2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-provider.h"
26 #include "byte-order.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
36 #include "mac-learning.h"
37 #include "meta-flow.h"
38 #include "multipath.h"
45 #include "ofp-actions.h"
46 #include "ofp-parse.h"
47 #include "ofp-print.h"
48 #include "ofproto-dpif-governor.h"
49 #include "ofproto-dpif-sflow.h"
50 #include "poll-loop.h"
54 #include "unaligned.h"
56 #include "vlan-bitmap.h"
59 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
61 COVERAGE_DEFINE(ofproto_dpif_expired);
62 COVERAGE_DEFINE(ofproto_dpif_xlate);
63 COVERAGE_DEFINE(facet_changed_rule);
64 COVERAGE_DEFINE(facet_revalidate);
65 COVERAGE_DEFINE(facet_unexpected);
66 COVERAGE_DEFINE(facet_suppress);
68 /* Maximum depth of flow table recursion (due to resubmit actions) in a
69 * flow translation. */
70 #define MAX_RESUBMIT_RECURSION 64
72 /* Number of implemented OpenFlow tables. */
73 enum { N_TABLES = 255 };
74 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
75 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
86 * - Do include packets and bytes from facets that have been deleted or
87 * whose own statistics have been folded into the rule.
89 * - Do include packets and bytes sent "by hand" that were accounted to
90 * the rule without any facet being involved (this is a rare corner
91 * case in rule_execute()).
93 * - Do not include packet or bytes that can be obtained from any facet's
94 * packet_count or byte_count member or that can be obtained from the
95 * datapath by, e.g., dpif_flow_get() for any subfacet.
97 uint64_t packet_count; /* Number of packets received. */
98 uint64_t byte_count; /* Number of bytes received. */
100 tag_type tag; /* Caches rule_calculate_tag() result. */
102 struct list facets; /* List of "struct facet"s. */
105 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
107 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
110 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
111 const struct flow *);
112 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
115 static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto,
116 const struct flow *flow);
118 static void rule_credit_stats(struct rule_dpif *,
119 const struct dpif_flow_stats *);
120 static void flow_push_stats(struct rule_dpif *, const struct flow *,
121 const struct dpif_flow_stats *);
122 static tag_type rule_calculate_tag(const struct flow *,
123 const struct minimask *, uint32_t basis);
124 static void rule_invalidate(const struct rule_dpif *);
126 #define MAX_MIRRORS 32
127 typedef uint32_t mirror_mask_t;
128 #define MIRROR_MASK_C(X) UINT32_C(X)
129 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
131 struct ofproto_dpif *ofproto; /* Owning ofproto. */
132 size_t idx; /* In ofproto's "mirrors" array. */
133 void *aux; /* Key supplied by ofproto's client. */
134 char *name; /* Identifier for log messages. */
136 /* Selection criteria. */
137 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
138 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
139 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
141 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
142 struct ofbundle *out; /* Output port or NULL. */
143 int out_vlan; /* Output VLAN or -1. */
144 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
147 int64_t packet_count; /* Number of packets sent. */
148 int64_t byte_count; /* Number of bytes sent. */
151 static void mirror_destroy(struct ofmirror *);
152 static void update_mirror_stats(struct ofproto_dpif *ofproto,
153 mirror_mask_t mirrors,
154 uint64_t packets, uint64_t bytes);
157 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
158 struct ofproto_dpif *ofproto; /* Owning ofproto. */
159 void *aux; /* Key supplied by ofproto's client. */
160 char *name; /* Identifier for log messages. */
163 struct list ports; /* Contains "struct ofport"s. */
164 enum port_vlan_mode vlan_mode; /* VLAN mode */
165 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
166 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
167 * NULL if all VLANs are trunked. */
168 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
169 struct bond *bond; /* Nonnull iff more than one port. */
170 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
173 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
175 /* Port mirroring info. */
176 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
177 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
178 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
181 static void bundle_remove(struct ofport *);
182 static void bundle_update(struct ofbundle *);
183 static void bundle_destroy(struct ofbundle *);
184 static void bundle_del_port(struct ofport_dpif *);
185 static void bundle_run(struct ofbundle *);
186 static void bundle_wait(struct ofbundle *);
187 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
188 uint16_t in_port, bool warn,
189 struct ofport_dpif **in_ofportp);
191 /* A controller may use OFPP_NONE as the ingress port to indicate that
192 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
193 * when an input bundle is needed for validation (e.g., mirroring or
194 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
195 * any 'port' structs, so care must be taken when dealing with it. */
196 static struct ofbundle ofpp_none_bundle = {
198 .vlan_mode = PORT_VLAN_TRUNK
201 static void stp_run(struct ofproto_dpif *ofproto);
202 static void stp_wait(struct ofproto_dpif *ofproto);
203 static int set_stp_port(struct ofport *,
204 const struct ofproto_port_stp_settings *);
206 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
208 struct action_xlate_ctx {
209 /* action_xlate_ctx_init() initializes these members. */
212 struct ofproto_dpif *ofproto;
214 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
215 * this flow when actions change header fields. */
218 /* The packet corresponding to 'flow', or a null pointer if we are
219 * revalidating without a packet to refer to. */
220 const struct ofpbuf *packet;
222 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
223 * actions update the flow table?
225 * We want to update these tables if we are actually processing a packet,
226 * or if we are accounting for packets that the datapath has processed, but
227 * not if we are just revalidating. */
230 /* The rule that we are currently translating, or NULL. */
231 struct rule_dpif *rule;
233 /* Union of the set of TCP flags seen so far in this flow. (Used only by
234 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
238 /* If nonnull, flow translation calls this function just before executing a
239 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
240 * when the recursion depth is exceeded.
242 * 'rule' is the rule being submitted into. It will be null if the
243 * resubmit or OFPP_TABLE action didn't find a matching rule.
245 * This is normally null so the client has to set it manually after
246 * calling action_xlate_ctx_init(). */
247 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule);
249 /* If nonnull, flow translation calls this function to report some
250 * significant decision, e.g. to explain why OFPP_NORMAL translation
251 * dropped a packet. */
252 void (*report_hook)(struct action_xlate_ctx *, const char *s);
254 /* If nonnull, flow translation credits the specified statistics to each
255 * rule reached through a resubmit or OFPP_TABLE action.
257 * This is normally null so the client has to set it manually after
258 * calling action_xlate_ctx_init(). */
259 const struct dpif_flow_stats *resubmit_stats;
261 /* xlate_actions() initializes and uses these members. The client might want
262 * to look at them after it returns. */
264 struct ofpbuf *odp_actions; /* Datapath actions. */
265 tag_type tags; /* Tags associated with actions. */
266 enum slow_path_reason slow; /* 0 if fast path may be used. */
267 bool has_learn; /* Actions include NXAST_LEARN? */
268 bool has_normal; /* Actions output to OFPP_NORMAL? */
269 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
270 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
271 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
273 /* xlate_actions() initializes and uses these members, but the client has no
274 * reason to look at them. */
276 int recurse; /* Recursion level, via xlate_table_action. */
277 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
278 struct flow base_flow; /* Flow at the last commit. */
279 uint32_t orig_skb_priority; /* Priority when packet arrived. */
280 uint8_t table_id; /* OpenFlow table ID where flow was found. */
281 uint32_t sflow_n_outputs; /* Number of output ports. */
282 uint32_t sflow_odp_port; /* Output port for composing sFlow action. */
283 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
284 bool exit; /* No further actions should be processed. */
285 struct flow orig_flow; /* Copy of original flow. */
288 static void action_xlate_ctx_init(struct action_xlate_ctx *,
289 struct ofproto_dpif *, const struct flow *,
290 ovs_be16 initial_tci, struct rule_dpif *,
291 uint8_t tcp_flags, const struct ofpbuf *);
292 static void xlate_actions(struct action_xlate_ctx *,
293 const struct ofpact *ofpacts, size_t ofpacts_len,
294 struct ofpbuf *odp_actions);
295 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
296 const struct ofpact *ofpacts,
299 static size_t put_userspace_action(const struct ofproto_dpif *,
300 struct ofpbuf *odp_actions,
302 const union user_action_cookie *);
304 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
305 enum slow_path_reason,
306 uint64_t *stub, size_t stub_size,
307 const struct nlattr **actionsp,
308 size_t *actions_lenp);
310 static void xlate_report(struct action_xlate_ctx *ctx, const char *s);
312 /* A subfacet (see "struct subfacet" below) has three possible installation
315 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
316 * case just after the subfacet is created, just before the subfacet is
317 * destroyed, or if the datapath returns an error when we try to install a
320 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
322 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
323 * ofproto_dpif is installed in the datapath.
326 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
327 SF_FAST_PATH, /* Full actions are installed. */
328 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
331 static const char *subfacet_path_to_string(enum subfacet_path);
333 /* A dpif flow and actions associated with a facet.
335 * See also the large comment on struct facet. */
338 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
339 struct list list_node; /* In struct facet's 'facets' list. */
340 struct facet *facet; /* Owning facet. */
344 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
345 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
346 * regenerate the ODP flow key from ->facet->flow. */
347 enum odp_key_fitness key_fitness;
351 long long int used; /* Time last used; time created if not used. */
353 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
354 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
358 * These should be essentially identical for every subfacet in a facet, but
359 * may differ in trivial ways due to VLAN splinters. */
360 size_t actions_len; /* Number of bytes in actions[]. */
361 struct nlattr *actions; /* Datapath actions. */
363 enum slow_path_reason slow; /* 0 if fast path may be used. */
364 enum subfacet_path path; /* Installed in datapath? */
366 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
367 * splinters can cause it to differ. This value should be removed when
368 * the VLAN splinters feature is no longer needed. */
369 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
371 /* Datapath port the packet arrived on. This is needed to remove
372 * flows for ports that are no longer part of the bridge. Since the
373 * flow definition only has the OpenFlow port number and the port is
374 * no longer part of the bridge, we can't determine the datapath port
375 * number needed to delete the flow from the datapath. */
376 uint32_t odp_in_port;
379 #define SUBFACET_DESTROY_MAX_BATCH 50
381 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
383 static struct subfacet *subfacet_find(struct ofproto_dpif *,
384 const struct nlattr *key, size_t key_len,
386 const struct flow *flow);
387 static void subfacet_destroy(struct subfacet *);
388 static void subfacet_destroy__(struct subfacet *);
389 static void subfacet_destroy_batch(struct ofproto_dpif *,
390 struct subfacet **, int n);
391 static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *,
393 static void subfacet_reset_dp_stats(struct subfacet *,
394 struct dpif_flow_stats *);
395 static void subfacet_update_time(struct subfacet *, long long int used);
396 static void subfacet_update_stats(struct subfacet *,
397 const struct dpif_flow_stats *);
398 static void subfacet_make_actions(struct subfacet *,
399 const struct ofpbuf *packet,
400 struct ofpbuf *odp_actions);
401 static int subfacet_install(struct subfacet *,
402 const struct nlattr *actions, size_t actions_len,
403 struct dpif_flow_stats *, enum slow_path_reason);
404 static void subfacet_uninstall(struct subfacet *);
406 static enum subfacet_path subfacet_want_path(enum slow_path_reason);
408 /* An exact-match instantiation of an OpenFlow flow.
410 * A facet associates a "struct flow", which represents the Open vSwitch
411 * userspace idea of an exact-match flow, with one or more subfacets. Each
412 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
413 * the facet. When the kernel module (or other dpif implementation) and Open
414 * vSwitch userspace agree on the definition of a flow key, there is exactly
415 * one subfacet per facet. If the dpif implementation supports more-specific
416 * flow matching than userspace, however, a facet can have more than one
417 * subfacet, each of which corresponds to some distinction in flow that
418 * userspace simply doesn't understand.
420 * Flow expiration works in terms of subfacets, so a facet must have at least
421 * one subfacet or it will never expire, leaking memory. */
424 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
425 struct list list_node; /* In owning rule's 'facets' list. */
426 struct rule_dpif *rule; /* Owning rule. */
429 struct list subfacets;
430 long long int used; /* Time last used; time created if not used. */
437 * - Do include packets and bytes sent "by hand", e.g. with
440 * - Do include packets and bytes that were obtained from the datapath
441 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
442 * DPIF_FP_ZERO_STATS).
444 * - Do not include packets or bytes that can be obtained from the
445 * datapath for any existing subfacet.
447 uint64_t packet_count; /* Number of packets received. */
448 uint64_t byte_count; /* Number of bytes received. */
450 /* Resubmit statistics. */
451 uint64_t prev_packet_count; /* Number of packets from last stats push. */
452 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
453 long long int prev_used; /* Used time from last stats push. */
456 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
457 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
458 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
460 /* Properties of datapath actions.
462 * Every subfacet has its own actions because actions can differ slightly
463 * between splintered and non-splintered subfacets due to the VLAN tag
464 * being initially different (present vs. absent). All of them have these
465 * properties in common so we just store one copy of them here. */
466 bool has_learn; /* Actions include NXAST_LEARN? */
467 bool has_normal; /* Actions output to OFPP_NORMAL? */
468 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
469 tag_type tags; /* Tags that would require revalidation. */
470 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
472 /* Storage for a single subfacet, to reduce malloc() time and space
473 * overhead. (A facet always has at least one subfacet and in the common
474 * case has exactly one subfacet.) */
475 struct subfacet one_subfacet;
478 static struct facet *facet_create(struct rule_dpif *,
479 const struct flow *, uint32_t hash);
480 static void facet_remove(struct facet *);
481 static void facet_free(struct facet *);
483 static struct facet *facet_find(struct ofproto_dpif *,
484 const struct flow *, uint32_t hash);
485 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
486 const struct flow *, uint32_t hash);
487 static void facet_revalidate(struct facet *);
488 static bool facet_check_consistency(struct facet *);
490 static void facet_flush_stats(struct facet *);
492 static void facet_update_time(struct facet *, long long int used);
493 static void facet_reset_counters(struct facet *);
494 static void facet_push_stats(struct facet *);
495 static void facet_learn(struct facet *);
496 static void facet_account(struct facet *);
498 static bool facet_is_controller_flow(struct facet *);
501 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
505 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
506 struct list bundle_node; /* In struct ofbundle's "ports" list. */
507 struct cfm *cfm; /* Connectivity Fault Management, if any. */
508 tag_type tag; /* Tag associated with this port. */
509 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
510 bool may_enable; /* May be enabled in bonds. */
511 long long int carrier_seq; /* Carrier status changes. */
514 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
515 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
516 long long int stp_state_entered;
518 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
520 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
522 * This is deprecated. It is only for compatibility with broken device
523 * drivers in old versions of Linux that do not properly support VLANs when
524 * VLAN devices are not used. When broken device drivers are no longer in
525 * widespread use, we will delete these interfaces. */
526 uint16_t realdev_ofp_port;
530 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
531 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
532 * traffic egressing the 'ofport' with that priority should be marked with. */
533 struct priority_to_dscp {
534 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
535 uint32_t priority; /* Priority of this queue (see struct flow). */
537 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
540 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
542 * This is deprecated. It is only for compatibility with broken device drivers
543 * in old versions of Linux that do not properly support VLANs when VLAN
544 * devices are not used. When broken device drivers are no longer in
545 * widespread use, we will delete these interfaces. */
546 struct vlan_splinter {
547 struct hmap_node realdev_vid_node;
548 struct hmap_node vlandev_node;
549 uint16_t realdev_ofp_port;
550 uint16_t vlandev_ofp_port;
554 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
555 uint32_t realdev, ovs_be16 vlan_tci);
556 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
557 static void vsp_remove(struct ofport_dpif *);
558 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
560 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
562 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
565 static struct ofport_dpif *
566 ofport_dpif_cast(const struct ofport *ofport)
568 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
569 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
572 static void port_run(struct ofport_dpif *);
573 static void port_run_fast(struct ofport_dpif *);
574 static void port_wait(struct ofport_dpif *);
575 static int set_cfm(struct ofport *, const struct cfm_settings *);
576 static void ofport_clear_priorities(struct ofport_dpif *);
578 struct dpif_completion {
579 struct list list_node;
580 struct ofoperation *op;
583 /* Extra information about a classifier table.
584 * Currently used just for optimized flow revalidation. */
586 /* If either of these is nonnull, then this table has a form that allows
587 * flows to be tagged to avoid revalidating most flows for the most common
588 * kinds of flow table changes. */
589 struct cls_table *catchall_table; /* Table that wildcards all fields. */
590 struct cls_table *other_table; /* Table with any other wildcard set. */
591 uint32_t basis; /* Keeps each table's tags separate. */
594 /* Reasons that we might need to revalidate every facet, and corresponding
597 * A value of 0 means that there is no need to revalidate.
599 * It would be nice to have some cleaner way to integrate with coverage
600 * counters, but with only a few reasons I guess this is good enough for
602 enum revalidate_reason {
603 REV_RECONFIGURE = 1, /* Switch configuration changed. */
604 REV_STP, /* Spanning tree protocol port status change. */
605 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
606 REV_FLOW_TABLE, /* Flow table changed. */
607 REV_INCONSISTENCY /* Facet self-check failed. */
609 COVERAGE_DEFINE(rev_reconfigure);
610 COVERAGE_DEFINE(rev_stp);
611 COVERAGE_DEFINE(rev_port_toggled);
612 COVERAGE_DEFINE(rev_flow_table);
613 COVERAGE_DEFINE(rev_inconsistency);
615 /* All datapaths of a given type share a single dpif backer instance. */
620 struct timer next_expiration;
621 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
624 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
625 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
627 static struct ofport_dpif *
628 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
630 struct ofproto_dpif {
631 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
633 struct dpif_backer *backer;
635 /* Special OpenFlow rules. */
636 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
637 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
643 struct netflow *netflow;
644 struct dpif_sflow *sflow;
645 struct hmap bundles; /* Contains "struct ofbundle"s. */
646 struct mac_learning *ml;
647 struct ofmirror *mirrors[MAX_MIRRORS];
649 bool has_bonded_bundles;
653 struct hmap subfacets;
654 struct governor *governor;
657 struct table_dpif tables[N_TABLES];
658 enum revalidate_reason need_revalidate;
659 struct tag_set revalidate_set;
661 /* Support for debugging async flow mods. */
662 struct list completions;
664 bool has_bundle_action; /* True when the first bundle action appears. */
665 struct netdev_stats stats; /* To account packets generated and consumed in
670 long long int stp_last_tick;
672 /* VLAN splinters. */
673 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
674 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
677 struct sset ports; /* Set of port names. */
678 struct sset port_poll_set; /* Queued names for port_poll() reply. */
679 int port_poll_errno; /* Last errno for port_poll() reply. */
682 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
683 * for debugging the asynchronous flow_mod implementation.) */
686 /* All existing ofproto_dpif instances, indexed by ->up.name. */
687 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
689 static void ofproto_dpif_unixctl_init(void);
691 static struct ofproto_dpif *
692 ofproto_dpif_cast(const struct ofproto *ofproto)
694 assert(ofproto->ofproto_class == &ofproto_dpif_class);
695 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
698 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
700 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
702 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
703 const struct ofpbuf *, ovs_be16 initial_tci,
706 /* Packet processing. */
707 static void update_learning_table(struct ofproto_dpif *,
708 const struct flow *, int vlan,
711 #define FLOW_MISS_MAX_BATCH 50
712 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
714 /* Flow expiration. */
715 static int expire(struct dpif_backer *);
718 static void send_netflow_active_timeouts(struct ofproto_dpif *);
721 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
722 static size_t compose_sflow_action(const struct ofproto_dpif *,
723 struct ofpbuf *odp_actions,
724 const struct flow *, uint32_t odp_port);
725 static void add_mirror_actions(struct action_xlate_ctx *ctx,
726 const struct flow *flow);
727 /* Global variables. */
728 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
730 /* Initial mappings of port to bridge mappings. */
731 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
733 /* Factory functions. */
736 init(const struct shash *iface_hints)
738 struct shash_node *node;
740 /* Make a local copy, since we don't own 'iface_hints' elements. */
741 SHASH_FOR_EACH(node, iface_hints) {
742 const struct iface_hint *orig_hint = node->data;
743 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
745 new_hint->br_name = xstrdup(orig_hint->br_name);
746 new_hint->br_type = xstrdup(orig_hint->br_type);
747 new_hint->ofp_port = orig_hint->ofp_port;
749 shash_add(&init_ofp_ports, node->name, new_hint);
754 enumerate_types(struct sset *types)
756 dp_enumerate_types(types);
760 enumerate_names(const char *type, struct sset *names)
762 struct ofproto_dpif *ofproto;
765 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
766 if (strcmp(type, ofproto->up.type)) {
769 sset_add(names, ofproto->up.name);
776 del(const char *type, const char *name)
781 error = dpif_open(name, type, &dpif);
783 error = dpif_delete(dpif);
790 port_open_type(const char *datapath_type, const char *port_type)
792 return dpif_port_open_type(datapath_type, port_type);
795 /* Type functions. */
797 static struct ofproto_dpif *
798 lookup_ofproto_dpif_by_port_name(const char *name)
800 struct ofproto_dpif *ofproto;
802 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
803 if (sset_contains(&ofproto->ports, name)) {
812 type_run(const char *type)
814 struct dpif_backer *backer;
818 backer = shash_find_data(&all_dpif_backers, type);
820 /* This is not necessarily a problem, since backers are only
821 * created on demand. */
825 dpif_run(backer->dpif);
827 if (timer_expired(&backer->next_expiration)) {
828 int delay = expire(backer);
829 timer_set_duration(&backer->next_expiration, delay);
832 /* Check for port changes in the dpif. */
833 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
834 struct ofproto_dpif *ofproto;
835 struct dpif_port port;
837 /* Don't report on the datapath's device. */
838 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
842 ofproto = lookup_ofproto_dpif_by_port_name(devname);
843 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
844 /* The port was removed. If we know the datapath,
845 * report it through poll_set(). If we don't, it may be
846 * notifying us of a removal we initiated, so ignore it.
847 * If there's a pending ENOBUFS, let it stand, since
848 * everything will be reevaluated. */
849 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
850 sset_add(&ofproto->port_poll_set, devname);
851 ofproto->port_poll_errno = 0;
853 } else if (!ofproto) {
854 /* The port was added, but we don't know with which
855 * ofproto we should associate it. Delete it. */
856 dpif_port_del(backer->dpif, port.port_no);
858 dpif_port_destroy(&port);
864 if (error != EAGAIN) {
865 struct ofproto_dpif *ofproto;
867 /* There was some sort of error, so propagate it to all
868 * ofprotos that use this backer. */
869 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
870 &all_ofproto_dpifs) {
871 if (ofproto->backer == backer) {
872 sset_clear(&ofproto->port_poll_set);
873 ofproto->port_poll_errno = error;
882 type_run_fast(const char *type)
884 struct dpif_backer *backer;
887 backer = shash_find_data(&all_dpif_backers, type);
889 /* This is not necessarily a problem, since backers are only
890 * created on demand. */
894 /* Handle one or more batches of upcalls, until there's nothing left to do
895 * or until we do a fixed total amount of work.
