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. */
798 type_run(const char *type)
800 struct dpif_backer *backer;
804 backer = shash_find_data(&all_dpif_backers, type);
806 /* This is not necessarily a problem, since backers are only
807 * created on demand. */
811 dpif_run(backer->dpif);
813 if (timer_expired(&backer->next_expiration)) {
814 int delay = expire(backer);
815 timer_set_duration(&backer->next_expiration, delay);
818 /* Check for port changes in the dpif. */
819 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
820 struct ofproto_dpif *ofproto = NULL;
821 struct dpif_port port;
823 /* Don't report on the datapath's device. */
824 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
828 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
829 &all_ofproto_dpifs) {
830 if (sset_contains(&ofproto->ports, devname)) {
835 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
836 /* The port was removed. If we know the datapath,
837 * report it through poll_set(). If we don't, it may be
838 * notifying us of a removal we initiated, so ignore it.
839 * If there's a pending ENOBUFS, let it stand, since
840 * everything will be reevaluated. */
841 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
842 sset_add(&ofproto->port_poll_set, devname);
843 ofproto->port_poll_errno = 0;
845 dpif_port_destroy(&port);
846 } else if (!ofproto) {
847 /* The port was added, but we don't know with which
848 * ofproto we should associate it. Delete it. */
849 dpif_port_del(backer->dpif, port.port_no);
855 if (error != EAGAIN) {
856 struct ofproto_dpif *ofproto;
858 /* There was some sort of error, so propagate it to all
859 * ofprotos that use this backer. */
860 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
861 &all_ofproto_dpifs) {
862 if (ofproto->backer == backer) {
863 sset_clear(&ofproto->port_poll_set);
864 ofproto->port_poll_errno = error;
873 type_run_fast(const char *type)
875 struct dpif_backer *backer;
878 backer = shash_find_data(&all_dpif_backers, type);
880 /* This is not necessarily a problem, since backers are only
881 * created on demand. */
885 /* Handle one or more batches of upcalls, until there's nothing left to do
886 * or until we do a fixed total amount of work.
888 * We do work in batches because it can be much cheaper to set up a number
889 * of flows and fire off their patches all at once. We do multiple batches
890 * because in some cases handling a packet can cause another packet to be
891 * queued almost immediately as part of the return flow. Both
892 * optimizations can make major improvements on some benchmarks and
893 * presumably for real traffic as well. */
895 while (work < FLOW_MISS_MAX_BATCH) {
896 int retval = handle_upcalls(backer, FLOW_MISS_MAX_BATCH - work);
907 type_wait(const char *type)
909 struct dpif_backer *backer;
911 backer = shash_find_data(&all_dpif_backers, type);
913 /* This is not necessarily a problem, since backers are only
914 * created on demand. */
918 timer_wait(&backer->next_expiration);
921 /* Basic life-cycle. */
923 static int add_internal_flows(struct ofproto_dpif *);
925 static struct ofproto *
928 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
933 dealloc(struct ofproto *ofproto_)
935 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
940 close_dpif_backer(struct dpif_backer *backer)
942 struct shash_node *node;
944 assert(backer->refcount > 0);
946 if (--backer->refcount) {
950 hmap_destroy(&backer->odp_to_ofport_map);
951 node = shash_find(&all_dpif_backers, backer->type);
953 shash_delete(&all_dpif_backers, node);
954 dpif_close(backer->dpif);
959 /* Datapath port slated for removal from datapath. */
961 struct list list_node;
966 open_dpif_backer(const char *type, struct dpif_backer **backerp)
968 struct dpif_backer *backer;
969 struct dpif_port_dump port_dump;
970 struct dpif_port port;
971 struct shash_node *node;
972 struct list garbage_list;
973 struct odp_garbage *garbage, *next;
979 backer = shash_find_data(&all_dpif_backers, type);
986 backer_name = xasprintf("ovs-%s", type);
988 /* Remove any existing datapaths, since we assume we're the only
989 * userspace controlling the datapath. */
991 dp_enumerate_names(type, &names);
992 SSET_FOR_EACH(name, &names) {
993 struct dpif *old_dpif;
995 /* Don't remove our backer if it exists. */
996 if (!strcmp(name, backer_name)) {
1000 if (dpif_open(name, type, &old_dpif)) {
1001 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1003 dpif_delete(old_dpif);
1004 dpif_close(old_dpif);
1007 sset_destroy(&names);
1009 backer = xmalloc(sizeof *backer);
1011 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1014 VLOG_ERR("failed to open datapath of type %s: %s", type,
1019 backer->type = xstrdup(type);
1020 backer->refcount = 1;
1021 hmap_init(&backer->odp_to_ofport_map);
1022 timer_set_duration(&backer->next_expiration, 1000);
1025 dpif_flow_flush(backer->dpif);
1027 /* Loop through the ports already on the datapath and remove any
1028 * that we don't need anymore. */
1029 list_init(&garbage_list);
1030 dpif_port_dump_start(&port_dump, backer->dpif);
1031 while (dpif_port_dump_next(&port_dump, &port)) {
1032 node = shash_find(&init_ofp_ports, port.name);
1033 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1034 garbage = xmalloc(sizeof *garbage);
1035 garbage->odp_port = port.port_no;
1036 list_push_front(&garbage_list, &garbage->list_node);
1039 dpif_port_dump_done(&port_dump);
1041 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1042 dpif_port_del(backer->dpif, garbage->odp_port);
1043 list_remove(&garbage->list_node);
1047 shash_add(&all_dpif_backers, type, backer);
1049 error = dpif_recv_set(backer->dpif, true);
1051 VLOG_ERR("failed to listen on datapath of type %s: %s",
1052 type, strerror(error));
1053 close_dpif_backer(backer);
1061 construct(struct ofproto *ofproto_)
1063 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1064 struct shash_node *node, *next;
1069 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1074 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1075 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1077 ofproto->n_matches = 0;
1079 ofproto->netflow = NULL;
1080 ofproto->sflow = NULL;
1081 ofproto->stp = NULL;
1082 hmap_init(&ofproto->bundles);
1083 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1084 for (i = 0; i < MAX_MIRRORS; i++) {
1085 ofproto->mirrors[i] = NULL;
1087 ofproto->has_bonded_bundles = false;
1089 hmap_init(&ofproto->facets);
1090 hmap_init(&ofproto->subfacets);
1091 ofproto->governor = NULL;
1093 for (i = 0; i < N_TABLES; i++) {
1094 struct table_dpif *table = &ofproto->tables[i];
1096 table->catchall_table = NULL;
1097 table->other_table = NULL;
1098 table->basis = random_uint32();
1100 ofproto->need_revalidate = 0;
1101 tag_set_init(&ofproto->revalidate_set);
1103 list_init(&ofproto->completions);
1105 ofproto_dpif_unixctl_init();
1107 ofproto->has_mirrors = false;
1108 ofproto->has_bundle_action = false;
1110 hmap_init(&ofproto->vlandev_map);
1111 hmap_init(&ofproto->realdev_vid_map);
1113 sset_init(&ofproto->ports);
1114 sset_init(&ofproto->port_poll_set);
1115 ofproto->port_poll_errno = 0;
1117 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1118 const struct iface_hint *iface_hint = node->data;
1120 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1121 /* Check if the datapath already has this port. */
1122 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1123 sset_add(&ofproto->ports, node->name);
1126 free(iface_hint->br_name);
1127 free(iface_hint->br_type);
1128 shash_delete(&init_ofp_ports, node);
1132 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1133 hash_string(ofproto->up.name, 0));
1134 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1136 ofproto_init_tables(ofproto_, N_TABLES);
1137 error = add_internal_flows(ofproto);
1138 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1144 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1145 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1147 struct ofputil_flow_mod fm;
1150 match_init_catchall(&fm.match);
1152 match_set_reg(&fm.match, 0, id);
1153 fm.new_cookie = htonll(0);
1154 fm.cookie = htonll(0);
1155 fm.cookie_mask = htonll(0);
1156 fm.table_id = TBL_INTERNAL;
1157 fm.command = OFPFC_ADD;
1158 fm.idle_timeout = 0;
1159 fm.hard_timeout = 0;
1163 fm.ofpacts = ofpacts->data;
1164 fm.ofpacts_len = ofpacts->size;
1166 error = ofproto_flow_mod(&ofproto->up, &fm);
1168 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1169 id, ofperr_to_string(error));
1173 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1174 assert(*rulep != NULL);
1180 add_internal_flows(struct ofproto_dpif *ofproto)
1182 struct ofpact_controller *controller;
1183 uint64_t ofpacts_stub[128 / 8];
1184 struct ofpbuf ofpacts;
1188 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1191 controller = ofpact_put_CONTROLLER(&ofpacts);
1192 controller->max_len = UINT16_MAX;
1193 controller->controller_id = 0;
1194 controller->reason = OFPR_NO_MATCH;
1195 ofpact_pad(&ofpacts);
1197 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1202 ofpbuf_clear(&ofpacts);
1203 error = add_internal_flow(ofproto, id++, &ofpacts,
1204 &ofproto->no_packet_in_rule);
1209 complete_operations(struct ofproto_dpif *ofproto)
1211 struct dpif_completion *c, *next;
1213 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1214 ofoperation_complete(c->op, 0);
1215 list_remove(&c->list_node);
1221 destruct(struct ofproto *ofproto_)
1223 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1224 struct rule_dpif *rule, *next_rule;
1225 struct oftable *table;
1228 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1229 complete_operations(ofproto);
1231 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1232 struct cls_cursor cursor;
1234 cls_cursor_init(&cursor, &table->cls, NULL);
1235 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1236 ofproto_rule_destroy(&rule->up);
1240 for (i = 0; i < MAX_MIRRORS; i++) {
1241 mirror_destroy(ofproto->mirrors[i]);
1244 netflow_destroy(ofproto->netflow);
1245 dpif_sflow_destroy(ofproto->sflow);
1246 hmap_destroy(&ofproto->bundles);
1247 mac_learning_destroy(ofproto->ml);
1249 hmap_destroy(&ofproto->facets);
1250 hmap_destroy(&ofproto->subfacets);
1251 governor_destroy(ofproto->governor);
1253 hmap_destroy(&ofproto->vlandev_map);
1254 hmap_destroy(&ofproto->realdev_vid_map);
1256 sset_destroy(&ofproto->ports);
1257 sset_destroy(&ofproto->port_poll_set);
1259 close_dpif_backer(ofproto->backer);
1263 run_fast(struct ofproto *ofproto_)
1265 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1266 struct ofport_dpif *ofport;
1268 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1269 port_run_fast(ofport);
1276 run(struct ofproto *ofproto_)
1278 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1279 struct ofport_dpif *ofport;
1280 struct ofbundle *bundle;
1284 complete_operations(ofproto);
1287 error = run_fast(ofproto_);
1292 if (ofproto->netflow) {
1293 if (netflow_run(ofproto->netflow)) {
1294 send_netflow_active_timeouts(ofproto);
1297 if (ofproto->sflow) {
1298 dpif_sflow_run(ofproto->sflow);
1301 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1304 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1309 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
1311 /* Now revalidate if there's anything to do. */
1312 if (ofproto->need_revalidate
1313 || !tag_set_is_empty(&ofproto->revalidate_set)) {
1314 struct tag_set revalidate_set = ofproto->revalidate_set;
1315 bool revalidate_all = ofproto->need_revalidate;
1316 struct facet *facet;
1318 switch (ofproto->need_revalidate) {
1319 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
1320 case REV_STP: COVERAGE_INC(rev_stp); break;
1321 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
1322 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
1323 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
1326 /* Clear the revalidation flags. */
1327 tag_set_init(&ofproto->revalidate_set);
1328 ofproto->need_revalidate = 0;
1330 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
1332 || tag_set_intersects(&revalidate_set, facet->tags)) {
1333 facet_revalidate(facet);
1338 /* Check the consistency of a random facet, to aid debugging. */
1339 if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) {
1340 struct facet *facet;
1342 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1343 struct facet, hmap_node);
1344 if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) {
1345 if (!facet_check_consistency(facet)) {
1346 ofproto->need_revalidate = REV_INCONSISTENCY;
1351 if (ofproto->governor) {
1354 governor_run(ofproto->governor);
1356 /* If the governor has shrunk to its minimum size and the number of
1357 * subfacets has dwindled, then drop the governor entirely.
1359 * For hysteresis, the number of subfacets to drop the governor is
1360 * smaller than the number needed to trigger its creation. */
1361 n_subfacets = hmap_count(&ofproto->subfacets);
1362 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1363 && governor_is_idle(ofproto->governor)) {
1364 governor_destroy(ofproto->governor);
1365 ofproto->governor = NULL;
1373 wait(struct ofproto *ofproto_)
1375 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1376 struct ofport_dpif *ofport;
1377 struct ofbundle *bundle;
1379 if (!clogged && !list_is_empty(&ofproto->completions)) {
1380 poll_immediate_wake();
1383 dpif_wait(ofproto->backer->dpif);
1384 dpif_recv_wait(ofproto->backer->dpif);
1385 if (ofproto->sflow) {
1386 dpif_sflow_wait(ofproto->sflow);
1388 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
1389 poll_immediate_wake();
1391 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1394 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1395 bundle_wait(bundle);
1397 if (ofproto->netflow) {
1398 netflow_wait(ofproto->netflow);
1400 mac_learning_wait(ofproto->ml);
1402 if (ofproto->need_revalidate) {
1403 /* Shouldn't happen, but if it does just go around again. */
1404 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1405 poll_immediate_wake();
1407 if (ofproto->governor) {
1408 governor_wait(ofproto->governor);
1413 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1415 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1417 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1418 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1422 flush(struct ofproto *ofproto_)
1424 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1425 struct subfacet *subfacet, *next_subfacet;
1426 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1430 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1431 &ofproto->subfacets) {
1432 if (subfacet->path != SF_NOT_INSTALLED) {
1433 batch[n_batch++] = subfacet;
1434 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1435 subfacet_destroy_batch(ofproto, batch, n_batch);
1439 subfacet_destroy(subfacet);
1444 subfacet_destroy_batch(ofproto, batch, n_batch);
1449 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1450 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1452 *arp_match_ip = true;
1453 *actions = (OFPUTIL_A_OUTPUT |
1454 OFPUTIL_A_SET_VLAN_VID |
1455 OFPUTIL_A_SET_VLAN_PCP |
1456 OFPUTIL_A_STRIP_VLAN |
1457 OFPUTIL_A_SET_DL_SRC |
1458 OFPUTIL_A_SET_DL_DST |
1459 OFPUTIL_A_SET_NW_SRC |
1460 OFPUTIL_A_SET_NW_DST |
1461 OFPUTIL_A_SET_NW_TOS |
1462 OFPUTIL_A_SET_TP_SRC |
1463 OFPUTIL_A_SET_TP_DST |
1468 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1470 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1471 struct dpif_dp_stats s;
1473 strcpy(ots->name, "classifier");
1475 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1477 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1478 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1481 static struct ofport *
1484 struct ofport_dpif *port = xmalloc(sizeof *port);
1489 port_dealloc(struct ofport *port_)
1491 struct ofport_dpif *port = ofport_dpif_cast(port_);
1496 port_construct(struct ofport *port_)
1498 struct ofport_dpif *port = ofport_dpif_cast(port_);
1499 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1500 struct dpif_port dpif_port;
1503 ofproto->need_revalidate = REV_RECONFIGURE;
1504 port->bundle = NULL;
1506 port->tag = tag_create_random();
1507 port->may_enable = true;
1508 port->stp_port = NULL;
1509 port->stp_state = STP_DISABLED;
1510 hmap_init(&port->priorities);
1511 port->realdev_ofp_port = 0;
1512 port->vlandev_vid = 0;
1513 port->carrier_seq = netdev_get_carrier_resets(port->up.netdev);
1515 error = dpif_port_query_by_name(ofproto->backer->dpif,
1516 netdev_get_name(port->up.netdev),
1522 port->odp_port = dpif_port.port_no;
1524 /* Sanity-check that a mapping doesn't already exist. This
1525 * shouldn't happen. */
1526 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1527 VLOG_ERR("port %s already has an OpenFlow port number\n",
1532 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1533 hash_int(port->odp_port, 0));
1535 if (ofproto->sflow) {
1536 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1543 port_destruct(struct ofport *port_)
1545 struct ofport_dpif *port = ofport_dpif_cast(port_);
1546 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1547 const char *devname = netdev_get_name(port->up.netdev);
1549 if (dpif_port_exists(ofproto->backer->dpif, devname)) {
1550 /* The underlying device is still there, so delete it. This
1551 * happens when the ofproto is being destroyed, since the caller
1552 * assumes that removal of attached ports will happen as part of
1554 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1557 sset_find_and_delete(&ofproto->ports, devname);
1558 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1559 ofproto->need_revalidate = REV_RECONFIGURE;
1560 bundle_remove(port_);
1561 set_cfm(port_, NULL);
1562 if (ofproto->sflow) {
1563 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1566 ofport_clear_priorities(port);
1567 hmap_destroy(&port->priorities);
1571 port_modified(struct ofport *port_)
1573 struct ofport_dpif *port = ofport_dpif_cast(port_);
1575 if (port->bundle && port->bundle->bond) {
1576 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1581 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1583 struct ofport_dpif *port = ofport_dpif_cast(port_);
1584 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1585 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1587 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1588 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1589 OFPUTIL_PC_NO_PACKET_IN)) {
1590 ofproto->need_revalidate = REV_RECONFIGURE;
1592 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1593 bundle_update(port->bundle);
1599 set_sflow(struct ofproto *ofproto_,
1600 const struct ofproto_sflow_options *sflow_options)
1602 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1603 struct dpif_sflow *ds = ofproto->sflow;
1605 if (sflow_options) {
1607 struct ofport_dpif *ofport;
1609 ds = ofproto->sflow = dpif_sflow_create();
1610 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1611 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1613 ofproto->need_revalidate = REV_RECONFIGURE;
1615 dpif_sflow_set_options(ds, sflow_options);
1618 dpif_sflow_destroy(ds);
1619 ofproto->need_revalidate = REV_RECONFIGURE;
1620 ofproto->sflow = NULL;
1627 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1629 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1636 struct ofproto_dpif *ofproto;
1638 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1639 ofproto->need_revalidate = REV_RECONFIGURE;
1640 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1643 if (cfm_configure(ofport->cfm, s)) {
1649 cfm_destroy(ofport->cfm);
1655 get_cfm_fault(const struct ofport *ofport_)
1657 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1659 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1663 get_cfm_opup(const struct ofport *ofport_)
1665 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1667 return ofport->cfm ? cfm_get_opup(ofport->cfm) : -1;
1671 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1674 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1677 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1685 get_cfm_health(const struct ofport *ofport_)
1687 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1689 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1692 /* Spanning Tree. */
1695 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1697 struct ofproto_dpif *ofproto = ofproto_;
1698 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1699 struct ofport_dpif *ofport;
1701 ofport = stp_port_get_aux(sp);
1703 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1704 ofproto->up.name, port_num);
1706 struct eth_header *eth = pkt->l2;
1708 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1709 if (eth_addr_is_zero(eth->eth_src)) {
1710 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1711 "with unknown MAC", ofproto->up.name, port_num);
1713 send_packet(ofport, pkt);
1719 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1721 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1723 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1725 /* Only revalidate flows if the configuration changed. */
1726 if (!s != !ofproto->stp) {
1727 ofproto->need_revalidate = REV_RECONFIGURE;
1731 if (!ofproto->stp) {
1732 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1733 send_bpdu_cb, ofproto);
1734 ofproto->stp_last_tick = time_msec();
1737 stp_set_bridge_id(ofproto->stp, s->system_id);
1738 stp_set_bridge_priority(ofproto->stp, s->priority);
1739 stp_set_hello_time(ofproto->stp, s->hello_time);
1740 stp_set_max_age(ofproto->stp, s->max_age);
1741 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1743 struct ofport *ofport;
1745 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1746 set_stp_port(ofport, NULL);
1749 stp_destroy(ofproto->stp);
1750 ofproto->stp = NULL;
1757 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1759 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1763 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1764 s->designated_root = stp_get_designated_root(ofproto->stp);
1765 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1774 update_stp_port_state(struct ofport_dpif *ofport)
1776 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1777 enum stp_state state;
1779 /* Figure out new state. */
1780 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1784 if (ofport->stp_state != state) {
1785 enum ofputil_port_state of_state;
1788 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1789 netdev_get_name(ofport->up.netdev),
1790 stp_state_name(ofport->stp_state),
1791 stp_state_name(state));
1792 if (stp_learn_in_state(ofport->stp_state)
1793 != stp_learn_in_state(state)) {
1794 /* xxx Learning action flows should also be flushed. */
1795 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1797 fwd_change = stp_forward_in_state(ofport->stp_state)
1798 != stp_forward_in_state(state);
1800 ofproto->need_revalidate = REV_STP;
1801 ofport->stp_state = state;
1802 ofport->stp_state_entered = time_msec();
1804 if (fwd_change && ofport->bundle) {
1805 bundle_update(ofport->bundle);
1808 /* Update the STP state bits in the OpenFlow port description. */
1809 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1810 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1811 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1812 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1813 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1815 ofproto_port_set_state(&ofport->up, of_state);
1819 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1820 * caller is responsible for assigning STP port numbers and ensuring
1821 * there are no duplicates. */
1823 set_stp_port(struct ofport *ofport_,
1824 const struct ofproto_port_stp_settings *s)
1826 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1827 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1828 struct stp_port *sp = ofport->stp_port;
1830 if (!s || !s->enable) {
1832 ofport->stp_port = NULL;
1833 stp_port_disable(sp);
1834 update_stp_port_state(ofport);
1837 } else if (sp && stp_port_no(sp) != s->port_num
1838 && ofport == stp_port_get_aux(sp)) {
1839 /* The port-id changed, so disable the old one if it's not
1840 * already in use by another port. */
1841 stp_port_disable(sp);
1844 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1845 stp_port_enable(sp);
1847 stp_port_set_aux(sp, ofport);
1848 stp_port_set_priority(sp, s->priority);
1849 stp_port_set_path_cost(sp, s->path_cost);
1851 update_stp_port_state(ofport);
1857 get_stp_port_status(struct ofport *ofport_,
1858 struct ofproto_port_stp_status *s)
1860 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1861 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1862 struct stp_port *sp = ofport->stp_port;
1864 if (!ofproto->stp || !sp) {
1870 s->port_id = stp_port_get_id(sp);
1871 s->state = stp_port_get_state(sp);
1872 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1873 s->role = stp_port_get_role(sp);
1874 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1880 stp_run(struct ofproto_dpif *ofproto)
1883 long long int now = time_msec();
1884 long long int elapsed = now - ofproto->stp_last_tick;
1885 struct stp_port *sp;
1888 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1889 ofproto->stp_last_tick = now;
1891 while (stp_get_changed_port(ofproto->stp, &sp)) {
1892 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1895 update_stp_port_state(ofport);
1899 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1900 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1906 stp_wait(struct ofproto_dpif *ofproto)
1909 poll_timer_wait(1000);
1913 /* Returns true if STP should process 'flow'. */
1915 stp_should_process_flow(const struct flow *flow)
1917 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1921 stp_process_packet(const struct ofport_dpif *ofport,
1922 const struct ofpbuf *packet)
1924 struct ofpbuf payload = *packet;
1925 struct eth_header *eth = payload.data;
1926 struct stp_port *sp = ofport->stp_port;
1928 /* Sink packets on ports that have STP disabled when the bridge has
1930 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1934 /* Trim off padding on payload. */
1935 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1936 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1939 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1940 stp_received_bpdu(sp, payload.data, payload.size);
1944 static struct priority_to_dscp *
1945 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1947 struct priority_to_dscp *pdscp;
1950 hash = hash_int(priority, 0);
1951 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1952 if (pdscp->priority == priority) {
1960 ofport_clear_priorities(struct ofport_dpif *ofport)
1962 struct priority_to_dscp *pdscp, *next;
1964 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1965 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1971 set_queues(struct ofport *ofport_,
1972 const struct ofproto_port_queue *qdscp_list,
1975 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1976 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1977 struct hmap new = HMAP_INITIALIZER(&new);
1980 for (i = 0; i < n_qdscp; i++) {
1981 struct priority_to_dscp *pdscp;
1985 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1986 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
1991 pdscp = get_priority(ofport, priority);
1993 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1995 pdscp = xmalloc(sizeof *pdscp);
1996 pdscp->priority = priority;
1998 ofproto->need_revalidate = REV_RECONFIGURE;
2001 if (pdscp->dscp != dscp) {
2003 ofproto->need_revalidate = REV_RECONFIGURE;
2006 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2009 if (!hmap_is_empty(&ofport->priorities)) {
2010 ofport_clear_priorities(ofport);
2011 ofproto->need_revalidate = REV_RECONFIGURE;
2014 hmap_swap(&new, &ofport->priorities);
2022 /* Expires all MAC learning entries associated with 'bundle' and forces its
2023 * ofproto to revalidate every flow.
