, uint16_t *timeout)
5836 if (max && (!*timeout || *timeout > max)) {
5842 xlate_fin_timeout(struct action_xlate_ctx *ctx,
5843 const struct ofpact_fin_timeout *oft)
5845 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
5846 struct rule_dpif *rule = ctx->rule;
5848 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
5849 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
5854 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
5856 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
5857 ? OFPUTIL_PC_NO_RECV_STP
5858 : OFPUTIL_PC_NO_RECV)) {
5862 /* Only drop packets here if both forwarding and learning are
5863 * disabled. If just learning is enabled, we need to have
5864 * OFPP_NORMAL and the learning action have a look at the packet
5865 * before we can drop it. */
5866 if (!stp_forward_in_state(port->stp_state)
5867 && !stp_learn_in_state(port->stp_state)) {
5875 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
5876 struct action_xlate_ctx *ctx)
5878 const struct ofport_dpif *port;
5879 bool was_evictable = true;
5880 const struct ofpact *a;
5882 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5883 if (port && !may_receive(port, ctx)) {
5884 /* Drop this flow. */
5889 /* Don't let the rule we're working on get evicted underneath us. */
5890 was_evictable = ctx->rule->up.evictable;
5891 ctx->rule->up.evictable = false;
5893 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
5894 struct ofpact_controller *controller;
5895 const struct ofpact_metadata *metadata;
5903 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
5904 ofpact_get_OUTPUT(a)->max_len, true);
5907 case OFPACT_CONTROLLER:
5908 controller = ofpact_get_CONTROLLER(a);
5909 execute_controller_action(ctx, controller->max_len,
5911 controller->controller_id);
5914 case OFPACT_ENQUEUE:
5915 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
5918 case OFPACT_SET_VLAN_VID:
5919 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
5920 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
5924 case OFPACT_SET_VLAN_PCP:
5925 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
5926 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
5931 case OFPACT_STRIP_VLAN:
5932 ctx->flow.vlan_tci = htons(0);
5935 case OFPACT_PUSH_VLAN:
5936 /* TODO:XXX 802.1AD(QinQ) */
5937 ctx->flow.vlan_tci = htons(VLAN_CFI);
5940 case OFPACT_SET_ETH_SRC:
5941 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
5945 case OFPACT_SET_ETH_DST:
5946 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
5950 case OFPACT_SET_IPV4_SRC:
5951 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
5954 case OFPACT_SET_IPV4_DST:
5955 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
5958 case OFPACT_SET_IPV4_DSCP:
5959 /* OpenFlow 1.0 only supports IPv4. */
5960 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5961 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5962 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
5966 case OFPACT_SET_L4_SRC_PORT:
5967 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
5970 case OFPACT_SET_L4_DST_PORT:
5971 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
5974 case OFPACT_RESUBMIT:
5975 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
5978 case OFPACT_SET_TUNNEL:
5979 ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
5982 case OFPACT_SET_QUEUE:
5983 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
5986 case OFPACT_POP_QUEUE:
5987 ctx->flow.skb_priority = ctx->orig_skb_priority;
5990 case OFPACT_REG_MOVE:
5991 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
5994 case OFPACT_REG_LOAD:
5995 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
5998 case OFPACT_DEC_TTL:
5999 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6005 /* Nothing to do. */
6008 case OFPACT_MULTIPATH:
6009 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
6012 case OFPACT_AUTOPATH:
6013 xlate_autopath(ctx, ofpact_get_AUTOPATH(a));
6017 ctx->ofproto->has_bundle_action = true;
6018 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6021 case OFPACT_OUTPUT_REG:
6022 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6026 ctx->has_learn = true;
6027 if (ctx->may_learn) {
6028 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6036 case OFPACT_FIN_TIMEOUT:
6037 ctx->has_fin_timeout = true;
6038 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6041 case OFPACT_CLEAR_ACTIONS:
6043 * Nothing to do because writa-actions is not supported for now.
6044 * When writa-actions is supported, clear-actions also must
6045 * be supported at the same time.
6049 case OFPACT_WRITE_METADATA:
6050 metadata = ofpact_get_WRITE_METADATA(a);
6051 ctx->flow.metadata &= ~metadata->mask;
6052 ctx->flow.metadata |= metadata->metadata & metadata->mask;
6055 case OFPACT_GOTO_TABLE: {
6056 /* TODO:XXX remove recursion */
6057 /* It is assumed that goto-table is last action */
6058 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6059 assert(ctx->table_id < ogt->table_id);
6060 xlate_table_action(ctx, ctx->flow.in_port, ogt->table_id, true);
6067 /* We've let OFPP_NORMAL and the learning action look at the packet,
6068 * so drop it now if forwarding is disabled. */
6069 if (port && !stp_forward_in_state(port->stp_state)) {
6070 ofpbuf_clear(ctx->odp_actions);
6071 add_sflow_action(ctx);
6074 ctx->rule->up.evictable = was_evictable;
6079 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
6080 struct ofproto_dpif *ofproto, const struct flow *flow,
6081 ovs_be16 initial_tci, struct rule_dpif *rule,
6082 uint8_t tcp_flags, const struct ofpbuf *packet)
6084 ctx->ofproto = ofproto;
6086 ctx->base_flow = ctx->flow;
6087 memset(&ctx->base_flow.tunnel, 0, sizeof ctx->base_flow.tunnel);
6088 ctx->base_flow.vlan_tci = initial_tci;
6090 ctx->packet = packet;
6091 ctx->may_learn = packet != NULL;
6092 ctx->tcp_flags = tcp_flags;
6093 ctx->resubmit_hook = NULL;
6094 ctx->report_hook = NULL;
6095 ctx->resubmit_stats = NULL;
6098 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6099 * into datapath actions in 'odp_actions', using 'ctx'. */
6101 xlate_actions(struct action_xlate_ctx *ctx,
6102 const struct ofpact *ofpacts, size_t ofpacts_len,
6103 struct ofpbuf *odp_actions)
6105 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6106 * that in the future we always keep a copy of the original flow for
6107 * tracing purposes. */
6108 static bool hit_resubmit_limit;
6110 enum slow_path_reason special;
6112 COVERAGE_INC(ofproto_dpif_xlate);
6114 ofpbuf_clear(odp_actions);
6115 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
6117 ctx->odp_actions = odp_actions;
6120 ctx->has_learn = false;
6121 ctx->has_normal = false;
6122 ctx->has_fin_timeout = false;
6123 ctx->nf_output_iface = NF_OUT_DROP;
6126 ctx->max_resubmit_trigger = false;
6127 ctx->orig_skb_priority = ctx->flow.skb_priority;
6131 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
6132 /* Do this conditionally because the copy is expensive enough that it
6133 * shows up in profiles.
