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"
39 #include "netdev-vport.h"
46 #include "ofp-actions.h"
47 #include "ofp-parse.h"
48 #include "ofp-print.h"
49 #include "ofproto-dpif-governor.h"
50 #include "ofproto-dpif-sflow.h"
51 #include "poll-loop.h"
56 #include "unaligned.h"
58 #include "vlan-bitmap.h"
61 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
63 COVERAGE_DEFINE(ofproto_dpif_expired);
64 COVERAGE_DEFINE(ofproto_dpif_xlate);
65 COVERAGE_DEFINE(facet_changed_rule);
66 COVERAGE_DEFINE(facet_revalidate);
67 COVERAGE_DEFINE(facet_unexpected);
68 COVERAGE_DEFINE(facet_suppress);
70 /* Maximum depth of flow table recursion (due to resubmit actions) in a
71 * flow translation. */
72 #define MAX_RESUBMIT_RECURSION 64
74 /* Number of implemented OpenFlow tables. */
75 enum { N_TABLES = 255 };
76 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
77 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
89 * - Do include packets and bytes from facets that have been deleted or
90 * whose own statistics have been folded into the rule.
92 * - Do include packets and bytes sent "by hand" that were accounted to
93 * the rule without any facet being involved (this is a rare corner
94 * case in rule_execute()).
96 * - Do not include packet or bytes that can be obtained from any facet's
97 * packet_count or byte_count member or that can be obtained from the
98 * datapath by, e.g., dpif_flow_get() for any subfacet.
100 uint64_t packet_count; /* Number of packets received. */
101 uint64_t byte_count; /* Number of bytes received. */
103 tag_type tag; /* Caches rule_calculate_tag() result. */
105 struct list facets; /* List of "struct facet"s. */
108 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
110 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
113 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
114 const struct flow *);
115 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
118 static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto,
119 const struct flow *flow);
121 static void rule_credit_stats(struct rule_dpif *,
122 const struct dpif_flow_stats *);
123 static void flow_push_stats(struct facet *, const struct dpif_flow_stats *);
124 static tag_type rule_calculate_tag(const struct flow *,
125 const struct minimask *, uint32_t basis);
126 static void rule_invalidate(const struct rule_dpif *);
128 #define MAX_MIRRORS 32
129 typedef uint32_t mirror_mask_t;
130 #define MIRROR_MASK_C(X) UINT32_C(X)
131 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
133 struct ofproto_dpif *ofproto; /* Owning ofproto. */
134 size_t idx; /* In ofproto's "mirrors" array. */
135 void *aux; /* Key supplied by ofproto's client. */
136 char *name; /* Identifier for log messages. */
138 /* Selection criteria. */
139 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
140 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
141 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
143 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
144 struct ofbundle *out; /* Output port or NULL. */
145 int out_vlan; /* Output VLAN or -1. */
146 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
149 int64_t packet_count; /* Number of packets sent. */
150 int64_t byte_count; /* Number of bytes sent. */
153 static void mirror_destroy(struct ofmirror *);
154 static void update_mirror_stats(struct ofproto_dpif *ofproto,
155 mirror_mask_t mirrors,
156 uint64_t packets, uint64_t bytes);
159 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
160 struct ofproto_dpif *ofproto; /* Owning ofproto. */
161 void *aux; /* Key supplied by ofproto's client. */
162 char *name; /* Identifier for log messages. */
165 struct list ports; /* Contains "struct ofport"s. */
166 enum port_vlan_mode vlan_mode; /* VLAN mode */
167 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
168 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
169 * NULL if all VLANs are trunked. */
170 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
171 struct bond *bond; /* Nonnull iff more than one port. */
172 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
175 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
177 /* Port mirroring info. */
178 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
179 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
180 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
183 static void bundle_remove(struct ofport *);
184 static void bundle_update(struct ofbundle *);
185 static void bundle_destroy(struct ofbundle *);
186 static void bundle_del_port(struct ofport_dpif *);
187 static void bundle_run(struct ofbundle *);
188 static void bundle_wait(struct ofbundle *);
189 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
190 uint16_t in_port, bool warn,
191 struct ofport_dpif **in_ofportp);
193 /* A controller may use OFPP_NONE as the ingress port to indicate that
194 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
195 * when an input bundle is needed for validation (e.g., mirroring or
196 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
197 * any 'port' structs, so care must be taken when dealing with it. */
198 static struct ofbundle ofpp_none_bundle = {
200 .vlan_mode = PORT_VLAN_TRUNK
203 static void stp_run(struct ofproto_dpif *ofproto);
204 static void stp_wait(struct ofproto_dpif *ofproto);
205 static int set_stp_port(struct ofport *,
206 const struct ofproto_port_stp_settings *);
208 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
210 struct action_xlate_ctx {
211 /* action_xlate_ctx_init() initializes these members. */
214 struct ofproto_dpif *ofproto;
216 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
217 * this flow when actions change header fields. */
220 /* The packet corresponding to 'flow', or a null pointer if we are
221 * revalidating without a packet to refer to. */
222 const struct ofpbuf *packet;
224 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
225 * actions update the flow table?
227 * We want to update these tables if we are actually processing a packet,
228 * or if we are accounting for packets that the datapath has processed, but
229 * not if we are just revalidating. */
232 /* The rule that we are currently translating, or NULL. */
233 struct rule_dpif *rule;
235 /* Union of the set of TCP flags seen so far in this flow. (Used only by
236 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
240 /* If nonnull, flow translation calls this function just before executing a
241 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
242 * when the recursion depth is exceeded.
244 * 'rule' is the rule being submitted into. It will be null if the
245 * resubmit or OFPP_TABLE action didn't find a matching rule.
247 * This is normally null so the client has to set it manually after
248 * calling action_xlate_ctx_init(). */
249 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule);
251 /* If nonnull, flow translation calls this function to report some
252 * significant decision, e.g. to explain why OFPP_NORMAL translation
253 * dropped a packet. */
254 void (*report_hook)(struct action_xlate_ctx *, const char *s);
256 /* If nonnull, flow translation credits the specified statistics to each
257 * rule reached through a resubmit or OFPP_TABLE action.
259 * This is normally null so the client has to set it manually after
260 * calling action_xlate_ctx_init(). */
261 const struct dpif_flow_stats *resubmit_stats;
263 /* xlate_actions() initializes and uses these members. The client might want
264 * to look at them after it returns. */
266 struct ofpbuf *odp_actions; /* Datapath actions. */
267 tag_type tags; /* Tags associated with actions. */
268 enum slow_path_reason slow; /* 0 if fast path may be used. */
269 bool has_learn; /* Actions include NXAST_LEARN? */
270 bool has_normal; /* Actions output to OFPP_NORMAL? */
271 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
272 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
273 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
275 /* xlate_actions() initializes and uses these members, but the client has no
276 * reason to look at them. */
278 int recurse; /* Recursion level, via xlate_table_action. */
279 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
280 struct flow base_flow; /* Flow at the last commit. */
281 uint32_t orig_skb_priority; /* Priority when packet arrived. */
282 uint8_t table_id; /* OpenFlow table ID where flow was found. */
283 uint32_t sflow_n_outputs; /* Number of output ports. */
284 uint32_t sflow_odp_port; /* Output port for composing sFlow action. */
285 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
286 bool exit; /* No further actions should be processed. */
287 struct flow orig_flow; /* Copy of original flow. */
290 /* Initial values of fields of the packet that may be changed during
291 * flow processing and needed later. */
292 struct initial_vals {
293 /* This is the value of vlan_tci in the packet as actually received from
294 * dpif. This is the same as the facet's flow.vlan_tci unless the packet
295 * was received via a VLAN splinter. In that case, this value is 0
296 * (because the packet as actually received from the dpif had no 802.1Q
297 * tag) but the facet's flow.vlan_tci is set to the VLAN that the splinter
300 * This member should be removed when the VLAN splinters feature is no
304 /* If received on a tunnel, the IP TOS value of the tunnel. */
305 uint8_t tunnel_ip_tos;
308 static void action_xlate_ctx_init(struct action_xlate_ctx *,
309 struct ofproto_dpif *, const struct flow *,
310 const struct initial_vals *initial_vals,
312 uint8_t tcp_flags, const struct ofpbuf *);
313 static void xlate_actions(struct action_xlate_ctx *,
314 const struct ofpact *ofpacts, size_t ofpacts_len,
315 struct ofpbuf *odp_actions);
316 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
317 const struct ofpact *ofpacts,
319 static void xlate_table_action(struct action_xlate_ctx *, uint16_t in_port,
320 uint8_t table_id, bool may_packet_in);
322 static size_t put_userspace_action(const struct ofproto_dpif *,
323 struct ofpbuf *odp_actions,
325 const union user_action_cookie *);
327 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
328 enum slow_path_reason,
329 uint64_t *stub, size_t stub_size,
330 const struct nlattr **actionsp,
331 size_t *actions_lenp);
333 static void xlate_report(struct action_xlate_ctx *ctx, const char *s);
335 /* A subfacet (see "struct subfacet" below) has three possible installation
338 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
339 * case just after the subfacet is created, just before the subfacet is
340 * destroyed, or if the datapath returns an error when we try to install a
343 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
345 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
346 * ofproto_dpif is installed in the datapath.
349 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
350 SF_FAST_PATH, /* Full actions are installed. */
351 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
354 static const char *subfacet_path_to_string(enum subfacet_path);
356 /* A dpif flow and actions associated with a facet.
358 * See also the large comment on struct facet. */
361 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
362 struct list list_node; /* In struct facet's 'facets' list. */
363 struct facet *facet; /* Owning facet. */
365 enum odp_key_fitness key_fitness;
369 long long int used; /* Time last used; time created if not used. */
371 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
372 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
376 * These should be essentially identical for every subfacet in a facet, but
377 * may differ in trivial ways due to VLAN splinters. */
378 size_t actions_len; /* Number of bytes in actions[]. */
379 struct nlattr *actions; /* Datapath actions. */
381 enum slow_path_reason slow; /* 0 if fast path may be used. */
382 enum subfacet_path path; /* Installed in datapath? */
384 /* Initial values of the packet that may be needed later. */
385 struct initial_vals initial_vals;
387 /* Datapath port the packet arrived on. This is needed to remove
388 * flows for ports that are no longer part of the bridge. Since the
389 * flow definition only has the OpenFlow port number and the port is
390 * no longer part of the bridge, we can't determine the datapath port
391 * number needed to delete the flow from the datapath. */
392 uint32_t odp_in_port;
395 #define SUBFACET_DESTROY_MAX_BATCH 50
397 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
399 static struct subfacet *subfacet_find(struct ofproto_dpif *,
400 const struct nlattr *key, size_t key_len,
402 static void subfacet_destroy(struct subfacet *);
403 static void subfacet_destroy__(struct subfacet *);
404 static void subfacet_destroy_batch(struct ofproto_dpif *,
405 struct subfacet **, int n);
406 static void subfacet_reset_dp_stats(struct subfacet *,
407 struct dpif_flow_stats *);
408 static void subfacet_update_time(struct subfacet *, long long int used);
409 static void subfacet_update_stats(struct subfacet *,
410 const struct dpif_flow_stats *);
411 static void subfacet_make_actions(struct subfacet *,
412 const struct ofpbuf *packet,
413 struct ofpbuf *odp_actions);
414 static int subfacet_install(struct subfacet *,
415 const struct nlattr *actions, size_t actions_len,
416 struct dpif_flow_stats *, enum slow_path_reason);
417 static void subfacet_uninstall(struct subfacet *);
419 static enum subfacet_path subfacet_want_path(enum slow_path_reason);
421 /* An exact-match instantiation of an OpenFlow flow.
423 * A facet associates a "struct flow", which represents the Open vSwitch
424 * userspace idea of an exact-match flow, with one or more subfacets. Each
425 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
426 * the facet. When the kernel module (or other dpif implementation) and Open
427 * vSwitch userspace agree on the definition of a flow key, there is exactly
428 * one subfacet per facet. If the dpif implementation supports more-specific
429 * flow matching than userspace, however, a facet can have more than one
430 * subfacet, each of which corresponds to some distinction in flow that
431 * userspace simply doesn't understand.
433 * Flow expiration works in terms of subfacets, so a facet must have at least
434 * one subfacet or it will never expire, leaking memory. */
437 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
438 struct list list_node; /* In owning rule's 'facets' list. */
439 struct rule_dpif *rule; /* Owning rule. */
442 struct list subfacets;
443 long long int used; /* Time last used; time created if not used. */
450 * - Do include packets and bytes sent "by hand", e.g. with
453 * - Do include packets and bytes that were obtained from the datapath
454 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
455 * DPIF_FP_ZERO_STATS).
457 * - Do not include packets or bytes that can be obtained from the
458 * datapath for any existing subfacet.
460 uint64_t packet_count; /* Number of packets received. */
461 uint64_t byte_count; /* Number of bytes received. */
463 /* Resubmit statistics. */
464 uint64_t prev_packet_count; /* Number of packets from last stats push. */
465 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
466 long long int prev_used; /* Used time from last stats push. */
469 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
470 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
471 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
473 /* Properties of datapath actions.