2025 * Normally MAC learning entries are removed only from the ofproto associated
2026 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2027 * are removed from every ofproto. When patch ports and SLB bonds are in use
2028 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2029 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2030 * with the host from which it migrated. */
2032 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2034 struct ofproto_dpif *ofproto = bundle->ofproto;
2035 struct mac_learning *ml = ofproto->ml;
2036 struct mac_entry *mac, *next_mac;
2038 ofproto->need_revalidate = REV_RECONFIGURE;
2039 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2040 if (mac->port.p == bundle) {
2042 struct ofproto_dpif *o;
2044 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2046 struct mac_entry *e;
2048 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2051 tag_set_add(&o->revalidate_set, e->tag);
2052 mac_learning_expire(o->ml, e);
2058 mac_learning_expire(ml, mac);
2063 static struct ofbundle *
2064 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2066 struct ofbundle *bundle;
2068 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2069 &ofproto->bundles) {
2070 if (bundle->aux == aux) {
2077 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2078 * ones that are found to 'bundles'. */
2080 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2081 void **auxes, size_t n_auxes,
2082 struct hmapx *bundles)
2086 hmapx_init(bundles);
2087 for (i = 0; i < n_auxes; i++) {
2088 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2090 hmapx_add(bundles, bundle);
2096 bundle_update(struct ofbundle *bundle)
2098 struct ofport_dpif *port;
2100 bundle->floodable = true;
2101 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2102 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2103 || !stp_forward_in_state(port->stp_state)) {
2104 bundle->floodable = false;
2111 bundle_del_port(struct ofport_dpif *port)
2113 struct ofbundle *bundle = port->bundle;
2115 bundle->ofproto->need_revalidate = REV_RECONFIGURE;
2117 list_remove(&port->bundle_node);
2118 port->bundle = NULL;
2121 lacp_slave_unregister(bundle->lacp, port);
2124 bond_slave_unregister(bundle->bond, port);
2127 bundle_update(bundle);
2131 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2132 struct lacp_slave_settings *lacp,
2133 uint32_t bond_stable_id)
2135 struct ofport_dpif *port;
2137 port = get_ofp_port(bundle->ofproto, ofp_port);
2142 if (port->bundle != bundle) {
2143 bundle->ofproto->need_revalidate = REV_RECONFIGURE;
2145 bundle_del_port(port);
2148 port->bundle = bundle;
2149 list_push_back(&bundle->ports, &port->bundle_node);
2150 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2151 || !stp_forward_in_state(port->stp_state)) {
2152 bundle->floodable = false;
2156 port->bundle->ofproto->need_revalidate = REV_RECONFIGURE;
2157 lacp_slave_register(bundle->lacp, port, lacp);
2160 port->bond_stable_id = bond_stable_id;
2166 bundle_destroy(struct ofbundle *bundle)
2168 struct ofproto_dpif *ofproto;
2169 struct ofport_dpif *port, *next_port;
2176 ofproto = bundle->ofproto;
2177 for (i = 0; i < MAX_MIRRORS; i++) {
2178 struct ofmirror *m = ofproto->mirrors[i];
2180 if (m->out == bundle) {
2182 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2183 || hmapx_find_and_delete(&m->dsts, bundle)) {
2184 ofproto->need_revalidate = REV_RECONFIGURE;
2189 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2190 bundle_del_port(port);
2193 bundle_flush_macs(bundle, true);
2194 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2196 free(bundle->trunks);
2197 lacp_destroy(bundle->lacp);
2198 bond_destroy(bundle->bond);
2203 bundle_set(struct ofproto *ofproto_, void *aux,
2204 const struct ofproto_bundle_settings *s)
2206 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2207 bool need_flush = false;
2208 struct ofport_dpif *port;
2209 struct ofbundle *bundle;
2210 unsigned long *trunks;
2216 bundle_destroy(bundle_lookup(ofproto, aux));
2220 assert(s->n_slaves == 1 || s->bond != NULL);
2221 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2223 bundle = bundle_lookup(ofproto, aux);
2225 bundle = xmalloc(sizeof *bundle);
2227 bundle->ofproto = ofproto;
2228 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2229 hash_pointer(aux, 0));
2231 bundle->name = NULL;
2233 list_init(&bundle->ports);
2234 bundle->vlan_mode = PORT_VLAN_TRUNK;
2236 bundle->trunks = NULL;
2237 bundle->use_priority_tags = s->use_priority_tags;
2238 bundle->lacp = NULL;
2239 bundle->bond = NULL;
2241 bundle->floodable = true;
2243 bundle->src_mirrors = 0;
2244 bundle->dst_mirrors = 0;
2245 bundle->mirror_out = 0;
2248 if (!bundle->name || strcmp(s->name, bundle->name)) {
2250 bundle->name = xstrdup(s->name);
2255 if (!bundle->lacp) {
2256 ofproto->need_revalidate = REV_RECONFIGURE;
2257 bundle->lacp = lacp_create();
2259 lacp_configure(bundle->lacp, s->lacp);
2261 lacp_destroy(bundle->lacp);
2262 bundle->lacp = NULL;
2265 /* Update set of ports. */
2267 for (i = 0; i < s->n_slaves; i++) {
2268 if (!bundle_add_port(bundle, s->slaves[i],
2269 s->lacp ? &s->lacp_slaves[i] : NULL,
2270 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
2274 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2275 struct ofport_dpif *next_port;
2277 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2278 for (i = 0; i < s->n_slaves; i++) {
2279 if (s->slaves[i] == port->up.ofp_port) {
2284 bundle_del_port(port);
2288 assert(list_size(&bundle->ports) <= s->n_slaves);
2290 if (list_is_empty(&bundle->ports)) {
2291 bundle_destroy(bundle);
2295 /* Set VLAN tagging mode */
2296 if (s->vlan_mode != bundle->vlan_mode
2297 || s->use_priority_tags != bundle->use_priority_tags) {
2298 bundle->vlan_mode = s->vlan_mode;
2299 bundle->use_priority_tags = s->use_priority_tags;
2304 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2305 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2307 if (vlan != bundle->vlan) {
2308 bundle->vlan = vlan;
2312 /* Get trunked VLANs. */
2313 switch (s->vlan_mode) {
2314 case PORT_VLAN_ACCESS:
2318 case PORT_VLAN_TRUNK:
2319 trunks = CONST_CAST(unsigned long *, s->trunks);
2322 case PORT_VLAN_NATIVE_UNTAGGED:
2323 case PORT_VLAN_NATIVE_TAGGED:
2324 if (vlan != 0 && (!s->trunks
2325 || !bitmap_is_set(s->trunks, vlan)
2326 || bitmap_is_set(s->trunks, 0))) {
2327 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2329 trunks = bitmap_clone(s->trunks, 4096);
2331 trunks = bitmap_allocate1(4096);
2333 bitmap_set1(trunks, vlan);
2334 bitmap_set0(trunks, 0);
2336 trunks = CONST_CAST(unsigned long *, s->trunks);
2343 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2344 free(bundle->trunks);
2345 if (trunks == s->trunks) {
2346 bundle->trunks = vlan_bitmap_clone(trunks);
2348 bundle->trunks = trunks;
2353 if (trunks != s->trunks) {
2358 if (!list_is_short(&bundle->ports)) {
2359 bundle->ofproto->has_bonded_bundles = true;
2361 if (bond_reconfigure(bundle->bond, s->bond)) {
2362 ofproto->need_revalidate = REV_RECONFIGURE;
2365 bundle->bond = bond_create(s->bond);
2366 ofproto->need_revalidate = REV_RECONFIGURE;
2369 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2370 bond_slave_register(bundle->bond, port, port->bond_stable_id,
2374 bond_destroy(bundle->bond);
2375 bundle->bond = NULL;
2378 /* If we changed something that would affect MAC learning, un-learn
2379 * everything on this port and force flow revalidation. */
2381 bundle_flush_macs(bundle, false);
2388 bundle_remove(struct ofport *port_)
2390 struct ofport_dpif *port = ofport_dpif_cast(port_);
2391 struct ofbundle *bundle = port->bundle;
2394 bundle_del_port(port);
2395 if (list_is_empty(&bundle->ports)) {
2396 bundle_destroy(bundle);
2397 } else if (list_is_short(&bundle->ports)) {
2398 bond_destroy(bundle->bond);
2399 bundle->bond = NULL;
2405 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2407 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2408 struct ofport_dpif *port = port_;
2409 uint8_t ea[ETH_ADDR_LEN];
2412 error = netdev_get_etheraddr(port->up.netdev, ea);
2414 struct ofpbuf packet;
2417 ofpbuf_init(&packet, 0);
2418 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2420 memcpy(packet_pdu, pdu, pdu_size);
2422 send_packet(port, &packet);
2423 ofpbuf_uninit(&packet);
2425 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2426 "%s (%s)", port->bundle->name,
2427 netdev_get_name(port->up.netdev), strerror(error));
2432 bundle_send_learning_packets(struct ofbundle *bundle)
2434 struct ofproto_dpif *ofproto = bundle->ofproto;
2435 int error, n_packets, n_errors;
2436 struct mac_entry *e;
2438 error = n_packets = n_errors = 0;
2439 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2440 if (e->port.p != bundle) {
2441 struct ofpbuf *learning_packet;
2442 struct ofport_dpif *port;
2446 /* The assignment to "port" is unnecessary but makes "grep"ing for
2447 * struct ofport_dpif more effective. */
2448 learning_packet = bond_compose_learning_packet(bundle->bond,
2452 ret = send_packet(port, learning_packet);
2453 ofpbuf_delete(learning_packet);
2463 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2464 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2465 "packets, last error was: %s",
2466 bundle->name, n_errors, n_packets, strerror(error));
2468 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2469 bundle->name, n_packets);
2474 bundle_run(struct ofbundle *bundle)
2477 lacp_run(bundle->lacp, send_pdu_cb);
2480 struct ofport_dpif *port;
2482 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2483 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2486 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
2487 lacp_status(bundle->lacp));
2488 if (bond_should_send_learning_packets(bundle->bond)) {
2489 bundle_send_learning_packets(bundle);
2495 bundle_wait(struct ofbundle *bundle)
2498 lacp_wait(bundle->lacp);
2501 bond_wait(bundle->bond);
2508 mirror_scan(struct ofproto_dpif *ofproto)
2512 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2513 if (!ofproto->mirrors[idx]) {
2520 static struct ofmirror *
2521 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2525 for (i = 0; i < MAX_MIRRORS; i++) {
2526 struct ofmirror *mirror = ofproto->mirrors[i];
2527 if (mirror && mirror->aux == aux) {
2535 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2537 mirror_update_dups(struct ofproto_dpif *ofproto)
2541 for (i = 0; i < MAX_MIRRORS; i++) {
2542 struct ofmirror *m = ofproto->mirrors[i];
2545 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2549 for (i = 0; i < MAX_MIRRORS; i++) {
2550 struct ofmirror *m1 = ofproto->mirrors[i];
2557 for (j = i + 1; j < MAX_MIRRORS; j++) {
2558 struct ofmirror *m2 = ofproto->mirrors[j];
2560 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2561 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2562 m2->dup_mirrors |= m1->dup_mirrors;
2569 mirror_set(struct ofproto *ofproto_, void *aux,
2570 const struct ofproto_mirror_settings *s)
2572 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2573 mirror_mask_t mirror_bit;
2574 struct ofbundle *bundle;
2575 struct ofmirror *mirror;
2576 struct ofbundle *out;
2577 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2578 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2581 mirror = mirror_lookup(ofproto, aux);
2583 mirror_destroy(mirror);
2589 idx = mirror_scan(ofproto);
2591 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2593 ofproto->up.name, MAX_MIRRORS, s->name);
2597 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2598 mirror->ofproto = ofproto;
2601 mirror->out_vlan = -1;
2602 mirror->name = NULL;
2605 if (!mirror->name || strcmp(s->name, mirror->name)) {
2607 mirror->name = xstrdup(s->name);
2610 /* Get the new configuration. */
2611 if (s->out_bundle) {
2612 out = bundle_lookup(ofproto, s->out_bundle);
2614 mirror_destroy(mirror);
2620 out_vlan = s->out_vlan;
2622 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2623 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2625 /* If the configuration has not changed, do nothing. */
2626 if (hmapx_equals(&srcs, &mirror->srcs)
2627 && hmapx_equals(&dsts, &mirror->dsts)
2628 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2629 && mirror->out == out
2630 && mirror->out_vlan == out_vlan)
2632 hmapx_destroy(&srcs);
2633 hmapx_destroy(&dsts);
2637 hmapx_swap(&srcs, &mirror->srcs);
2638 hmapx_destroy(&srcs);
2640 hmapx_swap(&dsts, &mirror->dsts);
2641 hmapx_destroy(&dsts);
2643 free(mirror->vlans);
2644 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2647 mirror->out_vlan = out_vlan;
2649 /* Update bundles. */
2650 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2651 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2652 if (hmapx_contains(&mirror->srcs, bundle)) {
2653 bundle->src_mirrors |= mirror_bit;
2655 bundle->src_mirrors &= ~mirror_bit;
2658 if (hmapx_contains(&mirror->dsts, bundle)) {
2659 bundle->dst_mirrors |= mirror_bit;
2661 bundle->dst_mirrors &= ~mirror_bit;
2664 if (mirror->out == bundle) {
2665 bundle->mirror_out |= mirror_bit;
2667 bundle->mirror_out &= ~mirror_bit;
2671 ofproto->need_revalidate = REV_RECONFIGURE;
2672 ofproto->has_mirrors = true;
2673 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2674 mirror_update_dups(ofproto);
2680 mirror_destroy(struct ofmirror *mirror)
2682 struct ofproto_dpif *ofproto;
2683 mirror_mask_t mirror_bit;
2684 struct ofbundle *bundle;
2691 ofproto = mirror->ofproto;
2692 ofproto->need_revalidate = REV_RECONFIGURE;
2693 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2695 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2696 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2697 bundle->src_mirrors &= ~mirror_bit;
2698 bundle->dst_mirrors &= ~mirror_bit;
2699 bundle->mirror_out &= ~mirror_bit;
2702 hmapx_destroy(&mirror->srcs);
2703 hmapx_destroy(&mirror->dsts);
2704 free(mirror->vlans);
2706 ofproto->mirrors[mirror->idx] = NULL;
2710 mirror_update_dups(ofproto);
2712 ofproto->has_mirrors = false;
2713 for (i = 0; i < MAX_MIRRORS; i++) {
2714 if (ofproto->mirrors[i]) {
2715 ofproto->has_mirrors = true;
2722 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2723 uint64_t *packets, uint64_t *bytes)
2725 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2726 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2729 *packets = *bytes = UINT64_MAX;
2733 *packets = mirror->packet_count;
2734 *bytes = mirror->byte_count;
2740 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2742 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2743 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2744 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2750 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2752 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2753 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2754 return bundle && bundle->mirror_out != 0;
2758 forward_bpdu_changed(struct ofproto *ofproto_)
2760 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2761 ofproto->need_revalidate = REV_RECONFIGURE;
2765 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2768 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2769 mac_learning_set_idle_time(ofproto->ml, idle_time);
2770 mac_learning_set_max_entries(ofproto->ml, max_entries);
2775 static struct ofport_dpif *
2776 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2778 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2779 return ofport ? ofport_dpif_cast(ofport) : NULL;
2782 static struct ofport_dpif *
2783 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2785 return get_ofp_port(ofproto, odp_port_to_ofp_port(ofproto, odp_port));
2789 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2790 struct ofproto_port *ofproto_port,
2791 struct dpif_port *dpif_port)
2793 ofproto_port->name = dpif_port->name;
2794 ofproto_port->type = dpif_port->type;
2795 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2799 port_run_fast(struct ofport_dpif *ofport)
2801 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2802 struct ofpbuf packet;
2804 ofpbuf_init(&packet, 0);
2805 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2806 send_packet(ofport, &packet);
2807 ofpbuf_uninit(&packet);
2812 port_run(struct ofport_dpif *ofport)
2814 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2815 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2816 bool enable = netdev_get_carrier(ofport->up.netdev);
2818 ofport->carrier_seq = carrier_seq;
2820 port_run_fast(ofport);
2822 int cfm_opup = cfm_get_opup(ofport->cfm);
2824 cfm_run(ofport->cfm);
2825 enable = enable && !cfm_get_fault(ofport->cfm);
2827 if (cfm_opup >= 0) {
2828 enable = enable && cfm_opup;
2832 if (ofport->bundle) {
2833 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2834 if (carrier_changed) {
2835 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2839 if (ofport->may_enable != enable) {
2840 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2842 if (ofproto->has_bundle_action) {
2843 ofproto->need_revalidate = REV_PORT_TOGGLED;
2847 ofport->may_enable = enable;
2851 port_wait(struct ofport_dpif *ofport)
2854 cfm_wait(ofport->cfm);
2859 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2860 struct ofproto_port *ofproto_port)
2862 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2863 struct dpif_port dpif_port;
2866 if (!sset_contains(&ofproto->ports, devname)) {
2869 error = dpif_port_query_by_name(ofproto->backer->dpif,
2870 devname, &dpif_port);
2872 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2878 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2880 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2881 uint32_t odp_port = UINT32_MAX;
2884 error = dpif_port_add(ofproto->backer->dpif, netdev, &odp_port);
2886 sset_add(&ofproto->ports, netdev_get_name(netdev));
2892 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2894 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2895 uint32_t odp_port = ofp_port_to_odp_port(ofproto, ofp_port);
2898 if (odp_port != OFPP_NONE) {
2899 error = dpif_port_del(ofproto->backer->dpif, odp_port);
2902 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2904 /* The caller is going to close ofport->up.netdev. If this is a
2905 * bonded port, then the bond is using that netdev, so remove it
2906 * from the bond. The client will need to reconfigure everything
2907 * after deleting ports, so then the slave will get re-added. */
2908 bundle_remove(&ofport->up);
2915 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2917 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2920 error = netdev_get_stats(ofport->up.netdev, stats);
2922 if (!error && ofport->odp_port == OVSP_LOCAL) {
2923 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2925 /* ofproto->stats.tx_packets represents packets that we created
2926 * internally and sent to some port (e.g. packets sent with
2927 * send_packet()). Account for them as if they had come from
2928 * OFPP_LOCAL and got forwarded. */
2930 if (stats->rx_packets != UINT64_MAX) {
2931 stats->rx_packets += ofproto->stats.tx_packets;
2934 if (stats->rx_bytes != UINT64_MAX) {
2935 stats->rx_bytes += ofproto->stats.tx_bytes;
2938 /* ofproto->stats.rx_packets represents packets that were received on
2939 * some port and we processed internally and dropped (e.g. STP).