6135 * We keep orig_flow in 'ctx' only because I couldn't make GCC 4.4
6136 * believe that I wasn't using it without initializing it if I kept it
6137 * in a local variable. */
6138 ctx->orig_flow = ctx->flow;
6141 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6142 switch (ctx->ofproto->up.frag_handling) {
6143 case OFPC_FRAG_NORMAL:
6144 /* We must pretend that transport ports are unavailable. */
6145 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
6146 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
6149 case OFPC_FRAG_DROP:
6152 case OFPC_FRAG_REASM:
6155 case OFPC_FRAG_NX_MATCH:
6156 /* Nothing to do. */
6159 case OFPC_INVALID_TTL_TO_CONTROLLER:
6164 special = process_special(ctx->ofproto, &ctx->flow, ctx->packet);
6166 ctx->slow |= special;
6168 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6169 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
6171 add_sflow_action(ctx);
6172 do_xlate_actions(ofpacts, ofpacts_len, ctx);
6174 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
6175 if (!hit_resubmit_limit) {
6176 /* We didn't record the original flow. Make sure we do from
6178 hit_resubmit_limit = true;
6179 } else if (!VLOG_DROP_ERR(&trace_rl)) {
6180 struct ds ds = DS_EMPTY_INITIALIZER;
6182 ofproto_trace(ctx->ofproto, &ctx->orig_flow, ctx->packet,
6184 VLOG_ERR("Trace triggered by excessive resubmit "
6185 "recursion:\n%s", ds_cstr(&ds));
6190 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
6191 ctx->odp_actions->data,
6192 ctx->odp_actions->size)) {
6193 ctx->slow |= SLOW_IN_BAND;
6195 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
6197 compose_output_action(ctx, OFPP_LOCAL);
6200 if (ctx->ofproto->has_mirrors) {
6201 add_mirror_actions(ctx, &ctx->orig_flow);
6203 fix_sflow_action(ctx);
6207 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
6208 * into datapath actions, using 'ctx', and discards the datapath actions. */
6210 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
6211 const struct ofpact *ofpacts,
6214 uint64_t odp_actions_stub[1024 / 8];
6215 struct ofpbuf odp_actions;
6217 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
6218 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
6219 ofpbuf_uninit(&odp_actions);
6223 xlate_report(struct action_xlate_ctx *ctx, const char *s)
6225 if (ctx->report_hook) {
6226 ctx->report_hook(ctx, s);
6230 /* OFPP_NORMAL implementation. */
6232 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
6234 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
6235 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
6236 * the bundle on which the packet was received, returns the VLAN to which the
6239 * Both 'vid' and the return value are in the range 0...4095. */
6241 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
6243 switch (in_bundle->vlan_mode) {
6244 case PORT_VLAN_ACCESS:
6245 return in_bundle->vlan;
6248 case PORT_VLAN_TRUNK:
6251 case PORT_VLAN_NATIVE_UNTAGGED:
6252 case PORT_VLAN_NATIVE_TAGGED:
6253 return vid ? vid : in_bundle->vlan;
6260 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
6261 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
6264 * 'vid' should be the VID obtained from the 802.1Q header that was received as
6265 * part of a packet (specify 0 if there was no 802.1Q header), in the range
6268 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
6270 /* Allow any VID on the OFPP_NONE port. */
6271 if (in_bundle == &ofpp_none_bundle) {
6275 switch (in_bundle->vlan_mode) {
6276 case PORT_VLAN_ACCESS:
6279 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6280 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
6281 "packet received on port %s configured as VLAN "
6282 "%"PRIu16" access port",
6283 in_bundle->ofproto->up.name, vid,
6284 in_bundle->name, in_bundle->vlan);
6290 case PORT_VLAN_NATIVE_UNTAGGED:
6291 case PORT_VLAN_NATIVE_TAGGED:
6293 /* Port must always carry its native VLAN. */
6297 case PORT_VLAN_TRUNK:
6298 if (!ofbundle_includes_vlan(in_bundle, vid)) {
6300 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6301 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
6302 "received on port %s not configured for trunking "
6304 in_bundle->ofproto->up.name, vid,
6305 in_bundle->name, vid);
6317 /* Given 'vlan', the VLAN that a packet belongs to, and
6318 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
6319 * that should be included in the 802.1Q header. (If the return value is 0,
6320 * then the 802.1Q header should only be included in the packet if there is a
6323 * Both 'vlan' and the return value are in the range 0...4095. */
6325 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
6327 switch (out_bundle->vlan_mode) {
6328 case PORT_VLAN_ACCESS:
6331 case PORT_VLAN_TRUNK:
6332 case PORT_VLAN_NATIVE_TAGGED:
6335 case PORT_VLAN_NATIVE_UNTAGGED:
6336 return vlan == out_bundle->vlan ? 0 : vlan;
6344 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
6347 struct ofport_dpif *port;
6349 ovs_be16 tci, old_tci;
6351 vid = output_vlan_to_vid(out_bundle, vlan);
6352 if (!out_bundle->bond) {
6353 port = ofbundle_get_a_port(out_bundle);
6355 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
6358 /* No slaves enabled, so drop packet. */
6363 old_tci = ctx->flow.vlan_tci;
6365 if (tci || out_bundle->use_priority_tags) {
6366 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
6368 tci |= htons(VLAN_CFI);
6371 ctx->flow.vlan_tci = tci;
6373 compose_output_action(ctx, port->up.ofp_port);
6374 ctx->flow.vlan_tci = old_tci;
6378 mirror_mask_ffs(mirror_mask_t mask)
6380 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
6385 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
6387 return (bundle->vlan_mode != PORT_VLAN_ACCESS
6388 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
6392 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
6394 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
6397 /* Returns an arbitrary interface within 'bundle'. */
6398 static struct ofport_dpif *
6399 ofbundle_get_a_port(const struct ofbundle *bundle)
6401 return CONTAINER_OF(list_front(&bundle->ports),