475 * Every subfacet has its own actions because actions can differ slightly
476 * between splintered and non-splintered subfacets due to the VLAN tag
477 * being initially different (present vs. absent). All of them have these
478 * properties in common so we just store one copy of them here. */
479 bool has_learn; /* Actions include NXAST_LEARN? */
480 bool has_normal; /* Actions output to OFPP_NORMAL? */
481 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
482 tag_type tags; /* Tags that would require revalidation. */
483 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
485 /* Storage for a single subfacet, to reduce malloc() time and space
486 * overhead. (A facet always has at least one subfacet and in the common
487 * case has exactly one subfacet.) */
488 struct subfacet one_subfacet;
491 static struct facet *facet_create(struct rule_dpif *,
492 const struct flow *, uint32_t hash);
493 static void facet_remove(struct facet *);
494 static void facet_free(struct facet *);
496 static struct facet *facet_find(struct ofproto_dpif *,
497 const struct flow *, uint32_t hash);
498 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
499 const struct flow *, uint32_t hash);
500 static void facet_revalidate(struct facet *);
501 static bool facet_check_consistency(struct facet *);
503 static void facet_flush_stats(struct facet *);
505 static void facet_update_time(struct facet *, long long int used);
506 static void facet_reset_counters(struct facet *);
507 static void facet_push_stats(struct facet *);
508 static void facet_learn(struct facet *);
509 static void facet_account(struct facet *);
511 static bool facet_is_controller_flow(struct facet *);
514 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
518 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
519 struct list bundle_node; /* In struct ofbundle's "ports" list. */
520 struct cfm *cfm; /* Connectivity Fault Management, if any. */
521 tag_type tag; /* Tag associated with this port. */
522 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
523 bool may_enable; /* May be enabled in bonds. */
524 long long int carrier_seq; /* Carrier status changes. */
525 struct tnl_port *tnl_port; /* Tunnel handle, or null. */
528 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
529 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
530 long long int stp_state_entered;
532 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
534 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
536 * This is deprecated. It is only for compatibility with broken device
537 * drivers in old versions of Linux that do not properly support VLANs when
538 * VLAN devices are not used. When broken device drivers are no longer in
539 * widespread use, we will delete these interfaces. */
540 uint16_t realdev_ofp_port;
544 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
545 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
546 * traffic egressing the 'ofport' with that priority should be marked with. */
547 struct priority_to_dscp {
548 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
549 uint32_t priority; /* Priority of this queue (see struct flow). */
551 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
554 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
556 * This is deprecated. It is only for compatibility with broken device drivers
557 * in old versions of Linux that do not properly support VLANs when VLAN
558 * devices are not used. When broken device drivers are no longer in
559 * widespread use, we will delete these interfaces. */
560 struct vlan_splinter {
561 struct hmap_node realdev_vid_node;
562 struct hmap_node vlandev_node;
563 uint16_t realdev_ofp_port;
564 uint16_t vlandev_ofp_port;
568 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
569 uint32_t realdev, ovs_be16 vlan_tci);
570 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
571 static void vsp_remove(struct ofport_dpif *);
572 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
574 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
576 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
579 static struct ofport_dpif *
580 ofport_dpif_cast(const struct ofport *ofport)
582 ovs_assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
583 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
586 static void port_run(struct ofport_dpif *);
587 static void port_run_fast(struct ofport_dpif *);
588 static void port_wait(struct ofport_dpif *);
589 static int set_cfm(struct ofport *, const struct cfm_settings *);
590 static void ofport_clear_priorities(struct ofport_dpif *);
592 struct dpif_completion {
593 struct list list_node;
594 struct ofoperation *op;
597 /* Extra information about a classifier table.
598 * Currently used just for optimized flow revalidation. */
600 /* If either of these is nonnull, then this table has a form that allows
601 * flows to be tagged to avoid revalidating most flows for the most common
602 * kinds of flow table changes. */
603 struct cls_table *catchall_table; /* Table that wildcards all fields. */
604 struct cls_table *other_table; /* Table with any other wildcard set. */
605 uint32_t basis; /* Keeps each table's tags separate. */
608 /* Reasons that we might need to revalidate every facet, and corresponding
611 * A value of 0 means that there is no need to revalidate.
613 * It would be nice to have some cleaner way to integrate with coverage
614 * counters, but with only a few reasons I guess this is good enough for
616 enum revalidate_reason {
617 REV_RECONFIGURE = 1, /* Switch configuration changed. */
618 REV_STP, /* Spanning tree protocol port status change. */
619 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
620 REV_FLOW_TABLE, /* Flow table changed. */
621 REV_INCONSISTENCY /* Facet self-check failed. */
623 COVERAGE_DEFINE(rev_reconfigure);
624 COVERAGE_DEFINE(rev_stp);
625 COVERAGE_DEFINE(rev_port_toggled);
626 COVERAGE_DEFINE(rev_flow_table);
627 COVERAGE_DEFINE(rev_inconsistency);
629 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
630 * These are datapath flows which have no associated ofproto, if they did we
631 * would use facets. */
633 struct hmap_node hmap_node;
638 /* All datapaths of a given type share a single dpif backer instance. */
643 struct timer next_expiration;
644 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
646 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
648 /* Facet revalidation flags applying to facets which use this backer. */
649 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
650 struct tag_set revalidate_set; /* Revalidate only matching facets. */
652 struct hmap drop_keys; /* Set of dropped odp keys. */
655 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
656 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
658 static void drop_key_clear(struct dpif_backer *);
659 static struct ofport_dpif *
660 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
662 struct ofproto_dpif {
663 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
665 struct dpif_backer *backer;
667 /* Special OpenFlow rules. */
668 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
669 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
675 struct netflow *netflow;
676 struct dpif_sflow *sflow;
677 struct hmap bundles; /* Contains "struct ofbundle"s. */
678 struct mac_learning *ml;
679 struct ofmirror *mirrors[MAX_MIRRORS];
681 bool has_bonded_bundles;
685 struct hmap subfacets;
686 struct governor *governor;
689 struct table_dpif tables[N_TABLES];
691 /* Support for debugging async flow mods. */
692 struct list completions;
694 bool has_bundle_action; /* True when the first bundle action appears. */
695 struct netdev_stats stats; /* To account packets generated and consumed in
700 long long int stp_last_tick;
702 /* VLAN splinters. */
703 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
704 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
707 struct sset ports; /* Set of standard port names. */
708 struct sset ghost_ports; /* Ports with no datapath port. */
709 struct sset port_poll_set; /* Queued names for port_poll() reply. */
710 int port_poll_errno; /* Last errno for port_poll() reply. */
713 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
714 * for debugging the asynchronous flow_mod implementation.) */
717 /* All existing ofproto_dpif instances, indexed by ->up.name. */
718 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
720 static void ofproto_dpif_unixctl_init(void);
722 static struct ofproto_dpif *
723 ofproto_dpif_cast(const struct ofproto *ofproto)
725 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
726 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
729 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
731 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
733 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
734 const struct ofpbuf *,
735 const struct initial_vals *, struct ds *);
737 /* Packet processing. */
738 static void update_learning_table(struct ofproto_dpif *,
739 const struct flow *, int vlan,
742 #define FLOW_MISS_MAX_BATCH 50
743 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
745 /* Flow expiration. */
746 static int expire(struct dpif_backer *);
749 static void send_netflow_active_timeouts(struct ofproto_dpif *);
752 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
753 static size_t compose_sflow_action(const struct ofproto_dpif *,
754 struct ofpbuf *odp_actions,
755 const struct flow *, uint32_t odp_port);
756 static void add_mirror_actions(struct action_xlate_ctx *ctx,
757 const struct flow *flow);
758 /* Global variables. */
759 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
761 /* Initial mappings of port to bridge mappings. */
762 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
764 /* Factory functions. */
767 init(const struct shash *iface_hints)
769 struct shash_node *node;
771 /* Make a local copy, since we don't own 'iface_hints' elements. */
772 SHASH_FOR_EACH(node, iface_hints) {
773 const struct iface_hint *orig_hint = node->data;
774 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
776 new_hint->br_name = xstrdup(orig_hint->br_name);
777 new_hint->br_type = xstrdup(orig_hint->br_type);
778 new_hint->ofp_port = orig_hint->ofp_port;
780 shash_add(&init_ofp_ports, node->name, new_hint);
785 enumerate_types(struct sset *types)
787 dp_enumerate_types(types);
791 enumerate_names(const char *type, struct sset *names)
793 struct ofproto_dpif *ofproto;
796 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
797 if (strcmp(type, ofproto->up.type)) {
800 sset_add(names, ofproto->up.name);
807 del(const char *type, const char *name)
812 error = dpif_open(name, type, &dpif);
814 error = dpif_delete(dpif);
821 port_open_type(const char *datapath_type, const char *port_type)
823 return dpif_port_open_type(datapath_type, port_type);
826 /* Type functions. */
828 static struct ofproto_dpif *
829 lookup_ofproto_dpif_by_port_name(const char *name)
831 struct ofproto_dpif *ofproto;
833 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
834 if (sset_contains(&ofproto->ports, name)) {
843 type_run(const char *type)
845 struct dpif_backer *backer;
849 backer = shash_find_data(&all_dpif_backers, type);
851 /* This is not necessarily a problem, since backers are only
852 * created on demand. */
856 dpif_run(backer->dpif);
858 if (backer->need_revalidate
859 || !tag_set_is_empty(&backer->revalidate_set)) {
860 struct tag_set revalidate_set = backer->revalidate_set;
861 bool need_revalidate = backer->need_revalidate;
862 struct ofproto_dpif *ofproto;
863 struct simap_node *node;
864 struct simap tmp_backers;
866 /* Handle tunnel garbage collection. */
867 simap_init(&tmp_backers);
868 simap_swap(&backer->tnl_backers, &tmp_backers);
870 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
871 struct ofport_dpif *iter;
873 if (backer != ofproto->backer) {
877 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
880 if (!iter->tnl_port) {
884 dp_port = netdev_vport_get_dpif_port(iter->up.netdev);
885 node = simap_find(&tmp_backers, dp_port);
887 simap_put(&backer->tnl_backers, dp_port, node->data);
888 simap_delete(&tmp_backers, node);
889 node = simap_find(&backer->tnl_backers, dp_port);
891 node = simap_find(&backer->tnl_backers, dp_port);
893 uint32_t odp_port = UINT32_MAX;
895 if (!dpif_port_add(backer->dpif, iter->up.netdev,
897 simap_put(&backer->tnl_backers, dp_port, odp_port);
898 node = simap_find(&backer->tnl_backers, dp_port);
903 iter->odp_port = node ? node->data : OVSP_NONE;
904 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
906 backer->need_revalidate = REV_RECONFIGURE;
911 SIMAP_FOR_EACH (node, &tmp_backers) {
912 dpif_port_del(backer->dpif, node->data);
914 simap_destroy(&tmp_backers);
916 switch (backer->need_revalidate) {
917 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
918 case REV_STP: COVERAGE_INC(rev_stp); break;
919 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
920 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
921 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
924 if (backer->need_revalidate) {
925 /* Clear the drop_keys in case we should now be accepting some
926 * formerly dropped flows. */
927 drop_key_clear(backer);
930 /* Clear the revalidation flags. */
931 tag_set_init(&backer->revalidate_set);
932 backer->need_revalidate = 0;
934 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
935 struct facet *facet, *next;
937 if (ofproto->backer != backer) {
941 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
943 || tag_set_intersects(&revalidate_set, facet->tags)) {
944 facet_revalidate(facet);
950 if (timer_expired(&backer->next_expiration)) {
951 int delay = expire(backer);
952 timer_set_duration(&backer->next_expiration, delay);
955 /* Check for port changes in the dpif. */
956 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
957 struct ofproto_dpif *ofproto;
958 struct dpif_port port;
960 /* Don't report on the datapath's device. */
961 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
965 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
966 &all_ofproto_dpifs) {
967 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
972 ofproto = lookup_ofproto_dpif_by_port_name(devname);
973 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
974 /* The port was removed. If we know the datapath,
975 * report it through poll_set(). If we don't, it may be
976 * notifying us of a removal we initiated, so ignore it.
977 * If there's a pending ENOBUFS, let it stand, since
978 * everything will be reevaluated. */
979 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
980 sset_add(&ofproto->port_poll_set, devname);
981 ofproto->port_poll_errno = 0;
983 } else if (!ofproto) {
984 /* The port was added, but we don't know with which
985 * ofproto we should associate it. Delete it. */
986 dpif_port_del(backer->dpif, port.port_no);
988 dpif_port_destroy(&port);
994 if (error != EAGAIN) {
995 struct ofproto_dpif *ofproto;
997 /* There was some sort of error, so propagate it to all
998 * ofprotos that use this backer. */
999 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1000 &all_ofproto_dpifs) {
1001 if (ofproto->backer == backer) {
1002 sset_clear(&ofproto->port_poll_set);
1003 ofproto->port_poll_errno = error;
1012 type_run_fast(const char *type)
1014 struct dpif_backer *backer;
1017 backer = shash_find_data(&all_dpif_backers, type);
1019 /* This is not necessarily a problem, since backers are only
1020 * created on demand. */
1024 /* Handle one or more batches of upcalls, until there's nothing left to do
1025 * or until we do a fixed total amount of work.