2940 * Account for them as if they had been forwarded to OFPP_LOCAL. */
2942 if (stats->tx_packets != UINT64_MAX) {
2943 stats->tx_packets += ofproto->stats.rx_packets;
2946 if (stats->tx_bytes != UINT64_MAX) {
2947 stats->tx_bytes += ofproto->stats.rx_bytes;
2954 /* Account packets for LOCAL port. */
2956 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2957 size_t tx_size, size_t rx_size)
2959 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2962 ofproto->stats.rx_packets++;
2963 ofproto->stats.rx_bytes += rx_size;
2966 ofproto->stats.tx_packets++;
2967 ofproto->stats.tx_bytes += tx_size;
2971 struct port_dump_state {
2977 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
2979 struct port_dump_state *state;
2981 *statep = state = xmalloc(sizeof *state);
2988 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2989 struct ofproto_port *port)
2991 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2992 struct port_dump_state *state = state_;
2993 struct sset_node *node;
2995 while ((node = sset_at_position(&ofproto->ports, &state->bucket,
2999 error = port_query_by_name(ofproto_, node->name, port);
3000 if (error != ENODEV) {
3009 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3011 struct port_dump_state *state = state_;
3018 port_poll(const struct ofproto *ofproto_, char **devnamep)
3020 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3022 if (ofproto->port_poll_errno) {
3023 int error = ofproto->port_poll_errno;
3024 ofproto->port_poll_errno = 0;
3028 if (sset_is_empty(&ofproto->port_poll_set)) {
3032 *devnamep = sset_pop(&ofproto->port_poll_set);
3037 port_poll_wait(const struct ofproto *ofproto_)
3039 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3040 dpif_port_poll_wait(ofproto->backer->dpif);
3044 port_is_lacp_current(const struct ofport *ofport_)
3046 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3047 return (ofport->bundle && ofport->bundle->lacp
3048 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3052 /* Upcall handling. */
3054 /* Flow miss batching.
3056 * Some dpifs implement operations faster when you hand them off in a batch.
3057 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3058 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3059 * more packets, plus possibly installing the flow in the dpif.
3061 * So far we only batch the operations that affect flow setup time the most.
3062 * It's possible to batch more than that, but the benefit might be minimal. */
3064 struct hmap_node hmap_node;
3065 struct ofproto_dpif *ofproto;
3067 enum odp_key_fitness key_fitness;
3068 const struct nlattr *key;
3070 ovs_be16 initial_tci;
3071 struct list packets;
3072 enum dpif_upcall_type upcall_type;
3073 uint32_t odp_in_port;
3076 struct flow_miss_op {
3077 struct dpif_op dpif_op;
3078 struct subfacet *subfacet; /* Subfacet */
3079 void *garbage; /* Pointer to pass to free(), NULL if none. */
3080 uint64_t stub[1024 / 8]; /* Temporary buffer. */
3083 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3084 * OpenFlow controller as necessary according to their individual
3085 * configurations. */
3087 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3088 const struct flow *flow)
3090 struct ofputil_packet_in pin;
3092 pin.packet = packet->data;
3093 pin.packet_len = packet->size;
3094 pin.reason = OFPR_NO_MATCH;
3095 pin.controller_id = 0;
3100 pin.send_len = 0; /* not used for flow table misses */
3102 flow_get_metadata(flow, &pin.fmd);
3104 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3107 static enum slow_path_reason
3108 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3109 const struct ofpbuf *packet)
3111 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
3117 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3119 cfm_process_heartbeat(ofport->cfm, packet);
3122 } else if (ofport->bundle && ofport->bundle->lacp
3123 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3125 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3128 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3130 stp_process_packet(ofport, packet);
3137 static struct flow_miss *
3138 flow_miss_find(struct hmap *todo, const struct flow *flow, uint32_t hash)
3140 struct flow_miss *miss;
3142 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3143 if (flow_equal(&miss->flow, flow)) {
3151 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3152 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3153 * 'miss' is associated with a subfacet the caller must also initialize the
3154 * returned op->subfacet, and if anything needs to be freed after processing
3155 * the op, the caller must initialize op->garbage also. */
3157 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3158 struct flow_miss_op *op)
3160 if (miss->flow.vlan_tci != miss->initial_tci) {
3161 /* This packet was received on a VLAN splinter port. We
3162 * added a VLAN to the packet to make the packet resemble
3163 * the flow, but the actions were composed assuming that
3164 * the packet contained no VLAN. So, we must remove the
3165 * VLAN header from the packet before trying to execute the
3167 eth_pop_vlan(packet);
3170 op->subfacet = NULL;
3172 op->dpif_op.type = DPIF_OP_EXECUTE;
3173 op->dpif_op.u.execute.key = miss->key;
3174 op->dpif_op.u.execute.key_len = miss->key_len;
3175 op->dpif_op.u.execute.packet = packet;
3178 /* Helper for handle_flow_miss_without_facet() and
3179 * handle_flow_miss_with_facet(). */
3181 handle_flow_miss_common(struct rule_dpif *rule,
3182 struct ofpbuf *packet, const struct flow *flow)
3184 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3186 ofproto->n_matches++;
3188 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3190 * Extra-special case for fail-open mode.
3192 * We are in fail-open mode and the packet matched the fail-open
3193 * rule, but we are connected to a controller too. We should send
3194 * the packet up to the controller in the hope that it will try to
3195 * set up a flow and thereby allow us to exit fail-open.
3197 * See the top-level comment in fail-open.c for more information.
3199 send_packet_in_miss(ofproto, packet, flow);
3203 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3204 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3205 * installing a datapath flow. The answer is usually "yes" (a return value of
3206 * true). However, for short flows the cost of bookkeeping is much higher than
3207 * the benefits, so when the datapath holds a large number of flows we impose
3208 * some heuristics to decide which flows are likely to be worth tracking. */
3210 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3211 struct flow_miss *miss, uint32_t hash)
3213 if (!ofproto->governor) {
3216 n_subfacets = hmap_count(&ofproto->subfacets);
3217 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3221 ofproto->governor = governor_create(ofproto->up.name);
3224 return governor_should_install_flow(ofproto->governor, hash,
3225 list_size(&miss->packets));
3228 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3229 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3230 * increment '*n_ops'. */
3232 handle_flow_miss_without_facet(struct flow_miss *miss,
3233 struct rule_dpif *rule,
3234 struct flow_miss_op *ops, size_t *n_ops)
3236 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3237 long long int now = time_msec();
3238 struct action_xlate_ctx ctx;
3239 struct ofpbuf *packet;
3241 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3242 struct flow_miss_op *op = &ops[*n_ops];
3243 struct dpif_flow_stats stats;
3244 struct ofpbuf odp_actions;
3246 COVERAGE_INC(facet_suppress);
3248 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3250 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3251 rule_credit_stats(rule, &stats);
3253 action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci,
3255 ctx.resubmit_stats = &stats;
3256 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3259 if (odp_actions.size) {
3260 struct dpif_execute *execute = &op->dpif_op.u.execute;
3262 init_flow_miss_execute_op(miss, packet, op);
3263 execute->actions = odp_actions.data;
3264 execute->actions_len = odp_actions.size;
3265 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3269 ofpbuf_uninit(&odp_actions);
3274 /* Handles 'miss', which matches 'facet'. May add any required datapath
3275 * operations to 'ops', incrementing '*n_ops' for each new op.
3277 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3278 * This is really important only for new facets: if we just called time_msec()
3279 * here, then the new subfacet or its packets could look (occasionally) as
3280 * though it was used some time after the facet was used. That can make a
3281 * one-packet flow look like it has a nonzero duration, which looks odd in
3282 * e.g. NetFlow statistics. */
3284 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3286 struct flow_miss_op *ops, size_t *n_ops)
3288 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3289 enum subfacet_path want_path;
3290 struct subfacet *subfacet;
3291 struct ofpbuf *packet;
3293 subfacet = subfacet_create(facet, miss, now);
3295 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3296 struct flow_miss_op *op = &ops[*n_ops];
3297 struct dpif_flow_stats stats;
3298 struct ofpbuf odp_actions;
3300 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3302 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3303 if (!subfacet->actions || subfacet->slow) {
3304 subfacet_make_actions(subfacet, packet, &odp_actions);
3307 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3308 subfacet_update_stats(subfacet, &stats);
3310 if (subfacet->actions_len) {
3311 struct dpif_execute *execute = &op->dpif_op.u.execute;
3313 init_flow_miss_execute_op(miss, packet, op);
3314 op->subfacet = subfacet;
3315 if (!subfacet->slow) {
3316 execute->actions = subfacet->actions;
3317 execute->actions_len = subfacet->actions_len;
3318 ofpbuf_uninit(&odp_actions);
3320 execute->actions = odp_actions.data;
3321 execute->actions_len = odp_actions.size;
3322 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3327 ofpbuf_uninit(&odp_actions);
3331 want_path = subfacet_want_path(subfacet->slow);
3332 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3333 struct flow_miss_op *op = &ops[(*n_ops)++];
3334 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3336 op->subfacet = subfacet;
3338 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3339 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3340 put->key = miss->key;
3341 put->key_len = miss->key_len;
3342 if (want_path == SF_FAST_PATH) {
3343 put->actions = subfacet->actions;
3344 put->actions_len = subfacet->actions_len;
3346 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
3347 op->stub, sizeof op->stub,
3348 &put->actions, &put->actions_len);
3354 /* Handles flow miss 'miss'. May add any required datapath operations
3355 * to 'ops', incrementing '*n_ops' for each new op. */
3357 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3360 struct ofproto_dpif *ofproto = miss->ofproto;
3361 struct facet *facet;
3365 /* The caller must ensure that miss->hmap_node.hash contains
3366 * flow_hash(miss->flow, 0). */
3367 hash = miss->hmap_node.hash;
3369 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3371 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
3373 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
3374 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
3378 facet = facet_create(rule, &miss->flow, hash);
3383 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3386 /* This function does post-processing on data returned from
3387 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the
3388 * rest of the upcall processing logic. In particular, if the extracted
3389 * in_port is a VLAN splinter port, it replaces flow->in_port by the "real"
3390 * port, sets flow->vlan_tci correctly for the VLAN of the VLAN splinter
3391 * port, and pushes a VLAN header onto 'packet' (if it is nonnull). The
3392 * caller must have called odp_flow_key_to_flow() and supply 'fitness' and
3393 * 'flow' from its output. The 'flow' argument must have had the "in_port"
3394 * member converted to the OpenFlow number.
3396 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
3397 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
3398 * (This differs from the value returned in flow->vlan_tci only for packets
3399 * received on VLAN splinters.) */
3400 static enum odp_key_fitness
3401 ofproto_dpif_vsp_adjust(const struct ofproto_dpif *ofproto,
3402 enum odp_key_fitness fitness,
3403 struct flow *flow, ovs_be16 *initial_tci,
3404 struct ofpbuf *packet)
3406 if (fitness == ODP_FIT_ERROR) {
3409 *initial_tci = flow->vlan_tci;
3411 if (vsp_adjust_flow(ofproto, flow)) {
3413 /* Make the packet resemble the flow, so that it gets sent to an
3414 * OpenFlow controller properly, so that it looks correct for
3415 * sFlow, and so that flow_extract() will get the correct vlan_tci
3416 * if it is called on 'packet'.
3418 * The allocated space inside 'packet' probably also contains
3419 * 'key', that is, both 'packet' and 'key' are probably part of a
3420 * struct dpif_upcall (see the large comment on that structure
3421 * definition), so pushing data on 'packet' is in general not a
3422 * good idea since it could overwrite 'key' or free it as a side
3423 * effect. However, it's OK in this special case because we know
3424 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
3425 * will just overwrite the 4-byte "struct nlattr", which is fine
3426 * since we don't need that header anymore. */
3427 eth_push_vlan(packet, flow->vlan_tci);
3430 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
3431 if (fitness == ODP_FIT_PERFECT) {
3432 fitness = ODP_FIT_TOO_MUCH;
3440 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3443 struct dpif_upcall *upcall;
3444 struct flow_miss *miss;
3445 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3446 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3447 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3457 /* Construct the to-do list.
3459 * This just amounts to extracting the flow from each packet and sticking
3460 * the packets that have the same flow in the same "flow_miss" structure so
3461 * that we can process them together. */
3464 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3465 struct flow_miss *miss = &misses[n_misses];
3466 struct flow_miss *existing_miss;
3467 enum odp_key_fitness fitness;
3468 struct ofproto_dpif *ofproto;
3469 struct ofport_dpif *port;
3470 uint32_t odp_in_port;
3474 fitness = odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
3475 port = odp_port_to_ofport(backer, flow.in_port);
3477 /* Received packet on port for which we couldn't associate
3478 * an ofproto. This can happen if a port is removed while
3479 * traffic is being received. Print a rate-limited message
3480 * in case it happens frequently. */
3481 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
3485 ofproto = ofproto_dpif_cast(port->up.ofproto);
3486 odp_in_port = flow.in_port;
3487 flow.in_port = port->up.ofp_port;
3489 /* Obtain metadata and check userspace/kernel agreement on flow match,
3490 * then set 'flow''s header pointers. */
3491 miss->key_fitness = ofproto_dpif_vsp_adjust(ofproto, fitness,
3492 &flow, &miss->initial_tci, upcall->packet);
3493 if (miss->key_fitness == ODP_FIT_ERROR) {
3496 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3497 &flow.tunnel, flow.in_port, &miss->flow);
3499 /* Add other packets to a to-do list. */
3500 hash = flow_hash(&miss->flow, 0);
3501 existing_miss = flow_miss_find(&todo, &miss->flow, hash);
3502 if (!existing_miss) {
3503 hmap_insert(&todo, &miss->hmap_node, hash);
3504 miss->ofproto = ofproto;
3505 miss->key = upcall->key;
3506 miss->key_len = upcall->key_len;
3507 miss->upcall_type = upcall->type;
3508 miss->odp_in_port = odp_in_port;
3509 list_init(&miss->packets);
3513 miss = existing_miss;
3515 list_push_back(&miss->packets, &upcall->packet->list_node);
3518 /* Process each element in the to-do list, constructing the set of
3519 * operations to batch. */
3521 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3522 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3524 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3526 /* Execute batch. */
3527 for (i = 0; i < n_ops; i++) {
3528 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3530 dpif_operate(backer->dpif, dpif_ops, n_ops);
3532 /* Free memory and update facets. */
3533 for (i = 0; i < n_ops; i++) {
3534 struct flow_miss_op *op = &flow_miss_ops[i];
3536 switch (op->dpif_op.type) {
3537 case DPIF_OP_EXECUTE:
3540 case DPIF_OP_FLOW_PUT:
3541 if (!op->dpif_op.error) {
3542 op->subfacet->path = subfacet_want_path(op->subfacet->slow);
3546 case DPIF_OP_FLOW_DEL:
3552 hmap_destroy(&todo);
3555 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
3556 classify_upcall(const struct dpif_upcall *upcall)
3558 union user_action_cookie cookie;
3560 /* First look at the upcall type. */
3561 switch (upcall->type) {
3562 case DPIF_UC_ACTION:
3568 case DPIF_N_UC_TYPES:
3570 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3574 /* "action" upcalls need a closer look. */
3575 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3576 switch (cookie.type) {
3577 case USER_ACTION_COOKIE_SFLOW:
3578 return SFLOW_UPCALL;
3580 case USER_ACTION_COOKIE_SLOW_PATH:
3583 case USER_ACTION_COOKIE_UNSPEC:
3585 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
3591 handle_sflow_upcall(struct dpif_backer *backer,
3592 const struct dpif_upcall *upcall)
3594 struct ofproto_dpif *ofproto;
3595 union user_action_cookie cookie;
3596 enum odp_key_fitness fitness;
3597 struct ofport_dpif *port;
3598 ovs_be16 initial_tci;
3600 uint32_t odp_in_port;
3602 fitness = odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
3604 port = odp_port_to_ofport(backer, flow.in_port);
3609 ofproto = ofproto_dpif_cast(port->up.ofproto);
3610 if (!ofproto->sflow) {
3614 odp_in_port = flow.in_port;
3615 flow.in_port = port->up.ofp_port;
3616 fitness = ofproto_dpif_vsp_adjust(ofproto, fitness, &flow,
3617 &initial_tci, upcall->packet);
3618 if (fitness == ODP_FIT_ERROR) {
3622 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3623 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3624 odp_in_port, &cookie);
3628 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3630 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3631 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3632 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3637 assert(max_batch <= FLOW_MISS_MAX_BATCH);
3640 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3641 struct dpif_upcall *upcall = &misses[n_misses];
3642 struct ofpbuf *buf = &miss_bufs[n_misses];
3645 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3646 sizeof miss_buf_stubs[n_misses]);
3647 error = dpif_recv(backer->dpif, upcall, buf);
3653 switch (classify_upcall(upcall)) {
3655 /* Handle it later. */
3660 handle_sflow_upcall(backer, upcall);
3670 /* Handle deferred MISS_UPCALL processing. */
3671 handle_miss_upcalls(backer, misses, n_misses);
3672 for (i = 0; i < n_misses; i++) {
3673 ofpbuf_uninit(&miss_bufs[i]);
3679 /* Flow expiration. */
3681 static int subfacet_max_idle(const struct ofproto_dpif *);
3682 static void update_stats(struct dpif_backer *);
3683 static void rule_expire(struct rule_dpif *);
3684 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
3686 /* This function is called periodically by run(). Its job is to collect
3687 * updates for the flows that have been installed into the datapath, most
3688 * importantly when they last were used, and then use that information to
3689 * expire flows that have not been used recently.