6402 struct ofport_dpif, bundle_node);
6406 vlan_is_mirrored(const struct ofmirror *m, int vlan)
6408 return !m->vlans || bitmap_is_set(m->vlans, vlan);
6412 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
6414 struct ofproto_dpif *ofproto = ctx->ofproto;
6415 mirror_mask_t mirrors;
6416 struct ofbundle *in_bundle;
6419 const struct nlattr *a;
6422 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
6423 ctx->packet != NULL, NULL);
6427 mirrors = in_bundle->src_mirrors;
6429 /* Drop frames on bundles reserved for mirroring. */
6430 if (in_bundle->mirror_out) {
6431 if (ctx->packet != NULL) {
6432 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6433 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6434 "%s, which is reserved exclusively for mirroring",
6435 ctx->ofproto->up.name, in_bundle->name);
6441 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
6442 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6445 vlan = input_vid_to_vlan(in_bundle, vid);
6447 /* Look at the output ports to check for destination selections. */
6449 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
6450 ctx->odp_actions->size) {
6451 enum ovs_action_attr type = nl_attr_type(a);
6452 struct ofport_dpif *ofport;
6454 if (type != OVS_ACTION_ATTR_OUTPUT) {
6458 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
6459 if (ofport && ofport->bundle) {
6460 mirrors |= ofport->bundle->dst_mirrors;
6468 /* Restore the original packet before adding the mirror actions. */
6469 ctx->flow = *orig_flow;
6474 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6476 if (!vlan_is_mirrored(m, vlan)) {
6477 mirrors = zero_rightmost_1bit(mirrors);
6481 mirrors &= ~m->dup_mirrors;
6482 ctx->mirrors |= m->dup_mirrors;
6484 output_normal(ctx, m->out, vlan);
6485 } else if (vlan != m->out_vlan
6486 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
6487 struct ofbundle *bundle;
6489 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
6490 if (ofbundle_includes_vlan(bundle, m->out_vlan)
6491 && !bundle->mirror_out) {
6492 output_normal(ctx, bundle, m->out_vlan);
6500 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
6501 uint64_t packets, uint64_t bytes)
6507 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
6510 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6513 /* In normal circumstances 'm' will not be NULL. However,
6514 * if mirrors are reconfigured, we can temporarily get out
6515 * of sync in facet_revalidate(). We could "correct" the
6516 * mirror list before reaching here, but doing that would
6517 * not properly account the traffic stats we've currently
6518 * accumulated for previous mirror configuration. */
6522 m->packet_count += packets;
6523 m->byte_count += bytes;
6527 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
6528 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
6529 * indicate this; newer upstream kernels use gratuitous ARP requests. */
6531 is_gratuitous_arp(const struct flow *flow)
6533 return (flow->dl_type == htons(ETH_TYPE_ARP)
6534 && eth_addr_is_broadcast(flow->dl_dst)
6535 && (flow->nw_proto == ARP_OP_REPLY
6536 || (flow->nw_proto == ARP_OP_REQUEST
6537 && flow->nw_src == flow->nw_dst)));
6541 update_learning_table(struct ofproto_dpif *ofproto,
6542 const struct flow *flow, int vlan,
6543 struct ofbundle *in_bundle)
6545 struct mac_entry *mac;
6547 /* Don't learn the OFPP_NONE port. */
6548 if (in_bundle == &ofpp_none_bundle) {
6552 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
6556 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
6557 if (is_gratuitous_arp(flow)) {
6558 /* We don't want to learn from gratuitous ARP packets that are
6559 * reflected back over bond slaves so we lock the learning table. */
6560 if (!in_bundle->bond) {
6561 mac_entry_set_grat_arp_lock(mac);
6562 } else if (mac_entry_is_grat_arp_locked(mac)) {
6567 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
6568 /* The log messages here could actually be useful in debugging,
6569 * so keep the rate limit relatively high. */
6570 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
6571 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
6572 "on port %s in VLAN %d",
6573 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
6574 in_bundle->name, vlan);
6576 mac->port.p = in_bundle;
6577 tag_set_add(&ofproto->revalidate_set,
6578 mac_learning_changed(ofproto->ml, mac));
6582 static struct ofbundle *
6583 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
6584 bool warn, struct ofport_dpif **in_ofportp)
6586 struct ofport_dpif *ofport;
6588 /* Find the port and bundle for the received packet. */
6589 ofport = get_ofp_port(ofproto, in_port);
6591 *in_ofportp = ofport;
6593 if (ofport && ofport->bundle) {
6594 return ofport->bundle;
6597 /* Special-case OFPP_NONE, which a controller may use as the ingress
6598 * port for traffic that it is sourcing. */
6599 if (in_port == OFPP_NONE) {
6600 return &ofpp_none_bundle;
6603 /* Odd. A few possible reasons here:
6605 * - We deleted a port but there are still a few packets queued up
6608 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
6609 * we don't know about.
6611 * - The ofproto client didn't configure the port as part of a bundle.
6612 * This is particularly likely to happen if a packet was received on the
6613 * port after it was created, but before the client had a chance to
6614 * configure its bundle.
6617 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6619 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
6620 "port %"PRIu16, ofproto->up.name, in_port);
6625 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
6626 * dropped. Returns true if they may be forwarded, false if they should be
6629 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
6630 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
6632 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
6633 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
6634 * checked by input_vid_is_valid().
6636 * May also add tags to '*tags', although the current implementation only does
6637 * so in one special case.