897 * We do work in batches because it can be much cheaper to set up a number
898 * of flows and fire off their patches all at once. We do multiple batches
899 * because in some cases handling a packet can cause another packet to be
900 * queued almost immediately as part of the return flow. Both
901 * optimizations can make major improvements on some benchmarks and
902 * presumably for real traffic as well. */
904 while (work < FLOW_MISS_MAX_BATCH) {
905 int retval = handle_upcalls(backer, FLOW_MISS_MAX_BATCH - work);
916 type_wait(const char *type)
918 struct dpif_backer *backer;
920 backer = shash_find_data(&all_dpif_backers, type);
922 /* This is not necessarily a problem, since backers are only
923 * created on demand. */
927 timer_wait(&backer->next_expiration);
930 /* Basic life-cycle. */
932 static int add_internal_flows(struct ofproto_dpif *);
934 static struct ofproto *
937 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
942 dealloc(struct ofproto *ofproto_)
944 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
949 close_dpif_backer(struct dpif_backer *backer)
951 struct shash_node *node;
953 assert(backer->refcount > 0);
955 if (--backer->refcount) {
959 hmap_destroy(&backer->odp_to_ofport_map);
960 node = shash_find(&all_dpif_backers, backer->type);
962 shash_delete(&all_dpif_backers, node);
963 dpif_close(backer->dpif);
968 /* Datapath port slated for removal from datapath. */
970 struct list list_node;
975 open_dpif_backer(const char *type, struct dpif_backer **backerp)
977 struct dpif_backer *backer;
978 struct dpif_port_dump port_dump;
979 struct dpif_port port;
980 struct shash_node *node;
981 struct list garbage_list;
982 struct odp_garbage *garbage, *next;
988 backer = shash_find_data(&all_dpif_backers, type);
995 backer_name = xasprintf("ovs-%s", type);
997 /* Remove any existing datapaths, since we assume we're the only
998 * userspace controlling the datapath. */
1000 dp_enumerate_names(type, &names);
1001 SSET_FOR_EACH(name, &names) {
1002 struct dpif *old_dpif;
1004 /* Don't remove our backer if it exists. */
1005 if (!strcmp(name, backer_name)) {
1009 if (dpif_open(name, type, &old_dpif)) {
1010 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1012 dpif_delete(old_dpif);
1013 dpif_close(old_dpif);
1016 sset_destroy(&names);
1018 backer = xmalloc(sizeof *backer);
1020 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1023 VLOG_ERR("failed to open datapath of type %s: %s", type,
1028 backer->type = xstrdup(type);
1029 backer->refcount = 1;
1030 hmap_init(&backer->odp_to_ofport_map);
1031 timer_set_duration(&backer->next_expiration, 1000);
1034 dpif_flow_flush(backer->dpif);
1036 /* Loop through the ports already on the datapath and remove any
1037 * that we don't need anymore. */
1038 list_init(&garbage_list);
1039 dpif_port_dump_start(&port_dump, backer->dpif);
1040 while (dpif_port_dump_next(&port_dump, &port)) {
1041 node = shash_find(&init_ofp_ports, port.name);
1042 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1043 garbage = xmalloc(sizeof *garbage);
1044 garbage->odp_port = port.port_no;
1045 list_push_front(&garbage_list, &garbage->list_node);
1048 dpif_port_dump_done(&port_dump);
1050 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1051 dpif_port_del(backer->dpif, garbage->odp_port);
1052 list_remove(&garbage->list_node);
1056 shash_add(&all_dpif_backers, type, backer);
1058 error = dpif_recv_set(backer->dpif, true);
1060 VLOG_ERR("failed to listen on datapath of type %s: %s",
1061 type, strerror(error));
1062 close_dpif_backer(backer);
1070 construct(struct ofproto *ofproto_)
1072 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1073 struct shash_node *node, *next;
1078 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1083 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1084 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1086 ofproto->n_matches = 0;
1088 ofproto->netflow = NULL;
1089 ofproto->sflow = NULL;
1090 ofproto->stp = NULL;
1091 hmap_init(&ofproto->bundles);
1092 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1093 for (i = 0; i < MAX_MIRRORS; i++) {
1094 ofproto->mirrors[i] = NULL;
1096 ofproto->has_bonded_bundles = false;
1098 hmap_init(&ofproto->facets);
1099 hmap_init(&ofproto->subfacets);
1100 ofproto->governor = NULL;
1102 for (i = 0; i < N_TABLES; i++) {
1103 struct table_dpif *table = &ofproto->tables[i];
1105 table->catchall_table = NULL;
1106 table->other_table = NULL;
1107 table->basis = random_uint32();
1109 ofproto->need_revalidate = 0;
1110 tag_set_init(&ofproto->revalidate_set);
1112 list_init(&ofproto->completions);
1114 ofproto_dpif_unixctl_init();
1116 ofproto->has_mirrors = false;
1117 ofproto->has_bundle_action = false;
1119 hmap_init(&ofproto->vlandev_map);
1120 hmap_init(&ofproto->realdev_vid_map);
1122 sset_init(&ofproto->ports);
1123 sset_init(&ofproto->port_poll_set);
1124 ofproto->port_poll_errno = 0;
1126 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1127 struct iface_hint *iface_hint = node->data;
1129 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1130 /* Check if the datapath already has this port. */
1131 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1132 sset_add(&ofproto->ports, node->name);
1135 free(iface_hint->br_name);
1136 free(iface_hint->br_type);
1138 shash_delete(&init_ofp_ports, node);
1142 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1143 hash_string(ofproto->up.name, 0));
1144 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1146 ofproto_init_tables(ofproto_, N_TABLES);
1147 error = add_internal_flows(ofproto);
1148 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1154 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1155 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1157 struct ofputil_flow_mod fm;
1160 match_init_catchall(&fm.match);
1162 match_set_reg(&fm.match, 0, id);
1163 fm.new_cookie = htonll(0);
1164 fm.cookie = htonll(0);
1165 fm.cookie_mask = htonll(0);
1166 fm.table_id = TBL_INTERNAL;
1167 fm.command = OFPFC_ADD;
1168 fm.idle_timeout = 0;
1169 fm.hard_timeout = 0;
1173 fm.ofpacts = ofpacts->data;
1174 fm.ofpacts_len = ofpacts->size;
1176 error = ofproto_flow_mod(&ofproto->up, &fm);
1178 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1179 id, ofperr_to_string(error));
1183 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1184 assert(*rulep != NULL);
1190 add_internal_flows(struct ofproto_dpif *ofproto)
1192 struct ofpact_controller *controller;
1193 uint64_t ofpacts_stub[128 / 8];
1194 struct ofpbuf ofpacts;
1198 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1201 controller = ofpact_put_CONTROLLER(&ofpacts);
1202 controller->max_len = UINT16_MAX;
1203 controller->controller_id = 0;
1204 controller->reason = OFPR_NO_MATCH;
1205 ofpact_pad(&ofpacts);
1207 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1212 ofpbuf_clear(&ofpacts);
1213 error = add_internal_flow(ofproto, id++, &ofpacts,
1214 &ofproto->no_packet_in_rule);
1219 complete_operations(struct ofproto_dpif *ofproto)
1221 struct dpif_completion *c, *next;
1223 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1224 ofoperation_complete(c->op, 0);
1225 list_remove(&c->list_node);
1231 destruct(struct ofproto *ofproto_)
1233 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1234 struct rule_dpif *rule, *next_rule;
1235 struct oftable *table;
1238 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1239 complete_operations(ofproto);
1241 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1242 struct cls_cursor cursor;
1244 cls_cursor_init(&cursor, &table->cls, NULL);
1245 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1246 ofproto_rule_destroy(&rule->up);
1250 for (i = 0; i < MAX_MIRRORS; i++) {
1251 mirror_destroy(ofproto->mirrors[i]);
1254 netflow_destroy(ofproto->netflow);
1255 dpif_sflow_destroy(ofproto->sflow);
1256 hmap_destroy(&ofproto->bundles);
1257 mac_learning_destroy(ofproto->ml);
1259 hmap_destroy(&ofproto->facets);
1260 hmap_destroy(&ofproto->subfacets);
1261 governor_destroy(ofproto->governor);
1263 hmap_destroy(&ofproto->vlandev_map);
1264 hmap_destroy(&ofproto->realdev_vid_map);
1266 sset_destroy(&ofproto->ports);
1267 sset_destroy(&ofproto->port_poll_set);
1269 close_dpif_backer(ofproto->backer);
1273 run_fast(struct ofproto *ofproto_)
1275 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1276 struct ofport_dpif *ofport;
1278 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1279 port_run_fast(ofport);
1286 run(struct ofproto *ofproto_)
1288 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1289 struct ofport_dpif *ofport;
1290 struct ofbundle *bundle;
1294 complete_operations(ofproto);
1297 error = run_fast(ofproto_);
1302 if (ofproto->netflow) {
1303 if (netflow_run(ofproto->netflow)) {
1304 send_netflow_active_timeouts(ofproto);
1307 if (ofproto->sflow) {
1308 dpif_sflow_run(ofproto->sflow);
1311 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1314 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1319 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
1321 /* Now revalidate if there's anything to do. */
1322 if (ofproto->need_revalidate
1323 || !tag_set_is_empty(&ofproto->revalidate_set)) {
1324 struct tag_set revalidate_set = ofproto->revalidate_set;
1325 bool revalidate_all = ofproto->need_revalidate;
1326 struct facet *facet;
1328 switch (ofproto->need_revalidate) {
1329 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
1330 case REV_STP: COVERAGE_INC(rev_stp); break;
1331 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
1332 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
1333 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
1336 /* Clear the revalidation flags. */
1337 tag_set_init(&ofproto->revalidate_set);
1338 ofproto->need_revalidate = 0;
1340 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
1342 || tag_set_intersects(&revalidate_set, facet->tags)) {
1343 facet_revalidate(facet);
1348 /* Check the consistency of a random facet, to aid debugging. */
1349 if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) {
1350 struct facet *facet;
1352 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1353 struct facet, hmap_node);
1354 if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) {
1355 if (!facet_check_consistency(facet)) {
1356 ofproto->need_revalidate = REV_INCONSISTENCY;
1361 if (ofproto->governor) {
1364 governor_run(ofproto->governor);
1366 /* If the governor has shrunk to its minimum size and the number of
1367 * subfacets has dwindled, then drop the governor entirely.
1369 * For hysteresis, the number of subfacets to drop the governor is
1370 * smaller than the number needed to trigger its creation. */
1371 n_subfacets = hmap_count(&ofproto->subfacets);
1372 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1373 && governor_is_idle(ofproto->governor)) {
1374 governor_destroy(ofproto->governor);
1375 ofproto->governor = NULL;
1383 wait(struct ofproto *ofproto_)
1385 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1386 struct ofport_dpif *ofport;
1387 struct ofbundle *bundle;
1389 if (!clogged && !list_is_empty(&ofproto->completions)) {
1390 poll_immediate_wake();
1393 dpif_wait(ofproto->backer->dpif);
1394 dpif_recv_wait(ofproto->backer->dpif);
1395 if (ofproto->sflow) {
1396 dpif_sflow_wait(ofproto->sflow);
1398 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
1399 poll_immediate_wake();
1401 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1404 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1405 bundle_wait(bundle);
1407 if (ofproto->netflow) {
1408 netflow_wait(ofproto->netflow);
1410 mac_learning_wait(ofproto->ml);
1412 if (ofproto->need_revalidate) {
1413 /* Shouldn't happen, but if it does just go around again. */
1414 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1415 poll_immediate_wake();
1417 if (ofproto->governor) {
1418 governor_wait(ofproto->governor);
1423 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1425 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1427 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1428 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1432 flush(struct ofproto *ofproto_)
1434 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1435 struct subfacet *subfacet, *next_subfacet;
1436 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1440 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1441 &ofproto->subfacets) {
1442 if (subfacet->path != SF_NOT_INSTALLED) {
1443 batch[n_batch++] = subfacet;
1444 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1445 subfacet_destroy_batch(ofproto, batch, n_batch);
1449 subfacet_destroy(subfacet);
1454 subfacet_destroy_batch(ofproto, batch, n_batch);
1459 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1460 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1462 *arp_match_ip = true;
1463 *actions = (OFPUTIL_A_OUTPUT |
1464 OFPUTIL_A_SET_VLAN_VID |
1465 OFPUTIL_A_SET_VLAN_PCP |
1466 OFPUTIL_A_STRIP_VLAN |
1467 OFPUTIL_A_SET_DL_SRC |
1468 OFPUTIL_A_SET_DL_DST |
1469 OFPUTIL_A_SET_NW_SRC |
1470 OFPUTIL_A_SET_NW_DST |
1471 OFPUTIL_A_SET_NW_TOS |
1472 OFPUTIL_A_SET_TP_SRC |
1473 OFPUTIL_A_SET_TP_DST |
1478 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1480 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1481 struct dpif_dp_stats s;
1483 strcpy(ots->name, "classifier");
1485 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1487 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1488 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1491 static struct ofport *
1494 struct ofport_dpif *port = xmalloc(sizeof *port);
1499 port_dealloc(struct ofport *port_)
1501 struct ofport_dpif *port = ofport_dpif_cast(port_);
1506 port_construct(struct ofport *port_)
1508 struct ofport_dpif *port = ofport_dpif_cast(port_);
1509 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1510 struct dpif_port dpif_port;
1513 ofproto->need_revalidate = REV_RECONFIGURE;
1514 port->bundle = NULL;
1516 port->tag = tag_create_random();
1517 port->may_enable = true;
1518 port->stp_port = NULL;
1519 port->stp_state = STP_DISABLED;
1520 hmap_init(&port->priorities);
1521 port->realdev_ofp_port = 0;
1522 port->vlandev_vid = 0;
1523 port->carrier_seq = netdev_get_carrier_resets(port->up.netdev);
1525 error = dpif_port_query_by_name(ofproto->backer->dpif,
1526 netdev_get_name(port->up.netdev),
1532 port->odp_port = dpif_port.port_no;
1534 /* Sanity-check that a mapping doesn't already exist. This
1535 * shouldn't happen. */
1536 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1537 VLOG_ERR("port %s already has an OpenFlow port number\n",
1542 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1543 hash_int(port->odp_port, 0));
1545 if (ofproto->sflow) {
1546 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1553 port_destruct(struct ofport *port_)
1555 struct ofport_dpif *port = ofport_dpif_cast(port_);
1556 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1557 const char *devname = netdev_get_name(port->up.netdev);
1559 if (dpif_port_exists(ofproto->backer->dpif, devname)) {
1560 /* The underlying device is still there, so delete it. This
1561 * happens when the ofproto is being destroyed, since the caller
1562 * assumes that removal of attached ports will happen as part of
1564 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1567 sset_find_and_delete(&ofproto->ports, devname);
1568 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1569 ofproto->need_revalidate = REV_RECONFIGURE;
1570 bundle_remove(port_);
1571 set_cfm(port_, NULL);
1572 if (ofproto->sflow) {
1573 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1576 ofport_clear_priorities(port);
1577 hmap_destroy(&port->priorities);
1581 port_modified(struct ofport *port_)
1583 struct ofport_dpif *port = ofport_dpif_cast(port_);
1585 if (port->bundle && port->bundle->bond) {
1586 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1591 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1593 struct ofport_dpif *port = ofport_dpif_cast(port_);
1594 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1595 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1597 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1598 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1599 OFPUTIL_PC_NO_PACKET_IN)) {
1600 ofproto->need_revalidate = REV_RECONFIGURE;
1602 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1603 bundle_update(port->bundle);
1609 set_sflow(struct ofproto *ofproto_,
1610 const struct ofproto_sflow_options *sflow_options)
1612 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1613 struct dpif_sflow *ds = ofproto->sflow;
1615 if (sflow_options) {
1617 struct ofport_dpif *ofport;
1619 ds = ofproto->sflow = dpif_sflow_create();
1620 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1621 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1623 ofproto->need_revalidate = REV_RECONFIGURE;
1625 dpif_sflow_set_options(ds, sflow_options);
1628 dpif_sflow_destroy(ds);
1629 ofproto->need_revalidate = REV_RECONFIGURE;
1630 ofproto->sflow = NULL;
1637 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1639 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1646 struct ofproto_dpif *ofproto;
1648 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1649 ofproto->need_revalidate = REV_RECONFIGURE;
1650 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1653 if (cfm_configure(ofport->cfm, s)) {
1659 cfm_destroy(ofport->cfm);
1665 get_cfm_fault(const struct ofport *ofport_)
1667 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1669 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1673 get_cfm_opup(const struct ofport *ofport_)
1675 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1677 return ofport->cfm ? cfm_get_opup(ofport->cfm) : -1;
1681 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1684 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1687 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1695 get_cfm_health(const struct ofport *ofport_)
1697 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1699 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1702 /* Spanning Tree. */
1705 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1707 struct ofproto_dpif *ofproto = ofproto_;
1708 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1709 struct ofport_dpif *ofport;
1711 ofport = stp_port_get_aux(sp);
1713 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1714 ofproto->up.name, port_num);
1716 struct eth_header *eth = pkt->l2;
1718 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1719 if (eth_addr_is_zero(eth->eth_src)) {
1720 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1721 "with unknown MAC", ofproto->up.name, port_num);
1723 send_packet(ofport, pkt);
1729 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1731 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1733 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1735 /* Only revalidate flows if the configuration changed. */
1736 if (!s != !ofproto->stp) {
1737 ofproto->need_revalidate = REV_RECONFIGURE;
1741 if (!ofproto->stp) {
1742 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1743 send_bpdu_cb, ofproto);
1744 ofproto->stp_last_tick = time_msec();
1747 stp_set_bridge_id(ofproto->stp, s->system_id);
1748 stp_set_bridge_priority(ofproto->stp, s->priority);
1749 stp_set_hello_time(ofproto->stp, s->hello_time);
1750 stp_set_max_age(ofproto->stp, s->max_age);
1751 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1753 struct ofport *ofport;
1755 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1756 set_stp_port(ofport, NULL);
1759 stp_destroy(ofproto->stp);
1760 ofproto->stp = NULL;
1767 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1769 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1773 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1774 s->designated_root = stp_get_designated_root(ofproto->stp);
1775 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1784 update_stp_port_state(struct ofport_dpif *ofport)
1786 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1787 enum stp_state state;
1789 /* Figure out new state. */
1790 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1794 if (ofport->stp_state != state) {
1795 enum ofputil_port_state of_state;
1798 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1799 netdev_get_name(ofport->up.netdev),
1800 stp_state_name(ofport->stp_state),
1801 stp_state_name(state));
1802 if (stp_learn_in_state(ofport->stp_state)
1803 != stp_learn_in_state(state)) {
1804 /* xxx Learning action flows should also be flushed. */
1805 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1807 fwd_change = stp_forward_in_state(ofport->stp_state)
1808 != stp_forward_in_state(state);
1810 ofproto->need_revalidate = REV_STP;
1811 ofport->stp_state = state;
1812 ofport->stp_state_entered = time_msec();
1814 if (fwd_change && ofport->bundle) {
1815 bundle_update(ofport->bundle);
1818 /* Update the STP state bits in the OpenFlow port description. */
1819 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1820 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1821 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1822 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1823 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1825 ofproto_port_set_state(&ofport->up, of_state);
1829 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1830 * caller is responsible for assigning STP port numbers and ensuring
1831 * there are no duplicates. */
1833 set_stp_port(struct ofport *ofport_,
1834 const struct ofproto_port_stp_settings *s)
1836 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1837 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1838 struct stp_port *sp = ofport->stp_port;
1840 if (!s || !s->enable) {
1842 ofport->stp_port = NULL;
1843 stp_port_disable(sp);
1844 update_stp_port_state(ofport);
1847 } else if (sp && stp_port_no(sp) != s->port_num
1848 && ofport == stp_port_get_aux(sp)) {
1849 /* The port-id changed, so disable the old one if it's not
1850 * already in use by another port. */
1851 stp_port_disable(sp);
1854 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1855 stp_port_enable(sp);
1857 stp_port_set_aux(sp, ofport);
1858 stp_port_set_priority(sp, s->priority);
1859 stp_port_set_path_cost(sp, s->path_cost);
1861 update_stp_port_state(ofport);
1867 get_stp_port_status(struct ofport *ofport_,
1868 struct ofproto_port_stp_status *s)
1870 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1871 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1872 struct stp_port *sp = ofport->stp_port;
1874 if (!ofproto->stp || !sp) {
1880 s->port_id = stp_port_get_id(sp);
1881 s->state = stp_port_get_state(sp);
1882 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1883 s->role = stp_port_get_role(sp);
1884 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1890 stp_run(struct ofproto_dpif *ofproto)
1893 long long int now = time_msec();
1894 long long int elapsed = now - ofproto->stp_last_tick;
1895 struct stp_port *sp;
1898 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1899 ofproto->stp_last_tick = now;
1901 while (stp_get_changed_port(ofproto->stp, &sp)) {
1902 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1905 update_stp_port_state(ofport);
1909 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1910 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1916 stp_wait(struct ofproto_dpif *ofproto)
1919 poll_timer_wait(1000);
1923 /* Returns true if STP should process 'flow'. */
1925 stp_should_process_flow(const struct flow *flow)
1927 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1931 stp_process_packet(const struct ofport_dpif *ofport,
1932 const struct ofpbuf *packet)
1934 struct ofpbuf payload = *packet;
1935 struct eth_header *eth = payload.data;
1936 struct stp_port *sp = ofport->stp_port;
1938 /* Sink packets on ports that have STP disabled when the bridge has
1940 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1944 /* Trim off padding on payload. */
1945 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1946 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1949 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1950 stp_received_bpdu(sp, payload.data, payload.size);
1954 static struct priority_to_dscp *
1955 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1957 struct priority_to_dscp *pdscp;
1960 hash = hash_int(priority, 0);
1961 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1962 if (pdscp->priority == priority) {
1970 ofport_clear_priorities(struct ofport_dpif *ofport)
1972 struct priority_to_dscp *pdscp, *next;
1974 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1975 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1981 set_queues(struct ofport *ofport_,
1982 const struct ofproto_port_queue *qdscp_list,
1985 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1986 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1987 struct hmap new = HMAP_INITIALIZER(&new);
1990 for (i = 0; i < n_qdscp; i++) {
1991 struct priority_to_dscp *pdscp;
1995 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1996 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2001 pdscp = get_priority(ofport, priority);
2003 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2005 pdscp = xmalloc(sizeof *pdscp);
2006 pdscp->priority = priority;
2008 ofproto->need_revalidate = REV_RECONFIGURE;
2011 if (pdscp->dscp != dscp) {
2013 ofproto->need_revalidate = REV_RECONFIGURE;
2016 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2019 if (!hmap_is_empty(&ofport->priorities)) {
2020 ofport_clear_priorities(ofport);
2021 ofproto->need_revalidate = REV_RECONFIGURE;
2024 hmap_swap(&new, &ofport->priorities);
2032 /* Expires all MAC learning entries associated with 'bundle' and forces its
2033 * ofproto to revalidate every flow.