3691 * Returns the number of milliseconds after which it should be called again. */
3693 expire(struct dpif_backer *backer)
3695 struct ofproto_dpif *ofproto;
3696 int max_idle = INT32_MAX;
3698 /* Update stats for each flow in the backer. */
3699 update_stats(backer);
3701 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3702 struct rule_dpif *rule, *next_rule;
3703 struct oftable *table;
3706 if (ofproto->backer != backer) {
3710 /* Expire subfacets that have been idle too long. */
3711 dp_max_idle = subfacet_max_idle(ofproto);
3712 expire_subfacets(ofproto, dp_max_idle);
3714 max_idle = MIN(max_idle, dp_max_idle);
3716 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3718 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
3719 struct cls_cursor cursor;
3721 cls_cursor_init(&cursor, &table->cls, NULL);
3722 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
3727 /* All outstanding data in existing flows has been accounted, so it's a
3728 * good time to do bond rebalancing. */
3729 if (ofproto->has_bonded_bundles) {
3730 struct ofbundle *bundle;
3732 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3734 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
3740 return MIN(max_idle, 1000);
3743 /* Updates flow table statistics given that the datapath just reported 'stats'
3744 * as 'subfacet''s statistics. */
3746 update_subfacet_stats(struct subfacet *subfacet,
3747 const struct dpif_flow_stats *stats)
3749 struct facet *facet = subfacet->facet;
3751 if (stats->n_packets >= subfacet->dp_packet_count) {
3752 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
3753 facet->packet_count += extra;
3755 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3758 if (stats->n_bytes >= subfacet->dp_byte_count) {
3759 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
3761 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3764 subfacet->dp_packet_count = stats->n_packets;
3765 subfacet->dp_byte_count = stats->n_bytes;
3767 facet->tcp_flags |= stats->tcp_flags;
3769 subfacet_update_time(subfacet, stats->used);
3770 if (facet->accounted_bytes < facet->byte_count) {
3772 facet_account(facet);
3773 facet->accounted_bytes = facet->byte_count;
3775 facet_push_stats(facet);
3778 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3779 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3781 delete_unexpected_flow(struct ofproto_dpif *ofproto,
3782 const struct nlattr *key, size_t key_len)
3784 if (!VLOG_DROP_WARN(&rl)) {
3788 odp_flow_key_format(key, key_len, &s);
3789 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
3793 COVERAGE_INC(facet_unexpected);
3794 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
3797 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3799 * This function also pushes statistics updates to rules which each facet
3800 * resubmits into. Generally these statistics will be accurate. However, if a
3801 * facet changes the rule it resubmits into at some time in between
3802 * update_stats() runs, it is possible that statistics accrued to the
3803 * old rule will be incorrectly attributed to the new rule. This could be
3804 * avoided by calling update_stats() whenever rules are created or
3805 * deleted. However, the performance impact of making so many calls to the
3806 * datapath do not justify the benefit of having perfectly accurate statistics.
3809 update_stats(struct dpif_backer *backer)
3811 const struct dpif_flow_stats *stats;
3812 struct dpif_flow_dump dump;
3813 const struct nlattr *key;
3816 dpif_flow_dump_start(&dump, backer->dpif);
3817 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
3819 struct subfacet *subfacet;
3820 enum odp_key_fitness fitness;
3821 struct ofproto_dpif *ofproto;
3822 struct ofport_dpif *port;
3825 fitness = odp_flow_key_to_flow(key, key_len, &flow);
3826 if (fitness == ODP_FIT_ERROR) {
3830 port = odp_port_to_ofport(backer, flow.in_port);
3832 /* This flow is for a port for which we couldn't associate an
3833 * ofproto. This can happen if a port is removed while
3834 * traffic is being received. Ignore this flow, since it
3835 * will get timed out. */
3839 ofproto = ofproto_dpif_cast(port->up.ofproto);
3840 flow.in_port = port->up.ofp_port;
3841 key_hash = odp_flow_key_hash(key, key_len);
3843 subfacet = subfacet_find(ofproto, key, key_len, key_hash, &flow);
3844 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3846 update_subfacet_stats(subfacet, stats);
3850 /* Stats are updated per-packet. */
3853 case SF_NOT_INSTALLED:
3855 delete_unexpected_flow(ofproto, key, key_len);
3859 dpif_flow_dump_done(&dump);
3862 /* Calculates and returns the number of milliseconds of idle time after which
3863 * subfacets should expire from the datapath. When a subfacet expires, we fold
3864 * its statistics into its facet, and when a facet's last subfacet expires, we
3865 * fold its statistic into its rule. */
3867 subfacet_max_idle(const struct ofproto_dpif *ofproto)
3870 * Idle time histogram.
3872 * Most of the time a switch has a relatively small number of subfacets.
3873 * When this is the case we might as well keep statistics for all of them
3874 * in userspace and to cache them in the kernel datapath for performance as
3877 * As the number of subfacets increases, the memory required to maintain
3878 * statistics about them in userspace and in the kernel becomes
3879 * significant. However, with a large number of subfacets it is likely
3880 * that only a few of them are "heavy hitters" that consume a large amount
3881 * of bandwidth. At this point, only heavy hitters are worth caching in
3882 * the kernel and maintaining in userspaces; other subfacets we can
3885 * The technique used to compute the idle time is to build a histogram with
3886 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3887 * that is installed in the kernel gets dropped in the appropriate bucket.
3888 * After the histogram has been built, we compute the cutoff so that only
3889 * the most-recently-used 1% of subfacets (but at least
3890 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3891 * the most-recently-used bucket of subfacets is kept, so actually an
3892 * arbitrary number of subfacets can be kept in any given expiration run
3893 * (though the next run will delete most of those unless they receive
3896 * This requires a second pass through the subfacets, in addition to the
3897 * pass made by update_stats(), because the former function never looks at
3898 * uninstallable subfacets.
3900 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3901 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3902 int buckets[N_BUCKETS] = { 0 };
3903 int total, subtotal, bucket;
3904 struct subfacet *subfacet;
3908 total = hmap_count(&ofproto->subfacets);
3909 if (total <= ofproto->up.flow_eviction_threshold) {
3910 return N_BUCKETS * BUCKET_WIDTH;
3913 /* Build histogram. */
3915 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3916 long long int idle = now - subfacet->used;
3917 int bucket = (idle <= 0 ? 0
3918 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3919 : (unsigned int) idle / BUCKET_WIDTH);
3923 /* Find the first bucket whose flows should be expired. */
3924 subtotal = bucket = 0;
3926 subtotal += buckets[bucket++];
3927 } while (bucket < N_BUCKETS &&
3928 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3930 if (VLOG_IS_DBG_ENABLED()) {
3934 ds_put_cstr(&s, "keep");
3935 for (i = 0; i < N_BUCKETS; i++) {
3937 ds_put_cstr(&s, ", drop");
3940 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3943 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3947 return bucket * BUCKET_WIDTH;
3951 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3953 /* Cutoff time for most flows. */
3954 long long int normal_cutoff = time_msec() - dp_max_idle;
3956 /* We really want to keep flows for special protocols around, so use a more
3957 * conservative cutoff. */
3958 long long int special_cutoff = time_msec() - 10000;
3960 struct subfacet *subfacet, *next_subfacet;
3961 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3965 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3966 &ofproto->subfacets) {
3967 long long int cutoff;
3969 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
3972 if (subfacet->used < cutoff) {
3973 if (subfacet->path != SF_NOT_INSTALLED) {
3974 batch[n_batch++] = subfacet;
3975 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
3976 subfacet_destroy_batch(ofproto, batch, n_batch);
3980 subfacet_destroy(subfacet);
3986 subfacet_destroy_batch(ofproto, batch, n_batch);
3990 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3991 * then delete it entirely. */
3993 rule_expire(struct rule_dpif *rule)
3995 struct facet *facet, *next_facet;
3999 if (rule->up.pending) {
4000 /* We'll have to expire it later. */
4004 /* Has 'rule' expired? */
4006 if (rule->up.hard_timeout
4007 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4008 reason = OFPRR_HARD_TIMEOUT;
4009 } else if (rule->up.idle_timeout
4010 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4011 reason = OFPRR_IDLE_TIMEOUT;
4016 COVERAGE_INC(ofproto_dpif_expired);
4018 /* Update stats. (This is a no-op if the rule expired due to an idle
4019 * timeout, because that only happens when the rule has no facets left.) */
4020 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4021 facet_remove(facet);
4024 /* Get rid of the rule. */
4025 ofproto_rule_expire(&rule->up, reason);
4030 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
4032 * The caller must already have determined that no facet with an identical
4033 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
4034 * the ofproto's classifier table.
4036 * 'hash' must be the return value of flow_hash(flow, 0).
4038 * The facet will initially have no subfacets. The caller should create (at
4039 * least) one subfacet with subfacet_create(). */
4040 static struct facet *
4041 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
4043 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4044 struct facet *facet;
4046 facet = xzalloc(sizeof *facet);
4047 facet->used = time_msec();
4048 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4049 list_push_back(&rule->facets, &facet->list_node);
4051 facet->flow = *flow;
4052 list_init(&facet->subfacets);
4053 netflow_flow_init(&facet->nf_flow);
4054 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4060 facet_free(struct facet *facet)
4065 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4066 * 'packet', which arrived on 'in_port'.
4068 * Takes ownership of 'packet'. */
4070 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4071 const struct nlattr *odp_actions, size_t actions_len,
4072 struct ofpbuf *packet)
4074 struct odputil_keybuf keybuf;
4078 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4079 odp_flow_key_from_flow(&key, flow,
4080 ofp_port_to_odp_port(ofproto, flow->in_port));
4082 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4083 odp_actions, actions_len, packet);
4085 ofpbuf_delete(packet);
4089 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4091 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4092 * rule's statistics, via subfacet_uninstall().
4094 * - Removes 'facet' from its rule and from ofproto->facets.
4097 facet_remove(struct facet *facet)
4099 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4100 struct subfacet *subfacet, *next_subfacet;
4102 assert(!list_is_empty(&facet->subfacets));
4104 /* First uninstall all of the subfacets to get final statistics. */
4105 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4106 subfacet_uninstall(subfacet);
4109 /* Flush the final stats to the rule.
4111 * This might require us to have at least one subfacet around so that we
4112 * can use its actions for accounting in facet_account(), which is why we
4113 * have uninstalled but not yet destroyed the subfacets. */
4114 facet_flush_stats(facet);
4116 /* Now we're really all done so destroy everything. */
4117 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4118 &facet->subfacets) {
4119 subfacet_destroy__(subfacet);
4121 hmap_remove(&ofproto->facets, &facet->hmap_node);
4122 list_remove(&facet->list_node);
4126 /* Feed information from 'facet' back into the learning table to keep it in
4127 * sync with what is actually flowing through the datapath. */
4129 facet_learn(struct facet *facet)
4131 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4132 struct action_xlate_ctx ctx;
4134 if (!facet->has_learn
4135 && !facet->has_normal
4136 && (!facet->has_fin_timeout
4137 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4141 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4142 facet->flow.vlan_tci,
4143 facet->rule, facet->tcp_flags, NULL);
4144 ctx.may_learn = true;
4145 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
4146 facet->rule->up.ofpacts_len);
4150 facet_account(struct facet *facet)
4152 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4153 struct subfacet *subfacet;
4154 const struct nlattr *a;
4159 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
4162 n_bytes = facet->byte_count - facet->accounted_bytes;
4164 /* This loop feeds byte counters to bond_account() for rebalancing to use
4165 * as a basis. We also need to track the actual VLAN on which the packet
4166 * is going to be sent to ensure that it matches the one passed to
4167 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4170 * We use the actions from an arbitrary subfacet because they should all
4171 * be equally valid for our purpose. */
4172 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
4173 struct subfacet, list_node);
4174 vlan_tci = facet->flow.vlan_tci;
4175 NL_ATTR_FOR_EACH_UNSAFE (a, left,
4176 subfacet->actions, subfacet->actions_len) {
4177 const struct ovs_action_push_vlan *vlan;
4178 struct ofport_dpif *port;
4180 switch (nl_attr_type(a)) {
4181 case OVS_ACTION_ATTR_OUTPUT:
4182 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4183 if (port && port->bundle && port->bundle->bond) {
4184 bond_account(port->bundle->bond, &facet->flow,
4185 vlan_tci_to_vid(vlan_tci), n_bytes);
4189 case OVS_ACTION_ATTR_POP_VLAN:
4190 vlan_tci = htons(0);
4193 case OVS_ACTION_ATTR_PUSH_VLAN:
4194 vlan = nl_attr_get(a);
4195 vlan_tci = vlan->vlan_tci;
4201 /* Returns true if the only action for 'facet' is to send to the controller.
4202 * (We don't report NetFlow expiration messages for such facets because they
4203 * are just part of the control logic for the network, not real traffic). */
4205 facet_is_controller_flow(struct facet *facet)
4208 const struct rule *rule = &facet->rule->up;
4209 const struct ofpact *ofpacts = rule->ofpacts;
4210 size_t ofpacts_len = rule->ofpacts_len;
4212 if (ofpacts_len > 0 &&
4213 ofpacts->type == OFPACT_CONTROLLER &&
4214 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4221 /* Folds all of 'facet''s statistics into its rule. Also updates the
4222 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4223 * 'facet''s statistics in the datapath should have been zeroed and folded into
4224 * its packet and byte counts before this function is called. */
4226 facet_flush_stats(struct facet *facet)
4228 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4229 struct subfacet *subfacet;
4231 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4232 assert(!subfacet->dp_byte_count);
4233 assert(!subfacet->dp_packet_count);
4236 facet_push_stats(facet);
4237 if (facet->accounted_bytes < facet->byte_count) {
4238 facet_account(facet);
4239 facet->accounted_bytes = facet->byte_count;
4242 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4243 struct ofexpired expired;
4244 expired.flow = facet->flow;
4245 expired.packet_count = facet->packet_count;
4246 expired.byte_count = facet->byte_count;
4247 expired.used = facet->used;
4248 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4251 facet->rule->packet_count += facet->packet_count;
4252 facet->rule->byte_count += facet->byte_count;
4254 /* Reset counters to prevent double counting if 'facet' ever gets
4256 facet_reset_counters(facet);
4258 netflow_flow_clear(&facet->nf_flow);
4259 facet->tcp_flags = 0;
4262 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4263 * Returns it if found, otherwise a null pointer.
4265 * 'hash' must be the return value of flow_hash(flow, 0).
4267 * The returned facet might need revalidation; use facet_lookup_valid()
4268 * instead if that is important. */
4269 static struct facet *
4270 facet_find(struct ofproto_dpif *ofproto,
4271 const struct flow *flow, uint32_t hash)
4273 struct facet *facet;
4275 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4276 if (flow_equal(flow, &facet->flow)) {
4284 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4285 * Returns it if found, otherwise a null pointer.
4287 * 'hash' must be the return value of flow_hash(flow, 0).
4289 * The returned facet is guaranteed to be valid. */
4290 static struct facet *
4291 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4294 struct facet *facet;
4296 facet = facet_find(ofproto, flow, hash);
4298 && (ofproto->need_revalidate
4299 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))) {
4300 facet_revalidate(facet);
4307 subfacet_path_to_string(enum subfacet_path path)
4310 case SF_NOT_INSTALLED:
4311 return "not installed";
4313 return "in fast path";
4315 return "in slow path";
4321 /* Returns the path in which a subfacet should be installed if its 'slow'
4322 * member has the specified value. */
4323 static enum subfacet_path
4324 subfacet_want_path(enum slow_path_reason slow)
4326 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
4329 /* Returns true if 'subfacet' needs to have its datapath flow updated,
4330 * supposing that its actions have been recalculated as 'want_actions' and that
4331 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
4333 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
4334 const struct ofpbuf *want_actions)
4336 enum subfacet_path want_path = subfacet_want_path(slow);
4337 return (want_path != subfacet->path
4338 || (want_path == SF_FAST_PATH
4339 && (subfacet->actions_len != want_actions->size
4340 || memcmp(subfacet->actions, want_actions->data,
4341 subfacet->actions_len))));
4345 facet_check_consistency(struct facet *facet)
4347 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4349 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4351 uint64_t odp_actions_stub[1024 / 8];
4352 struct ofpbuf odp_actions;
4354 struct rule_dpif *rule;
4355 struct subfacet *subfacet;
4356 bool may_log = false;
4359 /* Check the rule for consistency. */
4360 rule = rule_dpif_lookup(ofproto, &facet->flow);
4361 ok = rule == facet->rule;
4363 may_log = !VLOG_DROP_WARN(&rl);
4368 flow_format(&s, &facet->flow);
4369 ds_put_format(&s, ": facet associated with wrong rule (was "
4370 "table=%"PRIu8",", facet->rule->up.table_id);
4371 cls_rule_format(&facet->rule->up.cr, &s);
4372 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4374 cls_rule_format(&rule->up.cr, &s);
4375 ds_put_char(&s, ')');
4377 VLOG_WARN("%s", ds_cstr(&s));
4382 /* Check the datapath actions for consistency. */
4383 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4384 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4385 enum subfacet_path want_path;
4386 struct odputil_keybuf keybuf;
4387 struct action_xlate_ctx ctx;
4391 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4392 subfacet->initial_tci, rule, 0, NULL);
4393 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
4396 if (subfacet->path == SF_NOT_INSTALLED) {
4397 /* This only happens if the datapath reported an error when we
4398 * tried to install the flow. Don't flag another error here. */
4402 want_path = subfacet_want_path(subfacet->slow);
4403 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
4404 /* The actions for slow-path flows may legitimately vary from one
4405 * packet to the next. We're done. */
4409 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
4413 /* Inconsistency! */
4415 may_log = !VLOG_DROP_WARN(&rl);
4419 /* Rate-limited, skip reporting. */
4424 subfacet_get_key(subfacet, &keybuf, &key);
4425 odp_flow_key_format(key.data, key.size, &s);
4427 ds_put_cstr(&s, ": inconsistency in subfacet");
4428 if (want_path != subfacet->path) {
4429 enum odp_key_fitness fitness = subfacet->key_fitness;
4431 ds_put_format(&s, " (%s, fitness=%s)",
4432 subfacet_path_to_string(subfacet->path),
4433 odp_key_fitness_to_string(fitness));
4434 ds_put_format(&s, " (should have been %s)",
4435 subfacet_path_to_string(want_path));
4436 } else if (want_path == SF_FAST_PATH) {
4437 ds_put_cstr(&s, " (actions were: ");
4438 format_odp_actions(&s, subfacet->actions,
4439 subfacet->actions_len);
4440 ds_put_cstr(&s, ") (correct actions: ");
4441 format_odp_actions(&s, odp_actions.data, odp_actions.size);
4442 ds_put_char(&s, ')');
4444 ds_put_cstr(&s, " (actions: ");
4445 format_odp_actions(&s, subfacet->actions,
4446 subfacet->actions_len);
4447 ds_put_char(&s, ')');
4449 VLOG_WARN("%s", ds_cstr(&s));
4452 ofpbuf_uninit(&odp_actions);
4457 /* Re-searches the classifier for 'facet':
4459 * - If the rule found is different from 'facet''s current rule, moves
4460 * 'facet' to the new rule and recompiles its actions.