1027 * We do work in batches because it can be much cheaper to set up a number
1028 * of flows and fire off their patches all at once. We do multiple batches
1029 * because in some cases handling a packet can cause another packet to be
1030 * queued almost immediately as part of the return flow. Both
1031 * optimizations can make major improvements on some benchmarks and
1032 * presumably for real traffic as well. */
1034 while (work < FLOW_MISS_MAX_BATCH) {
1035 int retval = handle_upcalls(backer, FLOW_MISS_MAX_BATCH - work);
1046 type_wait(const char *type)
1048 struct dpif_backer *backer;
1050 backer = shash_find_data(&all_dpif_backers, type);
1052 /* This is not necessarily a problem, since backers are only
1053 * created on demand. */
1057 timer_wait(&backer->next_expiration);
1060 /* Basic life-cycle. */
1062 static int add_internal_flows(struct ofproto_dpif *);
1064 static struct ofproto *
1067 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1068 return &ofproto->up;
1072 dealloc(struct ofproto *ofproto_)
1074 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1079 close_dpif_backer(struct dpif_backer *backer)
1081 struct shash_node *node;
1083 ovs_assert(backer->refcount > 0);
1085 if (--backer->refcount) {
1089 drop_key_clear(backer);
1090 hmap_destroy(&backer->drop_keys);
1092 simap_destroy(&backer->tnl_backers);
1093 hmap_destroy(&backer->odp_to_ofport_map);
1094 node = shash_find(&all_dpif_backers, backer->type);
1096 shash_delete(&all_dpif_backers, node);
1097 dpif_close(backer->dpif);
1102 /* Datapath port slated for removal from datapath. */
1103 struct odp_garbage {
1104 struct list list_node;
1109 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1111 struct dpif_backer *backer;
1112 struct dpif_port_dump port_dump;
1113 struct dpif_port port;
1114 struct shash_node *node;
1115 struct list garbage_list;
1116 struct odp_garbage *garbage, *next;
1122 backer = shash_find_data(&all_dpif_backers, type);
1129 backer_name = xasprintf("ovs-%s", type);
1131 /* Remove any existing datapaths, since we assume we're the only
1132 * userspace controlling the datapath. */
1134 dp_enumerate_names(type, &names);
1135 SSET_FOR_EACH(name, &names) {
1136 struct dpif *old_dpif;
1138 /* Don't remove our backer if it exists. */
1139 if (!strcmp(name, backer_name)) {
1143 if (dpif_open(name, type, &old_dpif)) {
1144 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1146 dpif_delete(old_dpif);
1147 dpif_close(old_dpif);
1150 sset_destroy(&names);
1152 backer = xmalloc(sizeof *backer);
1154 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1157 VLOG_ERR("failed to open datapath of type %s: %s", type,
1163 backer->type = xstrdup(type);
1164 backer->refcount = 1;
1165 hmap_init(&backer->odp_to_ofport_map);
1166 hmap_init(&backer->drop_keys);
1167 timer_set_duration(&backer->next_expiration, 1000);
1168 backer->need_revalidate = 0;
1169 simap_init(&backer->tnl_backers);
1170 tag_set_init(&backer->revalidate_set);
1173 dpif_flow_flush(backer->dpif);
1175 /* Loop through the ports already on the datapath and remove any
1176 * that we don't need anymore. */
1177 list_init(&garbage_list);
1178 dpif_port_dump_start(&port_dump, backer->dpif);
1179 while (dpif_port_dump_next(&port_dump, &port)) {
1180 node = shash_find(&init_ofp_ports, port.name);
1181 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1182 garbage = xmalloc(sizeof *garbage);
1183 garbage->odp_port = port.port_no;
1184 list_push_front(&garbage_list, &garbage->list_node);
1187 dpif_port_dump_done(&port_dump);
1189 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1190 dpif_port_del(backer->dpif, garbage->odp_port);
1191 list_remove(&garbage->list_node);
1195 shash_add(&all_dpif_backers, type, backer);
1197 error = dpif_recv_set(backer->dpif, true);
1199 VLOG_ERR("failed to listen on datapath of type %s: %s",
1200 type, strerror(error));
1201 close_dpif_backer(backer);
1209 construct(struct ofproto *ofproto_)
1211 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1212 struct shash_node *node, *next;
1217 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1222 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1223 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1225 ofproto->n_matches = 0;
1227 ofproto->netflow = NULL;
1228 ofproto->sflow = NULL;
1229 ofproto->stp = NULL;
1230 hmap_init(&ofproto->bundles);
1231 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1232 for (i = 0; i < MAX_MIRRORS; i++) {
1233 ofproto->mirrors[i] = NULL;
1235 ofproto->has_bonded_bundles = false;
1237 hmap_init(&ofproto->facets);
1238 hmap_init(&ofproto->subfacets);
1239 ofproto->governor = NULL;
1241 for (i = 0; i < N_TABLES; i++) {
1242 struct table_dpif *table = &ofproto->tables[i];
1244 table->catchall_table = NULL;
1245 table->other_table = NULL;
1246 table->basis = random_uint32();
1249 list_init(&ofproto->completions);
1251 ofproto_dpif_unixctl_init();
1253 ofproto->has_mirrors = false;
1254 ofproto->has_bundle_action = false;
1256 hmap_init(&ofproto->vlandev_map);
1257 hmap_init(&ofproto->realdev_vid_map);
1259 sset_init(&ofproto->ports);
1260 sset_init(&ofproto->ghost_ports);
1261 sset_init(&ofproto->port_poll_set);
1262 ofproto->port_poll_errno = 0;
1264 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1265 struct iface_hint *iface_hint = node->data;
1267 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1268 /* Check if the datapath already has this port. */
1269 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1270 sset_add(&ofproto->ports, node->name);
1273 free(iface_hint->br_name);
1274 free(iface_hint->br_type);
1276 shash_delete(&init_ofp_ports, node);
1280 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1281 hash_string(ofproto->up.name, 0));
1282 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1284 ofproto_init_tables(ofproto_, N_TABLES);
1285 error = add_internal_flows(ofproto);
1286 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1292 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1293 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1295 struct ofputil_flow_mod fm;
1298 match_init_catchall(&fm.match);
1300 match_set_reg(&fm.match, 0, id);
1301 fm.new_cookie = htonll(0);
1302 fm.cookie = htonll(0);
1303 fm.cookie_mask = htonll(0);
1304 fm.table_id = TBL_INTERNAL;
1305 fm.command = OFPFC_ADD;
1306 fm.idle_timeout = 0;
1307 fm.hard_timeout = 0;
1311 fm.ofpacts = ofpacts->data;
1312 fm.ofpacts_len = ofpacts->size;
1314 error = ofproto_flow_mod(&ofproto->up, &fm);
1316 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1317 id, ofperr_to_string(error));
1321 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1322 ovs_assert(*rulep != NULL);
1328 add_internal_flows(struct ofproto_dpif *ofproto)
1330 struct ofpact_controller *controller;
1331 uint64_t ofpacts_stub[128 / 8];
1332 struct ofpbuf ofpacts;
1336 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1339 controller = ofpact_put_CONTROLLER(&ofpacts);
1340 controller->max_len = UINT16_MAX;
1341 controller->controller_id = 0;
1342 controller->reason = OFPR_NO_MATCH;
1343 ofpact_pad(&ofpacts);
1345 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1350 ofpbuf_clear(&ofpacts);
1351 error = add_internal_flow(ofproto, id++, &ofpacts,
1352 &ofproto->no_packet_in_rule);
1357 complete_operations(struct ofproto_dpif *ofproto)
1359 struct dpif_completion *c, *next;
1361 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1362 ofoperation_complete(c->op, 0);
1363 list_remove(&c->list_node);
1369 destruct(struct ofproto *ofproto_)
1371 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1372 struct rule_dpif *rule, *next_rule;
1373 struct oftable *table;
1376 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1377 complete_operations(ofproto);
1379 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1380 struct cls_cursor cursor;
1382 cls_cursor_init(&cursor, &table->cls, NULL);
1383 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1384 ofproto_rule_destroy(&rule->up);
1388 for (i = 0; i < MAX_MIRRORS; i++) {
1389 mirror_destroy(ofproto->mirrors[i]);
1392 netflow_destroy(ofproto->netflow);
1393 dpif_sflow_destroy(ofproto->sflow);
1394 hmap_destroy(&ofproto->bundles);
1395 mac_learning_destroy(ofproto->ml);
1397 hmap_destroy(&ofproto->facets);
1398 hmap_destroy(&ofproto->subfacets);
1399 governor_destroy(ofproto->governor);
1401 hmap_destroy(&ofproto->vlandev_map);
1402 hmap_destroy(&ofproto->realdev_vid_map);
1404 sset_destroy(&ofproto->ports);
1405 sset_destroy(&ofproto->ghost_ports);
1406 sset_destroy(&ofproto->port_poll_set);
1408 close_dpif_backer(ofproto->backer);
1412 run_fast(struct ofproto *ofproto_)
1414 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1415 struct ofport_dpif *ofport;
1417 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1418 port_run_fast(ofport);
1425 run(struct ofproto *ofproto_)
1427 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1428 struct ofport_dpif *ofport;
1429 struct ofbundle *bundle;
1433 complete_operations(ofproto);
1436 error = run_fast(ofproto_);
1441 if (ofproto->netflow) {
1442 if (netflow_run(ofproto->netflow)) {
1443 send_netflow_active_timeouts(ofproto);
1446 if (ofproto->sflow) {
1447 dpif_sflow_run(ofproto->sflow);
1450 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1453 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1458 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1460 /* Check the consistency of a random facet, to aid debugging. */
1461 if (!hmap_is_empty(&ofproto->facets)
1462 && !ofproto->backer->need_revalidate) {
1463 struct facet *facet;
1465 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1466 struct facet, hmap_node);
1467 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1469 if (!facet_check_consistency(facet)) {
1470 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1475 if (ofproto->governor) {
1478 governor_run(ofproto->governor);
1480 /* If the governor has shrunk to its minimum size and the number of
1481 * subfacets has dwindled, then drop the governor entirely.
1483 * For hysteresis, the number of subfacets to drop the governor is
1484 * smaller than the number needed to trigger its creation. */
1485 n_subfacets = hmap_count(&ofproto->subfacets);
1486 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1487 && governor_is_idle(ofproto->governor)) {
1488 governor_destroy(ofproto->governor);
1489 ofproto->governor = NULL;
1497 wait(struct ofproto *ofproto_)
1499 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1500 struct ofport_dpif *ofport;
1501 struct ofbundle *bundle;
1503 if (!clogged && !list_is_empty(&ofproto->completions)) {
1504 poll_immediate_wake();
1507 dpif_wait(ofproto->backer->dpif);
1508 dpif_recv_wait(ofproto->backer->dpif);
1509 if (ofproto->sflow) {
1510 dpif_sflow_wait(ofproto->sflow);
1512 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1513 poll_immediate_wake();
1515 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1518 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1519 bundle_wait(bundle);
1521 if (ofproto->netflow) {
1522 netflow_wait(ofproto->netflow);
1524 mac_learning_wait(ofproto->ml);
1526 if (ofproto->backer->need_revalidate) {
1527 /* Shouldn't happen, but if it does just go around again. */
1528 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1529 poll_immediate_wake();
1531 if (ofproto->governor) {
1532 governor_wait(ofproto->governor);
1537 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1539 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1541 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1542 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1546 flush(struct ofproto *ofproto_)
1548 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1549 struct subfacet *subfacet, *next_subfacet;
1550 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1554 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1555 &ofproto->subfacets) {
1556 if (subfacet->path != SF_NOT_INSTALLED) {
1557 batch[n_batch++] = subfacet;
1558 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1559 subfacet_destroy_batch(ofproto, batch, n_batch);
1563 subfacet_destroy(subfacet);
1568 subfacet_destroy_batch(ofproto, batch, n_batch);
1573 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1574 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1576 *arp_match_ip = true;
1577 *actions = (OFPUTIL_A_OUTPUT |
1578 OFPUTIL_A_SET_VLAN_VID |
1579 OFPUTIL_A_SET_VLAN_PCP |
1580 OFPUTIL_A_STRIP_VLAN |
1581 OFPUTIL_A_SET_DL_SRC |
1582 OFPUTIL_A_SET_DL_DST |
1583 OFPUTIL_A_SET_NW_SRC |
1584 OFPUTIL_A_SET_NW_DST |
1585 OFPUTIL_A_SET_NW_TOS |
1586 OFPUTIL_A_SET_TP_SRC |
1587 OFPUTIL_A_SET_TP_DST |
1592 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1594 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1595 struct dpif_dp_stats s;
1597 strcpy(ots->name, "classifier");
1599 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1601 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1602 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1605 static struct ofport *
1608 struct ofport_dpif *port = xmalloc(sizeof *port);
1613 port_dealloc(struct ofport *port_)
1615 struct ofport_dpif *port = ofport_dpif_cast(port_);
1620 port_construct(struct ofport *port_)
1622 struct ofport_dpif *port = ofport_dpif_cast(port_);
1623 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1624 const struct netdev *netdev = port->up.netdev;
1625 struct dpif_port dpif_port;
1628 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1629 port->bundle = NULL;
1631 port->tag = tag_create_random();
1632 port->may_enable = true;
1633 port->stp_port = NULL;
1634 port->stp_state = STP_DISABLED;
1635 port->tnl_port = NULL;
1636 hmap_init(&port->priorities);
1637 port->realdev_ofp_port = 0;
1638 port->vlandev_vid = 0;
1639 port->carrier_seq = netdev_get_carrier_resets(netdev);
1641 if (netdev_vport_is_patch(netdev)) {
1642 /* XXX By bailing out here, we don't do required sFlow work. */
1643 port->odp_port = OVSP_NONE;
1647 error = dpif_port_query_by_name(ofproto->backer->dpif,
1648 netdev_vport_get_dpif_port(netdev),
1654 port->odp_port = dpif_port.port_no;
1656 if (netdev_get_tunnel_config(netdev)) {
1657 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1659 /* Sanity-check that a mapping doesn't already exist. This
1660 * shouldn't happen for non-tunnel ports. */
1661 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1662 VLOG_ERR("port %s already has an OpenFlow port number",
1664 dpif_port_destroy(&dpif_port);
1668 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1669 hash_int(port->odp_port, 0));
1671 dpif_port_destroy(&dpif_port);
1673 if (ofproto->sflow) {
1674 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1681 port_destruct(struct ofport *port_)
1683 struct ofport_dpif *port = ofport_dpif_cast(port_);
1684 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1685 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1686 const char *devname = netdev_get_name(port->up.netdev);
1688 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1689 /* The underlying device is still there, so delete it. This
1690 * happens when the ofproto is being destroyed, since the caller
1691 * assumes that removal of attached ports will happen as part of
1693 if (!port->tnl_port) {
1694 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1696 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1699 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1700 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1703 tnl_port_del(port->tnl_port);
1704 sset_find_and_delete(&ofproto->ports, devname);
1705 sset_find_and_delete(&ofproto->ghost_ports, devname);
1706 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1707 bundle_remove(port_);
1708 set_cfm(port_, NULL);
1709 if (ofproto->sflow) {
1710 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1713 ofport_clear_priorities(port);
1714 hmap_destroy(&port->priorities);
1718 port_modified(struct ofport *port_)
1720 struct ofport_dpif *port = ofport_dpif_cast(port_);
1722 if (port->bundle && port->bundle->bond) {
1723 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1728 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1730 struct ofport_dpif *port = ofport_dpif_cast(port_);
1731 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1732 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1734 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1735 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1736 OFPUTIL_PC_NO_PACKET_IN)) {
1737 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1739 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1740 bundle_update(port->bundle);
1746 set_sflow(struct ofproto *ofproto_,