2035 * Normally MAC learning entries are removed only from the ofproto associated
2036 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2037 * are removed from every ofproto. When patch ports and SLB bonds are in use
2038 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2039 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2040 * with the host from which it migrated. */
2042 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2044 struct ofproto_dpif *ofproto = bundle->ofproto;
2045 struct mac_learning *ml = ofproto->ml;
2046 struct mac_entry *mac, *next_mac;
2048 ofproto->need_revalidate = REV_RECONFIGURE;
2049 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2050 if (mac->port.p == bundle) {
2052 struct ofproto_dpif *o;
2054 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2056 struct mac_entry *e;
2058 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2061 tag_set_add(&o->revalidate_set, e->tag);
2062 mac_learning_expire(o->ml, e);
2068 mac_learning_expire(ml, mac);
2073 static struct ofbundle *
2074 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2076 struct ofbundle *bundle;
2078 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2079 &ofproto->bundles) {
2080 if (bundle->aux == aux) {
2087 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2088 * ones that are found to 'bundles'. */
2090 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2091 void **auxes, size_t n_auxes,
2092 struct hmapx *bundles)
2096 hmapx_init(bundles);
2097 for (i = 0; i < n_auxes; i++) {
2098 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2100 hmapx_add(bundles, bundle);
2106 bundle_update(struct ofbundle *bundle)
2108 struct ofport_dpif *port;
2110 bundle->floodable = true;
2111 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2112 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2113 || !stp_forward_in_state(port->stp_state)) {
2114 bundle->floodable = false;
2121 bundle_del_port(struct ofport_dpif *port)
2123 struct ofbundle *bundle = port->bundle;
2125 bundle->ofproto->need_revalidate = REV_RECONFIGURE;
2127 list_remove(&port->bundle_node);
2128 port->bundle = NULL;
2131 lacp_slave_unregister(bundle->lacp, port);
2134 bond_slave_unregister(bundle->bond, port);
2137 bundle_update(bundle);
2141 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2142 struct lacp_slave_settings *lacp,
2143 uint32_t bond_stable_id)
2145 struct ofport_dpif *port;
2147 port = get_ofp_port(bundle->ofproto, ofp_port);
2152 if (port->bundle != bundle) {
2153 bundle->ofproto->need_revalidate = REV_RECONFIGURE;
2155 bundle_del_port(port);
2158 port->bundle = bundle;
2159 list_push_back(&bundle->ports, &port->bundle_node);
2160 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2161 || !stp_forward_in_state(port->stp_state)) {
2162 bundle->floodable = false;
2166 port->bundle->ofproto->need_revalidate = REV_RECONFIGURE;
2167 lacp_slave_register(bundle->lacp, port, lacp);
2170 port->bond_stable_id = bond_stable_id;
2176 bundle_destroy(struct ofbundle *bundle)
2178 struct ofproto_dpif *ofproto;
2179 struct ofport_dpif *port, *next_port;
2186 ofproto = bundle->ofproto;
2187 for (i = 0; i < MAX_MIRRORS; i++) {
2188 struct ofmirror *m = ofproto->mirrors[i];
2190 if (m->out == bundle) {
2192 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2193 || hmapx_find_and_delete(&m->dsts, bundle)) {
2194 ofproto->need_revalidate = REV_RECONFIGURE;
2199 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2200 bundle_del_port(port);
2203 bundle_flush_macs(bundle, true);
2204 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2206 free(bundle->trunks);
2207 lacp_destroy(bundle->lacp);
2208 bond_destroy(bundle->bond);
2213 bundle_set(struct ofproto *ofproto_, void *aux,
2214 const struct ofproto_bundle_settings *s)
2216 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2217 bool need_flush = false;
2218 struct ofport_dpif *port;
2219 struct ofbundle *bundle;
2220 unsigned long *trunks;
2226 bundle_destroy(bundle_lookup(ofproto, aux));
2230 assert(s->n_slaves == 1 || s->bond != NULL);
2231 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2233 bundle = bundle_lookup(ofproto, aux);
2235 bundle = xmalloc(sizeof *bundle);
2237 bundle->ofproto = ofproto;
2238 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2239 hash_pointer(aux, 0));
2241 bundle->name = NULL;
2243 list_init(&bundle->ports);
2244 bundle->vlan_mode = PORT_VLAN_TRUNK;
2246 bundle->trunks = NULL;
2247 bundle->use_priority_tags = s->use_priority_tags;
2248 bundle->lacp = NULL;
2249 bundle->bond = NULL;
2251 bundle->floodable = true;
2253 bundle->src_mirrors = 0;
2254 bundle->dst_mirrors = 0;
2255 bundle->mirror_out = 0;
2258 if (!bundle->name || strcmp(s->name, bundle->name)) {
2260 bundle->name = xstrdup(s->name);
2265 if (!bundle->lacp) {
2266 ofproto->need_revalidate = REV_RECONFIGURE;
2267 bundle->lacp = lacp_create();
2269 lacp_configure(bundle->lacp, s->lacp);
2271 lacp_destroy(bundle->lacp);
2272 bundle->lacp = NULL;
2275 /* Update set of ports. */
2277 for (i = 0; i < s->n_slaves; i++) {
2278 if (!bundle_add_port(bundle, s->slaves[i],
2279 s->lacp ? &s->lacp_slaves[i] : NULL,
2280 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
2284 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2285 struct ofport_dpif *next_port;
2287 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2288 for (i = 0; i < s->n_slaves; i++) {
2289 if (s->slaves[i] == port->up.ofp_port) {
2294 bundle_del_port(port);
2298 assert(list_size(&bundle->ports) <= s->n_slaves);
2300 if (list_is_empty(&bundle->ports)) {
2301 bundle_destroy(bundle);
2305 /* Set VLAN tagging mode */
2306 if (s->vlan_mode != bundle->vlan_mode
2307 || s->use_priority_tags != bundle->use_priority_tags) {
2308 bundle->vlan_mode = s->vlan_mode;
2309 bundle->use_priority_tags = s->use_priority_tags;
2314 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2315 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2317 if (vlan != bundle->vlan) {
2318 bundle->vlan = vlan;
2322 /* Get trunked VLANs. */
2323 switch (s->vlan_mode) {
2324 case PORT_VLAN_ACCESS:
2328 case PORT_VLAN_TRUNK:
2329 trunks = CONST_CAST(unsigned long *, s->trunks);
2332 case PORT_VLAN_NATIVE_UNTAGGED:
2333 case PORT_VLAN_NATIVE_TAGGED:
2334 if (vlan != 0 && (!s->trunks
2335 || !bitmap_is_set(s->trunks, vlan)
2336 || bitmap_is_set(s->trunks, 0))) {
2337 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2339 trunks = bitmap_clone(s->trunks, 4096);
2341 trunks = bitmap_allocate1(4096);
2343 bitmap_set1(trunks, vlan);
2344 bitmap_set0(trunks, 0);
2346 trunks = CONST_CAST(unsigned long *, s->trunks);
2353 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2354 free(bundle->trunks);
2355 if (trunks == s->trunks) {
2356 bundle->trunks = vlan_bitmap_clone(trunks);
2358 bundle->trunks = trunks;
2363 if (trunks != s->trunks) {
2368 if (!list_is_short(&bundle->ports)) {
2369 bundle->ofproto->has_bonded_bundles = true;
2371 if (bond_reconfigure(bundle->bond, s->bond)) {
2372 ofproto->need_revalidate = REV_RECONFIGURE;
2375 bundle->bond = bond_create(s->bond);
2376 ofproto->need_revalidate = REV_RECONFIGURE;
2379 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2380 bond_slave_register(bundle->bond, port, port->bond_stable_id,
2384 bond_destroy(bundle->bond);
2385 bundle->bond = NULL;
2388 /* If we changed something that would affect MAC learning, un-learn
2389 * everything on this port and force flow revalidation. */
2391 bundle_flush_macs(bundle, false);
2398 bundle_remove(struct ofport *port_)
2400 struct ofport_dpif *port = ofport_dpif_cast(port_);
2401 struct ofbundle *bundle = port->bundle;
2404 bundle_del_port(port);
2405 if (list_is_empty(&bundle->ports)) {
2406 bundle_destroy(bundle);
2407 } else if (list_is_short(&bundle->ports)) {
2408 bond_destroy(bundle->bond);
2409 bundle->bond = NULL;
2415 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2417 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2418 struct ofport_dpif *port = port_;
2419 uint8_t ea[ETH_ADDR_LEN];
2422 error = netdev_get_etheraddr(port->up.netdev, ea);
2424 struct ofpbuf packet;
2427 ofpbuf_init(&packet, 0);
2428 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2430 memcpy(packet_pdu, pdu, pdu_size);
2432 send_packet(port, &packet);
2433 ofpbuf_uninit(&packet);
2435 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2436 "%s (%s)", port->bundle->name,
2437 netdev_get_name(port->up.netdev), strerror(error));
2442 bundle_send_learning_packets(struct ofbundle *bundle)
2444 struct ofproto_dpif *ofproto = bundle->ofproto;
2445 int error, n_packets, n_errors;
2446 struct mac_entry *e;
2448 error = n_packets = n_errors = 0;
2449 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2450 if (e->port.p != bundle) {
2451 struct ofpbuf *learning_packet;
2452 struct ofport_dpif *port;
2456 /* The assignment to "port" is unnecessary but makes "grep"ing for
2457 * struct ofport_dpif more effective. */
2458 learning_packet = bond_compose_learning_packet(bundle->bond,
2462 ret = send_packet(port, learning_packet);
2463 ofpbuf_delete(learning_packet);
2473 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2474 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2475 "packets, last error was: %s",
2476 bundle->name, n_errors, n_packets, strerror(error));
2478 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2479 bundle->name, n_packets);
2484 bundle_run(struct ofbundle *bundle)
2487 lacp_run(bundle->lacp, send_pdu_cb);
2490 struct ofport_dpif *port;
2492 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2493 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2496 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
2497 lacp_status(bundle->lacp));
2498 if (bond_should_send_learning_packets(bundle->bond)) {
2499 bundle_send_learning_packets(bundle);
2505 bundle_wait(struct ofbundle *bundle)
2508 lacp_wait(bundle->lacp);
2511 bond_wait(bundle->bond);
2518 mirror_scan(struct ofproto_dpif *ofproto)
2522 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2523 if (!ofproto->mirrors[idx]) {
2530 static struct ofmirror *
2531 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2535 for (i = 0; i < MAX_MIRRORS; i++) {
2536 struct ofmirror *mirror = ofproto->mirrors[i];
2537 if (mirror && mirror->aux == aux) {
2545 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2547 mirror_update_dups(struct ofproto_dpif *ofproto)
2551 for (i = 0; i < MAX_MIRRORS; i++) {
2552 struct ofmirror *m = ofproto->mirrors[i];
2555 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2559 for (i = 0; i < MAX_MIRRORS; i++) {
2560 struct ofmirror *m1 = ofproto->mirrors[i];
2567 for (j = i + 1; j < MAX_MIRRORS; j++) {
2568 struct ofmirror *m2 = ofproto->mirrors[j];
2570 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2571 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2572 m2->dup_mirrors |= m1->dup_mirrors;
2579 mirror_set(struct ofproto *ofproto_, void *aux,
2580 const struct ofproto_mirror_settings *s)
2582 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2583 mirror_mask_t mirror_bit;
2584 struct ofbundle *bundle;
2585 struct ofmirror *mirror;
2586 struct ofbundle *out;
2587 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2588 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2591 mirror = mirror_lookup(ofproto, aux);
2593 mirror_destroy(mirror);
2599 idx = mirror_scan(ofproto);
2601 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2603 ofproto->up.name, MAX_MIRRORS, s->name);
2607 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2608 mirror->ofproto = ofproto;
2611 mirror->out_vlan = -1;
2612 mirror->name = NULL;
2615 if (!mirror->name || strcmp(s->name, mirror->name)) {
2617 mirror->name = xstrdup(s->name);
2620 /* Get the new configuration. */
2621 if (s->out_bundle) {
2622 out = bundle_lookup(ofproto, s->out_bundle);
2624 mirror_destroy(mirror);
2630 out_vlan = s->out_vlan;
2632 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2633 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2635 /* If the configuration has not changed, do nothing. */
2636 if (hmapx_equals(&srcs, &mirror->srcs)
2637 && hmapx_equals(&dsts, &mirror->dsts)
2638 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2639 && mirror->out == out
2640 && mirror->out_vlan == out_vlan)
2642 hmapx_destroy(&srcs);
2643 hmapx_destroy(&dsts);
2647 hmapx_swap(&srcs, &mirror->srcs);
2648 hmapx_destroy(&srcs);
2650 hmapx_swap(&dsts, &mirror->dsts);
2651 hmapx_destroy(&dsts);
2653 free(mirror->vlans);
2654 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2657 mirror->out_vlan = out_vlan;
2659 /* Update bundles. */
2660 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2661 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2662 if (hmapx_contains(&mirror->srcs, bundle)) {
2663 bundle->src_mirrors |= mirror_bit;
2665 bundle->src_mirrors &= ~mirror_bit;
2668 if (hmapx_contains(&mirror->dsts, bundle)) {
2669 bundle->dst_mirrors |= mirror_bit;
2671 bundle->dst_mirrors &= ~mirror_bit;
2674 if (mirror->out == bundle) {
2675 bundle->mirror_out |= mirror_bit;
2677 bundle->mirror_out &= ~mirror_bit;
2681 ofproto->need_revalidate = REV_RECONFIGURE;
2682 ofproto->has_mirrors = true;
2683 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2684 mirror_update_dups(ofproto);
2690 mirror_destroy(struct ofmirror *mirror)
2692 struct ofproto_dpif *ofproto;
2693 mirror_mask_t mirror_bit;
2694 struct ofbundle *bundle;
2701 ofproto = mirror->ofproto;
2702 ofproto->need_revalidate = REV_RECONFIGURE;
2703 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2705 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2706 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2707 bundle->src_mirrors &= ~mirror_bit;
2708 bundle->dst_mirrors &= ~mirror_bit;
2709 bundle->mirror_out &= ~mirror_bit;
2712 hmapx_destroy(&mirror->srcs);
2713 hmapx_destroy(&mirror->dsts);
2714 free(mirror->vlans);
2716 ofproto->mirrors[mirror->idx] = NULL;
2720 mirror_update_dups(ofproto);
2722 ofproto->has_mirrors = false;
2723 for (i = 0; i < MAX_MIRRORS; i++) {
2724 if (ofproto->mirrors[i]) {
2725 ofproto->has_mirrors = true;
2732 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2733 uint64_t *packets, uint64_t *bytes)
2735 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2736 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2739 *packets = *bytes = UINT64_MAX;
2743 *packets = mirror->packet_count;
2744 *bytes = mirror->byte_count;
2750 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2752 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2753 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2754 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2760 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2762 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2763 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2764 return bundle && bundle->mirror_out != 0;
2768 forward_bpdu_changed(struct ofproto *ofproto_)
2770 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2771 ofproto->need_revalidate = REV_RECONFIGURE;
2775 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2778 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2779 mac_learning_set_idle_time(ofproto->ml, idle_time);
2780 mac_learning_set_max_entries(ofproto->ml, max_entries);
2785 static struct ofport_dpif *
2786 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2788 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2789 return ofport ? ofport_dpif_cast(ofport) : NULL;
2792 static struct ofport_dpif *
2793 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2795 return get_ofp_port(ofproto, odp_port_to_ofp_port(ofproto, odp_port));
2799 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2800 struct ofproto_port *ofproto_port,
2801 struct dpif_port *dpif_port)
2803 ofproto_port->name = dpif_port->name;
2804 ofproto_port->type = dpif_port->type;
2805 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2809 port_run_fast(struct ofport_dpif *ofport)
2811 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2812 struct ofpbuf packet;
2814 ofpbuf_init(&packet, 0);
2815 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2816 send_packet(ofport, &packet);
2817 ofpbuf_uninit(&packet);
2822 port_run(struct ofport_dpif *ofport)
2824 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2825 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2826 bool enable = netdev_get_carrier(ofport->up.netdev);
2828 ofport->carrier_seq = carrier_seq;
2830 port_run_fast(ofport);
2832 int cfm_opup = cfm_get_opup(ofport->cfm);
2834 cfm_run(ofport->cfm);
2835 enable = enable && !cfm_get_fault(ofport->cfm);
2837 if (cfm_opup >= 0) {
2838 enable = enable && cfm_opup;
2842 if (ofport->bundle) {
2843 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2844 if (carrier_changed) {
2845 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2849 if (ofport->may_enable != enable) {
2850 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2852 if (ofproto->has_bundle_action) {
2853 ofproto->need_revalidate = REV_PORT_TOGGLED;
2857 ofport->may_enable = enable;
2861 port_wait(struct ofport_dpif *ofport)
2864 cfm_wait(ofport->cfm);
2869 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2870 struct ofproto_port *ofproto_port)
2872 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2873 struct dpif_port dpif_port;
2876 if (!sset_contains(&ofproto->ports, devname)) {
2879 error = dpif_port_query_by_name(ofproto->backer->dpif,
2880 devname, &dpif_port);
2882 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2888 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2890 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2891 uint32_t odp_port = UINT32_MAX;
2894 error = dpif_port_add(ofproto->backer->dpif, netdev, &odp_port);
2896 sset_add(&ofproto->ports, netdev_get_name(netdev));
2902 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2904 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2905 uint32_t odp_port = ofp_port_to_odp_port(ofproto, ofp_port);
2908 if (odp_port != OFPP_NONE) {
2909 error = dpif_port_del(ofproto->backer->dpif, odp_port);
2912 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2914 /* The caller is going to close ofport->up.netdev. If this is a
2915 * bonded port, then the bond is using that netdev, so remove it
2916 * from the bond. The client will need to reconfigure everything
2917 * after deleting ports, so then the slave will get re-added. */
2918 bundle_remove(&ofport->up);
2925 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2927 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2930 error = netdev_get_stats(ofport->up.netdev, stats);
2932 if (!error && ofport->odp_port == OVSP_LOCAL) {
2933 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2935 /* ofproto->stats.tx_packets represents packets that we created
2936 * internally and sent to some port (e.g. packets sent with
2937 * send_packet()). Account for them as if they had come from
2938 * OFPP_LOCAL and got forwarded. */
2940 if (stats->rx_packets != UINT64_MAX) {
2941 stats->rx_packets += ofproto->stats.tx_packets;
2944 if (stats->rx_bytes != UINT64_MAX) {
2945 stats->rx_bytes += ofproto->stats.tx_bytes;
2948 /* ofproto->stats.rx_packets represents packets that were received on
2949 * some port and we processed internally and dropped (e.g. STP).
2950 * Account for them as if they had been forwarded to OFPP_LOCAL. */
2952 if (stats->tx_packets != UINT64_MAX) {
2953 stats->tx_packets += ofproto->stats.rx_packets;
2956 if (stats->tx_bytes != UINT64_MAX) {
2957 stats->tx_bytes += ofproto->stats.rx_bytes;
2964 /* Account packets for LOCAL port. */
2966 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2967 size_t tx_size, size_t rx_size)
2969 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2972 ofproto->stats.rx_packets++;
2973 ofproto->stats.rx_bytes += rx_size;
2976 ofproto->stats.tx_packets++;
2977 ofproto->stats.tx_bytes += tx_size;
2981 struct port_dump_state {
2987 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
2989 struct port_dump_state *state;
2991 *statep = state = xmalloc(sizeof *state);
2998 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2999 struct ofproto_port *port)
3001 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3002 struct port_dump_state *state = state_;
3003 struct sset_node *node;
3005 while ((node = sset_at_position(&ofproto->ports, &state->bucket,
3009 error = port_query_by_name(ofproto_, node->name, port);
3010 if (error != ENODEV) {
3019 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3021 struct port_dump_state *state = state_;
3028 port_poll(const struct ofproto *ofproto_, char **devnamep)
3030 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3032 if (ofproto->port_poll_errno) {
3033 int error = ofproto->port_poll_errno;
3034 ofproto->port_poll_errno = 0;
3038 if (sset_is_empty(&ofproto->port_poll_set)) {
3042 *devnamep = sset_pop(&ofproto->port_poll_set);
3047 port_poll_wait(const struct ofproto *ofproto_)
3049 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3050 dpif_port_poll_wait(ofproto->backer->dpif);
3054 port_is_lacp_current(const struct ofport *ofport_)
3056 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3057 return (ofport->bundle && ofport->bundle->lacp
3058 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3062 /* Upcall handling. */
3064 /* Flow miss batching.
3066 * Some dpifs implement operations faster when you hand them off in a batch.
3067 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3068 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3069 * more packets, plus possibly installing the flow in the dpif.
3071 * So far we only batch the operations that affect flow setup time the most.