4462 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4463 * where it is and recompiles its actions anyway. */
4465 facet_revalidate(struct facet *facet)
4467 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4469 struct nlattr *odp_actions;
4472 struct actions *new_actions;
4474 struct action_xlate_ctx ctx;
4475 uint64_t odp_actions_stub[1024 / 8];
4476 struct ofpbuf odp_actions;
4478 struct rule_dpif *new_rule;
4479 struct subfacet *subfacet;
4482 COVERAGE_INC(facet_revalidate);
4484 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
4486 /* Calculate new datapath actions.
4488 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4489 * emit a NetFlow expiration and, if so, we need to have the old state
4490 * around to properly compose it. */
4492 /* If the datapath actions changed or the installability changed,
4493 * then we need to talk to the datapath. */
4496 memset(&ctx, 0, sizeof ctx);
4497 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4498 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4499 enum slow_path_reason slow;
4501 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4502 subfacet->initial_tci, new_rule, 0, NULL);
4503 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
4506 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4507 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
4508 struct dpif_flow_stats stats;
4510 subfacet_install(subfacet,
4511 odp_actions.data, odp_actions.size, &stats, slow);
4512 subfacet_update_stats(subfacet, &stats);
4515 new_actions = xcalloc(list_size(&facet->subfacets),
4516 sizeof *new_actions);
4518 new_actions[i].odp_actions = xmemdup(odp_actions.data,
4520 new_actions[i].actions_len = odp_actions.size;
4525 ofpbuf_uninit(&odp_actions);
4528 facet_flush_stats(facet);
4531 /* Update 'facet' now that we've taken care of all the old state. */
4532 facet->tags = ctx.tags;
4533 facet->nf_flow.output_iface = ctx.nf_output_iface;
4534 facet->has_learn = ctx.has_learn;
4535 facet->has_normal = ctx.has_normal;
4536 facet->has_fin_timeout = ctx.has_fin_timeout;
4537 facet->mirrors = ctx.mirrors;
4540 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4541 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4543 if (new_actions && new_actions[i].odp_actions) {
4544 free(subfacet->actions);
4545 subfacet->actions = new_actions[i].odp_actions;
4546 subfacet->actions_len = new_actions[i].actions_len;
4552 if (facet->rule != new_rule) {
4553 COVERAGE_INC(facet_changed_rule);
4554 list_remove(&facet->list_node);
4555 list_push_back(&new_rule->facets, &facet->list_node);
4556 facet->rule = new_rule;
4557 facet->used = new_rule->up.created;
4558 facet->prev_used = facet->used;
4562 /* Updates 'facet''s used time. Caller is responsible for calling
4563 * facet_push_stats() to update the flows which 'facet' resubmits into. */
4565 facet_update_time(struct facet *facet, long long int used)
4567 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4568 if (used > facet->used) {
4570 ofproto_rule_update_used(&facet->rule->up, used);
4571 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
4576 facet_reset_counters(struct facet *facet)
4578 facet->packet_count = 0;
4579 facet->byte_count = 0;
4580 facet->prev_packet_count = 0;
4581 facet->prev_byte_count = 0;
4582 facet->accounted_bytes = 0;
4586 facet_push_stats(struct facet *facet)
4588 struct dpif_flow_stats stats;
4590 assert(facet->packet_count >= facet->prev_packet_count);
4591 assert(facet->byte_count >= facet->prev_byte_count);
4592 assert(facet->used >= facet->prev_used);
4594 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4595 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4596 stats.used = facet->used;
4597 stats.tcp_flags = 0;
4599 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
4600 facet->prev_packet_count = facet->packet_count;
4601 facet->prev_byte_count = facet->byte_count;
4602 facet->prev_used = facet->used;
4604 flow_push_stats(facet->rule, &facet->flow, &stats);
4606 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
4607 facet->mirrors, stats.n_packets, stats.n_bytes);
4612 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4614 rule->packet_count += stats->n_packets;
4615 rule->byte_count += stats->n_bytes;
4616 ofproto_rule_update_used(&rule->up, stats->used);
4619 /* Pushes flow statistics to the rules which 'flow' resubmits into given
4620 * 'rule''s actions and mirrors. */
4622 flow_push_stats(struct rule_dpif *rule,
4623 const struct flow *flow, const struct dpif_flow_stats *stats)
4625 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4626 struct action_xlate_ctx ctx;
4628 ofproto_rule_update_used(&rule->up, stats->used);
4630 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule,
4632 ctx.resubmit_stats = stats;
4633 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
4634 rule->up.ofpacts_len);
4639 static struct subfacet *
4640 subfacet_find(struct ofproto_dpif *ofproto,
4641 const struct nlattr *key, size_t key_len, uint32_t key_hash,
4642 const struct flow *flow)
4644 struct subfacet *subfacet;
4646 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4647 &ofproto->subfacets) {
4649 ? (subfacet->key_len == key_len
4650 && !memcmp(key, subfacet->key, key_len))
4651 : flow_equal(flow, &subfacet->facet->flow)) {
4659 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4660 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4661 * existing subfacet if there is one, otherwise creates and returns a
4664 * If the returned subfacet is new, then subfacet->actions will be NULL, in
4665 * which case the caller must populate the actions with
4666 * subfacet_make_actions(). */
4667 static struct subfacet *
4668 subfacet_create(struct facet *facet, struct flow_miss *miss,
4671 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4672 enum odp_key_fitness key_fitness = miss->key_fitness;
4673 const struct nlattr *key = miss->key;
4674 size_t key_len = miss->key_len;
4676 struct subfacet *subfacet;
4678 key_hash = odp_flow_key_hash(key, key_len);
4680 if (list_is_empty(&facet->subfacets)) {
4681 subfacet = &facet->one_subfacet;
4683 subfacet = subfacet_find(ofproto, key, key_len, key_hash,
4686 if (subfacet->facet == facet) {
4690 /* This shouldn't happen. */
4691 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4692 subfacet_destroy(subfacet);
4695 subfacet = xmalloc(sizeof *subfacet);
4698 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
4699 list_push_back(&facet->subfacets, &subfacet->list_node);
4700 subfacet->facet = facet;
4701 subfacet->key_fitness = key_fitness;
4702 if (key_fitness != ODP_FIT_PERFECT) {
4703 subfacet->key = xmemdup(key, key_len);
4704 subfacet->key_len = key_len;
4706 subfacet->key = NULL;
4707 subfacet->key_len = 0;
4709 subfacet->used = now;
4710 subfacet->dp_packet_count = 0;
4711 subfacet->dp_byte_count = 0;
4712 subfacet->actions_len = 0;
4713 subfacet->actions = NULL;
4714 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
4717 subfacet->path = SF_NOT_INSTALLED;
4718 subfacet->initial_tci = miss->initial_tci;
4719 subfacet->odp_in_port = miss->odp_in_port;
4724 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4725 * its facet within 'ofproto', and frees it. */
4727 subfacet_destroy__(struct subfacet *subfacet)
4729 struct facet *facet = subfacet->facet;
4730 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4732 subfacet_uninstall(subfacet);
4733 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
4734 list_remove(&subfacet->list_node);
4735 free(subfacet->key);
4736 free(subfacet->actions);
4737 if (subfacet != &facet->one_subfacet) {
4742 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4743 * last remaining subfacet in its facet destroys the facet too. */
4745 subfacet_destroy(struct subfacet *subfacet)
4747 struct facet *facet = subfacet->facet;
4749 if (list_is_singleton(&facet->subfacets)) {
4750 /* facet_remove() needs at least one subfacet (it will remove it). */
4751 facet_remove(facet);
4753 subfacet_destroy__(subfacet);
4758 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
4759 struct subfacet **subfacets, int n)
4761 struct odputil_keybuf keybufs[SUBFACET_DESTROY_MAX_BATCH];
4762 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4763 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4764 struct ofpbuf keys[SUBFACET_DESTROY_MAX_BATCH];
4765 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4768 for (i = 0; i < n; i++) {
4769 ops[i].type = DPIF_OP_FLOW_DEL;
4770 subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]);
4771 ops[i].u.flow_del.key = keys[i].data;
4772 ops[i].u.flow_del.key_len = keys[i].size;
4773 ops[i].u.flow_del.stats = &stats[i];
4777 dpif_operate(ofproto->backer->dpif, opsp, n);
4778 for (i = 0; i < n; i++) {
4779 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4780 subfacets[i]->path = SF_NOT_INSTALLED;
4781 subfacet_destroy(subfacets[i]);
4785 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
4786 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
4787 * for use as temporary storage. */
4789 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
4793 if (!subfacet->key) {
4794 struct flow *flow = &subfacet->facet->flow;
4796 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
4797 odp_flow_key_from_flow(key, flow, subfacet->odp_in_port);
4799 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
4803 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
4804 * Translates the actions into 'odp_actions', which the caller must have
4805 * initialized and is responsible for uninitializing. */
4807 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
4808 struct ofpbuf *odp_actions)
4810 struct facet *facet = subfacet->facet;
4811 struct rule_dpif *rule = facet->rule;
4812 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4814 struct action_xlate_ctx ctx;
4816 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
4818 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
4819 facet->tags = ctx.tags;
4820 facet->has_learn = ctx.has_learn;
4821 facet->has_normal = ctx.has_normal;
4822 facet->has_fin_timeout = ctx.has_fin_timeout;
4823 facet->nf_flow.output_iface = ctx.nf_output_iface;
4824 facet->mirrors = ctx.mirrors;
4826 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4827 if (subfacet->actions_len != odp_actions->size
4828 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
4829 free(subfacet->actions);
4830 subfacet->actions_len = odp_actions->size;
4831 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
4835 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4836 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4837 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4838 * since 'subfacet' was last updated.
4840 * Returns 0 if successful, otherwise a positive errno value. */
4842 subfacet_install(struct subfacet *subfacet,
4843 const struct nlattr *actions, size_t actions_len,
4844 struct dpif_flow_stats *stats,
4845 enum slow_path_reason slow)
4847 struct facet *facet = subfacet->facet;
4848 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4849 enum subfacet_path path = subfacet_want_path(slow);
4850 uint64_t slow_path_stub[128 / 8];
4851 struct odputil_keybuf keybuf;
4852 enum dpif_flow_put_flags flags;
4856 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
4858 flags |= DPIF_FP_ZERO_STATS;
4861 if (path == SF_SLOW_PATH) {
4862 compose_slow_path(ofproto, &facet->flow, slow,
4863 slow_path_stub, sizeof slow_path_stub,
4864 &actions, &actions_len);
4867 subfacet_get_key(subfacet, &keybuf, &key);
4868 ret = dpif_flow_put(ofproto->backer->dpif, flags, key.data, key.size,
4869 actions, actions_len, stats);
4872 subfacet_reset_dp_stats(subfacet, stats);
4876 subfacet->path = path;
4882 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
4884 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
4885 stats, subfacet->slow);
4888 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4890 subfacet_uninstall(struct subfacet *subfacet)
4892 if (subfacet->path != SF_NOT_INSTALLED) {
4893 struct rule_dpif *rule = subfacet->facet->rule;
4894 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4895 struct odputil_keybuf keybuf;
4896 struct dpif_flow_stats stats;
4900 subfacet_get_key(subfacet, &keybuf, &key);
4901 error = dpif_flow_del(ofproto->backer->dpif,
4902 key.data, key.size, &stats);
4903 subfacet_reset_dp_stats(subfacet, &stats);
4905 subfacet_update_stats(subfacet, &stats);
4907 subfacet->path = SF_NOT_INSTALLED;
4909 assert(subfacet->dp_packet_count == 0);
4910 assert(subfacet->dp_byte_count == 0);
4914 /* Resets 'subfacet''s datapath statistics counters. This should be called
4915 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4916 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4917 * was reset in the datapath. 'stats' will be modified to include only
4918 * statistics new since 'subfacet' was last updated. */
4920 subfacet_reset_dp_stats(struct subfacet *subfacet,
4921 struct dpif_flow_stats *stats)
4924 && subfacet->dp_packet_count <= stats->n_packets
4925 && subfacet->dp_byte_count <= stats->n_bytes) {
4926 stats->n_packets -= subfacet->dp_packet_count;
4927 stats->n_bytes -= subfacet->dp_byte_count;
4930 subfacet->dp_packet_count = 0;
4931 subfacet->dp_byte_count = 0;
4934 /* Updates 'subfacet''s used time. The caller is responsible for calling
4935 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
4937 subfacet_update_time(struct subfacet *subfacet, long long int used)
4939 if (used > subfacet->used) {
4940 subfacet->used = used;
4941 facet_update_time(subfacet->facet, used);
4945 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4947 * Because of the meaning of a subfacet's counters, it only makes sense to do
4948 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4949 * represents a packet that was sent by hand or if it represents statistics
4950 * that have been cleared out of the datapath. */
4952 subfacet_update_stats(struct subfacet *subfacet,
4953 const struct dpif_flow_stats *stats)
4955 if (stats->n_packets || stats->used > subfacet->used) {
4956 struct facet *facet = subfacet->facet;
4958 subfacet_update_time(subfacet, stats->used);
4959 facet->packet_count += stats->n_packets;
4960 facet->byte_count += stats->n_bytes;
4961 facet->tcp_flags |= stats->tcp_flags;
4962 facet_push_stats(facet);
4963 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
4969 static struct rule_dpif *
4970 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
4972 struct rule_dpif *rule;
4974 rule = rule_dpif_lookup__(ofproto, flow, 0);
4979 return rule_dpif_miss_rule(ofproto, flow);
4982 static struct rule_dpif *
4983 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
4986 struct cls_rule *cls_rule;
4987 struct classifier *cls;
4989 if (table_id >= N_TABLES) {
4993 cls = &ofproto->up.tables[table_id].cls;
4994 if (flow->nw_frag & FLOW_NW_FRAG_ANY
4995 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4996 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
4997 * are unavailable. */
4998 struct flow ofpc_normal_flow = *flow;
4999 ofpc_normal_flow.tp_src = htons(0);
5000 ofpc_normal_flow.tp_dst = htons(0);
5001 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5003 cls_rule = classifier_lookup(cls, flow);
5005 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5008 static struct rule_dpif *
5009 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5011 struct ofport_dpif *port;
5013 port = get_ofp_port(ofproto, flow->in_port);
5015 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5016 return ofproto->miss_rule;
5019 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5020 return ofproto->no_packet_in_rule;
5022 return ofproto->miss_rule;
5026 complete_operation(struct rule_dpif *rule)
5028 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5030 rule_invalidate(rule);
5032 struct dpif_completion *c = xmalloc(sizeof *c);
5033 c->op = rule->up.pending;
5034 list_push_back(&ofproto->completions, &c->list_node);
5036 ofoperation_complete(rule->up.pending, 0);
5040 static struct rule *
5043 struct rule_dpif *rule = xmalloc(sizeof *rule);
5048 rule_dealloc(struct rule *rule_)
5050 struct rule_dpif *rule = rule_dpif_cast(rule_);
5055 rule_construct(struct rule *rule_)
5057 struct rule_dpif *rule = rule_dpif_cast(rule_);
5058 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5059 struct rule_dpif *victim;
5062 rule->packet_count = 0;
5063 rule->byte_count = 0;
5065 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5066 if (victim && !list_is_empty(&victim->facets)) {
5067 struct facet *facet;
5069 rule->facets = victim->facets;
5070 list_moved(&rule->facets);
5071 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5072 /* XXX: We're only clearing our local counters here. It's possible
5073 * that quite a few packets are unaccounted for in the datapath
5074 * statistics. These will be accounted to the new rule instead of
5075 * cleared as required. This could be fixed by clearing out the
5076 * datapath statistics for this facet, but currently it doesn't
5078 facet_reset_counters(facet);
5082 /* Must avoid list_moved() in this case. */
5083 list_init(&rule->facets);
5086 table_id = rule->up.table_id;
5088 rule->tag = victim->tag;
5089 } else if (table_id == 0) {
5094 miniflow_expand(&rule->up.cr.match.flow, &flow);
5095 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5096 ofproto->tables[table_id].basis);
5099 complete_operation(rule);
5104 rule_destruct(struct rule *rule_)
5106 struct rule_dpif *rule = rule_dpif_cast(rule_);
5107 struct facet *facet, *next_facet;
5109 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5110 facet_revalidate(facet);
5113 complete_operation(rule);
5117 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5119 struct rule_dpif *rule = rule_dpif_cast(rule_);
5120 struct facet *facet;
5122 /* Start from historical data for 'rule' itself that are no longer tracked
5123 * in facets. This counts, for example, facets that have expired. */
5124 *packets = rule->packet_count;
5125 *bytes = rule->byte_count;
5127 /* Add any statistics that are tracked by facets. This includes
5128 * statistical data recently updated by ofproto_update_stats() as well as
5129 * stats for packets that were executed "by hand" via dpif_execute(). */
5130 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5131 *packets += facet->packet_count;
5132 *bytes += facet->byte_count;
5137 rule_execute(struct rule *rule_, const struct flow *flow,
5138 struct ofpbuf *packet)
5140 struct rule_dpif *rule = rule_dpif_cast(rule_);
5141 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5143 struct dpif_flow_stats stats;
5145 struct action_xlate_ctx ctx;
5146 uint64_t odp_actions_stub[1024 / 8];
5147 struct ofpbuf odp_actions;
5149 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5150 rule_credit_stats(rule, &stats);
5152 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5153 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
5154 rule, stats.tcp_flags, packet);
5155 ctx.resubmit_stats = &stats;
5156 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
5158 execute_odp_actions(ofproto, flow, odp_actions.data,
5159 odp_actions.size, packet);
5161 ofpbuf_uninit(&odp_actions);
5167 rule_modify_actions(struct rule *rule_)
5169 struct rule_dpif *rule = rule_dpif_cast(rule_);
5171 complete_operation(rule);
5174 /* Sends 'packet' out 'ofport'.
5175 * May modify 'packet'.
5176 * Returns 0 if successful, otherwise a positive errno value. */
5178 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5180 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5181 struct ofpbuf key, odp_actions;
5182 struct odputil_keybuf keybuf;
5187 flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow);
5188 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5190 if (odp_port != ofport->odp_port) {
5191 eth_pop_vlan(packet);
5192 flow.vlan_tci = htons(0);
5195 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5196 odp_flow_key_from_flow(&key, &flow,
5197 ofp_port_to_odp_port(ofproto, flow.in_port));
5199 ofpbuf_init(&odp_actions, 32);
5200 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5202 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5203 error = dpif_execute(ofproto->backer->dpif,
5205 odp_actions.data, odp_actions.size,
5207 ofpbuf_uninit(&odp_actions);
5210 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5211 ofproto->up.name, odp_port, strerror(error));
5213 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5217 /* OpenFlow to datapath action translation. */
5219 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5220 struct action_xlate_ctx *);
5221 static void xlate_normal(struct action_xlate_ctx *);
5223 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5224 * The action will state 'slow' as the reason that the action is in the slow
5225 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5226 * dump-flows" output to see why a flow is in the slow path.)
5228 * The 'stub_size' bytes in 'stub' will be used to store the action.
5229 * 'stub_size' must be large enough for the action.