1747 const struct ofproto_sflow_options *sflow_options)
1749 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1750 struct dpif_sflow *ds = ofproto->sflow;
1752 if (sflow_options) {
1754 struct ofport_dpif *ofport;
1756 ds = ofproto->sflow = dpif_sflow_create();
1757 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1758 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1760 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1762 dpif_sflow_set_options(ds, sflow_options);
1765 dpif_sflow_destroy(ds);
1766 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1767 ofproto->sflow = NULL;
1774 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1776 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1783 struct ofproto_dpif *ofproto;
1785 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1786 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1787 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1790 if (cfm_configure(ofport->cfm, s)) {
1796 cfm_destroy(ofport->cfm);
1802 get_cfm_fault(const struct ofport *ofport_)
1804 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1806 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1810 get_cfm_opup(const struct ofport *ofport_)
1812 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1814 return ofport->cfm ? cfm_get_opup(ofport->cfm) : -1;
1818 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1821 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1824 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1832 get_cfm_health(const struct ofport *ofport_)
1834 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1836 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1839 /* Spanning Tree. */
1842 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1844 struct ofproto_dpif *ofproto = ofproto_;
1845 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1846 struct ofport_dpif *ofport;
1848 ofport = stp_port_get_aux(sp);
1850 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1851 ofproto->up.name, port_num);
1853 struct eth_header *eth = pkt->l2;
1855 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1856 if (eth_addr_is_zero(eth->eth_src)) {
1857 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1858 "with unknown MAC", ofproto->up.name, port_num);
1860 send_packet(ofport, pkt);
1866 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1868 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1870 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1872 /* Only revalidate flows if the configuration changed. */
1873 if (!s != !ofproto->stp) {
1874 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1878 if (!ofproto->stp) {
1879 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1880 send_bpdu_cb, ofproto);
1881 ofproto->stp_last_tick = time_msec();
1884 stp_set_bridge_id(ofproto->stp, s->system_id);
1885 stp_set_bridge_priority(ofproto->stp, s->priority);
1886 stp_set_hello_time(ofproto->stp, s->hello_time);
1887 stp_set_max_age(ofproto->stp, s->max_age);
1888 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1890 struct ofport *ofport;
1892 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1893 set_stp_port(ofport, NULL);
1896 stp_destroy(ofproto->stp);
1897 ofproto->stp = NULL;
1904 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1906 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1910 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1911 s->designated_root = stp_get_designated_root(ofproto->stp);
1912 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1921 update_stp_port_state(struct ofport_dpif *ofport)
1923 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1924 enum stp_state state;
1926 /* Figure out new state. */
1927 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1931 if (ofport->stp_state != state) {
1932 enum ofputil_port_state of_state;
1935 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1936 netdev_get_name(ofport->up.netdev),
1937 stp_state_name(ofport->stp_state),
1938 stp_state_name(state));
1939 if (stp_learn_in_state(ofport->stp_state)
1940 != stp_learn_in_state(state)) {
1941 /* xxx Learning action flows should also be flushed. */
1942 mac_learning_flush(ofproto->ml,
1943 &ofproto->backer->revalidate_set);
1945 fwd_change = stp_forward_in_state(ofport->stp_state)
1946 != stp_forward_in_state(state);
1948 ofproto->backer->need_revalidate = REV_STP;
1949 ofport->stp_state = state;
1950 ofport->stp_state_entered = time_msec();
1952 if (fwd_change && ofport->bundle) {
1953 bundle_update(ofport->bundle);
1956 /* Update the STP state bits in the OpenFlow port description. */
1957 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1958 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1959 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1960 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1961 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1963 ofproto_port_set_state(&ofport->up, of_state);
1967 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1968 * caller is responsible for assigning STP port numbers and ensuring
1969 * there are no duplicates. */
1971 set_stp_port(struct ofport *ofport_,
1972 const struct ofproto_port_stp_settings *s)
1974 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1975 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1976 struct stp_port *sp = ofport->stp_port;
1978 if (!s || !s->enable) {
1980 ofport->stp_port = NULL;
1981 stp_port_disable(sp);
1982 update_stp_port_state(ofport);
1985 } else if (sp && stp_port_no(sp) != s->port_num
1986 && ofport == stp_port_get_aux(sp)) {
1987 /* The port-id changed, so disable the old one if it's not
1988 * already in use by another port. */
1989 stp_port_disable(sp);
1992 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1993 stp_port_enable(sp);
1995 stp_port_set_aux(sp, ofport);
1996 stp_port_set_priority(sp, s->priority);
1997 stp_port_set_path_cost(sp, s->path_cost);
1999 update_stp_port_state(ofport);
2005 get_stp_port_status(struct ofport *ofport_,
2006 struct ofproto_port_stp_status *s)
2008 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2009 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2010 struct stp_port *sp = ofport->stp_port;
2012 if (!ofproto->stp || !sp) {
2018 s->port_id = stp_port_get_id(sp);
2019 s->state = stp_port_get_state(sp);
2020 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2021 s->role = stp_port_get_role(sp);
2022 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2028 stp_run(struct ofproto_dpif *ofproto)
2031 long long int now = time_msec();
2032 long long int elapsed = now - ofproto->stp_last_tick;
2033 struct stp_port *sp;
2036 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2037 ofproto->stp_last_tick = now;
2039 while (stp_get_changed_port(ofproto->stp, &sp)) {
2040 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2043 update_stp_port_state(ofport);
2047 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2048 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2054 stp_wait(struct ofproto_dpif *ofproto)
2057 poll_timer_wait(1000);
2061 /* Returns true if STP should process 'flow'. */
2063 stp_should_process_flow(const struct flow *flow)
2065 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2069 stp_process_packet(const struct ofport_dpif *ofport,
2070 const struct ofpbuf *packet)
2072 struct ofpbuf payload = *packet;
2073 struct eth_header *eth = payload.data;
2074 struct stp_port *sp = ofport->stp_port;
2076 /* Sink packets on ports that have STP disabled when the bridge has
2078 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2082 /* Trim off padding on payload. */
2083 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2084 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2087 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2088 stp_received_bpdu(sp, payload.data, payload.size);
2092 static struct priority_to_dscp *
2093 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2095 struct priority_to_dscp *pdscp;
2098 hash = hash_int(priority, 0);
2099 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2100 if (pdscp->priority == priority) {
2108 ofport_clear_priorities(struct ofport_dpif *ofport)
2110 struct priority_to_dscp *pdscp, *next;
2112 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2113 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2119 set_queues(struct ofport *ofport_,
2120 const struct ofproto_port_queue *qdscp_list,
2123 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2124 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2125 struct hmap new = HMAP_INITIALIZER(&new);
2128 for (i = 0; i < n_qdscp; i++) {
2129 struct priority_to_dscp *pdscp;
2133 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2134 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2139 pdscp = get_priority(ofport, priority);
2141 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2143 pdscp = xmalloc(sizeof *pdscp);
2144 pdscp->priority = priority;
2146 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2149 if (pdscp->dscp != dscp) {
2151 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2154 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2157 if (!hmap_is_empty(&ofport->priorities)) {
2158 ofport_clear_priorities(ofport);
2159 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2162 hmap_swap(&new, &ofport->priorities);
2170 /* Expires all MAC learning entries associated with 'bundle' and forces its
2171 * ofproto to revalidate every flow.
2173 * Normally MAC learning entries are removed only from the ofproto associated
2174 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2175 * are removed from every ofproto. When patch ports and SLB bonds are in use
2176 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2177 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2178 * with the host from which it migrated. */
2180 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2182 struct ofproto_dpif *ofproto = bundle->ofproto;
2183 struct mac_learning *ml = ofproto->ml;
2184 struct mac_entry *mac, *next_mac;
2186 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2187 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2188 if (mac->port.p == bundle) {
2190 struct ofproto_dpif *o;
2192 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2194 struct mac_entry *e;
2196 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2199 mac_learning_expire(o->ml, e);
2205 mac_learning_expire(ml, mac);
2210 static struct ofbundle *
2211 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2213 struct ofbundle *bundle;
2215 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2216 &ofproto->bundles) {
2217 if (bundle->aux == aux) {
2224 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2225 * ones that are found to 'bundles'. */
2227 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2228 void **auxes, size_t n_auxes,
2229 struct hmapx *bundles)
2233 hmapx_init(bundles);
2234 for (i = 0; i < n_auxes; i++) {
2235 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2237 hmapx_add(bundles, bundle);
2243 bundle_update(struct ofbundle *bundle)
2245 struct ofport_dpif *port;
2247 bundle->floodable = true;
2248 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2249 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2250 || !stp_forward_in_state(port->stp_state)) {
2251 bundle->floodable = false;
2258 bundle_del_port(struct ofport_dpif *port)
2260 struct ofbundle *bundle = port->bundle;
2262 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2264 list_remove(&port->bundle_node);
2265 port->bundle = NULL;
2268 lacp_slave_unregister(bundle->lacp, port);
2271 bond_slave_unregister(bundle->bond, port);
2274 bundle_update(bundle);
2278 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2279 struct lacp_slave_settings *lacp,
2280 uint32_t bond_stable_id)
2282 struct ofport_dpif *port;
2284 port = get_ofp_port(bundle->ofproto, ofp_port);
2289 if (port->bundle != bundle) {
2290 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2292 bundle_del_port(port);
2295 port->bundle = bundle;
2296 list_push_back(&bundle->ports, &port->bundle_node);
2297 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2298 || !stp_forward_in_state(port->stp_state)) {
2299 bundle->floodable = false;
2303 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2304 lacp_slave_register(bundle->lacp, port, lacp);
2307 port->bond_stable_id = bond_stable_id;
2313 bundle_destroy(struct ofbundle *bundle)
2315 struct ofproto_dpif *ofproto;
2316 struct ofport_dpif *port, *next_port;
2323 ofproto = bundle->ofproto;
2324 for (i = 0; i < MAX_MIRRORS; i++) {
2325 struct ofmirror *m = ofproto->mirrors[i];
2327 if (m->out == bundle) {
2329 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2330 || hmapx_find_and_delete(&m->dsts, bundle)) {
2331 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2336 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2337 bundle_del_port(port);
2340 bundle_flush_macs(bundle, true);
2341 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2343 free(bundle->trunks);
2344 lacp_destroy(bundle->lacp);
2345 bond_destroy(bundle->bond);
2350 bundle_set(struct ofproto *ofproto_, void *aux,
2351 const struct ofproto_bundle_settings *s)
2353 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2354 bool need_flush = false;
2355 struct ofport_dpif *port;
2356 struct ofbundle *bundle;
2357 unsigned long *trunks;
2363 bundle_destroy(bundle_lookup(ofproto, aux));
2367 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2368 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2370 bundle = bundle_lookup(ofproto, aux);
2372 bundle = xmalloc(sizeof *bundle);
2374 bundle->ofproto = ofproto;
2375 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2376 hash_pointer(aux, 0));
2378 bundle->name = NULL;
2380 list_init(&bundle->ports);
2381 bundle->vlan_mode = PORT_VLAN_TRUNK;
2383 bundle->trunks = NULL;
2384 bundle->use_priority_tags = s->use_priority_tags;
2385 bundle->lacp = NULL;
2386 bundle->bond = NULL;
2388 bundle->floodable = true;
2390 bundle->src_mirrors = 0;
2391 bundle->dst_mirrors = 0;
2392 bundle->mirror_out = 0;
2395 if (!bundle->name || strcmp(s->name, bundle->name)) {
2397 bundle->name = xstrdup(s->name);
2402 if (!bundle->lacp) {
2403 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2404 bundle->lacp = lacp_create();
2406 lacp_configure(bundle->lacp, s->lacp);
2408 lacp_destroy(bundle->lacp);
2409 bundle->lacp = NULL;
2412 /* Update set of ports. */
2414 for (i = 0; i < s->n_slaves; i++) {
2415 if (!bundle_add_port(bundle, s->slaves[i],
2416 s->lacp ? &s->lacp_slaves[i] : NULL,
2417 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
2421 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2422 struct ofport_dpif *next_port;
2424 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2425 for (i = 0; i < s->n_slaves; i++) {
2426 if (s->slaves[i] == port->up.ofp_port) {
2431 bundle_del_port(port);
2435 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2437 if (list_is_empty(&bundle->ports)) {
2438 bundle_destroy(bundle);
2442 /* Set VLAN tagging mode */
2443 if (s->vlan_mode != bundle->vlan_mode
2444 || s->use_priority_tags != bundle->use_priority_tags) {
2445 bundle->vlan_mode = s->vlan_mode;
2446 bundle->use_priority_tags = s->use_priority_tags;
2451 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2452 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2454 if (vlan != bundle->vlan) {
2455 bundle->vlan = vlan;
2459 /* Get trunked VLANs. */
2460 switch (s->vlan_mode) {
2461 case PORT_VLAN_ACCESS:
2465 case PORT_VLAN_TRUNK:
2466 trunks = CONST_CAST(unsigned long *, s->trunks);
2469 case PORT_VLAN_NATIVE_UNTAGGED:
2470 case PORT_VLAN_NATIVE_TAGGED:
2471 if (vlan != 0 && (!s->trunks
2472 || !bitmap_is_set(s->trunks, vlan)
2473 || bitmap_is_set(s->trunks, 0))) {
2474 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2476 trunks = bitmap_clone(s->trunks, 4096);
2478 trunks = bitmap_allocate1(4096);
2480 bitmap_set1(trunks, vlan);
2481 bitmap_set0(trunks, 0);
2483 trunks = CONST_CAST(unsigned long *, s->trunks);
2490 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2491 free(bundle->trunks);
2492 if (trunks == s->trunks) {
2493 bundle->trunks = vlan_bitmap_clone(trunks);
2495 bundle->trunks = trunks;
2500 if (trunks != s->trunks) {
2505 if (!list_is_short(&bundle->ports)) {
2506 bundle->ofproto->has_bonded_bundles = true;
2508 if (bond_reconfigure(bundle->bond, s->bond)) {
2509 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2512 bundle->bond = bond_create(s->bond);
2513 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2516 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2517 bond_slave_register(bundle->bond, port, port->bond_stable_id,