3072 * It's possible to batch more than that, but the benefit might be minimal. */
3074 struct hmap_node hmap_node;
3075 struct ofproto_dpif *ofproto;
3077 enum odp_key_fitness key_fitness;
3078 const struct nlattr *key;
3080 ovs_be16 initial_tci;
3081 struct list packets;
3082 enum dpif_upcall_type upcall_type;
3083 uint32_t odp_in_port;
3086 struct flow_miss_op {
3087 struct dpif_op dpif_op;
3088 struct subfacet *subfacet; /* Subfacet */
3089 void *garbage; /* Pointer to pass to free(), NULL if none. */
3090 uint64_t stub[1024 / 8]; /* Temporary buffer. */
3093 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3094 * OpenFlow controller as necessary according to their individual
3095 * configurations. */
3097 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3098 const struct flow *flow)
3100 struct ofputil_packet_in pin;
3102 pin.packet = packet->data;
3103 pin.packet_len = packet->size;
3104 pin.reason = OFPR_NO_MATCH;
3105 pin.controller_id = 0;
3110 pin.send_len = 0; /* not used for flow table misses */
3112 flow_get_metadata(flow, &pin.fmd);
3114 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3117 static enum slow_path_reason
3118 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3119 const struct ofpbuf *packet)
3121 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
3127 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3129 cfm_process_heartbeat(ofport->cfm, packet);
3132 } else if (ofport->bundle && ofport->bundle->lacp
3133 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3135 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3138 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3140 stp_process_packet(ofport, packet);
3147 static struct flow_miss *
3148 flow_miss_find(struct hmap *todo, const struct flow *flow, uint32_t hash)
3150 struct flow_miss *miss;
3152 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3153 if (flow_equal(&miss->flow, flow)) {
3161 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3162 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3163 * 'miss' is associated with a subfacet the caller must also initialize the
3164 * returned op->subfacet, and if anything needs to be freed after processing
3165 * the op, the caller must initialize op->garbage also. */
3167 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3168 struct flow_miss_op *op)
3170 if (miss->flow.vlan_tci != miss->initial_tci) {
3171 /* This packet was received on a VLAN splinter port. We
3172 * added a VLAN to the packet to make the packet resemble
3173 * the flow, but the actions were composed assuming that
3174 * the packet contained no VLAN. So, we must remove the
3175 * VLAN header from the packet before trying to execute the
3177 eth_pop_vlan(packet);
3180 op->subfacet = NULL;
3182 op->dpif_op.type = DPIF_OP_EXECUTE;
3183 op->dpif_op.u.execute.key = miss->key;
3184 op->dpif_op.u.execute.key_len = miss->key_len;
3185 op->dpif_op.u.execute.packet = packet;
3188 /* Helper for handle_flow_miss_without_facet() and
3189 * handle_flow_miss_with_facet(). */
3191 handle_flow_miss_common(struct rule_dpif *rule,
3192 struct ofpbuf *packet, const struct flow *flow)
3194 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3196 ofproto->n_matches++;
3198 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3200 * Extra-special case for fail-open mode.
3202 * We are in fail-open mode and the packet matched the fail-open
3203 * rule, but we are connected to a controller too. We should send
3204 * the packet up to the controller in the hope that it will try to
3205 * set up a flow and thereby allow us to exit fail-open.
3207 * See the top-level comment in fail-open.c for more information.
3209 send_packet_in_miss(ofproto, packet, flow);
3213 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3214 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3215 * installing a datapath flow. The answer is usually "yes" (a return value of
3216 * true). However, for short flows the cost of bookkeeping is much higher than
3217 * the benefits, so when the datapath holds a large number of flows we impose
3218 * some heuristics to decide which flows are likely to be worth tracking. */
3220 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3221 struct flow_miss *miss, uint32_t hash)
3223 if (!ofproto->governor) {
3226 n_subfacets = hmap_count(&ofproto->subfacets);
3227 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3231 ofproto->governor = governor_create(ofproto->up.name);
3234 return governor_should_install_flow(ofproto->governor, hash,
3235 list_size(&miss->packets));
3238 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3239 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3240 * increment '*n_ops'. */
3242 handle_flow_miss_without_facet(struct flow_miss *miss,
3243 struct rule_dpif *rule,
3244 struct flow_miss_op *ops, size_t *n_ops)
3246 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3247 long long int now = time_msec();
3248 struct action_xlate_ctx ctx;
3249 struct ofpbuf *packet;
3251 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3252 struct flow_miss_op *op = &ops[*n_ops];
3253 struct dpif_flow_stats stats;
3254 struct ofpbuf odp_actions;
3256 COVERAGE_INC(facet_suppress);
3258 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3260 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3261 rule_credit_stats(rule, &stats);
3263 action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci,
3265 ctx.resubmit_stats = &stats;
3266 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3269 if (odp_actions.size) {
3270 struct dpif_execute *execute = &op->dpif_op.u.execute;
3272 init_flow_miss_execute_op(miss, packet, op);
3273 execute->actions = odp_actions.data;
3274 execute->actions_len = odp_actions.size;
3275 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3279 ofpbuf_uninit(&odp_actions);
3284 /* Handles 'miss', which matches 'facet'. May add any required datapath
3285 * operations to 'ops', incrementing '*n_ops' for each new op.
3287 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3288 * This is really important only for new facets: if we just called time_msec()
3289 * here, then the new subfacet or its packets could look (occasionally) as
3290 * though it was used some time after the facet was used. That can make a
3291 * one-packet flow look like it has a nonzero duration, which looks odd in
3292 * e.g. NetFlow statistics. */
3294 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3296 struct flow_miss_op *ops, size_t *n_ops)
3298 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3299 enum subfacet_path want_path;
3300 struct subfacet *subfacet;
3301 struct ofpbuf *packet;
3303 subfacet = subfacet_create(facet, miss, now);
3305 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3306 struct flow_miss_op *op = &ops[*n_ops];
3307 struct dpif_flow_stats stats;
3308 struct ofpbuf odp_actions;
3310 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3312 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3313 if (!subfacet->actions || subfacet->slow) {
3314 subfacet_make_actions(subfacet, packet, &odp_actions);
3317 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3318 subfacet_update_stats(subfacet, &stats);
3320 if (subfacet->actions_len) {
3321 struct dpif_execute *execute = &op->dpif_op.u.execute;
3323 init_flow_miss_execute_op(miss, packet, op);
3324 op->subfacet = subfacet;
3325 if (!subfacet->slow) {
3326 execute->actions = subfacet->actions;
3327 execute->actions_len = subfacet->actions_len;
3328 ofpbuf_uninit(&odp_actions);
3330 execute->actions = odp_actions.data;
3331 execute->actions_len = odp_actions.size;
3332 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3337 ofpbuf_uninit(&odp_actions);
3341 want_path = subfacet_want_path(subfacet->slow);
3342 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3343 struct flow_miss_op *op = &ops[(*n_ops)++];
3344 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3346 op->subfacet = subfacet;
3348 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3349 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3350 put->key = miss->key;
3351 put->key_len = miss->key_len;
3352 if (want_path == SF_FAST_PATH) {
3353 put->actions = subfacet->actions;
3354 put->actions_len = subfacet->actions_len;
3356 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
3357 op->stub, sizeof op->stub,
3358 &put->actions, &put->actions_len);
3364 /* Handles flow miss 'miss'. May add any required datapath operations
3365 * to 'ops', incrementing '*n_ops' for each new op. */
3367 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3370 struct ofproto_dpif *ofproto = miss->ofproto;
3371 struct facet *facet;
3375 /* The caller must ensure that miss->hmap_node.hash contains
3376 * flow_hash(miss->flow, 0). */
3377 hash = miss->hmap_node.hash;
3379 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3381 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
3383 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
3384 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
3388 facet = facet_create(rule, &miss->flow, hash);
3393 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3396 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3397 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3398 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3399 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3400 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3401 * 'packet' ingressed.
3403 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3404 * 'flow''s in_port to OFPP_NONE.
3406 * This function does post-processing on data returned from
3407 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3408 * of the upcall processing logic. In particular, if the extracted in_port is
3409 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3410 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3411 * a VLAN header onto 'packet' (if it is nonnull).
3413 * Optionally, if nonnull, sets '*initial_tci' to the VLAN TCI with which the
3414 * packet was really received, that is, the actual VLAN TCI extracted by
3415 * odp_flow_key_to_flow(). (This differs from the value returned in
3416 * flow->vlan_tci only for packets received on VLAN splinters.)
3418 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3419 * or some other positive errno if there are other problems. */
3421 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3422 const struct nlattr *key, size_t key_len,
3423 struct flow *flow, enum odp_key_fitness *fitnessp,
3424 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3425 ovs_be16 *initial_tci)
3427 const struct ofport_dpif *port;
3428 enum odp_key_fitness fitness;
3431 fitness = odp_flow_key_to_flow(key, key_len, flow);
3432 if (fitness == ODP_FIT_ERROR) {
3438 *initial_tci = flow->vlan_tci;
3442 *odp_in_port = flow->in_port;
3445 port = odp_port_to_ofport(backer, flow->in_port);
3447 flow->in_port = OFPP_NONE;
3448 error = ofproto ? ENODEV : 0;
3453 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3456 flow->in_port = port->up.ofp_port;
3457 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3459 /* Make the packet resemble the flow, so that it gets sent to an
3460 * OpenFlow controller properly, so that it looks correct for
3461 * sFlow, and so that flow_extract() will get the correct vlan_tci
3462 * if it is called on 'packet'.
3464 * The allocated space inside 'packet' probably also contains
3465 * 'key', that is, both 'packet' and 'key' are probably part of a
3466 * struct dpif_upcall (see the large comment on that structure
3467 * definition), so pushing data on 'packet' is in general not a
3468 * good idea since it could overwrite 'key' or free it as a side
3469 * effect. However, it's OK in this special case because we know
3470 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
3471 * will just overwrite the 4-byte "struct nlattr", which is fine
3472 * since we don't need that header anymore. */
3473 eth_push_vlan(packet, flow->vlan_tci);
3476 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
3477 if (fitness == ODP_FIT_PERFECT) {
3478 fitness = ODP_FIT_TOO_MUCH;
3485 *fitnessp = fitness;
3491 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3494 struct dpif_upcall *upcall;
3495 struct flow_miss *miss;
3496 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3497 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3498 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3508 /* Construct the to-do list.
3510 * This just amounts to extracting the flow from each packet and sticking
3511 * the packets that have the same flow in the same "flow_miss" structure so
3512 * that we can process them together. */
3515 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3516 struct flow_miss *miss = &misses[n_misses];
3517 struct flow_miss *existing_miss;
3518 struct ofproto_dpif *ofproto;
3519 uint32_t odp_in_port;
3524 error = ofproto_receive(backer, upcall->packet, upcall->key,
3525 upcall->key_len, &flow, &miss->key_fitness,
3526 &ofproto, &odp_in_port, &miss->initial_tci);
3527 if (error == ENODEV) {
3528 /* Received packet on port for which we couldn't associate
3529 * an ofproto. This can happen if a port is removed while
3530 * traffic is being received. Print a rate-limited message
3531 * in case it happens frequently. */
3532 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
3538 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3539 &flow.tunnel, flow.in_port, &miss->flow);
3541 /* Add other packets to a to-do list. */
3542 hash = flow_hash(&miss->flow, 0);
3543 existing_miss = flow_miss_find(&todo, &miss->flow, hash);
3544 if (!existing_miss) {
3545 hmap_insert(&todo, &miss->hmap_node, hash);
3546 miss->ofproto = ofproto;
3547 miss->key = upcall->key;
3548 miss->key_len = upcall->key_len;
3549 miss->upcall_type = upcall->type;
3550 miss->odp_in_port = odp_in_port;
3551 list_init(&miss->packets);
3555 miss = existing_miss;
3557 list_push_back(&miss->packets, &upcall->packet->list_node);
3560 /* Process each element in the to-do list, constructing the set of
3561 * operations to batch. */
3563 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3564 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3566 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3568 /* Execute batch. */
3569 for (i = 0; i < n_ops; i++) {
3570 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3572 dpif_operate(backer->dpif, dpif_ops, n_ops);
3574 /* Free memory and update facets. */
3575 for (i = 0; i < n_ops; i++) {
3576 struct flow_miss_op *op = &flow_miss_ops[i];
3578 switch (op->dpif_op.type) {
3579 case DPIF_OP_EXECUTE:
3582 case DPIF_OP_FLOW_PUT:
3583 if (!op->dpif_op.error) {
3584 op->subfacet->path = subfacet_want_path(op->subfacet->slow);
3588 case DPIF_OP_FLOW_DEL:
3594 hmap_destroy(&todo);
3597 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
3598 classify_upcall(const struct dpif_upcall *upcall)
3600 union user_action_cookie cookie;
3602 /* First look at the upcall type. */
3603 switch (upcall->type) {
3604 case DPIF_UC_ACTION:
3610 case DPIF_N_UC_TYPES:
3612 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3616 /* "action" upcalls need a closer look. */
3617 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3618 switch (cookie.type) {
3619 case USER_ACTION_COOKIE_SFLOW:
3620 return SFLOW_UPCALL;
3622 case USER_ACTION_COOKIE_SLOW_PATH:
3625 case USER_ACTION_COOKIE_UNSPEC:
3627 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
3633 handle_sflow_upcall(struct dpif_backer *backer,
3634 const struct dpif_upcall *upcall)
3636 struct ofproto_dpif *ofproto;
3637 union user_action_cookie cookie;
3639 uint32_t odp_in_port;
3641 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3642 &flow, NULL, &ofproto, &odp_in_port, NULL)
3643 || !ofproto->sflow) {
3647 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3648 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3649 odp_in_port, &cookie);
3653 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3655 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3656 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3657 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3662 assert(max_batch <= FLOW_MISS_MAX_BATCH);
3665 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3666 struct dpif_upcall *upcall = &misses[n_misses];
3667 struct ofpbuf *buf = &miss_bufs[n_misses];
3670 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3671 sizeof miss_buf_stubs[n_misses]);
3672 error = dpif_recv(backer->dpif, upcall, buf);
3678 switch (classify_upcall(upcall)) {
3680 /* Handle it later. */
3685 handle_sflow_upcall(backer, upcall);
3695 /* Handle deferred MISS_UPCALL processing. */
3696 handle_miss_upcalls(backer, misses, n_misses);
3697 for (i = 0; i < n_misses; i++) {
3698 ofpbuf_uninit(&miss_bufs[i]);
3704 /* Flow expiration. */
3706 static int subfacet_max_idle(const struct ofproto_dpif *);
3707 static void update_stats(struct dpif_backer *);
3708 static void rule_expire(struct rule_dpif *);
3709 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
3711 /* This function is called periodically by run(). Its job is to collect
3712 * updates for the flows that have been installed into the datapath, most
3713 * importantly when they last were used, and then use that information to
3714 * expire flows that have not been used recently.
3716 * Returns the number of milliseconds after which it should be called again. */
3718 expire(struct dpif_backer *backer)
3720 struct ofproto_dpif *ofproto;
3721 int max_idle = INT32_MAX;
3723 /* Update stats for each flow in the backer. */
3724 update_stats(backer);
3726 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3727 struct rule_dpif *rule, *next_rule;
3728 struct oftable *table;
3731 if (ofproto->backer != backer) {
3735 /* Expire subfacets that have been idle too long. */
3736 dp_max_idle = subfacet_max_idle(ofproto);
3737 expire_subfacets(ofproto, dp_max_idle);
3739 max_idle = MIN(max_idle, dp_max_idle);
3741 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3743 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
3744 struct cls_cursor cursor;
3746 cls_cursor_init(&cursor, &table->cls, NULL);
3747 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
3752 /* All outstanding data in existing flows has been accounted, so it's a
3753 * good time to do bond rebalancing. */
3754 if (ofproto->has_bonded_bundles) {
3755 struct ofbundle *bundle;
3757 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3759 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
3765 return MIN(max_idle, 1000);
3768 /* Updates flow table statistics given that the datapath just reported 'stats'
3769 * as 'subfacet''s statistics. */
3771 update_subfacet_stats(struct subfacet *subfacet,
3772 const struct dpif_flow_stats *stats)
3774 struct facet *facet = subfacet->facet;
3776 if (stats->n_packets >= subfacet->dp_packet_count) {
3777 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
3778 facet->packet_count += extra;
3780 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3783 if (stats->n_bytes >= subfacet->dp_byte_count) {
3784 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
3786 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3789 subfacet->dp_packet_count = stats->n_packets;
3790 subfacet->dp_byte_count = stats->n_bytes;
3792 facet->tcp_flags |= stats->tcp_flags;
3794 subfacet_update_time(subfacet, stats->used);
3795 if (facet->accounted_bytes < facet->byte_count) {
3797 facet_account(facet);
3798 facet->accounted_bytes = facet->byte_count;
3800 facet_push_stats(facet);
3803 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3804 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3806 delete_unexpected_flow(struct ofproto_dpif *ofproto,
3807 const struct nlattr *key, size_t key_len)
3809 if (!VLOG_DROP_WARN(&rl)) {
3813 odp_flow_key_format(key, key_len, &s);
3814 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
3818 COVERAGE_INC(facet_unexpected);
3819 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
3822 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3824 * This function also pushes statistics updates to rules which each facet
3825 * resubmits into. Generally these statistics will be accurate. However, if a
3826 * facet changes the rule it resubmits into at some time in between
3827 * update_stats() runs, it is possible that statistics accrued to the
3828 * old rule will be incorrectly attributed to the new rule. This could be
3829 * avoided by calling update_stats() whenever rules are created or
3830 * deleted. However, the performance impact of making so many calls to the
3831 * datapath do not justify the benefit of having perfectly accurate statistics.
3834 update_stats(struct dpif_backer *backer)
3836 const struct dpif_flow_stats *stats;
3837 struct dpif_flow_dump dump;
3838 const struct nlattr *key;
3841 dpif_flow_dump_start(&dump, backer->dpif);
3842 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
3844 struct subfacet *subfacet;
3845 enum odp_key_fitness fitness;
3846 struct ofproto_dpif *ofproto;
3847 struct ofport_dpif *port;
3850 fitness = odp_flow_key_to_flow(key, key_len, &flow);
3851 if (fitness == ODP_FIT_ERROR) {
3855 port = odp_port_to_ofport(backer, flow.in_port);
3857 /* This flow is for a port for which we couldn't associate an
3858 * ofproto. This can happen if a port is removed while
3859 * traffic is being received. Ignore this flow, since it
3860 * will get timed out. */
3864 ofproto = ofproto_dpif_cast(port->up.ofproto);
3865 flow.in_port = port->up.ofp_port;
3866 key_hash = odp_flow_key_hash(key, key_len);
3868 subfacet = subfacet_find(ofproto, key, key_len, key_hash, &flow);
3869 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3871 update_subfacet_stats(subfacet, stats);
3875 /* Stats are updated per-packet. */
3878 case SF_NOT_INSTALLED:
3880 delete_unexpected_flow(ofproto, key, key_len);
3884 dpif_flow_dump_done(&dump);
3887 /* Calculates and returns the number of milliseconds of idle time after which
3888 * subfacets should expire from the datapath. When a subfacet expires, we fold
3889 * its statistics into its facet, and when a facet's last subfacet expires, we
3890 * fold its statistic into its rule. */
3892 subfacet_max_idle(const struct ofproto_dpif *ofproto)
3895 * Idle time histogram.
3897 * Most of the time a switch has a relatively small number of subfacets.
3898 * When this is the case we might as well keep statistics for all of them
3899 * in userspace and to cache them in the kernel datapath for performance as
3902 * As the number of subfacets increases, the memory required to maintain
3903 * statistics about them in userspace and in the kernel becomes
3904 * significant. However, with a large number of subfacets it is likely
3905 * that only a few of them are "heavy hitters" that consume a large amount
3906 * of bandwidth. At this point, only heavy hitters are worth caching in
3907 * the kernel and maintaining in userspaces; other subfacets we can
3910 * The technique used to compute the idle time is to build a histogram with
3911 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3912 * that is installed in the kernel gets dropped in the appropriate bucket.
3913 * After the histogram has been built, we compute the cutoff so that only
3914 * the most-recently-used 1% of subfacets (but at least
3915 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3916 * the most-recently-used bucket of subfacets is kept, so actually an
3917 * arbitrary number of subfacets can be kept in any given expiration run
3918 * (though the next run will delete most of those unless they receive
3921 * This requires a second pass through the subfacets, in addition to the
3922 * pass made by update_stats(), because the former function never looks at
3923 * uninstallable subfacets.
3925 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3926 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3927 int buckets[N_BUCKETS] = { 0 };
3928 int total, subtotal, bucket;
3929 struct subfacet *subfacet;
3933 total = hmap_count(&ofproto->subfacets);
3934 if (total <= ofproto->up.flow_eviction_threshold) {
3935 return N_BUCKETS * BUCKET_WIDTH;
3938 /* Build histogram. */
3940 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3941 long long int idle = now - subfacet->used;
3942 int bucket = (idle <= 0 ? 0
3943 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3944 : (unsigned int) idle / BUCKET_WIDTH);
3948 /* Find the first bucket whose flows should be expired. */
3949 subtotal = bucket = 0;
3951 subtotal += buckets[bucket++];
3952 } while (bucket < N_BUCKETS &&
3953 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3955 if (VLOG_IS_DBG_ENABLED()) {
3959 ds_put_cstr(&s, "keep");
3960 for (i = 0; i < N_BUCKETS; i++) {
3962 ds_put_cstr(&s, ", drop");
3965 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3968 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3972 return bucket * BUCKET_WIDTH;
3976 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3978 /* Cutoff time for most flows. */
3979 long long int normal_cutoff = time_msec() - dp_max_idle;
3981 /* We really want to keep flows for special protocols around, so use a more
3982 * conservative cutoff. */
3983 long long int special_cutoff = time_msec() - 10000;
3985 struct subfacet *subfacet, *next_subfacet;
3986 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3990 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3991 &ofproto->subfacets) {
3992 long long int cutoff;
3994 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
3997 if (subfacet->used < cutoff) {
3998 if (subfacet->path != SF_NOT_INSTALLED) {
3999 batch[n_batch++] = subfacet;
4000 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4001 subfacet_destroy_batch(ofproto, batch, n_batch);
4005 subfacet_destroy(subfacet);
4011 subfacet_destroy_batch(ofproto, batch, n_batch);
4015 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4016 * then delete it entirely. */
4018 rule_expire(struct rule_dpif *rule)
4020 struct facet *facet, *next_facet;
4024 if (rule->up.pending) {
4025 /* We'll have to expire it later. */
4029 /* Has 'rule' expired? */
4031 if (rule->up.hard_timeout
4032 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4033 reason = OFPRR_HARD_TIMEOUT;
4034 } else if (rule->up.idle_timeout
4035 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4036 reason = OFPRR_IDLE_TIMEOUT;
4041 COVERAGE_INC(ofproto_dpif_expired);
4043 /* Update stats. (This is a no-op if the rule expired due to an idle
4044 * timeout, because that only happens when the rule has no facets left.) */
4045 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4046 facet_remove(facet);
4049 /* Get rid of the rule. */
4050 ofproto_rule_expire(&rule->up, reason);
4055 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
4057 * The caller must already have determined that no facet with an identical
4058 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
4059 * the ofproto's classifier table.
4061 * 'hash' must be the return value of flow_hash(flow, 0).