5231 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5234 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5235 enum slow_path_reason slow,
5236 uint64_t *stub, size_t stub_size,
5237 const struct nlattr **actionsp, size_t *actions_lenp)
5239 union user_action_cookie cookie;
5242 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5243 cookie.slow_path.unused = 0;
5244 cookie.slow_path.reason = slow;
5246 ofpbuf_use_stack(&buf, stub, stub_size);
5247 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
5248 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT16_MAX);
5249 odp_put_userspace_action(pid, &cookie, &buf);
5251 put_userspace_action(ofproto, &buf, flow, &cookie);
5253 *actionsp = buf.data;
5254 *actions_lenp = buf.size;
5258 put_userspace_action(const struct ofproto_dpif *ofproto,
5259 struct ofpbuf *odp_actions,
5260 const struct flow *flow,
5261 const union user_action_cookie *cookie)
5265 pid = dpif_port_get_pid(ofproto->backer->dpif,
5266 ofp_port_to_odp_port(ofproto, flow->in_port));
5268 return odp_put_userspace_action(pid, cookie, odp_actions);
5272 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5273 ovs_be16 vlan_tci, uint32_t odp_port,
5274 unsigned int n_outputs, union user_action_cookie *cookie)
5278 cookie->type = USER_ACTION_COOKIE_SFLOW;
5279 cookie->sflow.vlan_tci = vlan_tci;
5281 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5282 * port information") for the interpretation of cookie->output. */
5283 switch (n_outputs) {
5285 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5286 cookie->sflow.output = 0x40000000 | 256;
5290 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5292 cookie->sflow.output = ifindex;
5297 /* 0x80000000 means "multiple output ports. */
5298 cookie->sflow.output = 0x80000000 | n_outputs;
5303 /* Compose SAMPLE action for sFlow. */
5305 compose_sflow_action(const struct ofproto_dpif *ofproto,
5306 struct ofpbuf *odp_actions,
5307 const struct flow *flow,
5310 uint32_t probability;
5311 union user_action_cookie cookie;
5312 size_t sample_offset, actions_offset;
5315 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5319 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5321 /* Number of packets out of UINT_MAX to sample. */
5322 probability = dpif_sflow_get_probability(ofproto->sflow);
5323 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5325 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5326 compose_sflow_cookie(ofproto, htons(0), odp_port,
5327 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5328 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
5330 nl_msg_end_nested(odp_actions, actions_offset);
5331 nl_msg_end_nested(odp_actions, sample_offset);
5332 return cookie_offset;
5335 /* SAMPLE action must be first action in any given list of actions.
5336 * At this point we do not have all information required to build it. So try to
5337 * build sample action as complete as possible. */
5339 add_sflow_action(struct action_xlate_ctx *ctx)
5341 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5343 &ctx->flow, OVSP_NONE);
5344 ctx->sflow_odp_port = 0;
5345 ctx->sflow_n_outputs = 0;
5348 /* Fix SAMPLE action according to data collected while composing ODP actions.
5349 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5350 * USERSPACE action's user-cookie which is required for sflow. */
5352 fix_sflow_action(struct action_xlate_ctx *ctx)
5354 const struct flow *base = &ctx->base_flow;
5355 union user_action_cookie *cookie;
5357 if (!ctx->user_cookie_offset) {
5361 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
5363 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5365 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5366 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5370 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
5373 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5374 uint32_t odp_port = ofp_port_to_odp_port(ctx->ofproto, ofp_port);
5375 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
5376 uint8_t flow_nw_tos = ctx->flow.nw_tos;
5380 struct priority_to_dscp *pdscp;
5382 if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5383 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5385 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5386 xlate_report(ctx, "STP not in forwarding state, skipping output");
5390 pdscp = get_priority(ofport, ctx->flow.skb_priority);
5392 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5393 ctx->flow.nw_tos |= pdscp->dscp;
5396 /* We may not have an ofport record for this port, but it doesn't hurt
5397 * to allow forwarding to it anyhow. Maybe such a port will appear
5398 * later and we're pre-populating the flow table. */
5401 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
5402 ctx->flow.vlan_tci);
5403 if (out_port != odp_port) {
5404 ctx->flow.vlan_tci = htons(0);
5406 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
5407 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
5409 ctx->sflow_odp_port = odp_port;
5410 ctx->sflow_n_outputs++;
5411 ctx->nf_output_iface = ofp_port;
5412 ctx->flow.vlan_tci = flow_vlan_tci;
5413 ctx->flow.nw_tos = flow_nw_tos;
5417 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
5419 compose_output_action__(ctx, ofp_port, true);
5423 xlate_table_action(struct action_xlate_ctx *ctx,
5424 uint16_t in_port, uint8_t table_id, bool may_packet_in)
5426 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
5427 struct ofproto_dpif *ofproto = ctx->ofproto;
5428 struct rule_dpif *rule;
5429 uint16_t old_in_port;
5430 uint8_t old_table_id;
5432 old_table_id = ctx->table_id;
5433 ctx->table_id = table_id;
5435 /* Look up a flow with 'in_port' as the input port. */
5436 old_in_port = ctx->flow.in_port;
5437 ctx->flow.in_port = in_port;
5438 rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id);
5441 if (table_id > 0 && table_id < N_TABLES) {
5442 struct table_dpif *table = &ofproto->tables[table_id];
5443 if (table->other_table) {
5444 ctx->tags |= (rule && rule->tag
5446 : rule_calculate_tag(&ctx->flow,
5447 &table->other_table->mask,
5452 /* Restore the original input port. Otherwise OFPP_NORMAL and
5453 * OFPP_IN_PORT will have surprising behavior. */
5454 ctx->flow.in_port = old_in_port;
5456 if (ctx->resubmit_hook) {
5457 ctx->resubmit_hook(ctx, rule);
5460 if (rule == NULL && may_packet_in) {
5462 * check if table configuration flags
5463 * OFPTC_TABLE_MISS_CONTROLLER, default.
5464 * OFPTC_TABLE_MISS_CONTINUE,
5465 * OFPTC_TABLE_MISS_DROP
5466 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
5468 rule = rule_dpif_miss_rule(ofproto, &ctx->flow);
5472 struct rule_dpif *old_rule = ctx->rule;
5474 if (ctx->resubmit_stats) {
5475 rule_credit_stats(rule, ctx->resubmit_stats);
5480 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
5481 ctx->rule = old_rule;
5485 ctx->table_id = old_table_id;
5487 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5489 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
5490 MAX_RESUBMIT_RECURSION);
5491 ctx->max_resubmit_trigger = true;
5496 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
5497 const struct ofpact_resubmit *resubmit)
5502 in_port = resubmit->in_port;
5503 if (in_port == OFPP_IN_PORT) {
5504 in_port = ctx->flow.in_port;
5507 table_id = resubmit->table_id;
5508 if (table_id == 255) {
5509 table_id = ctx->table_id;
5512 xlate_table_action(ctx, in_port, table_id, false);
5516 flood_packets(struct action_xlate_ctx *ctx, bool all)
5518 struct ofport_dpif *ofport;
5520 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
5521 uint16_t ofp_port = ofport->up.ofp_port;
5523 if (ofp_port == ctx->flow.in_port) {
5528 compose_output_action__(ctx, ofp_port, false);
5529 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
5530 compose_output_action(ctx, ofp_port);
5534 ctx->nf_output_iface = NF_OUT_FLOOD;
5538 execute_controller_action(struct action_xlate_ctx *ctx, int len,
5539 enum ofp_packet_in_reason reason,
5540 uint16_t controller_id)
5542 struct ofputil_packet_in pin;
5543 struct ofpbuf *packet;
5545 ctx->slow |= SLOW_CONTROLLER;
5550 packet = ofpbuf_clone(ctx->packet);
5552 if (packet->l2 && packet->l3) {
5553 struct eth_header *eh;
5555 eth_pop_vlan(packet);
5558 /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
5559 * LLC frame. Calculating the Ethernet type of these frames is more
5560 * trouble than seems appropriate for a simple assertion. */
5561 assert(ntohs(eh->eth_type) < ETH_TYPE_MIN
5562 || eh->eth_type == ctx->flow.dl_type);
5564 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
5565 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
5567 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
5568 eth_push_vlan(packet, ctx->flow.vlan_tci);
5572 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5573 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
5574 ctx->flow.nw_tos, ctx->flow.nw_ttl);
5578 if (ctx->flow.nw_proto == IPPROTO_TCP) {
5579 packet_set_tcp_port(packet, ctx->flow.tp_src,
5581 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
5582 packet_set_udp_port(packet, ctx->flow.tp_src,
5589 pin.packet = packet->data;
5590 pin.packet_len = packet->size;
5591 pin.reason = reason;
5592 pin.controller_id = controller_id;
5593 pin.table_id = ctx->table_id;
5594 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
5597 flow_get_metadata(&ctx->flow, &pin.fmd);
5599 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
5600 ofpbuf_delete(packet);
5604 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
5606 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
5607 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
5611 if (ctx->flow.nw_ttl > 1) {
5617 for (i = 0; i < ids->n_controllers; i++) {
5618 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
5622 /* Stop processing for current table. */
5628 xlate_output_action(struct action_xlate_ctx *ctx,
5629 uint16_t port, uint16_t max_len, bool may_packet_in)
5631 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
5633 ctx->nf_output_iface = NF_OUT_DROP;
5637 compose_output_action(ctx, ctx->flow.in_port);
5640 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
5646 flood_packets(ctx, false);
5649 flood_packets(ctx, true);
5651 case OFPP_CONTROLLER:
5652 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
5658 if (port != ctx->flow.in_port) {
5659 compose_output_action(ctx, port);
5661 xlate_report(ctx, "skipping output to input port");
5666 if (prev_nf_output_iface == NF_OUT_FLOOD) {
5667 ctx->nf_output_iface = NF_OUT_FLOOD;
5668 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
5669 ctx->nf_output_iface = prev_nf_output_iface;
5670 } else if (prev_nf_output_iface != NF_OUT_DROP &&
5671 ctx->nf_output_iface != NF_OUT_FLOOD) {
5672 ctx->nf_output_iface = NF_OUT_MULTI;
5677 xlate_output_reg_action(struct action_xlate_ctx *ctx,
5678 const struct ofpact_output_reg *or)
5680 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
5681 if (port <= UINT16_MAX) {
5682 xlate_output_action(ctx, port, or->max_len, false);
5687 xlate_enqueue_action(struct action_xlate_ctx *ctx,
5688 const struct ofpact_enqueue *enqueue)
5690 uint16_t ofp_port = enqueue->port;
5691 uint32_t queue_id = enqueue->queue;
5692 uint32_t flow_priority, priority;
5695 /* Translate queue to priority. */
5696 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
5697 queue_id, &priority);
5699 /* Fall back to ordinary output action. */
5700 xlate_output_action(ctx, enqueue->port, 0, false);
5704 /* Check output port. */
5705 if (ofp_port == OFPP_IN_PORT) {
5706 ofp_port = ctx->flow.in_port;
5707 } else if (ofp_port == ctx->flow.in_port) {
5711 /* Add datapath actions. */
5712 flow_priority = ctx->flow.skb_priority;
5713 ctx->flow.skb_priority = priority;
5714 compose_output_action(ctx, ofp_port);
5715 ctx->flow.skb_priority = flow_priority;
5717 /* Update NetFlow output port. */
5718 if (ctx->nf_output_iface == NF_OUT_DROP) {
5719 ctx->nf_output_iface = ofp_port;
5720 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
5721 ctx->nf_output_iface = NF_OUT_MULTI;
5726 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
5728 uint32_t skb_priority;
5730 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
5731 queue_id, &skb_priority)) {
5732 ctx->flow.skb_priority = skb_priority;
5734 /* Couldn't translate queue to a priority. Nothing to do. A warning
5735 * has already been logged. */
5739 struct xlate_reg_state {
5745 xlate_autopath(struct action_xlate_ctx *ctx,
5746 const struct ofpact_autopath *ap)
5748 uint16_t ofp_port = ap->port;
5749 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
5751 if (!port || !port->bundle) {
5752 ofp_port = OFPP_NONE;
5753 } else if (port->bundle->bond) {
5754 /* Autopath does not support VLAN hashing. */
5755 struct ofport_dpif *slave = bond_choose_output_slave(
5756 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
5758 ofp_port = slave->up.ofp_port;
5761 nxm_reg_load(&ap->dst, ofp_port, &ctx->flow);
5765 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
5767 struct ofproto_dpif *ofproto = ofproto_;
5768 struct ofport_dpif *port;
5778 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
5781 port = get_ofp_port(ofproto, ofp_port);
5782 return port ? port->may_enable : false;
5787 xlate_bundle_action(struct action_xlate_ctx *ctx,
5788 const struct ofpact_bundle *bundle)
5792 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
5793 if (bundle->dst.field) {
5794 nxm_reg_load(&bundle->dst, port, &ctx->flow);
5796 xlate_output_action(ctx, port, 0, false);
5801 xlate_learn_action(struct action_xlate_ctx *ctx,
5802 const struct ofpact_learn *learn)
5804 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
5805 struct ofputil_flow_mod fm;
5806 uint64_t ofpacts_stub[1024 / 8];
5807 struct ofpbuf ofpacts;
5810 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
5811 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
5813 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
5814 if (error && !VLOG_DROP_WARN(&rl)) {
5815 VLOG_WARN("learning action failed to modify flow table (%s)",
5816 ofperr_get_name(error));
5819 ofpbuf_uninit(&ofpacts);
5822 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
5823 * means "infinite". */
5825 reduce_timeout(uint16_t max, uint16_t *timeout)
5827 if (max && (!*timeout || *timeout > max)) {
5833 xlate_fin_timeout(struct action_xlate_ctx *ctx,
5834 const struct ofpact_fin_timeout *oft)
5836 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
5837 struct rule_dpif *rule = ctx->rule;
5839 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
5840 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
5845 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
5847 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
5848 ? OFPUTIL_PC_NO_RECV_STP
5849 : OFPUTIL_PC_NO_RECV)) {
5853 /* Only drop packets here if both forwarding and learning are
5854 * disabled. If just learning is enabled, we need to have
5855 * OFPP_NORMAL and the learning action have a look at the packet
5856 * before we can drop it. */
5857 if (!stp_forward_in_state(port->stp_state)
5858 && !stp_learn_in_state(port->stp_state)) {
5866 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
5867 struct action_xlate_ctx *ctx)
5869 const struct ofport_dpif *port;
5870 bool was_evictable = true;
5871 const struct ofpact *a;
5873 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5874 if (port && !may_receive(port, ctx)) {
5875 /* Drop this flow. */
5880 /* Don't let the rule we're working on get evicted underneath us. */
5881 was_evictable = ctx->rule->up.evictable;
5882 ctx->rule->up.evictable = false;
5884 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
5885 struct ofpact_controller *controller;
5886 const struct ofpact_metadata *metadata;
5894 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
5895 ofpact_get_OUTPUT(a)->max_len, true);
5898 case OFPACT_CONTROLLER:
5899 controller = ofpact_get_CONTROLLER(a);
5900 execute_controller_action(ctx, controller->max_len,
5902 controller->controller_id);
5905 case OFPACT_ENQUEUE:
5906 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
5909 case OFPACT_SET_VLAN_VID:
5910 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
5911 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
5915 case OFPACT_SET_VLAN_PCP:
5916 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
5917 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
5922 case OFPACT_STRIP_VLAN:
5923 ctx->flow.vlan_tci = htons(0);
5926 case OFPACT_PUSH_VLAN:
5927 /* TODO:XXX 802.1AD(QinQ) */
5928 ctx->flow.vlan_tci = htons(VLAN_CFI);
5931 case OFPACT_SET_ETH_SRC:
5932 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
5936 case OFPACT_SET_ETH_DST:
5937 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
5941 case OFPACT_SET_IPV4_SRC:
5942 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
5945 case OFPACT_SET_IPV4_DST:
5946 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
5949 case OFPACT_SET_IPV4_DSCP:
5950 /* OpenFlow 1.0 only supports IPv4. */
5951 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5952 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5953 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
5957 case OFPACT_SET_L4_SRC_PORT:
5958 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
5961 case OFPACT_SET_L4_DST_PORT:
5962 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
5965 case OFPACT_RESUBMIT:
5966 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
5969 case OFPACT_SET_TUNNEL:
5970 ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
5973 case OFPACT_SET_QUEUE:
5974 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
5977 case OFPACT_POP_QUEUE:
5978 ctx->flow.skb_priority = ctx->orig_skb_priority;
5981 case OFPACT_REG_MOVE:
5982 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
5985 case OFPACT_REG_LOAD:
5986 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
5989 case OFPACT_DEC_TTL:
5990 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
5996 /* Nothing to do. */
5999 case OFPACT_MULTIPATH:
6000 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
6003 case OFPACT_AUTOPATH:
6004 xlate_autopath(ctx, ofpact_get_AUTOPATH(a));
6008 ctx->ofproto->has_bundle_action = true;
6009 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6012 case OFPACT_OUTPUT_REG:
6013 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6017 ctx->has_learn = true;
6018 if (ctx->may_learn) {
6019 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6027 case OFPACT_FIN_TIMEOUT:
6028 ctx->has_fin_timeout = true;
6029 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6032 case OFPACT_CLEAR_ACTIONS:
6034 * Nothing to do because writa-actions is not supported for now.
6035 * When writa-actions is supported, clear-actions also must
6036 * be supported at the same time.
6040 case OFPACT_WRITE_METADATA:
6041 metadata = ofpact_get_WRITE_METADATA(a);
6042 ctx->flow.metadata &= ~metadata->mask;
6043 ctx->flow.metadata |= metadata->metadata & metadata->mask;
6046 case OFPACT_GOTO_TABLE: {
6047 /* TODO:XXX remove recursion */
6048 /* It is assumed that goto-table is last action */
6049 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6050 assert(ctx->table_id < ogt->table_id);
6051 xlate_table_action(ctx, ctx->flow.in_port, ogt->table_id, true);
6058 /* We've let OFPP_NORMAL and the learning action look at the packet,
6059 * so drop it now if forwarding is disabled. */
6060 if (port && !stp_forward_in_state(port->stp_state)) {
6061 ofpbuf_clear(ctx->odp_actions);
6062 add_sflow_action(ctx);
6065 ctx->rule->up.evictable = was_evictable;
6070 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
6071 struct ofproto_dpif *ofproto, const struct flow *flow,
6072 ovs_be16 initial_tci, struct rule_dpif *rule,
6073 uint8_t tcp_flags, const struct ofpbuf *packet)
6075 ctx->ofproto = ofproto;
6077 ctx->base_flow = ctx->flow;
6078 memset(&ctx->base_flow.tunnel, 0, sizeof ctx->base_flow.tunnel);
6079 ctx->base_flow.vlan_tci = initial_tci;
6081 ctx->packet = packet;
6082 ctx->may_learn = packet != NULL;
6083 ctx->tcp_flags = tcp_flags;
6084 ctx->resubmit_hook = NULL;
6085 ctx->report_hook = NULL;
6086 ctx->resubmit_stats = NULL;
6089 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6090 * into datapath actions in 'odp_actions', using 'ctx'. */
6092 xlate_actions(struct action_xlate_ctx *ctx,
6093 const struct ofpact *ofpacts, size_t ofpacts_len,
6094 struct ofpbuf *odp_actions)
6096 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6097 * that in the future we always keep a copy of the original flow for
6098 * tracing purposes. */
6099 static bool hit_resubmit_limit;
6101 enum slow_path_reason special;
6103 COVERAGE_INC(ofproto_dpif_xlate);
6105 ofpbuf_clear(odp_actions);
6106 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
6108 ctx->odp_actions = odp_actions;
6111 ctx->has_learn = false;
6112 ctx->has_normal = false;
6113 ctx->has_fin_timeout = false;
6114 ctx->nf_output_iface = NF_OUT_DROP;
6117 ctx->max_resubmit_trigger = false;
6118 ctx->orig_skb_priority = ctx->flow.skb_priority;
6122 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
6123 /* Do this conditionally because the copy is expensive enough that it
6124 * shows up in profiles.