2521 bond_destroy(bundle->bond);
2522 bundle->bond = NULL;
2525 /* If we changed something that would affect MAC learning, un-learn
2526 * everything on this port and force flow revalidation. */
2528 bundle_flush_macs(bundle, false);
2535 bundle_remove(struct ofport *port_)
2537 struct ofport_dpif *port = ofport_dpif_cast(port_);
2538 struct ofbundle *bundle = port->bundle;
2541 bundle_del_port(port);
2542 if (list_is_empty(&bundle->ports)) {
2543 bundle_destroy(bundle);
2544 } else if (list_is_short(&bundle->ports)) {
2545 bond_destroy(bundle->bond);
2546 bundle->bond = NULL;
2552 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2554 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2555 struct ofport_dpif *port = port_;
2556 uint8_t ea[ETH_ADDR_LEN];
2559 error = netdev_get_etheraddr(port->up.netdev, ea);
2561 struct ofpbuf packet;
2564 ofpbuf_init(&packet, 0);
2565 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2567 memcpy(packet_pdu, pdu, pdu_size);
2569 send_packet(port, &packet);
2570 ofpbuf_uninit(&packet);
2572 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2573 "%s (%s)", port->bundle->name,
2574 netdev_get_name(port->up.netdev), strerror(error));
2579 bundle_send_learning_packets(struct ofbundle *bundle)
2581 struct ofproto_dpif *ofproto = bundle->ofproto;
2582 int error, n_packets, n_errors;
2583 struct mac_entry *e;
2585 error = n_packets = n_errors = 0;
2586 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2587 if (e->port.p != bundle) {
2588 struct ofpbuf *learning_packet;
2589 struct ofport_dpif *port;
2593 /* The assignment to "port" is unnecessary but makes "grep"ing for
2594 * struct ofport_dpif more effective. */
2595 learning_packet = bond_compose_learning_packet(bundle->bond,
2599 ret = send_packet(port, learning_packet);
2600 ofpbuf_delete(learning_packet);
2610 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2611 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2612 "packets, last error was: %s",
2613 bundle->name, n_errors, n_packets, strerror(error));
2615 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2616 bundle->name, n_packets);
2621 bundle_run(struct ofbundle *bundle)
2624 lacp_run(bundle->lacp, send_pdu_cb);
2627 struct ofport_dpif *port;
2629 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2630 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2633 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2634 lacp_status(bundle->lacp));
2635 if (bond_should_send_learning_packets(bundle->bond)) {
2636 bundle_send_learning_packets(bundle);
2642 bundle_wait(struct ofbundle *bundle)
2645 lacp_wait(bundle->lacp);
2648 bond_wait(bundle->bond);
2655 mirror_scan(struct ofproto_dpif *ofproto)
2659 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2660 if (!ofproto->mirrors[idx]) {
2667 static struct ofmirror *
2668 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2672 for (i = 0; i < MAX_MIRRORS; i++) {
2673 struct ofmirror *mirror = ofproto->mirrors[i];
2674 if (mirror && mirror->aux == aux) {
2682 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2684 mirror_update_dups(struct ofproto_dpif *ofproto)
2688 for (i = 0; i < MAX_MIRRORS; i++) {
2689 struct ofmirror *m = ofproto->mirrors[i];
2692 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2696 for (i = 0; i < MAX_MIRRORS; i++) {
2697 struct ofmirror *m1 = ofproto->mirrors[i];
2704 for (j = i + 1; j < MAX_MIRRORS; j++) {
2705 struct ofmirror *m2 = ofproto->mirrors[j];
2707 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2708 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2709 m2->dup_mirrors |= m1->dup_mirrors;
2716 mirror_set(struct ofproto *ofproto_, void *aux,
2717 const struct ofproto_mirror_settings *s)
2719 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2720 mirror_mask_t mirror_bit;
2721 struct ofbundle *bundle;
2722 struct ofmirror *mirror;
2723 struct ofbundle *out;
2724 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2725 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2728 mirror = mirror_lookup(ofproto, aux);
2730 mirror_destroy(mirror);
2736 idx = mirror_scan(ofproto);
2738 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2740 ofproto->up.name, MAX_MIRRORS, s->name);
2744 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2745 mirror->ofproto = ofproto;
2748 mirror->out_vlan = -1;
2749 mirror->name = NULL;
2752 if (!mirror->name || strcmp(s->name, mirror->name)) {
2754 mirror->name = xstrdup(s->name);
2757 /* Get the new configuration. */
2758 if (s->out_bundle) {
2759 out = bundle_lookup(ofproto, s->out_bundle);
2761 mirror_destroy(mirror);
2767 out_vlan = s->out_vlan;
2769 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2770 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2772 /* If the configuration has not changed, do nothing. */
2773 if (hmapx_equals(&srcs, &mirror->srcs)
2774 && hmapx_equals(&dsts, &mirror->dsts)
2775 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2776 && mirror->out == out
2777 && mirror->out_vlan == out_vlan)
2779 hmapx_destroy(&srcs);
2780 hmapx_destroy(&dsts);
2784 hmapx_swap(&srcs, &mirror->srcs);
2785 hmapx_destroy(&srcs);
2787 hmapx_swap(&dsts, &mirror->dsts);
2788 hmapx_destroy(&dsts);
2790 free(mirror->vlans);
2791 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2794 mirror->out_vlan = out_vlan;
2796 /* Update bundles. */
2797 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2798 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2799 if (hmapx_contains(&mirror->srcs, bundle)) {
2800 bundle->src_mirrors |= mirror_bit;
2802 bundle->src_mirrors &= ~mirror_bit;
2805 if (hmapx_contains(&mirror->dsts, bundle)) {
2806 bundle->dst_mirrors |= mirror_bit;
2808 bundle->dst_mirrors &= ~mirror_bit;
2811 if (mirror->out == bundle) {
2812 bundle->mirror_out |= mirror_bit;
2814 bundle->mirror_out &= ~mirror_bit;
2818 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2819 ofproto->has_mirrors = true;
2820 mac_learning_flush(ofproto->ml,
2821 &ofproto->backer->revalidate_set);
2822 mirror_update_dups(ofproto);
2828 mirror_destroy(struct ofmirror *mirror)
2830 struct ofproto_dpif *ofproto;
2831 mirror_mask_t mirror_bit;
2832 struct ofbundle *bundle;
2839 ofproto = mirror->ofproto;
2840 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2841 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2843 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2844 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2845 bundle->src_mirrors &= ~mirror_bit;
2846 bundle->dst_mirrors &= ~mirror_bit;
2847 bundle->mirror_out &= ~mirror_bit;
2850 hmapx_destroy(&mirror->srcs);
2851 hmapx_destroy(&mirror->dsts);
2852 free(mirror->vlans);
2854 ofproto->mirrors[mirror->idx] = NULL;
2858 mirror_update_dups(ofproto);
2860 ofproto->has_mirrors = false;
2861 for (i = 0; i < MAX_MIRRORS; i++) {
2862 if (ofproto->mirrors[i]) {
2863 ofproto->has_mirrors = true;
2870 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2871 uint64_t *packets, uint64_t *bytes)
2873 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2874 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2877 *packets = *bytes = UINT64_MAX;
2881 *packets = mirror->packet_count;
2882 *bytes = mirror->byte_count;
2888 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2890 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2891 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2892 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2898 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2900 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2901 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2902 return bundle && bundle->mirror_out != 0;
2906 forward_bpdu_changed(struct ofproto *ofproto_)
2908 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2909 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2913 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2916 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2917 mac_learning_set_idle_time(ofproto->ml, idle_time);
2918 mac_learning_set_max_entries(ofproto->ml, max_entries);
2923 static struct ofport_dpif *
2924 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2926 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2927 return ofport ? ofport_dpif_cast(ofport) : NULL;
2930 static struct ofport_dpif *
2931 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2933 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2934 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2938 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2939 struct ofproto_port *ofproto_port,
2940 struct dpif_port *dpif_port)
2942 ofproto_port->name = dpif_port->name;
2943 ofproto_port->type = dpif_port->type;
2944 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2947 static struct ofport_dpif *
2948 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
2950 const struct ofproto_dpif *ofproto;
2953 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
2958 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2959 struct ofport *ofport;
2961 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
2962 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
2963 return ofport_dpif_cast(ofport);
2970 port_run_fast(struct ofport_dpif *ofport)
2972 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2973 struct ofpbuf packet;
2975 ofpbuf_init(&packet, 0);
2976 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2977 send_packet(ofport, &packet);
2978 ofpbuf_uninit(&packet);
2983 port_run(struct ofport_dpif *ofport)
2985 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2986 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2987 bool enable = netdev_get_carrier(ofport->up.netdev);
2989 ofport->carrier_seq = carrier_seq;
2991 port_run_fast(ofport);
2993 if (ofport->tnl_port
2994 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
2995 &ofport->tnl_port)) {
2996 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
3000 int cfm_opup = cfm_get_opup(ofport->cfm);
3002 cfm_run(ofport->cfm);
3003 enable = enable && !cfm_get_fault(ofport->cfm);
3005 if (cfm_opup >= 0) {
3006 enable = enable && cfm_opup;
3010 if (ofport->bundle) {
3011 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3012 if (carrier_changed) {
3013 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3017 if (ofport->may_enable != enable) {
3018 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3020 if (ofproto->has_bundle_action) {
3021 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3025 ofport->may_enable = enable;
3029 port_wait(struct ofport_dpif *ofport)
3032 cfm_wait(ofport->cfm);
3037 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3038 struct ofproto_port *ofproto_port)
3040 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3041 struct dpif_port dpif_port;
3044 if (sset_contains(&ofproto->ghost_ports, devname)) {
3045 const char *type = netdev_get_type_from_name(devname);
3047 /* We may be called before ofproto->up.port_by_name is populated with
3048 * the appropriate ofport. For this reason, we must get the name and
3049 * type from the netdev layer directly. */
3051 const struct ofport *ofport;
3053 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3054 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3055 ofproto_port->name = xstrdup(devname);
3056 ofproto_port->type = xstrdup(type);
3062 if (!sset_contains(&ofproto->ports, devname)) {
3065 error = dpif_port_query_by_name(ofproto->backer->dpif,
3066 devname, &dpif_port);
3068 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3074 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3076 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3077 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3078 const char *devname = netdev_get_name(netdev);
3080 if (netdev_vport_is_patch(netdev)) {
3081 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3085 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3086 uint32_t port_no = UINT32_MAX;
3089 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3093 if (netdev_get_tunnel_config(netdev)) {
3094 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3098 if (netdev_get_tunnel_config(netdev)) {
3099 sset_add(&ofproto->ghost_ports, devname);
3101 sset_add(&ofproto->ports, devname);
3107 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3109 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3110 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3117 sset_find_and_delete(&ofproto->ghost_ports,
3118 netdev_get_name(ofport->up.netdev));
3119 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3120 if (!ofport->tnl_port) {
3121 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3123 /* The caller is going to close ofport->up.netdev. If this is a
3124 * bonded port, then the bond is using that netdev, so remove it
3125 * from the bond. The client will need to reconfigure everything
3126 * after deleting ports, so then the slave will get re-added. */
3127 bundle_remove(&ofport->up);
3134 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3136 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3139 error = netdev_get_stats(ofport->up.netdev, stats);
3141 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3142 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3144 /* ofproto->stats.tx_packets represents packets that we created
3145 * internally and sent to some port (e.g. packets sent with
3146 * send_packet()). Account for them as if they had come from
3147 * OFPP_LOCAL and got forwarded. */
3149 if (stats->rx_packets != UINT64_MAX) {
3150 stats->rx_packets += ofproto->stats.tx_packets;
3153 if (stats->rx_bytes != UINT64_MAX) {
3154 stats->rx_bytes += ofproto->stats.tx_bytes;
3157 /* ofproto->stats.rx_packets represents packets that were received on
3158 * some port and we processed internally and dropped (e.g. STP).
3159 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3161 if (stats->tx_packets != UINT64_MAX) {
3162 stats->tx_packets += ofproto->stats.rx_packets;
3165 if (stats->tx_bytes != UINT64_MAX) {
3166 stats->tx_bytes += ofproto->stats.rx_bytes;
3173 /* Account packets for LOCAL port. */
3175 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
3176 size_t tx_size, size_t rx_size)
3178 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3181 ofproto->stats.rx_packets++;
3182 ofproto->stats.rx_bytes += rx_size;
3185 ofproto->stats.tx_packets++;
3186 ofproto->stats.tx_bytes += tx_size;
3190 struct port_dump_state {
3195 struct ofproto_port port;
3200 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3202 *statep = xzalloc(sizeof(struct port_dump_state));
3207 port_dump_next(const struct ofproto *ofproto_, void *state_,
3208 struct ofproto_port *port)
3210 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3211 struct port_dump_state *state = state_;
3212 const struct sset *sset;
3213 struct sset_node *node;
3215 if (state->has_port) {
3216 ofproto_port_destroy(&state->port);
3217 state->has_port = false;
3219 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3220 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3223 error = port_query_by_name(ofproto_, node->name, &state->port);
3225 *port = state->port;
3226 state->has_port = true;
3228 } else if (error != ENODEV) {
3233 if (!state->ghost) {
3234 state->ghost = true;
3237 return port_dump_next(ofproto_, state_, port);
3244 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3246 struct port_dump_state *state = state_;
3248 if (state->has_port) {
3249 ofproto_port_destroy(&state->port);
3256 port_poll(const struct ofproto *ofproto_, char **devnamep)
3258 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3260 if (ofproto->port_poll_errno) {
3261 int error = ofproto->port_poll_errno;
3262 ofproto->port_poll_errno = 0;
3266 if (sset_is_empty(&ofproto->port_poll_set)) {
3270 *devnamep = sset_pop(&ofproto->port_poll_set);
3275 port_poll_wait(const struct ofproto *ofproto_)
3277 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3278 dpif_port_poll_wait(ofproto->backer->dpif);
3282 port_is_lacp_current(const struct ofport *ofport_)
3284 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3285 return (ofport->bundle && ofport->bundle->lacp
3286 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3290 /* Upcall handling. */
3292 /* Flow miss batching.
3294 * Some dpifs implement operations faster when you hand them off in a batch.
3295 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3296 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3297 * more packets, plus possibly installing the flow in the dpif.
3299 * So far we only batch the operations that affect flow setup time the most.