4063 * The facet will initially have no subfacets. The caller should create (at
4064 * least) one subfacet with subfacet_create(). */
4065 static struct facet *
4066 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
4068 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4069 struct facet *facet;
4071 facet = xzalloc(sizeof *facet);
4072 facet->used = time_msec();
4073 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4074 list_push_back(&rule->facets, &facet->list_node);
4076 facet->flow = *flow;
4077 list_init(&facet->subfacets);
4078 netflow_flow_init(&facet->nf_flow);
4079 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4085 facet_free(struct facet *facet)
4090 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4091 * 'packet', which arrived on 'in_port'.
4093 * Takes ownership of 'packet'. */
4095 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4096 const struct nlattr *odp_actions, size_t actions_len,
4097 struct ofpbuf *packet)
4099 struct odputil_keybuf keybuf;
4103 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4104 odp_flow_key_from_flow(&key, flow,
4105 ofp_port_to_odp_port(ofproto, flow->in_port));
4107 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4108 odp_actions, actions_len, packet);
4110 ofpbuf_delete(packet);
4114 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4116 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4117 * rule's statistics, via subfacet_uninstall().
4119 * - Removes 'facet' from its rule and from ofproto->facets.
4122 facet_remove(struct facet *facet)
4124 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4125 struct subfacet *subfacet, *next_subfacet;
4127 assert(!list_is_empty(&facet->subfacets));
4129 /* First uninstall all of the subfacets to get final statistics. */
4130 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4131 subfacet_uninstall(subfacet);
4134 /* Flush the final stats to the rule.
4136 * This might require us to have at least one subfacet around so that we
4137 * can use its actions for accounting in facet_account(), which is why we
4138 * have uninstalled but not yet destroyed the subfacets. */
4139 facet_flush_stats(facet);
4141 /* Now we're really all done so destroy everything. */
4142 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4143 &facet->subfacets) {
4144 subfacet_destroy__(subfacet);
4146 hmap_remove(&ofproto->facets, &facet->hmap_node);
4147 list_remove(&facet->list_node);
4151 /* Feed information from 'facet' back into the learning table to keep it in
4152 * sync with what is actually flowing through the datapath. */
4154 facet_learn(struct facet *facet)
4156 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4157 struct action_xlate_ctx ctx;
4159 if (!facet->has_learn
4160 && !facet->has_normal
4161 && (!facet->has_fin_timeout
4162 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4166 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4167 facet->flow.vlan_tci,
4168 facet->rule, facet->tcp_flags, NULL);
4169 ctx.may_learn = true;
4170 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
4171 facet->rule->up.ofpacts_len);
4175 facet_account(struct facet *facet)
4177 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4178 struct subfacet *subfacet;
4179 const struct nlattr *a;
4184 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
4187 n_bytes = facet->byte_count - facet->accounted_bytes;
4189 /* This loop feeds byte counters to bond_account() for rebalancing to use
4190 * as a basis. We also need to track the actual VLAN on which the packet
4191 * is going to be sent to ensure that it matches the one passed to
4192 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4195 * We use the actions from an arbitrary subfacet because they should all
4196 * be equally valid for our purpose. */
4197 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
4198 struct subfacet, list_node);
4199 vlan_tci = facet->flow.vlan_tci;
4200 NL_ATTR_FOR_EACH_UNSAFE (a, left,
4201 subfacet->actions, subfacet->actions_len) {
4202 const struct ovs_action_push_vlan *vlan;
4203 struct ofport_dpif *port;
4205 switch (nl_attr_type(a)) {
4206 case OVS_ACTION_ATTR_OUTPUT:
4207 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4208 if (port && port->bundle && port->bundle->bond) {
4209 bond_account(port->bundle->bond, &facet->flow,
4210 vlan_tci_to_vid(vlan_tci), n_bytes);
4214 case OVS_ACTION_ATTR_POP_VLAN:
4215 vlan_tci = htons(0);
4218 case OVS_ACTION_ATTR_PUSH_VLAN:
4219 vlan = nl_attr_get(a);
4220 vlan_tci = vlan->vlan_tci;
4226 /* Returns true if the only action for 'facet' is to send to the controller.
4227 * (We don't report NetFlow expiration messages for such facets because they
4228 * are just part of the control logic for the network, not real traffic). */
4230 facet_is_controller_flow(struct facet *facet)
4233 const struct rule *rule = &facet->rule->up;
4234 const struct ofpact *ofpacts = rule->ofpacts;
4235 size_t ofpacts_len = rule->ofpacts_len;
4237 if (ofpacts_len > 0 &&
4238 ofpacts->type == OFPACT_CONTROLLER &&
4239 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4246 /* Folds all of 'facet''s statistics into its rule. Also updates the
4247 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4248 * 'facet''s statistics in the datapath should have been zeroed and folded into
4249 * its packet and byte counts before this function is called. */
4251 facet_flush_stats(struct facet *facet)
4253 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4254 struct subfacet *subfacet;
4256 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4257 assert(!subfacet->dp_byte_count);
4258 assert(!subfacet->dp_packet_count);
4261 facet_push_stats(facet);
4262 if (facet->accounted_bytes < facet->byte_count) {
4263 facet_account(facet);
4264 facet->accounted_bytes = facet->byte_count;
4267 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4268 struct ofexpired expired;
4269 expired.flow = facet->flow;
4270 expired.packet_count = facet->packet_count;
4271 expired.byte_count = facet->byte_count;
4272 expired.used = facet->used;
4273 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4276 facet->rule->packet_count += facet->packet_count;
4277 facet->rule->byte_count += facet->byte_count;
4279 /* Reset counters to prevent double counting if 'facet' ever gets
4281 facet_reset_counters(facet);
4283 netflow_flow_clear(&facet->nf_flow);
4284 facet->tcp_flags = 0;
4287 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4288 * Returns it if found, otherwise a null pointer.
4290 * 'hash' must be the return value of flow_hash(flow, 0).
4292 * The returned facet might need revalidation; use facet_lookup_valid()
4293 * instead if that is important. */
4294 static struct facet *
4295 facet_find(struct ofproto_dpif *ofproto,
4296 const struct flow *flow, uint32_t hash)
4298 struct facet *facet;
4300 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4301 if (flow_equal(flow, &facet->flow)) {
4309 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4310 * Returns it if found, otherwise a null pointer.
4312 * 'hash' must be the return value of flow_hash(flow, 0).
4314 * The returned facet is guaranteed to be valid. */
4315 static struct facet *
4316 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4319 struct facet *facet;
4321 facet = facet_find(ofproto, flow, hash);
4323 && (ofproto->need_revalidate
4324 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))) {
4325 facet_revalidate(facet);
4332 subfacet_path_to_string(enum subfacet_path path)
4335 case SF_NOT_INSTALLED:
4336 return "not installed";
4338 return "in fast path";
4340 return "in slow path";
4346 /* Returns the path in which a subfacet should be installed if its 'slow'
4347 * member has the specified value. */
4348 static enum subfacet_path
4349 subfacet_want_path(enum slow_path_reason slow)
4351 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
4354 /* Returns true if 'subfacet' needs to have its datapath flow updated,
4355 * supposing that its actions have been recalculated as 'want_actions' and that
4356 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
4358 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
4359 const struct ofpbuf *want_actions)
4361 enum subfacet_path want_path = subfacet_want_path(slow);
4362 return (want_path != subfacet->path
4363 || (want_path == SF_FAST_PATH
4364 && (subfacet->actions_len != want_actions->size
4365 || memcmp(subfacet->actions, want_actions->data,
4366 subfacet->actions_len))));
4370 facet_check_consistency(struct facet *facet)
4372 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4374 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4376 uint64_t odp_actions_stub[1024 / 8];
4377 struct ofpbuf odp_actions;
4379 struct rule_dpif *rule;
4380 struct subfacet *subfacet;
4381 bool may_log = false;
4384 /* Check the rule for consistency. */
4385 rule = rule_dpif_lookup(ofproto, &facet->flow);
4386 ok = rule == facet->rule;
4388 may_log = !VLOG_DROP_WARN(&rl);
4393 flow_format(&s, &facet->flow);
4394 ds_put_format(&s, ": facet associated with wrong rule (was "
4395 "table=%"PRIu8",", facet->rule->up.table_id);
4396 cls_rule_format(&facet->rule->up.cr, &s);
4397 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4399 cls_rule_format(&rule->up.cr, &s);
4400 ds_put_char(&s, ')');
4402 VLOG_WARN("%s", ds_cstr(&s));
4407 /* Check the datapath actions for consistency. */
4408 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4409 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4410 enum subfacet_path want_path;
4411 struct odputil_keybuf keybuf;
4412 struct action_xlate_ctx ctx;
4416 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4417 subfacet->initial_tci, rule, 0, NULL);
4418 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
4421 if (subfacet->path == SF_NOT_INSTALLED) {
4422 /* This only happens if the datapath reported an error when we
4423 * tried to install the flow. Don't flag another error here. */
4427 want_path = subfacet_want_path(subfacet->slow);
4428 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
4429 /* The actions for slow-path flows may legitimately vary from one
4430 * packet to the next. We're done. */
4434 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
4438 /* Inconsistency! */
4440 may_log = !VLOG_DROP_WARN(&rl);
4444 /* Rate-limited, skip reporting. */
4449 subfacet_get_key(subfacet, &keybuf, &key);
4450 odp_flow_key_format(key.data, key.size, &s);
4452 ds_put_cstr(&s, ": inconsistency in subfacet");
4453 if (want_path != subfacet->path) {
4454 enum odp_key_fitness fitness = subfacet->key_fitness;
4456 ds_put_format(&s, " (%s, fitness=%s)",
4457 subfacet_path_to_string(subfacet->path),
4458 odp_key_fitness_to_string(fitness));
4459 ds_put_format(&s, " (should have been %s)",
4460 subfacet_path_to_string(want_path));
4461 } else if (want_path == SF_FAST_PATH) {
4462 ds_put_cstr(&s, " (actions were: ");
4463 format_odp_actions(&s, subfacet->actions,
4464 subfacet->actions_len);
4465 ds_put_cstr(&s, ") (correct actions: ");
4466 format_odp_actions(&s, odp_actions.data, odp_actions.size);
4467 ds_put_char(&s, ')');
4469 ds_put_cstr(&s, " (actions: ");
4470 format_odp_actions(&s, subfacet->actions,
4471 subfacet->actions_len);
4472 ds_put_char(&s, ')');
4474 VLOG_WARN("%s", ds_cstr(&s));
4477 ofpbuf_uninit(&odp_actions);
4482 /* Re-searches the classifier for 'facet':
4484 * - If the rule found is different from 'facet''s current rule, moves
4485 * 'facet' to the new rule and recompiles its actions.
4487 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4488 * where it is and recompiles its actions anyway. */
4490 facet_revalidate(struct facet *facet)
4492 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4494 struct nlattr *odp_actions;
4497 struct actions *new_actions;
4499 struct action_xlate_ctx ctx;
4500 uint64_t odp_actions_stub[1024 / 8];
4501 struct ofpbuf odp_actions;
4503 struct rule_dpif *new_rule;
4504 struct subfacet *subfacet;
4507 COVERAGE_INC(facet_revalidate);
4509 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
4511 /* Calculate new datapath actions.
4513 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4514 * emit a NetFlow expiration and, if so, we need to have the old state
4515 * around to properly compose it. */
4517 /* If the datapath actions changed or the installability changed,
4518 * then we need to talk to the datapath. */
4521 memset(&ctx, 0, sizeof ctx);
4522 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4523 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4524 enum slow_path_reason slow;
4526 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4527 subfacet->initial_tci, new_rule, 0, NULL);
4528 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
4531 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4532 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
4533 struct dpif_flow_stats stats;
4535 subfacet_install(subfacet,
4536 odp_actions.data, odp_actions.size, &stats, slow);
4537 subfacet_update_stats(subfacet, &stats);
4540 new_actions = xcalloc(list_size(&facet->subfacets),
4541 sizeof *new_actions);
4543 new_actions[i].odp_actions = xmemdup(odp_actions.data,
4545 new_actions[i].actions_len = odp_actions.size;
4550 ofpbuf_uninit(&odp_actions);
4553 facet_flush_stats(facet);
4556 /* Update 'facet' now that we've taken care of all the old state. */
4557 facet->tags = ctx.tags;
4558 facet->nf_flow.output_iface = ctx.nf_output_iface;
4559 facet->has_learn = ctx.has_learn;
4560 facet->has_normal = ctx.has_normal;
4561 facet->has_fin_timeout = ctx.has_fin_timeout;
4562 facet->mirrors = ctx.mirrors;
4565 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4566 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4568 if (new_actions && new_actions[i].odp_actions) {
4569 free(subfacet->actions);
4570 subfacet->actions = new_actions[i].odp_actions;
4571 subfacet->actions_len = new_actions[i].actions_len;
4577 if (facet->rule != new_rule) {
4578 COVERAGE_INC(facet_changed_rule);
4579 list_remove(&facet->list_node);
4580 list_push_back(&new_rule->facets, &facet->list_node);
4581 facet->rule = new_rule;
4582 facet->used = new_rule->up.created;
4583 facet->prev_used = facet->used;
4587 /* Updates 'facet''s used time. Caller is responsible for calling
4588 * facet_push_stats() to update the flows which 'facet' resubmits into. */
4590 facet_update_time(struct facet *facet, long long int used)
4592 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4593 if (used > facet->used) {
4595 ofproto_rule_update_used(&facet->rule->up, used);
4596 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
4601 facet_reset_counters(struct facet *facet)
4603 facet->packet_count = 0;
4604 facet->byte_count = 0;
4605 facet->prev_packet_count = 0;
4606 facet->prev_byte_count = 0;
4607 facet->accounted_bytes = 0;
4611 facet_push_stats(struct facet *facet)
4613 struct dpif_flow_stats stats;
4615 assert(facet->packet_count >= facet->prev_packet_count);
4616 assert(facet->byte_count >= facet->prev_byte_count);
4617 assert(facet->used >= facet->prev_used);
4619 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4620 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4621 stats.used = facet->used;
4622 stats.tcp_flags = 0;
4624 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
4625 facet->prev_packet_count = facet->packet_count;
4626 facet->prev_byte_count = facet->byte_count;
4627 facet->prev_used = facet->used;
4629 flow_push_stats(facet->rule, &facet->flow, &stats);
4631 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
4632 facet->mirrors, stats.n_packets, stats.n_bytes);
4637 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4639 rule->packet_count += stats->n_packets;
4640 rule->byte_count += stats->n_bytes;
4641 ofproto_rule_update_used(&rule->up, stats->used);
4644 /* Pushes flow statistics to the rules which 'flow' resubmits into given
4645 * 'rule''s actions and mirrors. */
4647 flow_push_stats(struct rule_dpif *rule,
4648 const struct flow *flow, const struct dpif_flow_stats *stats)
4650 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4651 struct action_xlate_ctx ctx;
4653 ofproto_rule_update_used(&rule->up, stats->used);
4655 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule,
4657 ctx.resubmit_stats = stats;
4658 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
4659 rule->up.ofpacts_len);
4664 static struct subfacet *
4665 subfacet_find(struct ofproto_dpif *ofproto,
4666 const struct nlattr *key, size_t key_len, uint32_t key_hash,
4667 const struct flow *flow)
4669 struct subfacet *subfacet;
4671 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4672 &ofproto->subfacets) {
4674 ? (subfacet->key_len == key_len
4675 && !memcmp(key, subfacet->key, key_len))
4676 : flow_equal(flow, &subfacet->facet->flow)) {
4684 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4685 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4686 * existing subfacet if there is one, otherwise creates and returns a
4689 * If the returned subfacet is new, then subfacet->actions will be NULL, in
4690 * which case the caller must populate the actions with
4691 * subfacet_make_actions(). */
4692 static struct subfacet *
4693 subfacet_create(struct facet *facet, struct flow_miss *miss,
4696 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4697 enum odp_key_fitness key_fitness = miss->key_fitness;
4698 const struct nlattr *key = miss->key;
4699 size_t key_len = miss->key_len;
4701 struct subfacet *subfacet;
4703 key_hash = odp_flow_key_hash(key, key_len);
4705 if (list_is_empty(&facet->subfacets)) {
4706 subfacet = &facet->one_subfacet;
4708 subfacet = subfacet_find(ofproto, key, key_len, key_hash,
4711 if (subfacet->facet == facet) {
4715 /* This shouldn't happen. */
4716 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4717 subfacet_destroy(subfacet);
4720 subfacet = xmalloc(sizeof *subfacet);
4723 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
4724 list_push_back(&facet->subfacets, &subfacet->list_node);
4725 subfacet->facet = facet;
4726 subfacet->key_fitness = key_fitness;
4727 if (key_fitness != ODP_FIT_PERFECT) {
4728 subfacet->key = xmemdup(key, key_len);
4729 subfacet->key_len = key_len;
4731 subfacet->key = NULL;
4732 subfacet->key_len = 0;
4734 subfacet->used = now;
4735 subfacet->dp_packet_count = 0;
4736 subfacet->dp_byte_count = 0;
4737 subfacet->actions_len = 0;
4738 subfacet->actions = NULL;
4739 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
4742 subfacet->path = SF_NOT_INSTALLED;
4743 subfacet->initial_tci = miss->initial_tci;
4744 subfacet->odp_in_port = miss->odp_in_port;
4749 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4750 * its facet within 'ofproto', and frees it. */
4752 subfacet_destroy__(struct subfacet *subfacet)
4754 struct facet *facet = subfacet->facet;
4755 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4757 subfacet_uninstall(subfacet);
4758 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
4759 list_remove(&subfacet->list_node);
4760 free(subfacet->key);
4761 free(subfacet->actions);
4762 if (subfacet != &facet->one_subfacet) {
4767 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4768 * last remaining subfacet in its facet destroys the facet too. */
4770 subfacet_destroy(struct subfacet *subfacet)
4772 struct facet *facet = subfacet->facet;
4774 if (list_is_singleton(&facet->subfacets)) {
4775 /* facet_remove() needs at least one subfacet (it will remove it). */
4776 facet_remove(facet);
4778 subfacet_destroy__(subfacet);
4783 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
4784 struct subfacet **subfacets, int n)
4786 struct odputil_keybuf keybufs[SUBFACET_DESTROY_MAX_BATCH];
4787 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4788 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4789 struct ofpbuf keys[SUBFACET_DESTROY_MAX_BATCH];
4790 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4793 for (i = 0; i < n; i++) {
4794 ops[i].type = DPIF_OP_FLOW_DEL;
4795 subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]);
4796 ops[i].u.flow_del.key = keys[i].data;
4797 ops[i].u.flow_del.key_len = keys[i].size;
4798 ops[i].u.flow_del.stats = &stats[i];
4802 dpif_operate(ofproto->backer->dpif, opsp, n);
4803 for (i = 0; i < n; i++) {
4804 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4805 subfacets[i]->path = SF_NOT_INSTALLED;
4806 subfacet_destroy(subfacets[i]);
4810 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
4811 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
4812 * for use as temporary storage. */
4814 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
4818 if (!subfacet->key) {
4819 struct flow *flow = &subfacet->facet->flow;
4821 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
4822 odp_flow_key_from_flow(key, flow, subfacet->odp_in_port);
4824 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
4828 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
4829 * Translates the actions into 'odp_actions', which the caller must have
4830 * initialized and is responsible for uninitializing. */
4832 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
4833 struct ofpbuf *odp_actions)
4835 struct facet *facet = subfacet->facet;
4836 struct rule_dpif *rule = facet->rule;
4837 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4839 struct action_xlate_ctx ctx;
4841 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
4843 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
4844 facet->tags = ctx.tags;
4845 facet->has_learn = ctx.has_learn;
4846 facet->has_normal = ctx.has_normal;
4847 facet->has_fin_timeout = ctx.has_fin_timeout;
4848 facet->nf_flow.output_iface = ctx.nf_output_iface;
4849 facet->mirrors = ctx.mirrors;
4851 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4852 if (subfacet->actions_len != odp_actions->size
4853 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
4854 free(subfacet->actions);
4855 subfacet->actions_len = odp_actions->size;
4856 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
4860 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4861 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4862 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4863 * since 'subfacet' was last updated.