6126 * We keep orig_flow in 'ctx' only because I couldn't make GCC 4.4
6127 * believe that I wasn't using it without initializing it if I kept it
6128 * in a local variable. */
6129 ctx->orig_flow = ctx->flow;
6132 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6133 switch (ctx->ofproto->up.frag_handling) {
6134 case OFPC_FRAG_NORMAL:
6135 /* We must pretend that transport ports are unavailable. */
6136 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
6137 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
6140 case OFPC_FRAG_DROP:
6143 case OFPC_FRAG_REASM:
6146 case OFPC_FRAG_NX_MATCH:
6147 /* Nothing to do. */
6150 case OFPC_INVALID_TTL_TO_CONTROLLER:
6155 special = process_special(ctx->ofproto, &ctx->flow, ctx->packet);
6157 ctx->slow |= special;
6159 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6160 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
6162 add_sflow_action(ctx);
6163 do_xlate_actions(ofpacts, ofpacts_len, ctx);
6165 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
6166 if (!hit_resubmit_limit) {
6167 /* We didn't record the original flow. Make sure we do from
6169 hit_resubmit_limit = true;
6170 } else if (!VLOG_DROP_ERR(&trace_rl)) {
6171 struct ds ds = DS_EMPTY_INITIALIZER;
6173 ofproto_trace(ctx->ofproto, &ctx->orig_flow, ctx->packet,
6175 VLOG_ERR("Trace triggered by excessive resubmit "
6176 "recursion:\n%s", ds_cstr(&ds));
6181 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
6182 ctx->odp_actions->data,
6183 ctx->odp_actions->size)) {
6184 ctx->slow |= SLOW_IN_BAND;
6186 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
6188 compose_output_action(ctx, OFPP_LOCAL);
6191 if (ctx->ofproto->has_mirrors) {
6192 add_mirror_actions(ctx, &ctx->orig_flow);
6194 fix_sflow_action(ctx);
6198 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
6199 * into datapath actions, using 'ctx', and discards the datapath actions. */
6201 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
6202 const struct ofpact *ofpacts,
6205 uint64_t odp_actions_stub[1024 / 8];
6206 struct ofpbuf odp_actions;
6208 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
6209 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
6210 ofpbuf_uninit(&odp_actions);
6214 xlate_report(struct action_xlate_ctx *ctx, const char *s)
6216 if (ctx->report_hook) {
6217 ctx->report_hook(ctx, s);
6221 /* OFPP_NORMAL implementation. */
6223 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
6225 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
6226 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
6227 * the bundle on which the packet was received, returns the VLAN to which the
6230 * Both 'vid' and the return value are in the range 0...4095. */
6232 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
6234 switch (in_bundle->vlan_mode) {
6235 case PORT_VLAN_ACCESS:
6236 return in_bundle->vlan;
6239 case PORT_VLAN_TRUNK:
6242 case PORT_VLAN_NATIVE_UNTAGGED:
6243 case PORT_VLAN_NATIVE_TAGGED:
6244 return vid ? vid : in_bundle->vlan;
6251 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
6252 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
6255 * 'vid' should be the VID obtained from the 802.1Q header that was received as
6256 * part of a packet (specify 0 if there was no 802.1Q header), in the range
6259 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
6261 /* Allow any VID on the OFPP_NONE port. */
6262 if (in_bundle == &ofpp_none_bundle) {
6266 switch (in_bundle->vlan_mode) {
6267 case PORT_VLAN_ACCESS:
6270 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6271 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
6272 "packet received on port %s configured as VLAN "
6273 "%"PRIu16" access port",
6274 in_bundle->ofproto->up.name, vid,
6275 in_bundle->name, in_bundle->vlan);
6281 case PORT_VLAN_NATIVE_UNTAGGED:
6282 case PORT_VLAN_NATIVE_TAGGED:
6284 /* Port must always carry its native VLAN. */
6288 case PORT_VLAN_TRUNK:
6289 if (!ofbundle_includes_vlan(in_bundle, vid)) {
6291 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6292 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
6293 "received on port %s not configured for trunking "
6295 in_bundle->ofproto->up.name, vid,
6296 in_bundle->name, vid);
6308 /* Given 'vlan', the VLAN that a packet belongs to, and
6309 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
6310 * that should be included in the 802.1Q header. (If the return value is 0,
6311 * then the 802.1Q header should only be included in the packet if there is a
6314 * Both 'vlan' and the return value are in the range 0...4095. */
6316 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
6318 switch (out_bundle->vlan_mode) {
6319 case PORT_VLAN_ACCESS:
6322 case PORT_VLAN_TRUNK:
6323 case PORT_VLAN_NATIVE_TAGGED:
6326 case PORT_VLAN_NATIVE_UNTAGGED:
6327 return vlan == out_bundle->vlan ? 0 : vlan;
6335 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
6338 struct ofport_dpif *port;
6340 ovs_be16 tci, old_tci;
6342 vid = output_vlan_to_vid(out_bundle, vlan);
6343 if (!out_bundle->bond) {
6344 port = ofbundle_get_a_port(out_bundle);
6346 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
6349 /* No slaves enabled, so drop packet. */
6354 old_tci = ctx->flow.vlan_tci;
6356 if (tci || out_bundle->use_priority_tags) {
6357 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
6359 tci |= htons(VLAN_CFI);
6362 ctx->flow.vlan_tci = tci;
6364 compose_output_action(ctx, port->up.ofp_port);
6365 ctx->flow.vlan_tci = old_tci;
6369 mirror_mask_ffs(mirror_mask_t mask)
6371 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
6376 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
6378 return (bundle->vlan_mode != PORT_VLAN_ACCESS
6379 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
6383 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
6385 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
6388 /* Returns an arbitrary interface within 'bundle'. */
6389 static struct ofport_dpif *
6390 ofbundle_get_a_port(const struct ofbundle *bundle)
6392 return CONTAINER_OF(list_front(&bundle->ports),
6393 struct ofport_dpif, bundle_node);
6397 vlan_is_mirrored(const struct ofmirror *m, int vlan)
6399 return !m->vlans || bitmap_is_set(m->vlans, vlan);
6403 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
6405 struct ofproto_dpif *ofproto = ctx->ofproto;
6406 mirror_mask_t mirrors;
6407 struct ofbundle *in_bundle;
6410 const struct nlattr *a;
6413 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
6414 ctx->packet != NULL, NULL);
6418 mirrors = in_bundle->src_mirrors;
6420 /* Drop frames on bundles reserved for mirroring. */
6421 if (in_bundle->mirror_out) {
6422 if (ctx->packet != NULL) {
6423 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6424 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6425 "%s, which is reserved exclusively for mirroring",
6426 ctx->ofproto->up.name, in_bundle->name);
6432 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
6433 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6436 vlan = input_vid_to_vlan(in_bundle, vid);
6438 /* Look at the output ports to check for destination selections. */
6440 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
6441 ctx->odp_actions->size) {
6442 enum ovs_action_attr type = nl_attr_type(a);
6443 struct ofport_dpif *ofport;
6445 if (type != OVS_ACTION_ATTR_OUTPUT) {
6449 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
6450 if (ofport && ofport->bundle) {
6451 mirrors |= ofport->bundle->dst_mirrors;
6459 /* Restore the original packet before adding the mirror actions. */
6460 ctx->flow = *orig_flow;
6465 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6467 if (!vlan_is_mirrored(m, vlan)) {
6468 mirrors = zero_rightmost_1bit(mirrors);
6472 mirrors &= ~m->dup_mirrors;
6473 ctx->mirrors |= m->dup_mirrors;
6475 output_normal(ctx, m->out, vlan);
6476 } else if (vlan != m->out_vlan
6477 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
6478 struct ofbundle *bundle;
6480 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
6481 if (ofbundle_includes_vlan(bundle, m->out_vlan)
6482 && !bundle->mirror_out) {
6483 output_normal(ctx, bundle, m->out_vlan);
6491 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
6492 uint64_t packets, uint64_t bytes)
6498 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
6501 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6504 /* In normal circumstances 'm' will not be NULL. However,
6505 * if mirrors are reconfigured, we can temporarily get out
6506 * of sync in facet_revalidate(). We could "correct" the
6507 * mirror list before reaching here, but doing that would
6508 * not properly account the traffic stats we've currently
6509 * accumulated for previous mirror configuration. */
6513 m->packet_count += packets;
6514 m->byte_count += bytes;
6518 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
6519 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
6520 * indicate this; newer upstream kernels use gratuitous ARP requests. */
6522 is_gratuitous_arp(const struct flow *flow)
6524 return (flow->dl_type == htons(ETH_TYPE_ARP)
6525 && eth_addr_is_broadcast(flow->dl_dst)
6526 && (flow->nw_proto == ARP_OP_REPLY
6527 || (flow->nw_proto == ARP_OP_REQUEST
6528 && flow->nw_src == flow->nw_dst)));
6532 update_learning_table(struct ofproto_dpif *ofproto,
6533 const struct flow *flow, int vlan,
6534 struct ofbundle *in_bundle)
6536 struct mac_entry *mac;
6538 /* Don't learn the OFPP_NONE port. */
6539 if (in_bundle == &ofpp_none_bundle) {
6543 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
6547 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
6548 if (is_gratuitous_arp(flow)) {
6549 /* We don't want to learn from gratuitous ARP packets that are
6550 * reflected back over bond slaves so we lock the learning table. */
6551 if (!in_bundle->bond) {
6552 mac_entry_set_grat_arp_lock(mac);
6553 } else if (mac_entry_is_grat_arp_locked(mac)) {
6558 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
6559 /* The log messages here could actually be useful in debugging,
6560 * so keep the rate limit relatively high. */
6561 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
6562 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
6563 "on port %s in VLAN %d",
6564 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
6565 in_bundle->name, vlan);
6567 mac->port.p = in_bundle;
6568 tag_set_add(&ofproto->revalidate_set,
6569 mac_learning_changed(ofproto->ml, mac));
6573 static struct ofbundle *
6574 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
6575 bool warn, struct ofport_dpif **in_ofportp)
6577 struct ofport_dpif *ofport;
6579 /* Find the port and bundle for the received packet. */
6580 ofport = get_ofp_port(ofproto, in_port);
6582 *in_ofportp = ofport;
6584 if (ofport && ofport->bundle) {
6585 return ofport->bundle;
6588 /* Special-case OFPP_NONE, which a controller may use as the ingress
6589 * port for traffic that it is sourcing. */
6590 if (in_port == OFPP_NONE) {
6591 return &ofpp_none_bundle;
6594 /* Odd. A few possible reasons here:
6596 * - We deleted a port but there are still a few packets queued up
6599 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
6600 * we don't know about.
6602 * - The ofproto client didn't configure the port as part of a bundle.
6603 * This is particularly likely to happen if a packet was received on the
6604 * port after it was created, but before the client had a chance to
6605 * configure its bundle.
6608 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6610 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
6611 "port %"PRIu16, ofproto->up.name, in_port);
6616 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
6617 * dropped. Returns true if they may be forwarded, false if they should be
6620 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
6621 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
6623 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
6624 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
6625 * checked by input_vid_is_valid().
6627 * May also add tags to '*tags', although the current implementation only does
6628 * so in one special case.
6631 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
6634 struct ofproto_dpif *ofproto = ctx->ofproto;
6635 struct flow *flow = &ctx->flow;
6636 struct ofbundle *in_bundle = in_port->bundle;
6638 /* Drop frames for reserved multicast addresses
6639 * only if forward_bpdu option is absent. */
6640 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
6641 xlate_report(ctx, "packet has reserved destination MAC, dropping");
6645 if (in_bundle->bond) {
6646 struct mac_entry *mac;
6648 switch (bond_check_admissibility(in_bundle->bond, in_port,
6649 flow->dl_dst, &ctx->tags)) {
6654 xlate_report(ctx, "bonding refused admissibility, dropping");
6657 case BV_DROP_IF_MOVED:
6658 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
6659 if (mac && mac->port.p != in_bundle &&
6660 (!is_gratuitous_arp(flow)
6661 || mac_entry_is_grat_arp_locked(mac))) {
6662 xlate_report(ctx, "SLB bond thinks this packet looped back, "
6674 xlate_normal(struct action_xlate_ctx *ctx)
6676 struct ofport_dpif *in_port;
6677 struct ofbundle *in_bundle;
6678 struct mac_entry *mac;
6682 ctx->has_normal = true;
6684 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
6685 ctx->packet != NULL, &in_port);
6687 xlate_report(ctx, "no input bundle, dropping");
6691 /* Drop malformed frames. */
6692 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
6693 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
6694 if (ctx->packet != NULL) {
6695 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6696 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
6697 "VLAN tag received on port %s",
6698 ctx->ofproto->up.name, in_bundle->name);
6700 xlate_report(ctx, "partial VLAN tag, dropping");
6704 /* Drop frames on bundles reserved for mirroring. */
6705 if (in_bundle->mirror_out) {
6706 if (ctx->packet != NULL) {
6707 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6708 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6709 "%s, which is reserved exclusively for mirroring",
6710 ctx->ofproto->up.name, in_bundle->name);
6712 xlate_report(ctx, "input port is mirror output port, dropping");
6717 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
6718 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6719 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
6722 vlan = input_vid_to_vlan(in_bundle, vid);
6724 /* Check other admissibility requirements. */
6725 if (in_port && !is_admissible(ctx, in_port, vlan)) {
6729 /* Learn source MAC. */
6730 if (ctx->may_learn) {
6731 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
6734 /* Determine output bundle. */
6735 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
6738 if (mac->port.p != in_bundle) {
6739 xlate_report(ctx, "forwarding to learned port");
6740 output_normal(ctx, mac->port.p, vlan);
6742 xlate_report(ctx, "learned port is input port, dropping");
6745 struct ofbundle *bundle;
6747 xlate_report(ctx, "no learned MAC for destination, flooding");
6748 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
6749 if (bundle != in_bundle
6750 && ofbundle_includes_vlan(bundle, vlan)
6751 && bundle->floodable
6752 && !bundle->mirror_out) {
6753 output_normal(ctx, bundle, vlan);
6756 ctx->nf_output_iface = NF_OUT_FLOOD;
6760 /* Optimized flow revalidation.
6762 * It's a difficult problem, in general, to tell which facets need to have
6763 * their actions recalculated whenever the OpenFlow flow table changes. We
6764 * don't try to solve that general problem: for most kinds of OpenFlow flow
6765 * table changes, we recalculate the actions for every facet. This is
6766 * relatively expensive, but it's good enough if the OpenFlow flow table
6767 * doesn't change very often.
6769 * However, we can expect one particular kind of OpenFlow flow table change to
6770 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
6771 * of CPU on revalidating every facet whenever MAC learning modifies the flow
6772 * table, we add a special case that applies to flow tables in which every rule
6773 * has the same form (that is, the same wildcards), except that the table is
6774 * also allowed to have a single "catch-all" flow that matches all packets. We
6775 * optimize this case by tagging all of the facets that resubmit into the table
6776 * and invalidating the same tag whenever a flow changes in that table. The
6777 * end result is that we revalidate just the facets that need it (and sometimes
6778 * a few more, but not all of the facets or even all of the facets that
6779 * resubmit to the table modified by MAC learning). */
6781 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
6782 * into an OpenFlow table with the given 'basis'. */
6784 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
6787 if (minimask_is_catchall(mask)) {
6790 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
6791 return tag_create_deterministic(hash);
6795 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
6796 * taggability of that table.
6798 * This function must be called after *each* change to a flow table. If you
6799 * skip calling it on some changes then the pointer comparisons at the end can
6800 * be invalid if you get unlucky. For example, if a flow removal causes a
6801 * cls_table to be destroyed and then a flow insertion causes a cls_table with
6802 * different wildcards to be created with the same address, then this function
6803 * will incorrectly skip revalidation. */
6805 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
6807 struct table_dpif *table = &ofproto->tables[table_id];
6808 const struct oftable *oftable = &ofproto->up.tables[table_id];
6809 struct cls_table *catchall, *other;
6810 struct cls_table *t;
6812 catchall = other = NULL;
6814 switch (hmap_count(&oftable->cls.tables)) {
6816 /* We could tag this OpenFlow table but it would make the logic a
6817 * little harder and it's a corner case that doesn't seem worth it
6823 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
6824 if (cls_table_is_catchall(t)) {
6826 } else if (!other) {
6829 /* Indicate that we can't tag this by setting both tables to
6830 * NULL. (We know that 'catchall' is already NULL.) */
6837 /* Can't tag this table. */
6841 if (table->catchall_table != catchall || table->other_table != other) {
6842 table->catchall_table = catchall;
6843 table->other_table = other;
6844 ofproto->need_revalidate = REV_FLOW_TABLE;
6848 /* Given 'rule' that has changed in some way (either it is a rule being
6849 * inserted, a rule being deleted, or a rule whose actions are being
6850 * modified), marks facets for revalidation to ensure that packets will be
6851 * forwarded correctly according to the new state of the flow table.
6853 * This function must be called after *each* change to a flow table. See
6854 * the comment on table_update_taggable() for more information. */
6856 rule_invalidate(const struct rule_dpif *rule)
6858 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
6860 table_update_taggable(ofproto, rule->up.table_id);
6862 if (!ofproto->need_revalidate) {
6863 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
6865 if (table->other_table && rule->tag) {
6866 tag_set_add(&ofproto->revalidate_set, rule->tag);
6868 ofproto->need_revalidate = REV_FLOW_TABLE;
6874 set_frag_handling(struct ofproto *ofproto_,
6875 enum ofp_config_flags frag_handling)
6877 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6879 if (frag_handling != OFPC_FRAG_REASM) {
6880 ofproto->need_revalidate = REV_RECONFIGURE;
6888 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
6889 const struct flow *flow,
6890 const struct ofpact *ofpacts, size_t ofpacts_len)
6892 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6893 struct odputil_keybuf keybuf;
6894 struct dpif_flow_stats stats;
6898 struct action_xlate_ctx ctx;
6899 uint64_t odp_actions_stub[1024 / 8];
6900 struct ofpbuf odp_actions;
6902 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
6903 odp_flow_key_from_flow(&key, flow,
6904 ofp_port_to_odp_port(ofproto, flow->in_port));
6906 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
6908 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL,
6909 packet_get_tcp_flags(packet, flow), packet);
6910 ctx.resubmit_stats = &stats;
6912 ofpbuf_use_stub(&odp_actions,
6913 odp_actions_stub, sizeof odp_actions_stub);
6914 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
6915 dpif_execute(ofproto->backer->dpif, key.data, key.size,
6916 odp_actions.data, odp_actions.size, packet);
6917 ofpbuf_uninit(&odp_actions);
6925 set_netflow(struct ofproto *ofproto_,
6926 const struct netflow_options *netflow_options)
6928 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6930 if (netflow_options) {
6931 if (!ofproto->netflow) {
6932 ofproto->netflow = netflow_create();
6934 return netflow_set_options(ofproto->netflow, netflow_options);
6936 netflow_destroy(ofproto->netflow);
6937 ofproto->netflow = NULL;
6943 get_netflow_ids(const struct ofproto *ofproto_,
6944 uint8_t *engine_type, uint8_t *engine_id)
6946 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6948 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
6952 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
6954 if (!facet_is_controller_flow(facet) &&
6955 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
6956 struct subfacet *subfacet;
6957 struct ofexpired expired;
6959 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
6960 if (subfacet->path == SF_FAST_PATH) {
6961 struct dpif_flow_stats stats;
6963 subfacet_reinstall(subfacet, &stats);
6964 subfacet_update_stats(subfacet, &stats);
6968 expired.flow = facet->flow;
6969 expired.packet_count = facet->packet_count;
6970 expired.byte_count = facet->byte_count;
6971 expired.used = facet->used;
6972 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
6977 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
6979 struct facet *facet;
6981 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6982 send_active_timeout(ofproto, facet);
6986 static struct ofproto_dpif *
6987 ofproto_dpif_lookup(const char *name)
6989 struct ofproto_dpif *ofproto;
6991 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
6992 hash_string(name, 0), &all_ofproto_dpifs) {
6993 if (!strcmp(ofproto->up.name, name)) {
7001 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7002 const char *argv[], void *aux OVS_UNUSED)
7004 struct ofproto_dpif *ofproto;
7007 ofproto = ofproto_dpif_lookup(argv[1]);
7009 unixctl_command_reply_error(conn, "no such bridge");
7012 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
7014 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7015 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
7019 unixctl_command_reply(conn, "table successfully flushed");
7023 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7024 const char *argv[], void *aux OVS_UNUSED)
7026 struct ds ds = DS_EMPTY_INITIALIZER;
7027 const struct ofproto_dpif *ofproto;
7028 const struct mac_entry *e;
7030 ofproto = ofproto_dpif_lookup(argv[1]);
7032 unixctl_command_reply_error(conn, "no such bridge");
7036 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7037 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7038 struct ofbundle *bundle = e->port.p;
7039 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7040 ofbundle_get_a_port(bundle)->odp_port,
7041 e->vlan, ETH_ADDR_ARGS(e->mac),
7042 mac_entry_age(ofproto->ml, e));
7044 unixctl_command_reply(conn, ds_cstr(&ds));
7049 struct action_xlate_ctx ctx;
7055 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7056 const struct rule_dpif *rule)
7058 ds_put_char_multiple(result, '\t', level);
7060 ds_put_cstr(result, "No match\n");
7064 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7065 table_id, ntohll(rule->up.flow_cookie));
7066 cls_rule_format(&rule->up.cr, result);
7067 ds_put_char(result, '\n');
7069 ds_put_char_multiple(result, '\t', level);
7070 ds_put_cstr(result, "OpenFlow ");
7071 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7072 ds_put_char(result, '\n');
7076 trace_format_flow(struct ds *result, int level, const char *title,
7077 struct trace_ctx *trace)
7079 ds_put_char_multiple(result, '\t', level);
7080 ds_put_format(result, "%s: ", title);
7081 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
7082 ds_put_cstr(result, "unchanged");
7084 flow_format(result, &trace->ctx.flow);
7085 trace->flow = trace->ctx.flow;
7087 ds_put_char(result, '\n');
7091 trace_format_regs(struct ds *result, int level, const char *title,
7092 struct trace_ctx *trace)
7096 ds_put_char_multiple(result, '\t', level);
7097 ds_put_format(result, "%s:", title);
7098 for (i = 0; i < FLOW_N_REGS; i++) {
7099 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7101 ds_put_char(result, '\n');
7105 trace_format_odp(struct ds *result, int level, const char *title,
7106 struct trace_ctx *trace)
7108 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
7110 ds_put_char_multiple(result, '\t', level);
7111 ds_put_format(result, "%s: ", title);
7112 format_odp_actions(result, odp_actions->data, odp_actions->size);
7113 ds_put_char(result, '\n');
7117 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
7119 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7120 struct ds *result = trace->result;
7122 ds_put_char(result, '\n');
7123 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7124 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
7125 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
7126 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
7130 trace_report(struct action_xlate_ctx *ctx, const char *s)
7132 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7133 struct ds *result = trace->result;
7135 ds_put_char_multiple(result, '\t', ctx->recurse);
7136 ds_put_cstr(result, s);
7137 ds_put_char(result, '\n');
7141 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
7142 void *aux OVS_UNUSED)
7144 const char *dpname = argv[1];
7145 struct ofproto_dpif *ofproto;
7146 struct ofpbuf odp_key;
7147 struct ofpbuf *packet;
7148 ovs_be16 initial_tci;
7154 ofpbuf_init(&odp_key, 0);
7157 ofproto = ofproto_dpif_lookup(dpname);
7159 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
7163 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
7164 /* ofproto/trace dpname flow [-generate] */
7165 const char *flow_s = argv[2];
7166 const char *generate_s = argv[3];
7168 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
7169 * flow. We guess which type it is based on whether 'flow_s' contains
7170 * an '(', since a datapath flow always contains '(') but an
7171 * OpenFlow-like flow should not (in fact it's allowed but I believe
7172 * that's not documented anywhere).