3300 * It's possible to batch more than that, but the benefit might be minimal. */
3302 struct hmap_node hmap_node;
3303 struct ofproto_dpif *ofproto;
3305 enum odp_key_fitness key_fitness;
3306 const struct nlattr *key;
3308 struct initial_vals initial_vals;
3309 struct list packets;
3310 enum dpif_upcall_type upcall_type;
3311 uint32_t odp_in_port;
3314 struct flow_miss_op {
3315 struct dpif_op dpif_op;
3316 void *garbage; /* Pointer to pass to free(), NULL if none. */
3317 uint64_t stub[1024 / 8]; /* Temporary buffer. */
3320 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3321 * OpenFlow controller as necessary according to their individual
3322 * configurations. */
3324 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3325 const struct flow *flow)
3327 struct ofputil_packet_in pin;
3329 pin.packet = packet->data;
3330 pin.packet_len = packet->size;
3331 pin.reason = OFPR_NO_MATCH;
3332 pin.controller_id = 0;
3337 pin.send_len = 0; /* not used for flow table misses */
3339 flow_get_metadata(flow, &pin.fmd);
3341 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3344 static enum slow_path_reason
3345 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3346 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3350 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3352 cfm_process_heartbeat(ofport->cfm, packet);
3355 } else if (ofport->bundle && ofport->bundle->lacp
3356 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3358 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3361 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3363 stp_process_packet(ofport, packet);
3371 static struct flow_miss *
3372 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3373 const struct flow *flow, uint32_t hash)
3375 struct flow_miss *miss;
3377 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3378 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3386 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3387 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3388 * 'miss' is associated with a subfacet the caller must also initialize the
3389 * returned op->subfacet, and if anything needs to be freed after processing
3390 * the op, the caller must initialize op->garbage also. */
3392 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3393 struct flow_miss_op *op)
3395 if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) {
3396 /* This packet was received on a VLAN splinter port. We
3397 * added a VLAN to the packet to make the packet resemble
3398 * the flow, but the actions were composed assuming that
3399 * the packet contained no VLAN. So, we must remove the
3400 * VLAN header from the packet before trying to execute the
3402 eth_pop_vlan(packet);
3406 op->dpif_op.type = DPIF_OP_EXECUTE;
3407 op->dpif_op.u.execute.key = miss->key;
3408 op->dpif_op.u.execute.key_len = miss->key_len;
3409 op->dpif_op.u.execute.packet = packet;
3412 /* Helper for handle_flow_miss_without_facet() and
3413 * handle_flow_miss_with_facet(). */
3415 handle_flow_miss_common(struct rule_dpif *rule,
3416 struct ofpbuf *packet, const struct flow *flow)
3418 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3420 ofproto->n_matches++;
3422 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3424 * Extra-special case for fail-open mode.
3426 * We are in fail-open mode and the packet matched the fail-open
3427 * rule, but we are connected to a controller too. We should send
3428 * the packet up to the controller in the hope that it will try to
3429 * set up a flow and thereby allow us to exit fail-open.
3431 * See the top-level comment in fail-open.c for more information.
3433 send_packet_in_miss(ofproto, packet, flow);
3437 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3438 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3439 * installing a datapath flow. The answer is usually "yes" (a return value of
3440 * true). However, for short flows the cost of bookkeeping is much higher than
3441 * the benefits, so when the datapath holds a large number of flows we impose
3442 * some heuristics to decide which flows are likely to be worth tracking. */
3444 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3445 struct flow_miss *miss, uint32_t hash)
3447 if (!ofproto->governor) {
3450 n_subfacets = hmap_count(&ofproto->subfacets);
3451 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3455 ofproto->governor = governor_create(ofproto->up.name);
3458 return governor_should_install_flow(ofproto->governor, hash,
3459 list_size(&miss->packets));
3462 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3463 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3464 * increment '*n_ops'. */
3466 handle_flow_miss_without_facet(struct flow_miss *miss,
3467 struct rule_dpif *rule,
3468 struct flow_miss_op *ops, size_t *n_ops)
3470 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3471 long long int now = time_msec();
3472 struct action_xlate_ctx ctx;
3473 struct ofpbuf *packet;
3475 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3476 struct flow_miss_op *op = &ops[*n_ops];
3477 struct dpif_flow_stats stats;
3478 struct ofpbuf odp_actions;
3480 COVERAGE_INC(facet_suppress);
3482 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3484 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3485 rule_credit_stats(rule, &stats);
3487 action_xlate_ctx_init(&ctx, ofproto, &miss->flow,
3488 &miss->initial_vals, rule, 0, packet);
3489 ctx.resubmit_stats = &stats;
3490 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3493 if (odp_actions.size) {
3494 struct dpif_execute *execute = &op->dpif_op.u.execute;
3496 init_flow_miss_execute_op(miss, packet, op);
3497 execute->actions = odp_actions.data;
3498 execute->actions_len = odp_actions.size;
3499 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3503 ofpbuf_uninit(&odp_actions);
3508 /* Handles 'miss', which matches 'facet'. May add any required datapath
3509 * operations to 'ops', incrementing '*n_ops' for each new op.
3511 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3512 * This is really important only for new facets: if we just called time_msec()
3513 * here, then the new subfacet or its packets could look (occasionally) as
3514 * though it was used some time after the facet was used. That can make a
3515 * one-packet flow look like it has a nonzero duration, which looks odd in
3516 * e.g. NetFlow statistics. */
3518 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3520 struct flow_miss_op *ops, size_t *n_ops)
3522 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3523 enum subfacet_path want_path;
3524 struct subfacet *subfacet;
3525 struct ofpbuf *packet;
3527 subfacet = subfacet_create(facet, miss, now);
3529 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3530 struct flow_miss_op *op = &ops[*n_ops];
3531 struct dpif_flow_stats stats;
3532 struct ofpbuf odp_actions;
3534 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3536 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3537 if (!subfacet->actions || subfacet->slow) {
3538 subfacet_make_actions(subfacet, packet, &odp_actions);
3541 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3542 subfacet_update_stats(subfacet, &stats);
3544 if (subfacet->actions_len) {
3545 struct dpif_execute *execute = &op->dpif_op.u.execute;
3547 init_flow_miss_execute_op(miss, packet, op);
3548 if (!subfacet->slow) {
3549 execute->actions = subfacet->actions;
3550 execute->actions_len = subfacet->actions_len;
3551 ofpbuf_uninit(&odp_actions);
3553 execute->actions = odp_actions.data;
3554 execute->actions_len = odp_actions.size;
3555 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3560 ofpbuf_uninit(&odp_actions);
3564 want_path = subfacet_want_path(subfacet->slow);
3565 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3566 struct flow_miss_op *op = &ops[(*n_ops)++];
3567 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3569 subfacet->path = want_path;
3572 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3573 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3574 put->key = miss->key;
3575 put->key_len = miss->key_len;
3576 if (want_path == SF_FAST_PATH) {
3577 put->actions = subfacet->actions;
3578 put->actions_len = subfacet->actions_len;
3580 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
3581 op->stub, sizeof op->stub,
3582 &put->actions, &put->actions_len);
3588 /* Handles flow miss 'miss'. May add any required datapath operations
3589 * to 'ops', incrementing '*n_ops' for each new op. */
3591 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3594 struct ofproto_dpif *ofproto = miss->ofproto;
3595 struct facet *facet;
3599 /* The caller must ensure that miss->hmap_node.hash contains
3600 * flow_hash(miss->flow, 0). */
3601 hash = miss->hmap_node.hash;
3603 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3605 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
3607 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
3608 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
3612 facet = facet_create(rule, &miss->flow, hash);
3617 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3620 static struct drop_key *
3621 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3624 struct drop_key *drop_key;
3626 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3627 &backer->drop_keys) {
3628 if (drop_key->key_len == key_len
3629 && !memcmp(drop_key->key, key, key_len)) {
3637 drop_key_clear(struct dpif_backer *backer)
3639 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3640 struct drop_key *drop_key, *next;
3642 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3645 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3647 if (error && !VLOG_DROP_WARN(&rl)) {
3648 struct ds ds = DS_EMPTY_INITIALIZER;
3649 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3650 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3655 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3656 free(drop_key->key);
3661 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3662 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3663 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3664 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3665 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3666 * 'packet' ingressed.
3668 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3669 * 'flow''s in_port to OFPP_NONE.
3671 * This function does post-processing on data returned from
3672 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3673 * of the upcall processing logic. In particular, if the extracted in_port is
3674 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3675 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3676 * a VLAN header onto 'packet' (if it is nonnull).
3678 * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci'
3679 * to the VLAN TCI with which the packet was really received, that is, the
3680 * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from
3681 * the value returned in flow->vlan_tci only for packets received on
3682 * VLAN splinters.) Also, if received on an IP tunnel, sets
3683 * 'initial_vals->tunnel_ip_tos' to the tunnel's IP TOS.
3685 * Similarly, this function also includes some logic to help with tunnels. It
3686 * may modify 'flow' as necessary to make the tunneling implementation
3687 * transparent to the upcall processing logic.
3689 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3690 * or some other positive errno if there are other problems. */
3692 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3693 const struct nlattr *key, size_t key_len,
3694 struct flow *flow, enum odp_key_fitness *fitnessp,
3695 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3696 struct initial_vals *initial_vals)
3698 const struct ofport_dpif *port;
3699 enum odp_key_fitness fitness;
3702 fitness = odp_flow_key_to_flow(key, key_len, flow);
3703 if (fitness == ODP_FIT_ERROR) {
3709 initial_vals->vlan_tci = flow->vlan_tci;
3710 initial_vals->tunnel_ip_tos = flow->tunnel.ip_tos;
3714 *odp_in_port = flow->in_port;
3717 if (tnl_port_should_receive(flow)) {
3718 const struct ofport *ofport = tnl_port_receive(flow);
3720 flow->in_port = OFPP_NONE;
3723 port = ofport_dpif_cast(ofport);
3725 /* We can't reproduce 'key' from 'flow'. */
3726 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3728 /* XXX: Since the tunnel module is not scoped per backer, it's
3729 * theoretically possible that we'll receive an ofport belonging to an
3730 * entirely different datapath. In practice, this can't happen because
3731 * no platforms has two separate datapaths which each support
3733 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3735 port = odp_port_to_ofport(backer, flow->in_port);
3737 flow->in_port = OFPP_NONE;
3741 flow->in_port = port->up.ofp_port;
3742 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3744 /* Make the packet resemble the flow, so that it gets sent to
3745 * an OpenFlow controller properly, so that it looks correct
3746 * for sFlow, and so that flow_extract() will get the correct
3747 * vlan_tci if it is called on 'packet'.
3749 * The allocated space inside 'packet' probably also contains
3750 * 'key', that is, both 'packet' and 'key' are probably part of
3751 * a struct dpif_upcall (see the large comment on that
3752 * structure definition), so pushing data on 'packet' is in
3753 * general not a good idea since it could overwrite 'key' or
3754 * free it as a side effect. However, it's OK in this special
3755 * case because we know that 'packet' is inside a Netlink
3756 * attribute: pushing 4 bytes will just overwrite the 4-byte
3757 * "struct nlattr", which is fine since we don't need that
3758 * header anymore. */
3759 eth_push_vlan(packet, flow->vlan_tci);
3761 /* We can't reproduce 'key' from 'flow'. */
3762 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3768 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3773 *fitnessp = fitness;
3779 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3782 struct dpif_upcall *upcall;
3783 struct flow_miss *miss;
3784 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3785 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3786 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3796 /* Construct the to-do list.
3798 * This just amounts to extracting the flow from each packet and sticking
3799 * the packets that have the same flow in the same "flow_miss" structure so
3800 * that we can process them together. */
3803 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3804 struct flow_miss *miss = &misses[n_misses];
3805 struct flow_miss *existing_miss;
3806 struct ofproto_dpif *ofproto;
3807 uint32_t odp_in_port;
3812 error = ofproto_receive(backer, upcall->packet, upcall->key,
3813 upcall->key_len, &flow, &miss->key_fitness,
3814 &ofproto, &odp_in_port, &miss->initial_vals);
3815 if (error == ENODEV) {
3816 struct drop_key *drop_key;
3818 /* Received packet on port for which we couldn't associate
3819 * an ofproto. This can happen if a port is removed while
3820 * traffic is being received. Print a rate-limited message
3821 * in case it happens frequently. Install a drop flow so
3822 * that future packets of the flow are inexpensively dropped
3824 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
3827 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3829 drop_key = xmalloc(sizeof *drop_key);
3830 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3831 drop_key->key_len = upcall->key_len;
3833 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3834 hash_bytes(drop_key->key, drop_key->key_len, 0));
3835 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3836 drop_key->key, drop_key->key_len, NULL, 0, NULL);
3843 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3844 &flow.tunnel, flow.in_port, &miss->flow);
3846 /* Add other packets to a to-do list. */
3847 hash = flow_hash(&miss->flow, 0);
3848 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
3849 if (!existing_miss) {
3850 hmap_insert(&todo, &miss->hmap_node, hash);
3851 miss->ofproto = ofproto;
3852 miss->key = upcall->key;
3853 miss->key_len = upcall->key_len;
3854 miss->upcall_type = upcall->type;
3855 miss->odp_in_port = odp_in_port;
3856 list_init(&miss->packets);
3860 miss = existing_miss;
3862 list_push_back(&miss->packets, &upcall->packet->list_node);
3865 /* Process each element in the to-do list, constructing the set of
3866 * operations to batch. */
3868 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3869 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3871 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3873 /* Execute batch. */
3874 for (i = 0; i < n_ops; i++) {
3875 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3877 dpif_operate(backer->dpif, dpif_ops, n_ops);
3880 for (i = 0; i < n_ops; i++) {
3881 free(flow_miss_ops[i].garbage);
3883 hmap_destroy(&todo);
3886 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
3887 classify_upcall(const struct dpif_upcall *upcall)
3889 union user_action_cookie cookie;
3891 /* First look at the upcall type. */
3892 switch (upcall->type) {
3893 case DPIF_UC_ACTION:
3899 case DPIF_N_UC_TYPES:
3901 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3905 /* "action" upcalls need a closer look. */
3906 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3907 switch (cookie.type) {
3908 case USER_ACTION_COOKIE_SFLOW:
3909 return SFLOW_UPCALL;
3911 case USER_ACTION_COOKIE_SLOW_PATH:
3914 case USER_ACTION_COOKIE_UNSPEC:
3916 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
3922 handle_sflow_upcall(struct dpif_backer *backer,
3923 const struct dpif_upcall *upcall)
3925 struct ofproto_dpif *ofproto;
3926 union user_action_cookie cookie;
3928 uint32_t odp_in_port;
3930 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3931 &flow, NULL, &ofproto, &odp_in_port, NULL)
3932 || !ofproto->sflow) {
3936 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3937 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3938 odp_in_port, &cookie);
3942 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3944 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3945 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3946 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3951 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
3954 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3955 struct dpif_upcall *upcall = &misses[n_misses];
3956 struct ofpbuf *buf = &miss_bufs[n_misses];
3959 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3960 sizeof miss_buf_stubs[n_misses]);
3961 error = dpif_recv(backer->dpif, upcall, buf);
3967 switch (classify_upcall(upcall)) {
3969 /* Handle it later. */
3974 handle_sflow_upcall(backer, upcall);
3984 /* Handle deferred MISS_UPCALL processing. */
3985 handle_miss_upcalls(backer, misses, n_misses);
3986 for (i = 0; i < n_misses; i++) {
3987 ofpbuf_uninit(&miss_bufs[i]);
3993 /* Flow expiration. */
3995 static int subfacet_max_idle(const struct ofproto_dpif *);
3996 static void update_stats(struct dpif_backer *);
3997 static void rule_expire(struct rule_dpif *);
3998 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
4000 /* This function is called periodically by run(). Its job is to collect
4001 * updates for the flows that have been installed into the datapath, most
4002 * importantly when they last were used, and then use that information to
4003 * expire flows that have not been used recently.