4865 * Returns 0 if successful, otherwise a positive errno value. */
4867 subfacet_install(struct subfacet *subfacet,
4868 const struct nlattr *actions, size_t actions_len,
4869 struct dpif_flow_stats *stats,
4870 enum slow_path_reason slow)
4872 struct facet *facet = subfacet->facet;
4873 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4874 enum subfacet_path path = subfacet_want_path(slow);
4875 uint64_t slow_path_stub[128 / 8];
4876 struct odputil_keybuf keybuf;
4877 enum dpif_flow_put_flags flags;
4881 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
4883 flags |= DPIF_FP_ZERO_STATS;
4886 if (path == SF_SLOW_PATH) {
4887 compose_slow_path(ofproto, &facet->flow, slow,
4888 slow_path_stub, sizeof slow_path_stub,
4889 &actions, &actions_len);
4892 subfacet_get_key(subfacet, &keybuf, &key);
4893 ret = dpif_flow_put(ofproto->backer->dpif, flags, key.data, key.size,
4894 actions, actions_len, stats);
4897 subfacet_reset_dp_stats(subfacet, stats);
4901 subfacet->path = path;
4907 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
4909 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
4910 stats, subfacet->slow);
4913 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4915 subfacet_uninstall(struct subfacet *subfacet)
4917 if (subfacet->path != SF_NOT_INSTALLED) {
4918 struct rule_dpif *rule = subfacet->facet->rule;
4919 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4920 struct odputil_keybuf keybuf;
4921 struct dpif_flow_stats stats;
4925 subfacet_get_key(subfacet, &keybuf, &key);
4926 error = dpif_flow_del(ofproto->backer->dpif,
4927 key.data, key.size, &stats);
4928 subfacet_reset_dp_stats(subfacet, &stats);
4930 subfacet_update_stats(subfacet, &stats);
4932 subfacet->path = SF_NOT_INSTALLED;
4934 assert(subfacet->dp_packet_count == 0);
4935 assert(subfacet->dp_byte_count == 0);
4939 /* Resets 'subfacet''s datapath statistics counters. This should be called
4940 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4941 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4942 * was reset in the datapath. 'stats' will be modified to include only
4943 * statistics new since 'subfacet' was last updated. */
4945 subfacet_reset_dp_stats(struct subfacet *subfacet,
4946 struct dpif_flow_stats *stats)
4949 && subfacet->dp_packet_count <= stats->n_packets
4950 && subfacet->dp_byte_count <= stats->n_bytes) {
4951 stats->n_packets -= subfacet->dp_packet_count;
4952 stats->n_bytes -= subfacet->dp_byte_count;
4955 subfacet->dp_packet_count = 0;
4956 subfacet->dp_byte_count = 0;
4959 /* Updates 'subfacet''s used time. The caller is responsible for calling
4960 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
4962 subfacet_update_time(struct subfacet *subfacet, long long int used)
4964 if (used > subfacet->used) {
4965 subfacet->used = used;
4966 facet_update_time(subfacet->facet, used);
4970 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4972 * Because of the meaning of a subfacet's counters, it only makes sense to do
4973 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4974 * represents a packet that was sent by hand or if it represents statistics
4975 * that have been cleared out of the datapath. */
4977 subfacet_update_stats(struct subfacet *subfacet,
4978 const struct dpif_flow_stats *stats)
4980 if (stats->n_packets || stats->used > subfacet->used) {
4981 struct facet *facet = subfacet->facet;
4983 subfacet_update_time(subfacet, stats->used);
4984 facet->packet_count += stats->n_packets;
4985 facet->byte_count += stats->n_bytes;
4986 facet->tcp_flags |= stats->tcp_flags;
4987 facet_push_stats(facet);
4988 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
4994 static struct rule_dpif *
4995 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
4997 struct rule_dpif *rule;
4999 rule = rule_dpif_lookup__(ofproto, flow, 0);
5004 return rule_dpif_miss_rule(ofproto, flow);
5007 static struct rule_dpif *
5008 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5011 struct cls_rule *cls_rule;
5012 struct classifier *cls;
5014 if (table_id >= N_TABLES) {
5018 cls = &ofproto->up.tables[table_id].cls;
5019 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5020 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5021 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5022 * are unavailable. */
5023 struct flow ofpc_normal_flow = *flow;
5024 ofpc_normal_flow.tp_src = htons(0);
5025 ofpc_normal_flow.tp_dst = htons(0);
5026 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5028 cls_rule = classifier_lookup(cls, flow);
5030 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5033 static struct rule_dpif *
5034 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5036 struct ofport_dpif *port;
5038 port = get_ofp_port(ofproto, flow->in_port);
5040 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5041 return ofproto->miss_rule;
5044 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5045 return ofproto->no_packet_in_rule;
5047 return ofproto->miss_rule;
5051 complete_operation(struct rule_dpif *rule)
5053 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5055 rule_invalidate(rule);
5057 struct dpif_completion *c = xmalloc(sizeof *c);
5058 c->op = rule->up.pending;
5059 list_push_back(&ofproto->completions, &c->list_node);
5061 ofoperation_complete(rule->up.pending, 0);
5065 static struct rule *
5068 struct rule_dpif *rule = xmalloc(sizeof *rule);
5073 rule_dealloc(struct rule *rule_)
5075 struct rule_dpif *rule = rule_dpif_cast(rule_);
5080 rule_construct(struct rule *rule_)
5082 struct rule_dpif *rule = rule_dpif_cast(rule_);
5083 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5084 struct rule_dpif *victim;
5087 rule->packet_count = 0;
5088 rule->byte_count = 0;
5090 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5091 if (victim && !list_is_empty(&victim->facets)) {
5092 struct facet *facet;
5094 rule->facets = victim->facets;
5095 list_moved(&rule->facets);
5096 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5097 /* XXX: We're only clearing our local counters here. It's possible
5098 * that quite a few packets are unaccounted for in the datapath
5099 * statistics. These will be accounted to the new rule instead of
5100 * cleared as required. This could be fixed by clearing out the
5101 * datapath statistics for this facet, but currently it doesn't
5103 facet_reset_counters(facet);
5107 /* Must avoid list_moved() in this case. */
5108 list_init(&rule->facets);
5111 table_id = rule->up.table_id;
5113 rule->tag = victim->tag;
5114 } else if (table_id == 0) {
5119 miniflow_expand(&rule->up.cr.match.flow, &flow);
5120 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5121 ofproto->tables[table_id].basis);
5124 complete_operation(rule);
5129 rule_destruct(struct rule *rule_)
5131 struct rule_dpif *rule = rule_dpif_cast(rule_);
5132 struct facet *facet, *next_facet;
5134 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5135 facet_revalidate(facet);
5138 complete_operation(rule);
5142 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5144 struct rule_dpif *rule = rule_dpif_cast(rule_);
5145 struct facet *facet;
5147 /* Start from historical data for 'rule' itself that are no longer tracked
5148 * in facets. This counts, for example, facets that have expired. */
5149 *packets = rule->packet_count;
5150 *bytes = rule->byte_count;
5152 /* Add any statistics that are tracked by facets. This includes
5153 * statistical data recently updated by ofproto_update_stats() as well as
5154 * stats for packets that were executed "by hand" via dpif_execute(). */
5155 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5156 *packets += facet->packet_count;
5157 *bytes += facet->byte_count;
5162 rule_execute(struct rule *rule_, const struct flow *flow,
5163 struct ofpbuf *packet)
5165 struct rule_dpif *rule = rule_dpif_cast(rule_);
5166 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5168 struct dpif_flow_stats stats;
5170 struct action_xlate_ctx ctx;
5171 uint64_t odp_actions_stub[1024 / 8];
5172 struct ofpbuf odp_actions;
5174 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5175 rule_credit_stats(rule, &stats);
5177 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5178 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
5179 rule, stats.tcp_flags, packet);
5180 ctx.resubmit_stats = &stats;
5181 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
5183 execute_odp_actions(ofproto, flow, odp_actions.data,
5184 odp_actions.size, packet);
5186 ofpbuf_uninit(&odp_actions);
5192 rule_modify_actions(struct rule *rule_)
5194 struct rule_dpif *rule = rule_dpif_cast(rule_);
5196 complete_operation(rule);
5199 /* Sends 'packet' out 'ofport'.
5200 * May modify 'packet'.
5201 * Returns 0 if successful, otherwise a positive errno value. */
5203 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5205 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5206 struct ofpbuf key, odp_actions;
5207 struct odputil_keybuf keybuf;
5212 flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow);
5213 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5215 if (odp_port != ofport->odp_port) {
5216 eth_pop_vlan(packet);
5217 flow.vlan_tci = htons(0);
5220 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5221 odp_flow_key_from_flow(&key, &flow,
5222 ofp_port_to_odp_port(ofproto, flow.in_port));
5224 ofpbuf_init(&odp_actions, 32);
5225 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5227 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5228 error = dpif_execute(ofproto->backer->dpif,
5230 odp_actions.data, odp_actions.size,
5232 ofpbuf_uninit(&odp_actions);
5235 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5236 ofproto->up.name, odp_port, strerror(error));
5238 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5242 /* OpenFlow to datapath action translation. */
5244 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5245 struct action_xlate_ctx *);
5246 static void xlate_normal(struct action_xlate_ctx *);
5248 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5249 * The action will state 'slow' as the reason that the action is in the slow
5250 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5251 * dump-flows" output to see why a flow is in the slow path.)
5253 * The 'stub_size' bytes in 'stub' will be used to store the action.
5254 * 'stub_size' must be large enough for the action.
5256 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5259 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5260 enum slow_path_reason slow,
5261 uint64_t *stub, size_t stub_size,
5262 const struct nlattr **actionsp, size_t *actions_lenp)
5264 union user_action_cookie cookie;
5267 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5268 cookie.slow_path.unused = 0;
5269 cookie.slow_path.reason = slow;
5271 ofpbuf_use_stack(&buf, stub, stub_size);
5272 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
5273 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT16_MAX);
5274 odp_put_userspace_action(pid, &cookie, &buf);
5276 put_userspace_action(ofproto, &buf, flow, &cookie);
5278 *actionsp = buf.data;
5279 *actions_lenp = buf.size;
5283 put_userspace_action(const struct ofproto_dpif *ofproto,
5284 struct ofpbuf *odp_actions,
5285 const struct flow *flow,
5286 const union user_action_cookie *cookie)
5290 pid = dpif_port_get_pid(ofproto->backer->dpif,
5291 ofp_port_to_odp_port(ofproto, flow->in_port));
5293 return odp_put_userspace_action(pid, cookie, odp_actions);
5297 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5298 ovs_be16 vlan_tci, uint32_t odp_port,
5299 unsigned int n_outputs, union user_action_cookie *cookie)
5303 cookie->type = USER_ACTION_COOKIE_SFLOW;
5304 cookie->sflow.vlan_tci = vlan_tci;
5306 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5307 * port information") for the interpretation of cookie->output. */
5308 switch (n_outputs) {
5310 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5311 cookie->sflow.output = 0x40000000 | 256;
5315 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5317 cookie->sflow.output = ifindex;
5322 /* 0x80000000 means "multiple output ports. */
5323 cookie->sflow.output = 0x80000000 | n_outputs;
5328 /* Compose SAMPLE action for sFlow. */
5330 compose_sflow_action(const struct ofproto_dpif *ofproto,
5331 struct ofpbuf *odp_actions,
5332 const struct flow *flow,
5335 uint32_t probability;
5336 union user_action_cookie cookie;
5337 size_t sample_offset, actions_offset;
5340 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5344 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5346 /* Number of packets out of UINT_MAX to sample. */
5347 probability = dpif_sflow_get_probability(ofproto->sflow);
5348 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5350 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5351 compose_sflow_cookie(ofproto, htons(0), odp_port,
5352 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5353 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
5355 nl_msg_end_nested(odp_actions, actions_offset);
5356 nl_msg_end_nested(odp_actions, sample_offset);
5357 return cookie_offset;
5360 /* SAMPLE action must be first action in any given list of actions.
5361 * At this point we do not have all information required to build it. So try to
5362 * build sample action as complete as possible. */
5364 add_sflow_action(struct action_xlate_ctx *ctx)
5366 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5368 &ctx->flow, OVSP_NONE);
5369 ctx->sflow_odp_port = 0;
5370 ctx->sflow_n_outputs = 0;
5373 /* Fix SAMPLE action according to data collected while composing ODP actions.
5374 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5375 * USERSPACE action's user-cookie which is required for sflow. */
5377 fix_sflow_action(struct action_xlate_ctx *ctx)
5379 const struct flow *base = &ctx->base_flow;
5380 union user_action_cookie *cookie;
5382 if (!ctx->user_cookie_offset) {
5386 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
5388 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5390 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5391 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5395 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
5398 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5399 uint32_t odp_port = ofp_port_to_odp_port(ctx->ofproto, ofp_port);
5400 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
5401 uint8_t flow_nw_tos = ctx->flow.nw_tos;
5402 struct priority_to_dscp *pdscp;
5406 xlate_report(ctx, "Nonexistent output port");
5408 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5409 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5411 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5412 xlate_report(ctx, "STP not in forwarding state, skipping output");
5416 pdscp = get_priority(ofport, ctx->flow.skb_priority);
5418 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5419 ctx->flow.nw_tos |= pdscp->dscp;
5422 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
5423 ctx->flow.vlan_tci);
5424 if (out_port != odp_port) {
5425 ctx->flow.vlan_tci = htons(0);
5427 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
5428 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
5430 ctx->sflow_odp_port = odp_port;
5431 ctx->sflow_n_outputs++;
5432 ctx->nf_output_iface = ofp_port;
5433 ctx->flow.vlan_tci = flow_vlan_tci;
5434 ctx->flow.nw_tos = flow_nw_tos;
5438 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
5440 compose_output_action__(ctx, ofp_port, true);
5444 xlate_table_action(struct action_xlate_ctx *ctx,
5445 uint16_t in_port, uint8_t table_id, bool may_packet_in)
5447 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
5448 struct ofproto_dpif *ofproto = ctx->ofproto;
5449 struct rule_dpif *rule;
5450 uint16_t old_in_port;
5451 uint8_t old_table_id;
5453 old_table_id = ctx->table_id;
5454 ctx->table_id = table_id;
5456 /* Look up a flow with 'in_port' as the input port. */
5457 old_in_port = ctx->flow.in_port;
5458 ctx->flow.in_port = in_port;
5459 rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id);
5462 if (table_id > 0 && table_id < N_TABLES) {
5463 struct table_dpif *table = &ofproto->tables[table_id];
5464 if (table->other_table) {
5465 ctx->tags |= (rule && rule->tag
5467 : rule_calculate_tag(&ctx->flow,
5468 &table->other_table->mask,
5473 /* Restore the original input port. Otherwise OFPP_NORMAL and
5474 * OFPP_IN_PORT will have surprising behavior. */
5475 ctx->flow.in_port = old_in_port;
5477 if (ctx->resubmit_hook) {
5478 ctx->resubmit_hook(ctx, rule);
5481 if (rule == NULL && may_packet_in) {
5483 * check if table configuration flags
5484 * OFPTC_TABLE_MISS_CONTROLLER, default.
5485 * OFPTC_TABLE_MISS_CONTINUE,
5486 * OFPTC_TABLE_MISS_DROP
5487 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
5489 rule = rule_dpif_miss_rule(ofproto, &ctx->flow);
5493 struct rule_dpif *old_rule = ctx->rule;
5495 if (ctx->resubmit_stats) {
5496 rule_credit_stats(rule, ctx->resubmit_stats);
5501 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
5502 ctx->rule = old_rule;
5506 ctx->table_id = old_table_id;
5508 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5510 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
5511 MAX_RESUBMIT_RECURSION);
5512 ctx->max_resubmit_trigger = true;
5517 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
5518 const struct ofpact_resubmit *resubmit)
5523 in_port = resubmit->in_port;
5524 if (in_port == OFPP_IN_PORT) {
5525 in_port = ctx->flow.in_port;
5528 table_id = resubmit->table_id;
5529 if (table_id == 255) {
5530 table_id = ctx->table_id;
5533 xlate_table_action(ctx, in_port, table_id, false);
5537 flood_packets(struct action_xlate_ctx *ctx, bool all)
5539 struct ofport_dpif *ofport;
5541 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
5542 uint16_t ofp_port = ofport->up.ofp_port;
5544 if (ofp_port == ctx->flow.in_port) {
5549 compose_output_action__(ctx, ofp_port, false);
5550 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
5551 compose_output_action(ctx, ofp_port);
5555 ctx->nf_output_iface = NF_OUT_FLOOD;
5559 execute_controller_action(struct action_xlate_ctx *ctx, int len,
5560 enum ofp_packet_in_reason reason,
5561 uint16_t controller_id)
5563 struct ofputil_packet_in pin;
5564 struct ofpbuf *packet;
5566 ctx->slow |= SLOW_CONTROLLER;
5571 packet = ofpbuf_clone(ctx->packet);
5573 if (packet->l2 && packet->l3) {
5574 struct eth_header *eh;
5576 eth_pop_vlan(packet);
5579 /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
5580 * LLC frame. Calculating the Ethernet type of these frames is more
5581 * trouble than seems appropriate for a simple assertion. */
5582 assert(ntohs(eh->eth_type) < ETH_TYPE_MIN
5583 || eh->eth_type == ctx->flow.dl_type);
5585 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
5586 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
5588 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
5589 eth_push_vlan(packet, ctx->flow.vlan_tci);
5593 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5594 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
5595 ctx->flow.nw_tos, ctx->flow.nw_ttl);
5599 if (ctx->flow.nw_proto == IPPROTO_TCP) {
5600 packet_set_tcp_port(packet, ctx->flow.tp_src,
5602 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
5603 packet_set_udp_port(packet, ctx->flow.tp_src,
5610 pin.packet = packet->data;
5611 pin.packet_len = packet->size;
5612 pin.reason = reason;
5613 pin.controller_id = controller_id;
5614 pin.table_id = ctx->table_id;
5615 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
5618 flow_get_metadata(&ctx->flow, &pin.fmd);
5620 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
5621 ofpbuf_delete(packet);
5625 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
5627 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
5628 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
5632 if (ctx->flow.nw_ttl > 1) {
5638 for (i = 0; i < ids->n_controllers; i++) {
5639 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
5643 /* Stop processing for current table. */
5649 xlate_output_action(struct action_xlate_ctx *ctx,
5650 uint16_t port, uint16_t max_len, bool may_packet_in)
5652 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
5654 ctx->nf_output_iface = NF_OUT_DROP;
5658 compose_output_action(ctx, ctx->flow.in_port);
5661 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
5667 flood_packets(ctx, false);
5670 flood_packets(ctx, true);
5672 case OFPP_CONTROLLER:
5673 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
5679 if (port != ctx->flow.in_port) {
5680 compose_output_action(ctx, port);
5682 xlate_report(ctx, "skipping output to input port");
5687 if (prev_nf_output_iface == NF_OUT_FLOOD) {
5688 ctx->nf_output_iface = NF_OUT_FLOOD;
5689 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
5690 ctx->nf_output_iface = prev_nf_output_iface;
5691 } else if (prev_nf_output_iface != NF_OUT_DROP &&
5692 ctx->nf_output_iface != NF_OUT_FLOOD) {
5693 ctx->nf_output_iface = NF_OUT_MULTI;
5698 xlate_output_reg_action(struct action_xlate_ctx *ctx,
5699 const struct ofpact_output_reg *or)
5701 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
5702 if (port <= UINT16_MAX) {
5703 xlate_output_action(ctx, port, or->max_len, false);
5708 xlate_enqueue_action(struct action_xlate_ctx *ctx,
5709 const struct ofpact_enqueue *enqueue)
5711 uint16_t ofp_port = enqueue->port;
5712 uint32_t queue_id = enqueue->queue;
5713 uint32_t flow_priority, priority;
5716 /* Translate queue to priority. */
5717 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
5718 queue_id, &priority);
5720 /* Fall back to ordinary output action. */
5721 xlate_output_action(ctx, enqueue->port, 0, false);
5725 /* Check output port. */
5726 if (ofp_port == OFPP_IN_PORT) {
5727 ofp_port = ctx->flow.in_port;
5728 } else if (ofp_port == ctx->flow.in_port) {
5732 /* Add datapath actions. */
5733 flow_priority = ctx->flow.skb_priority;
5734 ctx->flow.skb_priority = priority;
5735 compose_output_action(ctx, ofp_port);
5736 ctx->flow.skb_priority = flow_priority;
5738 /* Update NetFlow output port. */
5739 if (ctx->nf_output_iface == NF_OUT_DROP) {
5740 ctx->nf_output_iface = ofp_port;
5741 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
5742 ctx->nf_output_iface = NF_OUT_MULTI;
5747 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
5749 uint32_t skb_priority;
5751 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
5752 queue_id, &skb_priority)) {
5753 ctx->flow.skb_priority = skb_priority;
5755 /* Couldn't translate queue to a priority. Nothing to do. A warning
5756 * has already been logged. */
5760 struct xlate_reg_state {
5766 xlate_autopath(struct action_xlate_ctx *ctx,
5767 const struct ofpact_autopath *ap)
5769 uint16_t ofp_port = ap->port;
5770 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
5772 if (!port || !port->bundle) {
5773 ofp_port = OFPP_NONE;
5774 } else if (port->bundle->bond) {
5775 /* Autopath does not support VLAN hashing. */
5776 struct ofport_dpif *slave = bond_choose_output_slave(
5777 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
5779 ofp_port = slave->up.ofp_port;
5782 nxm_reg_load(&ap->dst, ofp_port, &ctx->flow);
5786 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
5788 struct ofproto_dpif *ofproto = ofproto_;
5789 struct ofport_dpif *port;
5799 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
5802 port = get_ofp_port(ofproto, ofp_port);
5803 return port ? port->may_enable : false;
5808 xlate_bundle_action(struct action_xlate_ctx *ctx,
5809 const struct ofpact_bundle *bundle)
5813 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
5814 if (bundle->dst.field) {
5815 nxm_reg_load(&bundle->dst, port, &ctx->flow);
5817 xlate_output_action(ctx, port, 0, false);
5822 xlate_learn_action(struct action_xlate_ctx *ctx,
5823 const struct ofpact_learn *learn)
5825 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
5826 struct ofputil_flow_mod fm;
5827 uint64_t ofpacts_stub[1024 / 8];
5828 struct ofpbuf ofpacts;
5831 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
5832 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
5834 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
5835 if (error && !VLOG_DROP_WARN(&rl)) {
5836 VLOG_WARN("learning action failed to modify flow table (%s)",
5837 ofperr_get_name(error));
5840 ofpbuf_uninit(&ofpacts);
5843 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
5844 * means "infinite". */
5846 reduce_timeout(uint16_t max, uint16_t *timeout)
5848 if (max && (!*timeout || *timeout > max)) {
5854 xlate_fin_timeout(struct action_xlate_ctx *ctx,
5855 const struct ofpact_fin_timeout *oft)
5857 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
5858 struct rule_dpif *rule = ctx->rule;
5860 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
5861 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
5866 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
5868 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
5869 ? OFPUTIL_PC_NO_RECV_STP
5870 : OFPUTIL_PC_NO_RECV)) {
5874 /* Only drop packets here if both forwarding and learning are
5875 * disabled. If just learning is enabled, we need to have
5876 * OFPP_NORMAL and the learning action have a look at the packet
5877 * before we can drop it. */
5878 if (!stp_forward_in_state(port->stp_state)
5879 && !stp_learn_in_state(port->stp_state)) {
5887 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
5888 struct action_xlate_ctx *ctx)
5890 const struct ofport_dpif *port;
5891 bool was_evictable = true;
5892 const struct ofpact *a;
5894 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5895 if (port && !may_receive(port, ctx)) {
5896 /* Drop this flow. */
5901 /* Don't let the rule we're working on get evicted underneath us. */
5902 was_evictable = ctx->rule->up.evictable;
5903 ctx->rule->up.evictable = false;
5905 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
5906 struct ofpact_controller *controller;
5907 const struct ofpact_metadata *metadata;
5915 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
5916 ofpact_get_OUTPUT(a)->max_len, true);
5919 case OFPACT_CONTROLLER:
5920 controller = ofpact_get_CONTROLLER(a);
5921 execute_controller_action(ctx, controller->max_len,
5923 controller->controller_id);
5926 case OFPACT_ENQUEUE:
5927 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
5930 case OFPACT_SET_VLAN_VID:
5931 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
5932 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
5936 case OFPACT_SET_VLAN_PCP:
5937 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
5938 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
5943 case OFPACT_STRIP_VLAN:
5944 ctx->flow.vlan_tci = htons(0);
5947 case OFPACT_PUSH_VLAN:
5948 /* XXX 802.1AD(QinQ) */
5949 ctx->flow.vlan_tci = htons(VLAN_CFI);
5952 case OFPACT_SET_ETH_SRC:
5953 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
5957 case OFPACT_SET_ETH_DST:
5958 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
5962 case OFPACT_SET_IPV4_SRC:
5963 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
5966 case OFPACT_SET_IPV4_DST:
5967 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
5970 case OFPACT_SET_IPV4_DSCP:
5971 /* OpenFlow 1.0 only supports IPv4. */
5972 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5973 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5974 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
5978 case OFPACT_SET_L4_SRC_PORT:
5979 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
5982 case OFPACT_SET_L4_DST_PORT:
5983 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
5986 case OFPACT_RESUBMIT:
5987 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
5990 case OFPACT_SET_TUNNEL:
5991 ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
5994 case OFPACT_SET_QUEUE:
5995 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
5998 case OFPACT_POP_QUEUE:
5999 ctx->flow.skb_priority = ctx->orig_skb_priority;
6002 case OFPACT_REG_MOVE:
6003 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
6006 case OFPACT_REG_LOAD:
6007 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
6010 case OFPACT_DEC_TTL:
6011 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6017 /* Nothing to do. */
6020 case OFPACT_MULTIPATH:
6021 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
6024 case OFPACT_AUTOPATH:
6025 xlate_autopath(ctx, ofpact_get_AUTOPATH(a));
6029 ctx->ofproto->has_bundle_action = true;
6030 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6033 case OFPACT_OUTPUT_REG:
6034 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6038 ctx->has_learn = true;
6039 if (ctx->may_learn) {
6040 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6048 case OFPACT_FIN_TIMEOUT:
6049 ctx->has_fin_timeout = true;
6050 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6053 case OFPACT_CLEAR_ACTIONS:
6055 * Nothing to do because writa-actions is not supported for now.