6640 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
6643 struct ofproto_dpif *ofproto = ctx->ofproto;
6644 struct flow *flow = &ctx->flow;
6645 struct ofbundle *in_bundle = in_port->bundle;
6647 /* Drop frames for reserved multicast addresses
6648 * only if forward_bpdu option is absent. */
6649 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
6650 xlate_report(ctx, "packet has reserved destination MAC, dropping");
6654 if (in_bundle->bond) {
6655 struct mac_entry *mac;
6657 switch (bond_check_admissibility(in_bundle->bond, in_port,
6658 flow->dl_dst, &ctx->tags)) {
6663 xlate_report(ctx, "bonding refused admissibility, dropping");
6666 case BV_DROP_IF_MOVED:
6667 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
6668 if (mac && mac->port.p != in_bundle &&
6669 (!is_gratuitous_arp(flow)
6670 || mac_entry_is_grat_arp_locked(mac))) {
6671 xlate_report(ctx, "SLB bond thinks this packet looped back, "
6683 xlate_normal(struct action_xlate_ctx *ctx)
6685 struct ofport_dpif *in_port;
6686 struct ofbundle *in_bundle;
6687 struct mac_entry *mac;
6691 ctx->has_normal = true;
6693 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
6694 ctx->packet != NULL, &in_port);
6696 xlate_report(ctx, "no input bundle, dropping");
6700 /* Drop malformed frames. */
6701 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
6702 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
6703 if (ctx->packet != NULL) {
6704 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6705 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
6706 "VLAN tag received on port %s",
6707 ctx->ofproto->up.name, in_bundle->name);
6709 xlate_report(ctx, "partial VLAN tag, dropping");
6713 /* Drop frames on bundles reserved for mirroring. */
6714 if (in_bundle->mirror_out) {
6715 if (ctx->packet != NULL) {
6716 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6717 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6718 "%s, which is reserved exclusively for mirroring",
6719 ctx->ofproto->up.name, in_bundle->name);
6721 xlate_report(ctx, "input port is mirror output port, dropping");
6726 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
6727 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6728 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
6731 vlan = input_vid_to_vlan(in_bundle, vid);
6733 /* Check other admissibility requirements. */
6734 if (in_port && !is_admissible(ctx, in_port, vlan)) {
6738 /* Learn source MAC. */
6739 if (ctx->may_learn) {
6740 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
6743 /* Determine output bundle. */
6744 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
6747 if (mac->port.p != in_bundle) {
6748 xlate_report(ctx, "forwarding to learned port");
6749 output_normal(ctx, mac->port.p, vlan);
6751 xlate_report(ctx, "learned port is input port, dropping");
6754 struct ofbundle *bundle;
6756 xlate_report(ctx, "no learned MAC for destination, flooding");
6757 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
6758 if (bundle != in_bundle
6759 && ofbundle_includes_vlan(bundle, vlan)
6760 && bundle->floodable
6761 && !bundle->mirror_out) {
6762 output_normal(ctx, bundle, vlan);
6765 ctx->nf_output_iface = NF_OUT_FLOOD;
6769 /* Optimized flow revalidation.
6771 * It's a difficult problem, in general, to tell which facets need to have
6772 * their actions recalculated whenever the OpenFlow flow table changes. We
6773 * don't try to solve that general problem: for most kinds of OpenFlow flow
6774 * table changes, we recalculate the actions for every facet. This is
6775 * relatively expensive, but it's good enough if the OpenFlow flow table
6776 * doesn't change very often.
6778 * However, we can expect one particular kind of OpenFlow flow table change to
6779 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
6780 * of CPU on revalidating every facet whenever MAC learning modifies the flow
6781 * table, we add a special case that applies to flow tables in which every rule
6782 * has the same form (that is, the same wildcards), except that the table is
6783 * also allowed to have a single "catch-all" flow that matches all packets. We
6784 * optimize this case by tagging all of the facets that resubmit into the table
6785 * and invalidating the same tag whenever a flow changes in that table. The
6786 * end result is that we revalidate just the facets that need it (and sometimes
6787 * a few more, but not all of the facets or even all of the facets that
6788 * resubmit to the table modified by MAC learning). */
6790 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
6791 * into an OpenFlow table with the given 'basis'. */
6793 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
6796 if (minimask_is_catchall(mask)) {
6799 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
6800 return tag_create_deterministic(hash);
6804 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
6805 * taggability of that table.
6807 * This function must be called after *each* change to a flow table. If you
6808 * skip calling it on some changes then the pointer comparisons at the end can
6809 * be invalid if you get unlucky. For example, if a flow removal causes a
6810 * cls_table to be destroyed and then a flow insertion causes a cls_table with
6811 * different wildcards to be created with the same address, then this function
6812 * will incorrectly skip revalidation. */
6814 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
6816 struct table_dpif *table = &ofproto->tables[table_id];
6817 const struct oftable *oftable = &ofproto->up.tables[table_id];
6818 struct cls_table *catchall, *other;
6819 struct cls_table *t;
6821 catchall = other = NULL;
6823 switch (hmap_count(&oftable->cls.tables)) {
6825 /* We could tag this OpenFlow table but it would make the logic a
6826 * little harder and it's a corner case that doesn't seem worth it
6832 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
6833 if (cls_table_is_catchall(t)) {
6835 } else if (!other) {
6838 /* Indicate that we can't tag this by setting both tables to
6839 * NULL. (We know that 'catchall' is already NULL.) */
6846 /* Can't tag this table. */
6850 if (table->catchall_table != catchall || table->other_table != other) {
6851 table->catchall_table = catchall;
6852 table->other_table = other;
6853 ofproto->need_revalidate = REV_FLOW_TABLE;
6857 /* Given 'rule' that has changed in some way (either it is a rule being
6858 * inserted, a rule being deleted, or a rule whose actions are being
6859 * modified), marks facets for revalidation to ensure that packets will be
6860 * forwarded correctly according to the new state of the flow table.