4005 * Returns the number of milliseconds after which it should be called again. */
4007 expire(struct dpif_backer *backer)
4009 struct ofproto_dpif *ofproto;
4010 int max_idle = INT32_MAX;
4012 /* Periodically clear out the drop keys in an effort to keep them
4013 * relatively few. */
4014 drop_key_clear(backer);
4016 /* Update stats for each flow in the backer. */
4017 update_stats(backer);
4019 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4020 struct rule *rule, *next_rule;
4023 if (ofproto->backer != backer) {
4027 /* Expire subfacets that have been idle too long. */
4028 dp_max_idle = subfacet_max_idle(ofproto);
4029 expire_subfacets(ofproto, dp_max_idle);
4031 max_idle = MIN(max_idle, dp_max_idle);
4033 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4035 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4036 &ofproto->up.expirable) {
4037 rule_expire(rule_dpif_cast(rule));
4040 /* All outstanding data in existing flows has been accounted, so it's a
4041 * good time to do bond rebalancing. */
4042 if (ofproto->has_bonded_bundles) {
4043 struct ofbundle *bundle;
4045 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4047 bond_rebalance(bundle->bond, &backer->revalidate_set);
4053 return MIN(max_idle, 1000);
4056 /* Updates flow table statistics given that the datapath just reported 'stats'
4057 * as 'subfacet''s statistics. */
4059 update_subfacet_stats(struct subfacet *subfacet,
4060 const struct dpif_flow_stats *stats)
4062 struct facet *facet = subfacet->facet;
4064 if (stats->n_packets >= subfacet->dp_packet_count) {
4065 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
4066 facet->packet_count += extra;
4068 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4071 if (stats->n_bytes >= subfacet->dp_byte_count) {
4072 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
4074 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4077 subfacet->dp_packet_count = stats->n_packets;
4078 subfacet->dp_byte_count = stats->n_bytes;
4080 facet->tcp_flags |= stats->tcp_flags;
4082 subfacet_update_time(subfacet, stats->used);
4083 if (facet->accounted_bytes < facet->byte_count) {
4085 facet_account(facet);
4086 facet->accounted_bytes = facet->byte_count;
4088 facet_push_stats(facet);
4091 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4092 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4094 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4095 const struct nlattr *key, size_t key_len)
4097 if (!VLOG_DROP_WARN(&rl)) {
4101 odp_flow_key_format(key, key_len, &s);
4102 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4106 COVERAGE_INC(facet_unexpected);
4107 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4110 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4112 * This function also pushes statistics updates to rules which each facet
4113 * resubmits into. Generally these statistics will be accurate. However, if a
4114 * facet changes the rule it resubmits into at some time in between
4115 * update_stats() runs, it is possible that statistics accrued to the
4116 * old rule will be incorrectly attributed to the new rule. This could be
4117 * avoided by calling update_stats() whenever rules are created or
4118 * deleted. However, the performance impact of making so many calls to the
4119 * datapath do not justify the benefit of having perfectly accurate statistics.
4122 update_stats(struct dpif_backer *backer)
4124 const struct dpif_flow_stats *stats;
4125 struct dpif_flow_dump dump;
4126 const struct nlattr *key;
4129 dpif_flow_dump_start(&dump, backer->dpif);
4130 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4132 struct subfacet *subfacet;
4133 struct ofproto_dpif *ofproto;
4134 struct ofport_dpif *ofport;
4137 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4142 ofport = get_ofp_port(ofproto, flow.in_port);
4143 if (ofport && ofport->tnl_port) {
4144 netdev_vport_inc_rx(ofport->up.netdev, stats);
4147 key_hash = odp_flow_key_hash(key, key_len);
4148 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4149 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4151 update_subfacet_stats(subfacet, stats);
4155 /* Stats are updated per-packet. */
4158 case SF_NOT_INSTALLED:
4160 delete_unexpected_flow(ofproto, key, key_len);
4164 dpif_flow_dump_done(&dump);
4167 /* Calculates and returns the number of milliseconds of idle time after which
4168 * subfacets should expire from the datapath. When a subfacet expires, we fold
4169 * its statistics into its facet, and when a facet's last subfacet expires, we
4170 * fold its statistic into its rule. */
4172 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4175 * Idle time histogram.
4177 * Most of the time a switch has a relatively small number of subfacets.
4178 * When this is the case we might as well keep statistics for all of them
4179 * in userspace and to cache them in the kernel datapath for performance as
4182 * As the number of subfacets increases, the memory required to maintain
4183 * statistics about them in userspace and in the kernel becomes
4184 * significant. However, with a large number of subfacets it is likely
4185 * that only a few of them are "heavy hitters" that consume a large amount
4186 * of bandwidth. At this point, only heavy hitters are worth caching in
4187 * the kernel and maintaining in userspaces; other subfacets we can
4190 * The technique used to compute the idle time is to build a histogram with
4191 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4192 * that is installed in the kernel gets dropped in the appropriate bucket.
4193 * After the histogram has been built, we compute the cutoff so that only
4194 * the most-recently-used 1% of subfacets (but at least
4195 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4196 * the most-recently-used bucket of subfacets is kept, so actually an
4197 * arbitrary number of subfacets can be kept in any given expiration run
4198 * (though the next run will delete most of those unless they receive
4201 * This requires a second pass through the subfacets, in addition to the
4202 * pass made by update_stats(), because the former function never looks at
4203 * uninstallable subfacets.
4205 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4206 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4207 int buckets[N_BUCKETS] = { 0 };
4208 int total, subtotal, bucket;
4209 struct subfacet *subfacet;
4213 total = hmap_count(&ofproto->subfacets);
4214 if (total <= ofproto->up.flow_eviction_threshold) {
4215 return N_BUCKETS * BUCKET_WIDTH;
4218 /* Build histogram. */
4220 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4221 long long int idle = now - subfacet->used;
4222 int bucket = (idle <= 0 ? 0
4223 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4224 : (unsigned int) idle / BUCKET_WIDTH);
4228 /* Find the first bucket whose flows should be expired. */
4229 subtotal = bucket = 0;
4231 subtotal += buckets[bucket++];
4232 } while (bucket < N_BUCKETS &&
4233 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4235 if (VLOG_IS_DBG_ENABLED()) {
4239 ds_put_cstr(&s, "keep");
4240 for (i = 0; i < N_BUCKETS; i++) {
4242 ds_put_cstr(&s, ", drop");
4245 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4248 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4252 return bucket * BUCKET_WIDTH;
4256 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4258 /* Cutoff time for most flows. */
4259 long long int normal_cutoff = time_msec() - dp_max_idle;
4261 /* We really want to keep flows for special protocols around, so use a more
4262 * conservative cutoff. */
4263 long long int special_cutoff = time_msec() - 10000;
4265 struct subfacet *subfacet, *next_subfacet;
4266 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4270 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4271 &ofproto->subfacets) {
4272 long long int cutoff;
4274 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
4277 if (subfacet->used < cutoff) {
4278 if (subfacet->path != SF_NOT_INSTALLED) {
4279 batch[n_batch++] = subfacet;
4280 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4281 subfacet_destroy_batch(ofproto, batch, n_batch);
4285 subfacet_destroy(subfacet);
4291 subfacet_destroy_batch(ofproto, batch, n_batch);
4295 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4296 * then delete it entirely. */
4298 rule_expire(struct rule_dpif *rule)
4300 struct facet *facet, *next_facet;
4304 if (rule->up.pending) {
4305 /* We'll have to expire it later. */
4309 /* Has 'rule' expired? */
4311 if (rule->up.hard_timeout
4312 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4313 reason = OFPRR_HARD_TIMEOUT;
4314 } else if (rule->up.idle_timeout
4315 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4316 reason = OFPRR_IDLE_TIMEOUT;
4321 COVERAGE_INC(ofproto_dpif_expired);
4323 /* Update stats. (This is a no-op if the rule expired due to an idle
4324 * timeout, because that only happens when the rule has no facets left.) */
4325 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4326 facet_remove(facet);
4329 /* Get rid of the rule. */
4330 ofproto_rule_expire(&rule->up, reason);
4335 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
4337 * The caller must already have determined that no facet with an identical
4338 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
4339 * the ofproto's classifier table.
4341 * 'hash' must be the return value of flow_hash(flow, 0).
4343 * The facet will initially have no subfacets. The caller should create (at
4344 * least) one subfacet with subfacet_create(). */
4345 static struct facet *
4346 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
4348 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4349 struct facet *facet;
4351 facet = xzalloc(sizeof *facet);
4352 facet->used = time_msec();
4353 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4354 list_push_back(&rule->facets, &facet->list_node);
4356 facet->flow = *flow;
4357 list_init(&facet->subfacets);
4358 netflow_flow_init(&facet->nf_flow);
4359 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4365 facet_free(struct facet *facet)
4370 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4371 * 'packet', which arrived on 'in_port'. */
4373 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4374 const struct nlattr *odp_actions, size_t actions_len,
4375 struct ofpbuf *packet)
4377 struct odputil_keybuf keybuf;
4381 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4382 odp_flow_key_from_flow(&key, flow,
4383 ofp_port_to_odp_port(ofproto, flow->in_port));
4385 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4386 odp_actions, actions_len, packet);
4390 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4392 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4393 * rule's statistics, via subfacet_uninstall().
4395 * - Removes 'facet' from its rule and from ofproto->facets.
4398 facet_remove(struct facet *facet)
4400 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4401 struct subfacet *subfacet, *next_subfacet;
4403 ovs_assert(!list_is_empty(&facet->subfacets));
4405 /* First uninstall all of the subfacets to get final statistics. */
4406 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4407 subfacet_uninstall(subfacet);
4410 /* Flush the final stats to the rule.
4412 * This might require us to have at least one subfacet around so that we
4413 * can use its actions for accounting in facet_account(), which is why we
4414 * have uninstalled but not yet destroyed the subfacets. */
4415 facet_flush_stats(facet);
4417 /* Now we're really all done so destroy everything. */
4418 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4419 &facet->subfacets) {
4420 subfacet_destroy__(subfacet);
4422 hmap_remove(&ofproto->facets, &facet->hmap_node);
4423 list_remove(&facet->list_node);
4427 /* Feed information from 'facet' back into the learning table to keep it in
4428 * sync with what is actually flowing through the datapath. */
4430 facet_learn(struct facet *facet)
4432 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4433 struct subfacet *subfacet= CONTAINER_OF(list_front(&facet->subfacets),
4434 struct subfacet, list_node);
4435 struct action_xlate_ctx ctx;
4437 if (!facet->has_learn
4438 && !facet->has_normal
4439 && (!facet->has_fin_timeout
4440 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4444 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4445 &subfacet->initial_vals,
4446 facet->rule, facet->tcp_flags, NULL);
4447 ctx.may_learn = true;
4448 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
4449 facet->rule->up.ofpacts_len);
4453 facet_account(struct facet *facet)
4455 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4456 struct subfacet *subfacet;
4457 const struct nlattr *a;
4462 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
4465 n_bytes = facet->byte_count - facet->accounted_bytes;
4467 /* This loop feeds byte counters to bond_account() for rebalancing to use
4468 * as a basis. We also need to track the actual VLAN on which the packet
4469 * is going to be sent to ensure that it matches the one passed to
4470 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4473 * We use the actions from an arbitrary subfacet because they should all
4474 * be equally valid for our purpose. */
4475 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
4476 struct subfacet, list_node);
4477 vlan_tci = facet->flow.vlan_tci;
4478 NL_ATTR_FOR_EACH_UNSAFE (a, left,
4479 subfacet->actions, subfacet->actions_len) {
4480 const struct ovs_action_push_vlan *vlan;
4481 struct ofport_dpif *port;
4483 switch (nl_attr_type(a)) {
4484 case OVS_ACTION_ATTR_OUTPUT:
4485 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4486 if (port && port->bundle && port->bundle->bond) {
4487 bond_account(port->bundle->bond, &facet->flow,
4488 vlan_tci_to_vid(vlan_tci), n_bytes);
4492 case OVS_ACTION_ATTR_POP_VLAN:
4493 vlan_tci = htons(0);
4496 case OVS_ACTION_ATTR_PUSH_VLAN:
4497 vlan = nl_attr_get(a);
4498 vlan_tci = vlan->vlan_tci;
4504 /* Returns true if the only action for 'facet' is to send to the controller.
4505 * (We don't report NetFlow expiration messages for such facets because they
4506 * are just part of the control logic for the network, not real traffic). */
4508 facet_is_controller_flow(struct facet *facet)
4511 const struct rule *rule = &facet->rule->up;
4512 const struct ofpact *ofpacts = rule->ofpacts;
4513 size_t ofpacts_len = rule->ofpacts_len;
4515 if (ofpacts_len > 0 &&
4516 ofpacts->type == OFPACT_CONTROLLER &&
4517 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4524 /* Folds all of 'facet''s statistics into its rule. Also updates the
4525 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4526 * 'facet''s statistics in the datapath should have been zeroed and folded into
4527 * its packet and byte counts before this function is called. */
4529 facet_flush_stats(struct facet *facet)
4531 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4532 struct subfacet *subfacet;
4534 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4535 ovs_assert(!subfacet->dp_byte_count);
4536 ovs_assert(!subfacet->dp_packet_count);
4539 facet_push_stats(facet);
4540 if (facet->accounted_bytes < facet->byte_count) {
4541 facet_account(facet);
4542 facet->accounted_bytes = facet->byte_count;
4545 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4546 struct ofexpired expired;
4547 expired.flow = facet->flow;
4548 expired.packet_count = facet->packet_count;
4549 expired.byte_count = facet->byte_count;
4550 expired.used = facet->used;
4551 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4554 facet->rule->packet_count += facet->packet_count;
4555 facet->rule->byte_count += facet->byte_count;
4557 /* Reset counters to prevent double counting if 'facet' ever gets
4559 facet_reset_counters(facet);
4561 netflow_flow_clear(&facet->nf_flow);
4562 facet->tcp_flags = 0;
4565 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4566 * Returns it if found, otherwise a null pointer.