6056 * When writa-actions is supported, clear-actions also must
6057 * be supported at the same time.
6061 case OFPACT_WRITE_METADATA:
6062 metadata = ofpact_get_WRITE_METADATA(a);
6063 ctx->flow.metadata &= ~metadata->mask;
6064 ctx->flow.metadata |= metadata->metadata & metadata->mask;
6067 case OFPACT_GOTO_TABLE: {
6068 /* XXX remove recursion */
6069 /* It is assumed that goto-table is last action */
6070 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6071 assert(ctx->table_id < ogt->table_id);
6072 xlate_table_action(ctx, ctx->flow.in_port, ogt->table_id, true);
6079 /* We've let OFPP_NORMAL and the learning action look at the packet,
6080 * so drop it now if forwarding is disabled. */
6081 if (port && !stp_forward_in_state(port->stp_state)) {
6082 ofpbuf_clear(ctx->odp_actions);
6083 add_sflow_action(ctx);
6086 ctx->rule->up.evictable = was_evictable;
6091 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
6092 struct ofproto_dpif *ofproto, const struct flow *flow,
6093 ovs_be16 initial_tci, struct rule_dpif *rule,
6094 uint8_t tcp_flags, const struct ofpbuf *packet)
6096 ctx->ofproto = ofproto;
6098 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
6099 ctx->base_flow = ctx->flow;
6100 ctx->base_flow.vlan_tci = initial_tci;
6102 ctx->packet = packet;
6103 ctx->may_learn = packet != NULL;
6104 ctx->tcp_flags = tcp_flags;
6105 ctx->resubmit_hook = NULL;
6106 ctx->report_hook = NULL;
6107 ctx->resubmit_stats = NULL;
6110 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6111 * into datapath actions in 'odp_actions', using 'ctx'. */
6113 xlate_actions(struct action_xlate_ctx *ctx,
6114 const struct ofpact *ofpacts, size_t ofpacts_len,
6115 struct ofpbuf *odp_actions)
6117 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6118 * that in the future we always keep a copy of the original flow for
6119 * tracing purposes. */
6120 static bool hit_resubmit_limit;
6122 enum slow_path_reason special;
6124 COVERAGE_INC(ofproto_dpif_xlate);
6126 ofpbuf_clear(odp_actions);
6127 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
6129 ctx->odp_actions = odp_actions;
6132 ctx->has_learn = false;
6133 ctx->has_normal = false;
6134 ctx->has_fin_timeout = false;
6135 ctx->nf_output_iface = NF_OUT_DROP;
6138 ctx->max_resubmit_trigger = false;
6139 ctx->orig_skb_priority = ctx->flow.skb_priority;
6143 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
6144 /* Do this conditionally because the copy is expensive enough that it
6145 * shows up in profiles.
6147 * We keep orig_flow in 'ctx' only because I couldn't make GCC 4.4
6148 * believe that I wasn't using it without initializing it if I kept it
6149 * in a local variable. */
6150 ctx->orig_flow = ctx->flow;
6153 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6154 switch (ctx->ofproto->up.frag_handling) {
6155 case OFPC_FRAG_NORMAL:
6156 /* We must pretend that transport ports are unavailable. */
6157 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
6158 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
6161 case OFPC_FRAG_DROP:
6164 case OFPC_FRAG_REASM:
6167 case OFPC_FRAG_NX_MATCH:
6168 /* Nothing to do. */
6171 case OFPC_INVALID_TTL_TO_CONTROLLER:
6176 special = process_special(ctx->ofproto, &ctx->flow, ctx->packet);
6178 ctx->slow |= special;
6180 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6181 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
6183 add_sflow_action(ctx);
6184 do_xlate_actions(ofpacts, ofpacts_len, ctx);
6186 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
6187 if (!hit_resubmit_limit) {
6188 /* We didn't record the original flow. Make sure we do from
6190 hit_resubmit_limit = true;
6191 } else if (!VLOG_DROP_ERR(&trace_rl)) {
6192 struct ds ds = DS_EMPTY_INITIALIZER;
6194 ofproto_trace(ctx->ofproto, &ctx->orig_flow, ctx->packet,
6196 VLOG_ERR("Trace triggered by excessive resubmit "
6197 "recursion:\n%s", ds_cstr(&ds));
6202 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
6203 ctx->odp_actions->data,
6204 ctx->odp_actions->size)) {
6205 ctx->slow |= SLOW_IN_BAND;
6207 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
6209 compose_output_action(ctx, OFPP_LOCAL);
6212 if (ctx->ofproto->has_mirrors) {
6213 add_mirror_actions(ctx, &ctx->orig_flow);
6215 fix_sflow_action(ctx);
6219 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
6220 * into datapath actions, using 'ctx', and discards the datapath actions. */
6222 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
6223 const struct ofpact *ofpacts,
6226 uint64_t odp_actions_stub[1024 / 8];
6227 struct ofpbuf odp_actions;
6229 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
6230 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
6231 ofpbuf_uninit(&odp_actions);
6235 xlate_report(struct action_xlate_ctx *ctx, const char *s)
6237 if (ctx->report_hook) {
6238 ctx->report_hook(ctx, s);
6242 /* OFPP_NORMAL implementation. */
6244 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
6246 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
6247 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
6248 * the bundle on which the packet was received, returns the VLAN to which the
6251 * Both 'vid' and the return value are in the range 0...4095. */
6253 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
6255 switch (in_bundle->vlan_mode) {
6256 case PORT_VLAN_ACCESS:
6257 return in_bundle->vlan;
6260 case PORT_VLAN_TRUNK:
6263 case PORT_VLAN_NATIVE_UNTAGGED:
6264 case PORT_VLAN_NATIVE_TAGGED:
6265 return vid ? vid : in_bundle->vlan;
6272 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
6273 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
6276 * 'vid' should be the VID obtained from the 802.1Q header that was received as
6277 * part of a packet (specify 0 if there was no 802.1Q header), in the range
6280 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
6282 /* Allow any VID on the OFPP_NONE port. */
6283 if (in_bundle == &ofpp_none_bundle) {
6287 switch (in_bundle->vlan_mode) {
6288 case PORT_VLAN_ACCESS:
6291 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6292 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
6293 "packet received on port %s configured as VLAN "
6294 "%"PRIu16" access port",
6295 in_bundle->ofproto->up.name, vid,
6296 in_bundle->name, in_bundle->vlan);
6302 case PORT_VLAN_NATIVE_UNTAGGED:
6303 case PORT_VLAN_NATIVE_TAGGED:
6305 /* Port must always carry its native VLAN. */
6309 case PORT_VLAN_TRUNK:
6310 if (!ofbundle_includes_vlan(in_bundle, vid)) {
6312 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6313 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
6314 "received on port %s not configured for trunking "
6316 in_bundle->ofproto->up.name, vid,
6317 in_bundle->name, vid);
6329 /* Given 'vlan', the VLAN that a packet belongs to, and
6330 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
6331 * that should be included in the 802.1Q header. (If the return value is 0,
6332 * then the 802.1Q header should only be included in the packet if there is a
6335 * Both 'vlan' and the return value are in the range 0...4095. */
6337 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
6339 switch (out_bundle->vlan_mode) {
6340 case PORT_VLAN_ACCESS:
6343 case PORT_VLAN_TRUNK:
6344 case PORT_VLAN_NATIVE_TAGGED:
6347 case PORT_VLAN_NATIVE_UNTAGGED:
6348 return vlan == out_bundle->vlan ? 0 : vlan;
6356 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
6359 struct ofport_dpif *port;
6361 ovs_be16 tci, old_tci;
6363 vid = output_vlan_to_vid(out_bundle, vlan);
6364 if (!out_bundle->bond) {
6365 port = ofbundle_get_a_port(out_bundle);
6367 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
6370 /* No slaves enabled, so drop packet. */
6375 old_tci = ctx->flow.vlan_tci;
6377 if (tci || out_bundle->use_priority_tags) {
6378 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
6380 tci |= htons(VLAN_CFI);
6383 ctx->flow.vlan_tci = tci;
6385 compose_output_action(ctx, port->up.ofp_port);
6386 ctx->flow.vlan_tci = old_tci;
6390 mirror_mask_ffs(mirror_mask_t mask)
6392 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
6397 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
6399 return (bundle->vlan_mode != PORT_VLAN_ACCESS
6400 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
6404 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
6406 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
6409 /* Returns an arbitrary interface within 'bundle'. */
6410 static struct ofport_dpif *
6411 ofbundle_get_a_port(const struct ofbundle *bundle)
6413 return CONTAINER_OF(list_front(&bundle->ports),
6414 struct ofport_dpif, bundle_node);
6418 vlan_is_mirrored(const struct ofmirror *m, int vlan)
6420 return !m->vlans || bitmap_is_set(m->vlans, vlan);
6424 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
6426 struct ofproto_dpif *ofproto = ctx->ofproto;
6427 mirror_mask_t mirrors;
6428 struct ofbundle *in_bundle;
6431 const struct nlattr *a;
6434 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
6435 ctx->packet != NULL, NULL);
6439 mirrors = in_bundle->src_mirrors;
6441 /* Drop frames on bundles reserved for mirroring. */
6442 if (in_bundle->mirror_out) {
6443 if (ctx->packet != NULL) {
6444 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6445 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6446 "%s, which is reserved exclusively for mirroring",
6447 ctx->ofproto->up.name, in_bundle->name);
6453 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
6454 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6457 vlan = input_vid_to_vlan(in_bundle, vid);
6459 /* Look at the output ports to check for destination selections. */
6461 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
6462 ctx->odp_actions->size) {
6463 enum ovs_action_attr type = nl_attr_type(a);
6464 struct ofport_dpif *ofport;
6466 if (type != OVS_ACTION_ATTR_OUTPUT) {
6470 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
6471 if (ofport && ofport->bundle) {
6472 mirrors |= ofport->bundle->dst_mirrors;
6480 /* Restore the original packet before adding the mirror actions. */
6481 ctx->flow = *orig_flow;
6486 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6488 if (!vlan_is_mirrored(m, vlan)) {
6489 mirrors = zero_rightmost_1bit(mirrors);
6493 mirrors &= ~m->dup_mirrors;
6494 ctx->mirrors |= m->dup_mirrors;
6496 output_normal(ctx, m->out, vlan);
6497 } else if (vlan != m->out_vlan
6498 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
6499 struct ofbundle *bundle;
6501 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
6502 if (ofbundle_includes_vlan(bundle, m->out_vlan)
6503 && !bundle->mirror_out) {
6504 output_normal(ctx, bundle, m->out_vlan);
6512 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
6513 uint64_t packets, uint64_t bytes)
6519 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
6522 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6525 /* In normal circumstances 'm' will not be NULL. However,
6526 * if mirrors are reconfigured, we can temporarily get out
6527 * of sync in facet_revalidate(). We could "correct" the
6528 * mirror list before reaching here, but doing that would
6529 * not properly account the traffic stats we've currently
6530 * accumulated for previous mirror configuration. */
6534 m->packet_count += packets;
6535 m->byte_count += bytes;
6539 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
6540 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
6541 * indicate this; newer upstream kernels use gratuitous ARP requests. */
6543 is_gratuitous_arp(const struct flow *flow)
6545 return (flow->dl_type == htons(ETH_TYPE_ARP)
6546 && eth_addr_is_broadcast(flow->dl_dst)
6547 && (flow->nw_proto == ARP_OP_REPLY
6548 || (flow->nw_proto == ARP_OP_REQUEST
6549 && flow->nw_src == flow->nw_dst)));
6553 update_learning_table(struct ofproto_dpif *ofproto,
6554 const struct flow *flow, int vlan,
6555 struct ofbundle *in_bundle)
6557 struct mac_entry *mac;
6559 /* Don't learn the OFPP_NONE port. */
6560 if (in_bundle == &ofpp_none_bundle) {
6564 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
6568 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
6569 if (is_gratuitous_arp(flow)) {
6570 /* We don't want to learn from gratuitous ARP packets that are
6571 * reflected back over bond slaves so we lock the learning table. */
6572 if (!in_bundle->bond) {
6573 mac_entry_set_grat_arp_lock(mac);
6574 } else if (mac_entry_is_grat_arp_locked(mac)) {
6579 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
6580 /* The log messages here could actually be useful in debugging,
6581 * so keep the rate limit relatively high. */
6582 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
6583 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
6584 "on port %s in VLAN %d",
6585 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
6586 in_bundle->name, vlan);
6588 mac->port.p = in_bundle;
6589 tag_set_add(&ofproto->revalidate_set,
6590 mac_learning_changed(ofproto->ml, mac));
6594 static struct ofbundle *
6595 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
6596 bool warn, struct ofport_dpif **in_ofportp)
6598 struct ofport_dpif *ofport;
6600 /* Find the port and bundle for the received packet. */
6601 ofport = get_ofp_port(ofproto, in_port);
6603 *in_ofportp = ofport;
6605 if (ofport && ofport->bundle) {
6606 return ofport->bundle;
6609 /* Special-case OFPP_NONE, which a controller may use as the ingress
6610 * port for traffic that it is sourcing. */
6611 if (in_port == OFPP_NONE) {
6612 return &ofpp_none_bundle;
6615 /* Odd. A few possible reasons here:
6617 * - We deleted a port but there are still a few packets queued up
6620 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
6621 * we don't know about.
6623 * - The ofproto client didn't configure the port as part of a bundle.
6624 * This is particularly likely to happen if a packet was received on the
6625 * port after it was created, but before the client had a chance to
6626 * configure its bundle.
6629 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6631 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
6632 "port %"PRIu16, ofproto->up.name, in_port);
6637 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
6638 * dropped. Returns true if they may be forwarded, false if they should be
6641 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
6642 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
6644 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
6645 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
6646 * checked by input_vid_is_valid().
6648 * May also add tags to '*tags', although the current implementation only does
6649 * so in one special case.
6652 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
6655 struct ofproto_dpif *ofproto = ctx->ofproto;
6656 struct flow *flow = &ctx->flow;
6657 struct ofbundle *in_bundle = in_port->bundle;
6659 /* Drop frames for reserved multicast addresses
6660 * only if forward_bpdu option is absent. */
6661 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
6662 xlate_report(ctx, "packet has reserved destination MAC, dropping");
6666 if (in_bundle->bond) {
6667 struct mac_entry *mac;
6669 switch (bond_check_admissibility(in_bundle->bond, in_port,
6670 flow->dl_dst, &ctx->tags)) {
6675 xlate_report(ctx, "bonding refused admissibility, dropping");
6678 case BV_DROP_IF_MOVED:
6679 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
6680 if (mac && mac->port.p != in_bundle &&
6681 (!is_gratuitous_arp(flow)
6682 || mac_entry_is_grat_arp_locked(mac))) {
6683 xlate_report(ctx, "SLB bond thinks this packet looped back, "
6695 xlate_normal(struct action_xlate_ctx *ctx)
6697 struct ofport_dpif *in_port;
6698 struct ofbundle *in_bundle;
6699 struct mac_entry *mac;
6703 ctx->has_normal = true;
6705 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
6706 ctx->packet != NULL, &in_port);
6708 xlate_report(ctx, "no input bundle, dropping");
6712 /* Drop malformed frames. */
6713 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
6714 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
6715 if (ctx->packet != NULL) {
6716 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6717 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
6718 "VLAN tag received on port %s",
6719 ctx->ofproto->up.name, in_bundle->name);
6721 xlate_report(ctx, "partial VLAN tag, dropping");
6725 /* Drop frames on bundles reserved for mirroring. */
6726 if (in_bundle->mirror_out) {
6727 if (ctx->packet != NULL) {
6728 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6729 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6730 "%s, which is reserved exclusively for mirroring",
6731 ctx->ofproto->up.name, in_bundle->name);
6733 xlate_report(ctx, "input port is mirror output port, dropping");
6738 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
6739 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6740 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
6743 vlan = input_vid_to_vlan(in_bundle, vid);
6745 /* Check other admissibility requirements. */
6746 if (in_port && !is_admissible(ctx, in_port, vlan)) {
6750 /* Learn source MAC. */
6751 if (ctx->may_learn) {
6752 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
6755 /* Determine output bundle. */
6756 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
6759 if (mac->port.p != in_bundle) {
6760 xlate_report(ctx, "forwarding to learned port");
6761 output_normal(ctx, mac->port.p, vlan);
6763 xlate_report(ctx, "learned port is input port, dropping");
6766 struct ofbundle *bundle;
6768 xlate_report(ctx, "no learned MAC for destination, flooding");
6769 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
6770 if (bundle != in_bundle
6771 && ofbundle_includes_vlan(bundle, vlan)
6772 && bundle->floodable
6773 && !bundle->mirror_out) {
6774 output_normal(ctx, bundle, vlan);
6777 ctx->nf_output_iface = NF_OUT_FLOOD;
6781 /* Optimized flow revalidation.
6783 * It's a difficult problem, in general, to tell which facets need to have
6784 * their actions recalculated whenever the OpenFlow flow table changes. We
6785 * don't try to solve that general problem: for most kinds of OpenFlow flow
6786 * table changes, we recalculate the actions for every facet. This is
6787 * relatively expensive, but it's good enough if the OpenFlow flow table
6788 * doesn't change very often.
6790 * However, we can expect one particular kind of OpenFlow flow table change to
6791 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
6792 * of CPU on revalidating every facet whenever MAC learning modifies the flow
6793 * table, we add a special case that applies to flow tables in which every rule
6794 * has the same form (that is, the same wildcards), except that the table is
6795 * also allowed to have a single "catch-all" flow that matches all packets. We
6796 * optimize this case by tagging all of the facets that resubmit into the table
6797 * and invalidating the same tag whenever a flow changes in that table. The
6798 * end result is that we revalidate just the facets that need it (and sometimes
6799 * a few more, but not all of the facets or even all of the facets that
6800 * resubmit to the table modified by MAC learning). */
6802 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
6803 * into an OpenFlow table with the given 'basis'. */
6805 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
6808 if (minimask_is_catchall(mask)) {
6811 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
6812 return tag_create_deterministic(hash);
6816 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
6817 * taggability of that table.
6819 * This function must be called after *each* change to a flow table. If you
6820 * skip calling it on some changes then the pointer comparisons at the end can
6821 * be invalid if you get unlucky. For example, if a flow removal causes a
6822 * cls_table to be destroyed and then a flow insertion causes a cls_table with
6823 * different wildcards to be created with the same address, then this function
6824 * will incorrectly skip revalidation. */
6826 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
6828 struct table_dpif *table = &ofproto->tables[table_id];
6829 const struct oftable *oftable = &ofproto->up.tables[table_id];
6830 struct cls_table *catchall, *other;
6831 struct cls_table *t;
6833 catchall = other = NULL;
6835 switch (hmap_count(&oftable->cls.tables)) {
6837 /* We could tag this OpenFlow table but it would make the logic a
6838 * little harder and it's a corner case that doesn't seem worth it
6844 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
6845 if (cls_table_is_catchall(t)) {
6847 } else if (!other) {
6850 /* Indicate that we can't tag this by setting both tables to
6851 * NULL. (We know that 'catchall' is already NULL.) */
6858 /* Can't tag this table. */
6862 if (table->catchall_table != catchall || table->other_table != other) {
6863 table->catchall_table = catchall;
6864 table->other_table = other;
6865 ofproto->need_revalidate = REV_FLOW_TABLE;
6869 /* Given 'rule' that has changed in some way (either it is a rule being
6870 * inserted, a rule being deleted, or a rule whose actions are being
6871 * modified), marks facets for revalidation to ensure that packets will be
6872 * forwarded correctly according to the new state of the flow table.