6862 * This function must be called after *each* change to a flow table. See
6863 * the comment on table_update_taggable() for more information. */
6865 rule_invalidate(const struct rule_dpif *rule)
6867 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
6869 table_update_taggable(ofproto, rule->up.table_id);
6871 if (!ofproto->need_revalidate) {
6872 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
6874 if (table->other_table && rule->tag) {
6875 tag_set_add(&ofproto->revalidate_set, rule->tag);
6877 ofproto->need_revalidate = REV_FLOW_TABLE;
6883 set_frag_handling(struct ofproto *ofproto_,
6884 enum ofp_config_flags frag_handling)
6886 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6888 if (frag_handling != OFPC_FRAG_REASM) {
6889 ofproto->need_revalidate = REV_RECONFIGURE;
6897 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
6898 const struct flow *flow,
6899 const struct ofpact *ofpacts, size_t ofpacts_len)
6901 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6902 struct odputil_keybuf keybuf;
6903 struct dpif_flow_stats stats;
6907 struct action_xlate_ctx ctx;
6908 uint64_t odp_actions_stub[1024 / 8];
6909 struct ofpbuf odp_actions;
6911 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
6912 odp_flow_key_from_flow(&key, flow,
6913 ofp_port_to_odp_port(ofproto, flow->in_port));
6915 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
6917 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL,
6918 packet_get_tcp_flags(packet, flow), packet);
6919 ctx.resubmit_stats = &stats;
6921 ofpbuf_use_stub(&odp_actions,
6922 odp_actions_stub, sizeof odp_actions_stub);
6923 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
6924 dpif_execute(ofproto->backer->dpif, key.data, key.size,
6925 odp_actions.data, odp_actions.size, packet);
6926 ofpbuf_uninit(&odp_actions);
6934 set_netflow(struct ofproto *ofproto_,
6935 const struct netflow_options *netflow_options)
6937 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6939 if (netflow_options) {
6940 if (!ofproto->netflow) {
6941 ofproto->netflow = netflow_create();
6943 return netflow_set_options(ofproto->netflow, netflow_options);
6945 netflow_destroy(ofproto->netflow);
6946 ofproto->netflow = NULL;
6952 get_netflow_ids(const struct ofproto *ofproto_,
6953 uint8_t *engine_type, uint8_t *engine_id)
6955 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6957 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
6961 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
6963 if (!facet_is_controller_flow(facet) &&
6964 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
6965 struct subfacet *subfacet;
6966 struct ofexpired expired;
6968 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
6969 if (subfacet->path == SF_FAST_PATH) {
6970 struct dpif_flow_stats stats;
6972 subfacet_reinstall(subfacet, &stats);
6973 subfacet_update_stats(subfacet, &stats);
6977 expired.flow = facet->flow;
6978 expired.packet_count = facet->packet_count;
6979 expired.byte_count = facet->byte_count;
6980 expired.used = facet->used;
6981 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
6986 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
6988 struct facet *facet;
6990 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6991 send_active_timeout(ofproto, facet);
6995 static struct ofproto_dpif *
6996 ofproto_dpif_lookup(const char *name)
6998 struct ofproto_dpif *ofproto;
7000 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7001 hash_string(name, 0), &all_ofproto_dpifs) {
7002 if (!strcmp(ofproto->up.name, name)) {
7010 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7011 const char *argv[], void *aux OVS_UNUSED)
7013 struct ofproto_dpif *ofproto;
7016 ofproto = ofproto_dpif_lookup(argv[1]);
7018 unixctl_command_reply_error(conn, "no such bridge");
7021 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
7023 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7024 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
7028 unixctl_command_reply(conn, "table successfully flushed");
7032 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7033 const char *argv[], void *aux OVS_UNUSED)
7035 struct ds ds = DS_EMPTY_INITIALIZER;
7036 const struct ofproto_dpif *ofproto;
7037 const struct mac_entry *e;
7039 ofproto = ofproto_dpif_lookup(argv[1]);
7041 unixctl_command_reply_error(conn, "no such bridge");
7045 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7046 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7047 struct ofbundle *bundle = e->port.p;
7048 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7049 ofbundle_get_a_port(bundle)->odp_port,
7050 e->vlan, ETH_ADDR_ARGS(e->mac),
7051 mac_entry_age(ofproto->ml, e));
7053 unixctl_command_reply(conn, ds_cstr(&ds));
7058 struct action_xlate_ctx ctx;
7064 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7065 const struct rule_dpif *rule)
7067 ds_put_char_multiple(result, '\t', level);
7069 ds_put_cstr(result, "No match\n");
7073 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7074 table_id, ntohll(rule->up.flow_cookie));
7075 cls_rule_format(&rule->up.cr, result);
7076 ds_put_char(result, '\n');
7078 ds_put_char_multiple(result, '\t', level);
7079 ds_put_cstr(result, "OpenFlow ");
7080 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7081 ds_put_char(result, '\n');
7085 trace_format_flow(struct ds *result, int level, const char *title,
7086 struct trace_ctx *trace)
7088 ds_put_char_multiple(result, '\t', level);
7089 ds_put_format(result, "%s: ", title);
7090 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
7091 ds_put_cstr(result, "unchanged");
7093 flow_format(result, &trace->ctx.flow);
7094 trace->flow = trace->ctx.flow;
7096 ds_put_char(result, '\n');
7100 trace_format_regs(struct ds *result, int level, const char *title,
7101 struct trace_ctx *trace)
7105 ds_put_char_multiple(result, '\t', level);
7106 ds_put_format(result, "%s:", title);
7107 for (i = 0; i < FLOW_N_REGS; i++) {
7108 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7110 ds_put_char(result, '\n');
7114 trace_format_odp(struct ds *result, int level, const char *title,
7115 struct trace_ctx *trace)
7117 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
7119 ds_put_char_multiple(result, '\t', level);
7120 ds_put_format(result, "%s: ", title);
7121 format_odp_actions(result, odp_actions->data, odp_actions->size);
7122 ds_put_char(result, '\n');
7126 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
7128 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7129 struct ds *result = trace->result;
7131 ds_put_char(result, '\n');
7132 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7133 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
7134 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
7135 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
7139 trace_report(struct action_xlate_ctx *ctx, const char *s)
7141 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7142 struct ds *result = trace->result;
7144 ds_put_char_multiple(result, '\t', ctx->recurse);
7145 ds_put_cstr(result, s);
7146 ds_put_char(result, '\n');
7150 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
7151 void *aux OVS_UNUSED)
7153 const char *dpname = argv[1];
7154 struct ofproto_dpif *ofproto;
7155 struct ofpbuf odp_key;
7156 struct ofpbuf *packet;
7157 ovs_be16 initial_tci;
7163 ofpbuf_init(&odp_key, 0);
7166 ofproto = ofproto_dpif_lookup(dpname);
7168 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
7172 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
7173 /* ofproto/trace dpname flow [-generate] */
7174 const char *flow_s = argv[2];
7175 const char *generate_s = argv[3];
7177 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
7178 * flow. We guess which type it is based on whether 'flow_s' contains
7179 * an '(', since a datapath flow always contains '(') but an
7180 * OpenFlow-like flow should not (in fact it's allowed but I believe
7181 * that's not documented anywhere).