7174 * An alternative would be to try to parse 'flow_s' both ways, but then
7175 * it would be tricky giving a sensible error message. After all, do
7176 * you just say "syntax error" or do you present both error messages?
7177 * Both choices seem lousy. */
7178 if (strchr(flow_s, '(')) {
7179 enum odp_key_fitness fitness;
7182 /* Convert string to datapath key. */
7183 ofpbuf_init(&odp_key, 0);
7184 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
7186 unixctl_command_reply_error(conn, "Bad flow syntax");
7190 fitness = odp_flow_key_to_flow(odp_key.data, odp_key.size, &flow);
7191 flow.in_port = odp_port_to_ofp_port(ofproto, flow.in_port);
7193 /* Convert odp_key to flow. */
7194 error = ofproto_dpif_vsp_adjust(ofproto, fitness, &flow,
7195 &initial_tci, NULL);
7196 if (error == ODP_FIT_ERROR) {
7197 unixctl_command_reply_error(conn, "Invalid flow");
7203 error_s = parse_ofp_exact_flow(&flow, argv[2]);
7205 unixctl_command_reply_error(conn, error_s);
7210 initial_tci = flow.vlan_tci;
7211 vsp_adjust_flow(ofproto, &flow);
7214 /* Generate a packet, if requested. */
7216 packet = ofpbuf_new(0);
7217 flow_compose(packet, &flow);
7219 } else if (argc == 7) {
7220 /* ofproto/trace dpname priority tun_id in_port mark packet */
7221 const char *priority_s = argv[2];
7222 const char *tun_id_s = argv[3];
7223 const char *in_port_s = argv[4];
7224 const char *mark_s = argv[5];
7225 const char *packet_s = argv[6];
7226 uint32_t in_port = atoi(in_port_s);
7227 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
7228 uint32_t priority = atoi(priority_s);
7229 uint32_t mark = atoi(mark_s);
7232 msg = eth_from_hex(packet_s, &packet);
7234 unixctl_command_reply_error(conn, msg);
7238 ds_put_cstr(&result, "Packet: ");
7239 s = ofp_packet_to_string(packet->data, packet->size);
7240 ds_put_cstr(&result, s);
7243 flow_extract(packet, priority, mark, NULL, in_port, &flow);
7244 flow.tunnel.tun_id = tun_id;
7245 initial_tci = flow.vlan_tci;
7247 unixctl_command_reply_error(conn, "Bad command syntax");
7251 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
7252 unixctl_command_reply(conn, ds_cstr(&result));
7255 ds_destroy(&result);
7256 ofpbuf_delete(packet);
7257 ofpbuf_uninit(&odp_key);
7261 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
7262 const struct ofpbuf *packet, ovs_be16 initial_tci,
7265 struct rule_dpif *rule;
7267 ds_put_cstr(ds, "Flow: ");
7268 flow_format(ds, flow);
7269 ds_put_char(ds, '\n');
7271 rule = rule_dpif_lookup(ofproto, flow);
7273 trace_format_rule(ds, 0, 0, rule);
7274 if (rule == ofproto->miss_rule) {
7275 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
7276 } else if (rule == ofproto->no_packet_in_rule) {
7277 ds_put_cstr(ds, "\nNo match, packets dropped because "
7278 "OFPPC_NO_PACKET_IN is set on in_port.\n");
7282 uint64_t odp_actions_stub[1024 / 8];
7283 struct ofpbuf odp_actions;
7285 struct trace_ctx trace;
7288 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
7291 ofpbuf_use_stub(&odp_actions,
7292 odp_actions_stub, sizeof odp_actions_stub);
7293 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
7294 rule, tcp_flags, packet);
7295 trace.ctx.resubmit_hook = trace_resubmit;
7296 trace.ctx.report_hook = trace_report;
7297 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
7300 ds_put_char(ds, '\n');
7301 trace_format_flow(ds, 0, "Final flow", &trace);
7302 ds_put_cstr(ds, "Datapath actions: ");
7303 format_odp_actions(ds, odp_actions.data, odp_actions.size);
7304 ofpbuf_uninit(&odp_actions);
7306 if (trace.ctx.slow) {
7307 enum slow_path_reason slow;
7309 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
7310 "slow path because it:");
7311 for (slow = trace.ctx.slow; slow; ) {
7312 enum slow_path_reason bit = rightmost_1bit(slow);
7316 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
7319 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
7322 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
7325 ds_put_cstr(ds, "\n\t- Needs in-band special case "
7328 ds_put_cstr(ds, "\n\t (The datapath actions are "
7329 "incomplete--for complete actions, "
7330 "please supply a packet.)");
7333 case SLOW_CONTROLLER:
7334 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
7335 "to the OpenFlow controller.");
7338 ds_put_cstr(ds, "\n\t- Needs more specific matching "
7339 "than the datapath supports.");
7346 if (slow & ~SLOW_MATCH) {
7347 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
7348 "the special slow-path processing.");
7355 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7356 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7359 unixctl_command_reply(conn, NULL);
7363 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7364 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7367 unixctl_command_reply(conn, NULL);
7370 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
7371 * 'reply' describing the results. */
7373 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
7375 struct facet *facet;
7379 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7380 if (!facet_check_consistency(facet)) {
7385 ofproto->need_revalidate = REV_INCONSISTENCY;
7389 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
7390 ofproto->up.name, errors);
7392 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
7397 ofproto_dpif_self_check(struct unixctl_conn *conn,
7398 int argc, const char *argv[], void *aux OVS_UNUSED)
7400 struct ds reply = DS_EMPTY_INITIALIZER;
7401 struct ofproto_dpif *ofproto;
7404 ofproto = ofproto_dpif_lookup(argv[1]);
7406 unixctl_command_reply_error(conn, "Unknown ofproto (use "
7407 "ofproto/list for help)");
7410 ofproto_dpif_self_check__(ofproto, &reply);
7412 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7413 ofproto_dpif_self_check__(ofproto, &reply);
7417 unixctl_command_reply(conn, ds_cstr(&reply));
7421 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
7422 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
7423 * to destroy 'ofproto_shash' and free the returned value. */
7424 static const struct shash_node **
7425 get_ofprotos(struct shash *ofproto_shash)
7427 const struct ofproto_dpif *ofproto;
7429 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7430 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
7431 shash_add_nocopy(ofproto_shash, name, ofproto);
7434 return shash_sort(ofproto_shash);
7438 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
7439 const char *argv[] OVS_UNUSED,
7440 void *aux OVS_UNUSED)
7442 struct ds ds = DS_EMPTY_INITIALIZER;
7443 struct shash ofproto_shash;
7444 const struct shash_node **sorted_ofprotos;
7447 shash_init(&ofproto_shash);
7448 sorted_ofprotos = get_ofprotos(&ofproto_shash);
7449 for (i = 0; i < shash_count(&ofproto_shash); i++) {
7450 const struct shash_node *node = sorted_ofprotos[i];
7451 ds_put_format(&ds, "%s\n", node->name);
7454 shash_destroy(&ofproto_shash);
7455 free(sorted_ofprotos);
7457 unixctl_command_reply(conn, ds_cstr(&ds));
7462 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
7464 struct dpif_dp_stats s;
7465 const struct shash_node **ports;
7468 dpif_get_dp_stats(ofproto->backer->dpif, &s);
7470 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
7471 dpif_name(ofproto->backer->dpif));
7472 /* xxx It would be better to show bridge-specific stats instead
7473 * xxx of dp ones. */
7475 "\tlookups: hit:%"PRIu64" missed:%"PRIu64" lost:%"PRIu64"\n",
7476 s.n_hit, s.n_missed, s.n_lost);
7477 ds_put_format(ds, "\tflows: %zu\n",
7478 hmap_count(&ofproto->subfacets));
7480 ports = shash_sort(&ofproto->up.port_by_name);
7481 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
7482 const struct shash_node *node = ports[i];
7483 struct ofport *ofport = node->data;
7484 const char *name = netdev_get_name(ofport->netdev);
7485 const char *type = netdev_get_type(ofport->netdev);
7487 ds_put_format(ds, "\t%s %u/%u:", name, ofport->ofp_port,
7488 ofp_port_to_odp_port(ofproto, ofport->ofp_port));
7489 if (strcmp(type, "system")) {
7490 struct netdev *netdev;
7493 ds_put_format(ds, " (%s", type);
7495 error = netdev_open(name, type, &netdev);
7500 error = netdev_get_config(netdev, &config);
7502 const struct smap_node **nodes;
7505 nodes = smap_sort(&config);
7506 for (i = 0; i < smap_count(&config); i++) {
7507 const struct smap_node *node = nodes[i];
7508 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
7509 node->key, node->value);
7513 smap_destroy(&config);
7515 netdev_close(netdev);
7517 ds_put_char(ds, ')');
7519 ds_put_char(ds, '\n');
7525 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
7526 const char *argv[], void *aux OVS_UNUSED)
7528 struct ds ds = DS_EMPTY_INITIALIZER;
7529 const struct ofproto_dpif *ofproto;
7533 for (i = 1; i < argc; i++) {
7534 ofproto = ofproto_dpif_lookup(argv[i]);
7536 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
7537 "for help)", argv[i]);
7538 unixctl_command_reply_error(conn, ds_cstr(&ds));
7541 show_dp_format(ofproto, &ds);
7544 struct shash ofproto_shash;
7545 const struct shash_node **sorted_ofprotos;
7548 shash_init(&ofproto_shash);
7549 sorted_ofprotos = get_ofprotos(&ofproto_shash);
7550 for (i = 0; i < shash_count(&ofproto_shash); i++) {
7551 const struct shash_node *node = sorted_ofprotos[i];
7552 show_dp_format(node->data, &ds);
7555 shash_destroy(&ofproto_shash);
7556 free(sorted_ofprotos);
7559 unixctl_command_reply(conn, ds_cstr(&ds));
7564 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
7565 int argc OVS_UNUSED, const char *argv[],
7566 void *aux OVS_UNUSED)
7568 struct ds ds = DS_EMPTY_INITIALIZER;
7569 const struct ofproto_dpif *ofproto;
7570 struct subfacet *subfacet;
7572 ofproto = ofproto_dpif_lookup(argv[1]);
7574 unixctl_command_reply_error(conn, "no such bridge");
7578 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
7579 struct odputil_keybuf keybuf;
7582 subfacet_get_key(subfacet, &keybuf, &key);
7583 odp_flow_key_format(key.data, key.size, &ds);
7585 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
7586 subfacet->dp_packet_count, subfacet->dp_byte_count);
7587 if (subfacet->used) {
7588 ds_put_format(&ds, "%.3fs",
7589 (time_msec() - subfacet->used) / 1000.0);
7591 ds_put_format(&ds, "never");
7593 if (subfacet->facet->tcp_flags) {
7594 ds_put_cstr(&ds, ", flags:");
7595 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
7598 ds_put_cstr(&ds, ", actions:");
7599 format_odp_actions(&ds, subfacet->actions, subfacet->actions_len);
7600 ds_put_char(&ds, '\n');
7603 unixctl_command_reply(conn, ds_cstr(&ds));
7608 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
7609 int argc OVS_UNUSED, const char *argv[],
7610 void *aux OVS_UNUSED)
7612 struct ds ds = DS_EMPTY_INITIALIZER;
7613 struct ofproto_dpif *ofproto;
7615 ofproto = ofproto_dpif_lookup(argv[1]);
7617 unixctl_command_reply_error(conn, "no such bridge");
7621 flush(&ofproto->up);
7623 unixctl_command_reply(conn, ds_cstr(&ds));
7628 ofproto_dpif_unixctl_init(void)
7630 static bool registered;
7636 unixctl_command_register(
7638 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
7639 2, 6, ofproto_unixctl_trace, NULL);
7640 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
7641 ofproto_unixctl_fdb_flush, NULL);
7642 unixctl_command_register("fdb/show", "bridge", 1, 1,
7643 ofproto_unixctl_fdb_show, NULL);
7644 unixctl_command_register("ofproto/clog", "", 0, 0,
7645 ofproto_dpif_clog, NULL);
7646 unixctl_command_register("ofproto/unclog", "", 0, 0,
7647 ofproto_dpif_unclog, NULL);
7648 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
7649 ofproto_dpif_self_check, NULL);
7650 unixctl_command_register("dpif/dump-dps", "", 0, 0,
7651 ofproto_unixctl_dpif_dump_dps, NULL);
7652 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
7653 ofproto_unixctl_dpif_show, NULL);
7654 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
7655 ofproto_unixctl_dpif_dump_flows, NULL);
7656 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
7657 ofproto_unixctl_dpif_del_flows, NULL);
7660 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
7662 * This is deprecated. It is only for compatibility with broken device drivers
7663 * in old versions of Linux that do not properly support VLANs when VLAN
7664 * devices are not used. When broken device drivers are no longer in
7665 * widespread use, we will delete these interfaces. */
7668 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
7670 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
7671 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
7673 if (realdev_ofp_port == ofport->realdev_ofp_port
7674 && vid == ofport->vlandev_vid) {
7678 ofproto->need_revalidate = REV_RECONFIGURE;
7680 if (ofport->realdev_ofp_port) {
7683 if (realdev_ofp_port && ofport->bundle) {
7684 /* vlandevs are enslaved to their realdevs, so they are not allowed to
7685 * themselves be part of a bundle. */
7686 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
7689 ofport->realdev_ofp_port = realdev_ofp_port;
7690 ofport->vlandev_vid = vid;
7692 if (realdev_ofp_port) {
7693 vsp_add(ofport, realdev_ofp_port, vid);
7700 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
7702 return hash_2words(realdev_ofp_port, vid);
7705 /* Returns the ODP port number of the Linux VLAN device that corresponds to
7706 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
7707 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
7708 * it would return the port number of eth0.9.
7710 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
7711 * function just returns its 'realdev_odp_port' argument. */
7713 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
7714 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
7716 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
7717 uint16_t realdev_ofp_port;
7718 int vid = vlan_tci_to_vid(vlan_tci);
7719 const struct vlan_splinter *vsp;
7721 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
7722 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
7723 hash_realdev_vid(realdev_ofp_port, vid),
7724 &ofproto->realdev_vid_map) {
7725 if (vsp->realdev_ofp_port == realdev_ofp_port
7726 && vsp->vid == vid) {
7727 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
7731 return realdev_odp_port;
7734 static struct vlan_splinter *
7735 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
7737 struct vlan_splinter *vsp;
7739 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
7740 &ofproto->vlandev_map) {
7741 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
7749 /* Returns the OpenFlow port number of the "real" device underlying the Linux
7750 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
7751 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
7752 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
7753 * eth0 and store 9 in '*vid'.
7755 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
7756 * VLAN device. Unless VLAN splinters are enabled, this is what this function
7759 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
7760 uint16_t vlandev_ofp_port, int *vid)
7762 if (!hmap_is_empty(&ofproto->vlandev_map)) {
7763 const struct vlan_splinter *vsp;
7765 vsp = vlandev_find(ofproto, vlandev_ofp_port);
7770 return vsp->realdev_ofp_port;
7776 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
7777 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
7778 * 'flow->in_port' to the "real" device backing the VLAN device, sets
7779 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
7780 * always the case unless VLAN splinters are enabled), returns false without
7781 * making any changes. */
7783 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
7788 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
7793 /* Cause the flow to be processed as if it came in on the real device with
7794 * the VLAN device's VLAN ID. */
7795 flow->in_port = realdev;
7796 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
7801 vsp_remove(struct ofport_dpif *port)
7803 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7804 struct vlan_splinter *vsp;
7806 vsp = vlandev_find(ofproto, port->up.ofp_port);
7808 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
7809 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
7812 port->realdev_ofp_port = 0;
7814 VLOG_ERR("missing vlan device record");
7819 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
7821 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7823 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
7824 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
7825 == realdev_ofp_port)) {
7826 struct vlan_splinter *vsp;
7828 vsp = xmalloc(sizeof *vsp);
7829 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
7830 hash_int(port->up.ofp_port, 0));
7831 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
7832 hash_realdev_vid(realdev_ofp_port, vid));
7833 vsp->realdev_ofp_port = realdev_ofp_port;
7834 vsp->vlandev_ofp_port = port->up.ofp_port;
7837 port->realdev_ofp_port = realdev_ofp_port;
7839 VLOG_ERR("duplicate vlan device record");
7844 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
7846 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
7847 return ofport ? ofport->odp_port : OVSP_NONE;
7850 static struct ofport_dpif *
7851 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
7853 struct ofport_dpif *port;
7855 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
7856 hash_int(odp_port, 0),
7857 &backer->odp_to_ofport_map) {
7858 if (port->odp_port == odp_port) {
7867 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
7869 struct ofport_dpif *port;
7871 port = odp_port_to_ofport(ofproto->backer, odp_port);
7872 if (port && ofproto == ofproto_dpif_cast(port->up.ofproto)) {
7873 return port->up.ofp_port;
7879 const struct ofproto_class ofproto_dpif_class = {
7914 port_is_lacp_current,
7915 NULL, /* rule_choose_table */
7922 rule_modify_actions,
7931 get_cfm_remote_mpids,
7936 get_stp_port_status,
7943 is_mirror_output_bundle,
7944 forward_bpdu_changed,
7945 set_mac_table_config,