6874 * This function must be called after *each* change to a flow table. See
6875 * the comment on table_update_taggable() for more information. */
6877 rule_invalidate(const struct rule_dpif *rule)
6879 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
6881 table_update_taggable(ofproto, rule->up.table_id);
6883 if (!ofproto->need_revalidate) {
6884 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
6886 if (table->other_table && rule->tag) {
6887 tag_set_add(&ofproto->revalidate_set, rule->tag);
6889 ofproto->need_revalidate = REV_FLOW_TABLE;
6895 set_frag_handling(struct ofproto *ofproto_,
6896 enum ofp_config_flags frag_handling)
6898 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6900 if (frag_handling != OFPC_FRAG_REASM) {
6901 ofproto->need_revalidate = REV_RECONFIGURE;
6909 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
6910 const struct flow *flow,
6911 const struct ofpact *ofpacts, size_t ofpacts_len)
6913 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6914 struct odputil_keybuf keybuf;
6915 struct dpif_flow_stats stats;
6919 struct action_xlate_ctx ctx;
6920 uint64_t odp_actions_stub[1024 / 8];
6921 struct ofpbuf odp_actions;
6923 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
6924 odp_flow_key_from_flow(&key, flow,
6925 ofp_port_to_odp_port(ofproto, flow->in_port));
6927 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
6929 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL,
6930 packet_get_tcp_flags(packet, flow), packet);
6931 ctx.resubmit_stats = &stats;
6933 ofpbuf_use_stub(&odp_actions,
6934 odp_actions_stub, sizeof odp_actions_stub);
6935 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
6936 dpif_execute(ofproto->backer->dpif, key.data, key.size,
6937 odp_actions.data, odp_actions.size, packet);
6938 ofpbuf_uninit(&odp_actions);
6946 set_netflow(struct ofproto *ofproto_,
6947 const struct netflow_options *netflow_options)
6949 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6951 if (netflow_options) {
6952 if (!ofproto->netflow) {
6953 ofproto->netflow = netflow_create();
6955 return netflow_set_options(ofproto->netflow, netflow_options);
6957 netflow_destroy(ofproto->netflow);
6958 ofproto->netflow = NULL;
6964 get_netflow_ids(const struct ofproto *ofproto_,
6965 uint8_t *engine_type, uint8_t *engine_id)
6967 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6969 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
6973 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
6975 if (!facet_is_controller_flow(facet) &&
6976 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
6977 struct subfacet *subfacet;
6978 struct ofexpired expired;
6980 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
6981 if (subfacet->path == SF_FAST_PATH) {
6982 struct dpif_flow_stats stats;
6984 subfacet_reinstall(subfacet, &stats);
6985 subfacet_update_stats(subfacet, &stats);
6989 expired.flow = facet->flow;
6990 expired.packet_count = facet->packet_count;
6991 expired.byte_count = facet->byte_count;
6992 expired.used = facet->used;
6993 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
6998 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7000 struct facet *facet;
7002 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7003 send_active_timeout(ofproto, facet);
7007 static struct ofproto_dpif *
7008 ofproto_dpif_lookup(const char *name)
7010 struct ofproto_dpif *ofproto;
7012 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7013 hash_string(name, 0), &all_ofproto_dpifs) {
7014 if (!strcmp(ofproto->up.name, name)) {
7022 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7023 const char *argv[], void *aux OVS_UNUSED)
7025 struct ofproto_dpif *ofproto;
7028 ofproto = ofproto_dpif_lookup(argv[1]);
7030 unixctl_command_reply_error(conn, "no such bridge");
7033 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
7035 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7036 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
7040 unixctl_command_reply(conn, "table successfully flushed");
7044 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7045 const char *argv[], void *aux OVS_UNUSED)
7047 struct ds ds = DS_EMPTY_INITIALIZER;
7048 const struct ofproto_dpif *ofproto;
7049 const struct mac_entry *e;
7051 ofproto = ofproto_dpif_lookup(argv[1]);
7053 unixctl_command_reply_error(conn, "no such bridge");
7057 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7058 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7059 struct ofbundle *bundle = e->port.p;
7060 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7061 ofbundle_get_a_port(bundle)->odp_port,
7062 e->vlan, ETH_ADDR_ARGS(e->mac),
7063 mac_entry_age(ofproto->ml, e));
7065 unixctl_command_reply(conn, ds_cstr(&ds));
7070 struct action_xlate_ctx ctx;
7076 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7077 const struct rule_dpif *rule)
7079 ds_put_char_multiple(result, '\t', level);
7081 ds_put_cstr(result, "No match\n");
7085 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7086 table_id, ntohll(rule->up.flow_cookie));
7087 cls_rule_format(&rule->up.cr, result);
7088 ds_put_char(result, '\n');
7090 ds_put_char_multiple(result, '\t', level);
7091 ds_put_cstr(result, "OpenFlow ");
7092 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7093 ds_put_char(result, '\n');
7097 trace_format_flow(struct ds *result, int level, const char *title,
7098 struct trace_ctx *trace)
7100 ds_put_char_multiple(result, '\t', level);
7101 ds_put_format(result, "%s: ", title);
7102 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
7103 ds_put_cstr(result, "unchanged");
7105 flow_format(result, &trace->ctx.flow);
7106 trace->flow = trace->ctx.flow;
7108 ds_put_char(result, '\n');
7112 trace_format_regs(struct ds *result, int level, const char *title,
7113 struct trace_ctx *trace)
7117 ds_put_char_multiple(result, '\t', level);
7118 ds_put_format(result, "%s:", title);
7119 for (i = 0; i < FLOW_N_REGS; i++) {
7120 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7122 ds_put_char(result, '\n');
7126 trace_format_odp(struct ds *result, int level, const char *title,
7127 struct trace_ctx *trace)
7129 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
7131 ds_put_char_multiple(result, '\t', level);
7132 ds_put_format(result, "%s: ", title);
7133 format_odp_actions(result, odp_actions->data, odp_actions->size);
7134 ds_put_char(result, '\n');
7138 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
7140 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7141 struct ds *result = trace->result;
7143 ds_put_char(result, '\n');
7144 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7145 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
7146 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
7147 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
7151 trace_report(struct action_xlate_ctx *ctx, const char *s)
7153 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7154 struct ds *result = trace->result;
7156 ds_put_char_multiple(result, '\t', ctx->recurse);
7157 ds_put_cstr(result, s);
7158 ds_put_char(result, '\n');
7162 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
7163 void *aux OVS_UNUSED)
7165 const char *dpname = argv[1];
7166 struct ofproto_dpif *ofproto;
7167 struct ofpbuf odp_key;
7168 struct ofpbuf *packet;
7169 ovs_be16 initial_tci;
7175 ofpbuf_init(&odp_key, 0);
7178 ofproto = ofproto_dpif_lookup(dpname);
7180 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
7184 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
7185 /* ofproto/trace dpname flow [-generate] */
7186 const char *flow_s = argv[2];
7187 const char *generate_s = argv[3];
7189 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
7190 * flow. We guess which type it is based on whether 'flow_s' contains
7191 * an '(', since a datapath flow always contains '(') but an
7192 * OpenFlow-like flow should not (in fact it's allowed but I believe
7193 * that's not documented anywhere).
7195 * An alternative would be to try to parse 'flow_s' both ways, but then
7196 * it would be tricky giving a sensible error message. After all, do
7197 * you just say "syntax error" or do you present both error messages?
7198 * Both choices seem lousy. */
7199 if (strchr(flow_s, '(')) {
7202 /* Convert string to datapath key. */
7203 ofpbuf_init(&odp_key, 0);
7204 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
7206 unixctl_command_reply_error(conn, "Bad flow syntax");
7210 /* XXX: Since we allow the user to specify an ofproto, it's
7211 * possible they will specify a different ofproto than the one the
7212 * port actually belongs too. Ideally we should simply remove the
7213 * ability to specify the ofproto. */
7214 if (ofproto_receive(ofproto->backer, NULL, odp_key.data,
7215 odp_key.size, &flow, NULL, NULL, NULL,
7217 unixctl_command_reply_error(conn, "Invalid flow");
7223 error_s = parse_ofp_exact_flow(&flow, argv[2]);
7225 unixctl_command_reply_error(conn, error_s);
7230 initial_tci = flow.vlan_tci;
7233 /* Generate a packet, if requested. */
7235 packet = ofpbuf_new(0);
7236 flow_compose(packet, &flow);
7238 } else if (argc == 7) {
7239 /* ofproto/trace dpname priority tun_id in_port mark packet */
7240 const char *priority_s = argv[2];
7241 const char *tun_id_s = argv[3];
7242 const char *in_port_s = argv[4];
7243 const char *mark_s = argv[5];
7244 const char *packet_s = argv[6];
7245 uint32_t in_port = atoi(in_port_s);
7246 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
7247 uint32_t priority = atoi(priority_s);
7248 uint32_t mark = atoi(mark_s);
7251 msg = eth_from_hex(packet_s, &packet);
7253 unixctl_command_reply_error(conn, msg);
7257 ds_put_cstr(&result, "Packet: ");
7258 s = ofp_packet_to_string(packet->data, packet->size);
7259 ds_put_cstr(&result, s);
7262 flow_extract(packet, priority, mark, NULL, in_port, &flow);
7263 flow.tunnel.tun_id = tun_id;
7264 initial_tci = flow.vlan_tci;
7266 unixctl_command_reply_error(conn, "Bad command syntax");
7270 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
7271 unixctl_command_reply(conn, ds_cstr(&result));
7274 ds_destroy(&result);
7275 ofpbuf_delete(packet);
7276 ofpbuf_uninit(&odp_key);
7280 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
7281 const struct ofpbuf *packet, ovs_be16 initial_tci,
7284 struct rule_dpif *rule;
7286 ds_put_cstr(ds, "Flow: ");
7287 flow_format(ds, flow);
7288 ds_put_char(ds, '\n');
7290 rule = rule_dpif_lookup(ofproto, flow);
7292 trace_format_rule(ds, 0, 0, rule);
7293 if (rule == ofproto->miss_rule) {
7294 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
7295 } else if (rule == ofproto->no_packet_in_rule) {
7296 ds_put_cstr(ds, "\nNo match, packets dropped because "
7297 "OFPPC_NO_PACKET_IN is set on in_port.\n");
7301 uint64_t odp_actions_stub[1024 / 8];
7302 struct ofpbuf odp_actions;
7304 struct trace_ctx trace;
7307 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
7310 ofpbuf_use_stub(&odp_actions,
7311 odp_actions_stub, sizeof odp_actions_stub);
7312 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
7313 rule, tcp_flags, packet);
7314 trace.ctx.resubmit_hook = trace_resubmit;
7315 trace.ctx.report_hook = trace_report;
7316 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
7319 ds_put_char(ds, '\n');
7320 trace_format_flow(ds, 0, "Final flow", &trace);
7321 ds_put_cstr(ds, "Datapath actions: ");
7322 format_odp_actions(ds, odp_actions.data, odp_actions.size);
7323 ofpbuf_uninit(&odp_actions);
7325 if (trace.ctx.slow) {
7326 enum slow_path_reason slow;
7328 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
7329 "slow path because it:");
7330 for (slow = trace.ctx.slow; slow; ) {
7331 enum slow_path_reason bit = rightmost_1bit(slow);
7335 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
7338 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
7341 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
7344 ds_put_cstr(ds, "\n\t- Needs in-band special case "
7347 ds_put_cstr(ds, "\n\t (The datapath actions are "
7348 "incomplete--for complete actions, "
7349 "please supply a packet.)");
7352 case SLOW_CONTROLLER:
7353 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
7354 "to the OpenFlow controller.");
7357 ds_put_cstr(ds, "\n\t- Needs more specific matching "
7358 "than the datapath supports.");
7365 if (slow & ~SLOW_MATCH) {
7366 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
7367 "the special slow-path processing.");
7374 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7375 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7378 unixctl_command_reply(conn, NULL);
7382 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7383 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7386 unixctl_command_reply(conn, NULL);
7389 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
7390 * 'reply' describing the results. */
7392 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
7394 struct facet *facet;
7398 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7399 if (!facet_check_consistency(facet)) {
7404 ofproto->need_revalidate = REV_INCONSISTENCY;
7408 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
7409 ofproto->up.name, errors);
7411 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
7416 ofproto_dpif_self_check(struct unixctl_conn *conn,
7417 int argc, const char *argv[], void *aux OVS_UNUSED)
7419 struct ds reply = DS_EMPTY_INITIALIZER;
7420 struct ofproto_dpif *ofproto;
7423 ofproto = ofproto_dpif_lookup(argv[1]);
7425 unixctl_command_reply_error(conn, "Unknown ofproto (use "
7426 "ofproto/list for help)");
7429 ofproto_dpif_self_check__(ofproto, &reply);
7431 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7432 ofproto_dpif_self_check__(ofproto, &reply);
7436 unixctl_command_reply(conn, ds_cstr(&reply));
7440 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
7441 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
7442 * to destroy 'ofproto_shash' and free the returned value. */
7443 static const struct shash_node **
7444 get_ofprotos(struct shash *ofproto_shash)
7446 const struct ofproto_dpif *ofproto;
7448 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7449 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
7450 shash_add_nocopy(ofproto_shash, name, ofproto);
7453 return shash_sort(ofproto_shash);
7457 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
7458 const char *argv[] OVS_UNUSED,
7459 void *aux OVS_UNUSED)
7461 struct ds ds = DS_EMPTY_INITIALIZER;
7462 struct shash ofproto_shash;
7463 const struct shash_node **sorted_ofprotos;
7466 shash_init(&ofproto_shash);
7467 sorted_ofprotos = get_ofprotos(&ofproto_shash);
7468 for (i = 0; i < shash_count(&ofproto_shash); i++) {
7469 const struct shash_node *node = sorted_ofprotos[i];
7470 ds_put_format(&ds, "%s\n", node->name);
7473 shash_destroy(&ofproto_shash);
7474 free(sorted_ofprotos);
7476 unixctl_command_reply(conn, ds_cstr(&ds));
7481 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
7483 struct dpif_dp_stats s;
7484 const struct shash_node **ports;
7487 dpif_get_dp_stats(ofproto->backer->dpif, &s);
7489 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
7490 dpif_name(ofproto->backer->dpif));
7491 /* xxx It would be better to show bridge-specific stats instead
7492 * xxx of dp ones. */
7494 "\tlookups: hit:%"PRIu64" missed:%"PRIu64" lost:%"PRIu64"\n",
7495 s.n_hit, s.n_missed, s.n_lost);
7496 ds_put_format(ds, "\tflows: %zu\n",
7497 hmap_count(&ofproto->subfacets));
7499 ports = shash_sort(&ofproto->up.port_by_name);
7500 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
7501 const struct shash_node *node = ports[i];
7502 struct ofport *ofport = node->data;
7503 const char *name = netdev_get_name(ofport->netdev);
7504 const char *type = netdev_get_type(ofport->netdev);
7506 ds_put_format(ds, "\t%s %u/%u:", name, ofport->ofp_port,
7507 ofp_port_to_odp_port(ofproto, ofport->ofp_port));
7508 if (strcmp(type, "system")) {
7509 struct netdev *netdev;
7512 ds_put_format(ds, " (%s", type);
7514 error = netdev_open(name, type, &netdev);
7519 error = netdev_get_config(netdev, &config);
7521 const struct smap_node **nodes;
7524 nodes = smap_sort(&config);
7525 for (i = 0; i < smap_count(&config); i++) {
7526 const struct smap_node *node = nodes[i];
7527 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
7528 node->key, node->value);
7532 smap_destroy(&config);
7534 netdev_close(netdev);
7536 ds_put_char(ds, ')');
7538 ds_put_char(ds, '\n');
7544 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
7545 const char *argv[], void *aux OVS_UNUSED)
7547 struct ds ds = DS_EMPTY_INITIALIZER;
7548 const struct ofproto_dpif *ofproto;
7552 for (i = 1; i < argc; i++) {
7553 ofproto = ofproto_dpif_lookup(argv[i]);
7555 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
7556 "for help)", argv[i]);
7557 unixctl_command_reply_error(conn, ds_cstr(&ds));
7560 show_dp_format(ofproto, &ds);
7563 struct shash ofproto_shash;
7564 const struct shash_node **sorted_ofprotos;
7567 shash_init(&ofproto_shash);
7568 sorted_ofprotos = get_ofprotos(&ofproto_shash);
7569 for (i = 0; i < shash_count(&ofproto_shash); i++) {
7570 const struct shash_node *node = sorted_ofprotos[i];
7571 show_dp_format(node->data, &ds);
7574 shash_destroy(&ofproto_shash);
7575 free(sorted_ofprotos);
7578 unixctl_command_reply(conn, ds_cstr(&ds));
7583 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
7584 int argc OVS_UNUSED, const char *argv[],
7585 void *aux OVS_UNUSED)
7587 struct ds ds = DS_EMPTY_INITIALIZER;
7588 const struct ofproto_dpif *ofproto;
7589 struct subfacet *subfacet;
7591 ofproto = ofproto_dpif_lookup(argv[1]);
7593 unixctl_command_reply_error(conn, "no such bridge");
7597 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
7598 struct odputil_keybuf keybuf;
7601 subfacet_get_key(subfacet, &keybuf, &key);
7602 odp_flow_key_format(key.data, key.size, &ds);
7604 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
7605 subfacet->dp_packet_count, subfacet->dp_byte_count);
7606 if (subfacet->used) {
7607 ds_put_format(&ds, "%.3fs",
7608 (time_msec() - subfacet->used) / 1000.0);
7610 ds_put_format(&ds, "never");
7612 if (subfacet->facet->tcp_flags) {
7613 ds_put_cstr(&ds, ", flags:");
7614 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
7617 ds_put_cstr(&ds, ", actions:");
7618 format_odp_actions(&ds, subfacet->actions, subfacet->actions_len);
7619 ds_put_char(&ds, '\n');
7622 unixctl_command_reply(conn, ds_cstr(&ds));
7627 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
7628 int argc OVS_UNUSED, const char *argv[],
7629 void *aux OVS_UNUSED)
7631 struct ds ds = DS_EMPTY_INITIALIZER;
7632 struct ofproto_dpif *ofproto;
7634 ofproto = ofproto_dpif_lookup(argv[1]);
7636 unixctl_command_reply_error(conn, "no such bridge");
7640 flush(&ofproto->up);
7642 unixctl_command_reply(conn, ds_cstr(&ds));
7647 ofproto_dpif_unixctl_init(void)
7649 static bool registered;
7655 unixctl_command_register(
7657 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
7658 2, 6, ofproto_unixctl_trace, NULL);
7659 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
7660 ofproto_unixctl_fdb_flush, NULL);
7661 unixctl_command_register("fdb/show", "bridge", 1, 1,
7662 ofproto_unixctl_fdb_show, NULL);
7663 unixctl_command_register("ofproto/clog", "", 0, 0,
7664 ofproto_dpif_clog, NULL);
7665 unixctl_command_register("ofproto/unclog", "", 0, 0,
7666 ofproto_dpif_unclog, NULL);
7667 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
7668 ofproto_dpif_self_check, NULL);
7669 unixctl_command_register("dpif/dump-dps", "", 0, 0,
7670 ofproto_unixctl_dpif_dump_dps, NULL);
7671 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
7672 ofproto_unixctl_dpif_show, NULL);
7673 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
7674 ofproto_unixctl_dpif_dump_flows, NULL);
7675 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
7676 ofproto_unixctl_dpif_del_flows, NULL);
7679 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
7681 * This is deprecated. It is only for compatibility with broken device drivers
7682 * in old versions of Linux that do not properly support VLANs when VLAN
7683 * devices are not used. When broken device drivers are no longer in
7684 * widespread use, we will delete these interfaces. */
7687 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
7689 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
7690 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
7692 if (realdev_ofp_port == ofport->realdev_ofp_port
7693 && vid == ofport->vlandev_vid) {
7697 ofproto->need_revalidate = REV_RECONFIGURE;
7699 if (ofport->realdev_ofp_port) {
7702 if (realdev_ofp_port && ofport->bundle) {
7703 /* vlandevs are enslaved to their realdevs, so they are not allowed to
7704 * themselves be part of a bundle. */
7705 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
7708 ofport->realdev_ofp_port = realdev_ofp_port;
7709 ofport->vlandev_vid = vid;
7711 if (realdev_ofp_port) {
7712 vsp_add(ofport, realdev_ofp_port, vid);
7719 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
7721 return hash_2words(realdev_ofp_port, vid);
7724 /* Returns the ODP port number of the Linux VLAN device that corresponds to
7725 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
7726 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
7727 * it would return the port number of eth0.9.
7729 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
7730 * function just returns its 'realdev_odp_port' argument. */
7732 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
7733 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
7735 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
7736 uint16_t realdev_ofp_port;
7737 int vid = vlan_tci_to_vid(vlan_tci);
7738 const struct vlan_splinter *vsp;
7740 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
7741 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
7742 hash_realdev_vid(realdev_ofp_port, vid),
7743 &ofproto->realdev_vid_map) {
7744 if (vsp->realdev_ofp_port == realdev_ofp_port
7745 && vsp->vid == vid) {
7746 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
7750 return realdev_odp_port;
7753 static struct vlan_splinter *
7754 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
7756 struct vlan_splinter *vsp;
7758 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
7759 &ofproto->vlandev_map) {
7760 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
7768 /* Returns the OpenFlow port number of the "real" device underlying the Linux
7769 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
7770 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
7771 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
7772 * eth0 and store 9 in '*vid'.
7774 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
7775 * VLAN device. Unless VLAN splinters are enabled, this is what this function
7778 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
7779 uint16_t vlandev_ofp_port, int *vid)
7781 if (!hmap_is_empty(&ofproto->vlandev_map)) {
7782 const struct vlan_splinter *vsp;
7784 vsp = vlandev_find(ofproto, vlandev_ofp_port);
7789 return vsp->realdev_ofp_port;
7795 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
7796 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
7797 * 'flow->in_port' to the "real" device backing the VLAN device, sets
7798 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
7799 * always the case unless VLAN splinters are enabled), returns false without
7800 * making any changes. */
7802 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
7807 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
7812 /* Cause the flow to be processed as if it came in on the real device with
7813 * the VLAN device's VLAN ID. */
7814 flow->in_port = realdev;
7815 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
7820 vsp_remove(struct ofport_dpif *port)
7822 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7823 struct vlan_splinter *vsp;
7825 vsp = vlandev_find(ofproto, port->up.ofp_port);
7827 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
7828 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
7831 port->realdev_ofp_port = 0;
7833 VLOG_ERR("missing vlan device record");
7838 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
7840 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7842 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
7843 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
7844 == realdev_ofp_port)) {
7845 struct vlan_splinter *vsp;
7847 vsp = xmalloc(sizeof *vsp);
7848 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
7849 hash_int(port->up.ofp_port, 0));
7850 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
7851 hash_realdev_vid(realdev_ofp_port, vid));
7852 vsp->realdev_ofp_port = realdev_ofp_port;
7853 vsp->vlandev_ofp_port = port->up.ofp_port;
7856 port->realdev_ofp_port = realdev_ofp_port;
7858 VLOG_ERR("duplicate vlan device record");
7863 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
7865 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
7866 return ofport ? ofport->odp_port : OVSP_NONE;
7869 static struct ofport_dpif *
7870 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
7872 struct ofport_dpif *port;
7874 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
7875 hash_int(odp_port, 0),
7876 &backer->odp_to_ofport_map) {
7877 if (port->odp_port == odp_port) {
7886 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
7888 struct ofport_dpif *port;
7890 port = odp_port_to_ofport(ofproto->backer, odp_port);
7891 if (port && ofproto == ofproto_dpif_cast(port->up.ofproto)) {
7892 return port->up.ofp_port;
7898 const struct ofproto_class ofproto_dpif_class = {
7933 port_is_lacp_current,
7934 NULL, /* rule_choose_table */
7941 rule_modify_actions,
7950 get_cfm_remote_mpids,
7955 get_stp_port_status,
7962 is_mirror_output_bundle,
7963 forward_bpdu_changed,
7964 set_mac_table_config,