7183 * An alternative would be to try to parse 'flow_s' both ways, but then
7184 * it would be tricky giving a sensible error message. After all, do
7185 * you just say "syntax error" or do you present both error messages?
7186 * Both choices seem lousy. */
7187 if (strchr(flow_s, '(')) {
7188 enum odp_key_fitness fitness;
7191 /* Convert string to datapath key. */
7192 ofpbuf_init(&odp_key, 0);
7193 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
7195 unixctl_command_reply_error(conn, "Bad flow syntax");
7199 fitness = odp_flow_key_to_flow(odp_key.data, odp_key.size, &flow);
7200 flow.in_port = odp_port_to_ofp_port(ofproto, flow.in_port);
7202 /* Convert odp_key to flow. */
7203 error = ofproto_dpif_vsp_adjust(ofproto, fitness, &flow,
7204 &initial_tci, NULL);
7205 if (error == ODP_FIT_ERROR) {
7206 unixctl_command_reply_error(conn, "Invalid flow");
7212 error_s = parse_ofp_exact_flow(&flow, argv[2]);
7214 unixctl_command_reply_error(conn, error_s);
7219 initial_tci = flow.vlan_tci;
7220 vsp_adjust_flow(ofproto, &flow);
7223 /* Generate a packet, if requested. */
7225 packet = ofpbuf_new(0);
7226 flow_compose(packet, &flow);
7228 } else if (argc == 7) {
7229 /* ofproto/trace dpname priority tun_id in_port mark packet */
7230 const char *priority_s = argv[2];
7231 const char *tun_id_s = argv[3];
7232 const char *in_port_s = argv[4];
7233 const char *mark_s = argv[5];
7234 const char *packet_s = argv[6];
7235 uint32_t in_port = atoi(in_port_s);
7236 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
7237 uint32_t priority = atoi(priority_s);
7238 uint32_t mark = atoi(mark_s);
7241 msg = eth_from_hex(packet_s, &packet);
7243 unixctl_command_reply_error(conn, msg);
7247 ds_put_cstr(&result, "Packet: ");
7248 s = ofp_packet_to_string(packet->data, packet->size);
7249 ds_put_cstr(&result, s);
7252 flow_extract(packet, priority, mark, NULL, in_port, &flow);
7253 flow.tunnel.tun_id = tun_id;
7254 initial_tci = flow.vlan_tci;
7256 unixctl_command_reply_error(conn, "Bad command syntax");
7260 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
7261 unixctl_command_reply(conn, ds_cstr(&result));
7264 ds_destroy(&result);
7265 ofpbuf_delete(packet);
7266 ofpbuf_uninit(&odp_key);
7270 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
7271 const struct ofpbuf *packet, ovs_be16 initial_tci,
7274 struct rule_dpif *rule;
7276 ds_put_cstr(ds, "Flow: ");
7277 flow_format(ds, flow);
7278 ds_put_char(ds, '\n');
7280 rule = rule_dpif_lookup(ofproto, flow);
7282 trace_format_rule(ds, 0, 0, rule);
7283 if (rule == ofproto->miss_rule) {
7284 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
7285 } else if (rule == ofproto->no_packet_in_rule) {
7286 ds_put_cstr(ds, "\nNo match, packets dropped because "
7287 "OFPPC_NO_PACKET_IN is set on in_port.\n");
7291 uint64_t odp_actions_stub[1024 / 8];
7292 struct ofpbuf odp_actions;
7294 struct trace_ctx trace;
7297 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
7300 ofpbuf_use_stub(&odp_actions,
7301 odp_actions_stub, sizeof odp_actions_stub);
7302 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
7303 rule, tcp_flags, packet);
7304 trace.ctx.resubmit_hook = trace_resubmit;
7305 trace.ctx.report_hook = trace_report;
7306 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
7309 ds_put_char(ds, '\n');
7310 trace_format_flow(ds, 0, "Final flow", &trace);
7311 ds_put_cstr(ds, "Datapath actions: ");
7312 format_odp_actions(ds, odp_actions.data, odp_actions.size);
7313 ofpbuf_uninit(&odp_actions);
7315 if (trace.ctx.slow) {
7316 enum slow_path_reason slow;
7318 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
7319 "slow path because it:");
7320 for (slow = trace.ctx.slow; slow; ) {
7321 enum slow_path_reason bit = rightmost_1bit(slow);
7325 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
7328 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
7331 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
7334 ds_put_cstr(ds, "\n\t- Needs in-band special case "
7337 ds_put_cstr(ds, "\n\t (The datapath actions are "
7338 "incomplete--for complete actions, "
7339 "please supply a packet.)");
7342 case SLOW_CONTROLLER:
7343 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
7344 "to the OpenFlow controller.");
7347 ds_put_cstr(ds, "\n\t- Needs more specific matching "
7348 "than the datapath supports.");
7355 if (slow & ~SLOW_MATCH) {
7356 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
7357 "the special slow-path processing.");
7364 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7365 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7368 unixctl_command_reply(conn, NULL);
7372 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7373 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7376 unixctl_command_reply(conn, NULL);
7379 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
7380 * 'reply' describing the results. */
7382 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
7384 struct facet *facet;
7388 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7389 if (!facet_check_consistency(facet)) {
7394 ofproto->need_revalidate = REV_INCONSISTENCY;
7398 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
7399 ofproto->up.name, errors);
7401 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
7406 ofproto_dpif_self_check(struct unixctl_conn *conn,
7407 int argc, const char *argv[], void *aux OVS_UNUSED)
7409 struct ds reply = DS_EMPTY_INITIALIZER;
7410 struct ofproto_dpif *ofproto;
7413 ofproto = ofproto_dpif_lookup(argv[1]);
7415 unixctl_command_reply_error(conn, "Unknown ofproto (use "
7416 "ofproto/list for help)");
7419 ofproto_dpif_self_check__(ofproto, &reply);
7421 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7422 ofproto_dpif_self_check__(ofproto, &reply);
7426 unixctl_command_reply(conn, ds_cstr(&reply));
7430 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
7431 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
7432 * to destroy 'ofproto_shash' and free the returned value. */
7433 static const struct shash_node **
7434 get_ofprotos(struct shash *ofproto_shash)
7436 const struct ofproto_dpif *ofproto;
7438 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7439 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
7440 shash_add_nocopy(ofproto_shash, name, ofproto);
7443 return shash_sort(ofproto_shash);
7447 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
7448 const char *argv[] OVS_UNUSED,
7449 void *aux OVS_UNUSED)
7451 struct ds ds = DS_EMPTY_INITIALIZER;
7452 struct shash ofproto_shash;
7453 const struct shash_node **sorted_ofprotos;
7456 shash_init(&ofproto_shash);
7457 sorted_ofprotos = get_ofprotos(&ofproto_shash);
7458 for (i = 0; i < shash_count(&ofproto_shash); i++) {
7459 const struct shash_node *node = sorted_ofprotos[i];
7460 ds_put_format(&ds, "%s\n", node->name);
7463 shash_destroy(&ofproto_shash);
7464 free(sorted_ofprotos);
7466 unixctl_command_reply(conn, ds_cstr(&ds));
7471 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
7473 struct dpif_dp_stats s;
7474 const struct shash_node **ports;
7477 dpif_get_dp_stats(ofproto->backer->dpif, &s);
7479 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
7480 dpif_name(ofproto->backer->dpif));
7481 /* xxx It would be better to show bridge-specific stats instead
7482 * xxx of dp ones. */
7484 "\tlookups: hit:%"PRIu64" missed:%"PRIu64" lost:%"PRIu64"\n",
7485 s.n_hit, s.n_missed, s.n_lost);
7486 ds_put_format(ds, "\tflows: %zu\n",
7487 hmap_count(&ofproto->subfacets));
7489 ports = shash_sort(&ofproto->up.port_by_name);
7490 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
7491 const struct shash_node *node = ports[i];
7492 struct ofport *ofport = node->data;
7493 const char *name = netdev_get_name(ofport->netdev);
7494 const char *type = netdev_get_type(ofport->netdev);
7496 ds_put_format(ds, "\t%s %u/%u:", name, ofport->ofp_port,
7497 ofp_port_to_odp_port(ofproto, ofport->ofp_port));
7498 if (strcmp(type, "system")) {
7499 struct netdev *netdev;
7502 ds_put_format(ds, " (%s", type);
7504 error = netdev_open(name, type, &netdev);
7509 error = netdev_get_config(netdev, &config);
7511 const struct smap_node **nodes;
7514 nodes = smap_sort(&config);
7515 for (i = 0; i < smap_count(&config); i++) {
7516 const struct smap_node *node = nodes[i];
7517 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
7518 node->key, node->value);
7522 smap_destroy(&config);
7524 netdev_close(netdev);
7526 ds_put_char(ds, ')');
7528 ds_put_char(ds, '\n');
7534 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
7535 const char *argv[], void *aux OVS_UNUSED)
7537 struct ds ds = DS_EMPTY_INITIALIZER;
7538 const struct ofproto_dpif *ofproto;
7542 for (i = 1; i < argc; i++) {
7543 ofproto = ofproto_dpif_lookup(argv[i]);
7545 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
7546 "for help)", argv[i]);
7547 unixctl_command_reply_error(conn, ds_cstr(&ds));
7550 show_dp_format(ofproto, &ds);
7553 struct shash ofproto_shash;
7554 const struct shash_node **sorted_ofprotos;
7557 shash_init(&ofproto_shash);
7558 sorted_ofprotos = get_ofprotos(&ofproto_shash);
7559 for (i = 0; i < shash_count(&ofproto_shash); i++) {
7560 const struct shash_node *node = sorted_ofprotos[i];
7561 show_dp_format(node->data, &ds);
7564 shash_destroy(&ofproto_shash);
7565 free(sorted_ofprotos);
7568 unixctl_command_reply(conn, ds_cstr(&ds));
7573 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
7574 int argc OVS_UNUSED, const char *argv[],
7575 void *aux OVS_UNUSED)
7577 struct ds ds = DS_EMPTY_INITIALIZER;
7578 const struct ofproto_dpif *ofproto;
7579 struct subfacet *subfacet;
7581 ofproto = ofproto_dpif_lookup(argv[1]);
7583 unixctl_command_reply_error(conn, "no such bridge");
7587 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
7588 struct odputil_keybuf keybuf;
7591 subfacet_get_key(subfacet, &keybuf, &key);
7592 odp_flow_key_format(key.data, key.size, &ds);
7594 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
7595 subfacet->dp_packet_count, subfacet->dp_byte_count);
7596 if (subfacet->used) {
7597 ds_put_format(&ds, "%.3fs",
7598 (time_msec() - subfacet->used) / 1000.0);
7600 ds_put_format(&ds, "never");
7602 if (subfacet->facet->tcp_flags) {
7603 ds_put_cstr(&ds, ", flags:");
7604 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
7607 ds_put_cstr(&ds, ", actions:");
7608 format_odp_actions(&ds, subfacet->actions, subfacet->actions_len);
7609 ds_put_char(&ds, '\n');
7612 unixctl_command_reply(conn, ds_cstr(&ds));
7617 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
7618 int argc OVS_UNUSED, const char *argv[],
7619 void *aux OVS_UNUSED)
7621 struct ds ds = DS_EMPTY_INITIALIZER;
7622 struct ofproto_dpif *ofproto;
7624 ofproto = ofproto_dpif_lookup(argv[1]);
7626 unixctl_command_reply_error(conn, "no such bridge");
7630 flush(&ofproto->up);
7632 unixctl_command_reply(conn, ds_cstr(&ds));
7637 ofproto_dpif_unixctl_init(void)
7639 static bool registered;
7645 unixctl_command_register(
7647 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
7648 2, 6, ofproto_unixctl_trace, NULL);
7649 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
7650 ofproto_unixctl_fdb_flush, NULL);
7651 unixctl_command_register("fdb/show", "bridge", 1, 1,
7652 ofproto_unixctl_fdb_show, NULL);
7653 unixctl_command_register("ofproto/clog", "", 0, 0,
7654 ofproto_dpif_clog, NULL);
7655 unixctl_command_register("ofproto/unclog", "", 0, 0,
7656 ofproto_dpif_unclog, NULL);
7657 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
7658 ofproto_dpif_self_check, NULL);
7659 unixctl_command_register("dpif/dump-dps", "", 0, 0,
7660 ofproto_unixctl_dpif_dump_dps, NULL);
7661 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
7662 ofproto_unixctl_dpif_show, NULL);
7663 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
7664 ofproto_unixctl_dpif_dump_flows, NULL);
7665 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
7666 ofproto_unixctl_dpif_del_flows, NULL);
7669 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
7671 * This is deprecated. It is only for compatibility with broken device drivers
7672 * in old versions of Linux that do not properly support VLANs when VLAN
7673 * devices are not used. When broken device drivers are no longer in
7674 * widespread use, we will delete these interfaces. */
7677 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
7679 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
7680 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
7682 if (realdev_ofp_port == ofport->realdev_ofp_port
7683 && vid == ofport->vlandev_vid) {
7687 ofproto->need_revalidate = REV_RECONFIGURE;
7689 if (ofport->realdev_ofp_port) {
7692 if (realdev_ofp_port && ofport->bundle) {
7693 /* vlandevs are enslaved to their realdevs, so they are not allowed to
7694 * themselves be part of a bundle. */
7695 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
7698 ofport->realdev_ofp_port = realdev_ofp_port;
7699 ofport->vlandev_vid = vid;
7701 if (realdev_ofp_port) {
7702 vsp_add(ofport, realdev_ofp_port, vid);
7709 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
7711 return hash_2words(realdev_ofp_port, vid);
7714 /* Returns the ODP port number of the Linux VLAN device that corresponds to
7715 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
7716 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
7717 * it would return the port number of eth0.9.
7719 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
7720 * function just returns its 'realdev_odp_port' argument. */
7722 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
7723 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
7725 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
7726 uint16_t realdev_ofp_port;
7727 int vid = vlan_tci_to_vid(vlan_tci);
7728 const struct vlan_splinter *vsp;
7730 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
7731 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
7732 hash_realdev_vid(realdev_ofp_port, vid),
7733 &ofproto->realdev_vid_map) {
7734 if (vsp->realdev_ofp_port == realdev_ofp_port
7735 && vsp->vid == vid) {
7736 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
7740 return realdev_odp_port;
7743 static struct vlan_splinter *
7744 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
7746 struct vlan_splinter *vsp;
7748 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
7749 &ofproto->vlandev_map) {
7750 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
7758 /* Returns the OpenFlow port number of the "real" device underlying the Linux
7759 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
7760 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
7761 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
7762 * eth0 and store 9 in '*vid'.
7764 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
7765 * VLAN device. Unless VLAN splinters are enabled, this is what this function
7768 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
7769 uint16_t vlandev_ofp_port, int *vid)
7771 if (!hmap_is_empty(&ofproto->vlandev_map)) {
7772 const struct vlan_splinter *vsp;
7774 vsp = vlandev_find(ofproto, vlandev_ofp_port);
7779 return vsp->realdev_ofp_port;
7785 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
7786 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
7787 * 'flow->in_port' to the "real" device backing the VLAN device, sets
7788 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
7789 * always the case unless VLAN splinters are enabled), returns false without
7790 * making any changes. */
7792 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
7797 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
7802 /* Cause the flow to be processed as if it came in on the real device with
7803 * the VLAN device's VLAN ID. */
7804 flow->in_port = realdev;
7805 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
7810 vsp_remove(struct ofport_dpif *port)
7812 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7813 struct vlan_splinter *vsp;
7815 vsp = vlandev_find(ofproto, port->up.ofp_port);
7817 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
7818 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
7821 port->realdev_ofp_port = 0;
7823 VLOG_ERR("missing vlan device record");
7828 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
7830 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7832 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
7833 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
7834 == realdev_ofp_port)) {
7835 struct vlan_splinter *vsp;
7837 vsp = xmalloc(sizeof *vsp);
7838 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
7839 hash_int(port->up.ofp_port, 0));
7840 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
7841 hash_realdev_vid(realdev_ofp_port, vid));
7842 vsp->realdev_ofp_port = realdev_ofp_port;
7843 vsp->vlandev_ofp_port = port->up.ofp_port;
7846 port->realdev_ofp_port = realdev_ofp_port;
7848 VLOG_ERR("duplicate vlan device record");
7853 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
7855 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
7856 return ofport ? ofport->odp_port : OVSP_NONE;
7859 static struct ofport_dpif *
7860 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
7862 struct ofport_dpif *port;
7864 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
7865 hash_int(odp_port, 0),
7866 &backer->odp_to_ofport_map) {
7867 if (port->odp_port == odp_port) {
7876 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
7878 struct ofport_dpif *port;
7880 port = odp_port_to_ofport(ofproto->backer, odp_port);
7881 if (port && ofproto == ofproto_dpif_cast(port->up.ofproto)) {
7882 return port->up.ofp_port;
7888 const struct ofproto_class ofproto_dpif_class = {
7923 port_is_lacp_current,
7924 NULL, /* rule_choose_table */
7931 rule_modify_actions,
7940 get_cfm_remote_mpids,
7945 get_stp_port_status,
7952 is_mirror_output_bundle,
7953 forward_bpdu_changed,
7954 set_mac_table_config,