4568 * 'hash' must be the return value of flow_hash(flow, 0).
4570 * The returned facet might need revalidation; use facet_lookup_valid()
4571 * instead if that is important. */
4572 static struct facet *
4573 facet_find(struct ofproto_dpif *ofproto,
4574 const struct flow *flow, uint32_t hash)
4576 struct facet *facet;
4578 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4579 if (flow_equal(flow, &facet->flow)) {
4587 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4588 * Returns it if found, otherwise a null pointer.
4590 * 'hash' must be the return value of flow_hash(flow, 0).
4592 * The returned facet is guaranteed to be valid. */
4593 static struct facet *
4594 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4597 struct facet *facet;
4599 facet = facet_find(ofproto, flow, hash);
4601 && (ofproto->backer->need_revalidate
4602 || tag_set_intersects(&ofproto->backer->revalidate_set,
4604 facet_revalidate(facet);
4606 /* facet_revalidate() may have destroyed 'facet'. */
4607 facet = facet_find(ofproto, flow, hash);
4614 subfacet_path_to_string(enum subfacet_path path)
4617 case SF_NOT_INSTALLED:
4618 return "not installed";
4620 return "in fast path";
4622 return "in slow path";
4628 /* Returns the path in which a subfacet should be installed if its 'slow'
4629 * member has the specified value. */
4630 static enum subfacet_path
4631 subfacet_want_path(enum slow_path_reason slow)
4633 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
4636 /* Returns true if 'subfacet' needs to have its datapath flow updated,
4637 * supposing that its actions have been recalculated as 'want_actions' and that
4638 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
4640 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
4641 const struct ofpbuf *want_actions)
4643 enum subfacet_path want_path = subfacet_want_path(slow);
4644 return (want_path != subfacet->path
4645 || (want_path == SF_FAST_PATH
4646 && (subfacet->actions_len != want_actions->size
4647 || memcmp(subfacet->actions, want_actions->data,
4648 subfacet->actions_len))));
4652 facet_check_consistency(struct facet *facet)
4654 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4656 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4658 uint64_t odp_actions_stub[1024 / 8];
4659 struct ofpbuf odp_actions;
4661 struct rule_dpif *rule;
4662 struct subfacet *subfacet;
4663 bool may_log = false;
4666 /* Check the rule for consistency. */
4667 rule = rule_dpif_lookup(ofproto, &facet->flow);
4668 ok = rule == facet->rule;
4670 may_log = !VLOG_DROP_WARN(&rl);
4675 flow_format(&s, &facet->flow);
4676 ds_put_format(&s, ": facet associated with wrong rule (was "
4677 "table=%"PRIu8",", facet->rule->up.table_id);
4678 cls_rule_format(&facet->rule->up.cr, &s);
4679 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4681 cls_rule_format(&rule->up.cr, &s);
4682 ds_put_char(&s, ')');
4684 VLOG_WARN("%s", ds_cstr(&s));
4689 /* Check the datapath actions for consistency. */
4690 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4691 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4692 enum subfacet_path want_path;
4693 struct action_xlate_ctx ctx;
4696 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4697 &subfacet->initial_vals, rule, 0, NULL);
4698 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
4701 if (subfacet->path == SF_NOT_INSTALLED) {
4702 /* This only happens if the datapath reported an error when we
4703 * tried to install the flow. Don't flag another error here. */
4707 want_path = subfacet_want_path(subfacet->slow);
4708 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
4709 /* The actions for slow-path flows may legitimately vary from one
4710 * packet to the next. We're done. */
4714 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
4718 /* Inconsistency! */
4720 may_log = !VLOG_DROP_WARN(&rl);
4724 /* Rate-limited, skip reporting. */
4729 odp_flow_key_format(subfacet->key, subfacet->key_len, &s);
4731 ds_put_cstr(&s, ": inconsistency in subfacet");
4732 if (want_path != subfacet->path) {
4733 enum odp_key_fitness fitness = subfacet->key_fitness;
4735 ds_put_format(&s, " (%s, fitness=%s)",
4736 subfacet_path_to_string(subfacet->path),
4737 odp_key_fitness_to_string(fitness));
4738 ds_put_format(&s, " (should have been %s)",
4739 subfacet_path_to_string(want_path));
4740 } else if (want_path == SF_FAST_PATH) {
4741 ds_put_cstr(&s, " (actions were: ");
4742 format_odp_actions(&s, subfacet->actions,
4743 subfacet->actions_len);
4744 ds_put_cstr(&s, ") (correct actions: ");
4745 format_odp_actions(&s, odp_actions.data, odp_actions.size);
4746 ds_put_char(&s, ')');
4748 ds_put_cstr(&s, " (actions: ");
4749 format_odp_actions(&s, subfacet->actions,
4750 subfacet->actions_len);
4751 ds_put_char(&s, ')');
4753 VLOG_WARN("%s", ds_cstr(&s));
4756 ofpbuf_uninit(&odp_actions);
4761 /* Re-searches the classifier for 'facet':
4763 * - If the rule found is different from 'facet''s current rule, moves
4764 * 'facet' to the new rule and recompiles its actions.
4766 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4767 * where it is and recompiles its actions anyway.
4769 * - If any of 'facet''s subfacets correspond to a new flow according to
4770 * ofproto_receive(), 'facet' is removed. */
4772 facet_revalidate(struct facet *facet)
4774 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4776 struct nlattr *odp_actions;
4779 struct actions *new_actions;
4781 struct action_xlate_ctx ctx;
4782 uint64_t odp_actions_stub[1024 / 8];
4783 struct ofpbuf odp_actions;
4785 struct rule_dpif *new_rule;
4786 struct subfacet *subfacet;
4789 COVERAGE_INC(facet_revalidate);
4791 /* Check that child subfacets still correspond to this facet. Tunnel
4792 * configuration changes could cause a subfacet's OpenFlow in_port to
4794 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4795 struct ofproto_dpif *recv_ofproto;
4796 struct flow recv_flow;
4799 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
4800 subfacet->key_len, &recv_flow, NULL,
4801 &recv_ofproto, NULL, NULL);
4803 || recv_ofproto != ofproto
4804 || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) {
4805 facet_remove(facet);
4810 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
4812 /* Calculate new datapath actions.
4814 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4815 * emit a NetFlow expiration and, if so, we need to have the old state
4816 * around to properly compose it. */
4818 /* If the datapath actions changed or the installability changed,
4819 * then we need to talk to the datapath. */
4822 memset(&ctx, 0, sizeof ctx);
4823 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4824 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4825 enum slow_path_reason slow;
4827 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4828 &subfacet->initial_vals, new_rule, 0, NULL);
4829 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
4832 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4833 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
4834 struct dpif_flow_stats stats;
4836 subfacet_install(subfacet,
4837 odp_actions.data, odp_actions.size, &stats, slow);
4838 subfacet_update_stats(subfacet, &stats);
4841 new_actions = xcalloc(list_size(&facet->subfacets),
4842 sizeof *new_actions);
4844 new_actions[i].odp_actions = xmemdup(odp_actions.data,
4846 new_actions[i].actions_len = odp_actions.size;
4851 ofpbuf_uninit(&odp_actions);
4854 facet_flush_stats(facet);
4857 /* Update 'facet' now that we've taken care of all the old state. */
4858 facet->tags = ctx.tags;
4859 facet->nf_flow.output_iface = ctx.nf_output_iface;
4860 facet->has_learn = ctx.has_learn;
4861 facet->has_normal = ctx.has_normal;
4862 facet->has_fin_timeout = ctx.has_fin_timeout;
4863 facet->mirrors = ctx.mirrors;
4866 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4867 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4869 if (new_actions && new_actions[i].odp_actions) {
4870 free(subfacet->actions);
4871 subfacet->actions = new_actions[i].odp_actions;
4872 subfacet->actions_len = new_actions[i].actions_len;
4878 if (facet->rule != new_rule) {
4879 COVERAGE_INC(facet_changed_rule);
4880 list_remove(&facet->list_node);
4881 list_push_back(&new_rule->facets, &facet->list_node);
4882 facet->rule = new_rule;
4883 facet->used = new_rule->up.created;
4884 facet->prev_used = facet->used;
4888 /* Updates 'facet''s used time. Caller is responsible for calling
4889 * facet_push_stats() to update the flows which 'facet' resubmits into. */
4891 facet_update_time(struct facet *facet, long long int used)
4893 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4894 if (used > facet->used) {
4896 ofproto_rule_update_used(&facet->rule->up, used);
4897 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
4902 facet_reset_counters(struct facet *facet)
4904 facet->packet_count = 0;
4905 facet->byte_count = 0;
4906 facet->prev_packet_count = 0;
4907 facet->prev_byte_count = 0;
4908 facet->accounted_bytes = 0;
4912 facet_push_stats(struct facet *facet)
4914 struct dpif_flow_stats stats;
4916 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4917 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4918 ovs_assert(facet->used >= facet->prev_used);
4920 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4921 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4922 stats.used = facet->used;
4923 stats.tcp_flags = 0;
4925 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
4926 facet->prev_packet_count = facet->packet_count;
4927 facet->prev_byte_count = facet->byte_count;
4928 facet->prev_used = facet->used;
4930 flow_push_stats(facet, &stats);
4932 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
4933 facet->mirrors, stats.n_packets, stats.n_bytes);
4938 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4940 rule->packet_count += stats->n_packets;
4941 rule->byte_count += stats->n_bytes;
4942 ofproto_rule_update_used(&rule->up, stats->used);
4945 /* Pushes flow statistics to the rules which 'facet->flow' resubmits
4946 * into given 'facet->rule''s actions and mirrors. */
4948 flow_push_stats(struct facet *facet, const struct dpif_flow_stats *stats)
4950 struct rule_dpif *rule = facet->rule;
4951 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4952 struct subfacet *subfacet = CONTAINER_OF(list_front(&facet->subfacets),
4953 struct subfacet, list_node);
4954 struct action_xlate_ctx ctx;
4956 ofproto_rule_update_used(&rule->up, stats->used);
4958 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4959 &subfacet->initial_vals, rule, 0, NULL);
4960 ctx.resubmit_stats = stats;
4961 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
4962 rule->up.ofpacts_len);
4967 static struct subfacet *
4968 subfacet_find(struct ofproto_dpif *ofproto,
4969 const struct nlattr *key, size_t key_len, uint32_t key_hash)
4971 struct subfacet *subfacet;
4973 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4974 &ofproto->subfacets) {
4975 if (subfacet->key_len == key_len
4976 && !memcmp(key, subfacet->key, key_len)) {
4984 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4985 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4986 * existing subfacet if there is one, otherwise creates and returns a
4989 * If the returned subfacet is new, then subfacet->actions will be NULL, in
4990 * which case the caller must populate the actions with
4991 * subfacet_make_actions(). */
4992 static struct subfacet *
4993 subfacet_create(struct facet *facet, struct flow_miss *miss,
4996 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4997 enum odp_key_fitness key_fitness = miss->key_fitness;
4998 const struct nlattr *key = miss->key;
4999 size_t key_len = miss->key_len;
5001 struct subfacet *subfacet;
5003 key_hash = odp_flow_key_hash(key, key_len);
5005 if (list_is_empty(&facet->subfacets)) {
5006 subfacet = &facet->one_subfacet;
5008 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
5010 if (subfacet->facet == facet) {
5014 /* This shouldn't happen. */
5015 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
5016 subfacet_destroy(subfacet);
5019 subfacet = xmalloc(sizeof *subfacet);
5022 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
5023 list_push_back(&facet->subfacets, &subfacet->list_node);
5024 subfacet->facet = facet;
5025 subfacet->key_fitness = key_fitness;
5026 subfacet->key = xmemdup(key, key_len);
5027 subfacet->key_len = key_len;
5028 subfacet->used = now;
5029 subfacet->dp_packet_count = 0;
5030 subfacet->dp_byte_count = 0;
5031 subfacet->actions_len = 0;
5032 subfacet->actions = NULL;
5033 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
5036 subfacet->path = SF_NOT_INSTALLED;
5037 subfacet->initial_vals = miss->initial_vals;
5038 subfacet->odp_in_port = miss->odp_in_port;
5043 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5044 * its facet within 'ofproto', and frees it. */
5046 subfacet_destroy__(struct subfacet *subfacet)
5048 struct facet *facet = subfacet->facet;
5049 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5051 subfacet_uninstall(subfacet);
5052 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5053 list_remove(&subfacet->list_node);
5054 free(subfacet->key);
5055 free(subfacet->actions);
5056 if (subfacet != &facet->one_subfacet) {
5061 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5062 * last remaining subfacet in its facet destroys the facet too. */
5064 subfacet_destroy(struct subfacet *subfacet)
5066 struct facet *facet = subfacet->facet;
5068 if (list_is_singleton(&facet->subfacets)) {
5069 /* facet_remove() needs at least one subfacet (it will remove it). */
5070 facet_remove(facet);
5072 subfacet_destroy__(subfacet);
5077 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5078 struct subfacet **subfacets, int n)
5080 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5081 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5082 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5085 for (i = 0; i < n; i++) {
5086 ops[i].type = DPIF_OP_FLOW_DEL;
5087 ops[i].u.flow_del.key = subfacets[i]->key;
5088 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5089 ops[i].u.flow_del.stats = &stats[i];
5093 dpif_operate(ofproto->backer->dpif, opsp, n);
5094 for (i = 0; i < n; i++) {
5095 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5096 subfacets[i]->path = SF_NOT_INSTALLED;
5097 subfacet_destroy(subfacets[i]);
5101 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
5102 * Translates the actions into 'odp_actions', which the caller must have
5103 * initialized and is responsible for uninitializing. */
5105 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
5106 struct ofpbuf *odp_actions)
5108 struct facet *facet = subfacet->facet;
5109 struct rule_dpif *rule = facet->rule;
5110 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5112 struct action_xlate_ctx ctx;
5114 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5115 &subfacet->initial_vals, rule, 0, packet);
5116 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
5117 facet->tags = ctx.tags;
5118 facet->has_learn = ctx.has_learn;
5119 facet->has_normal = ctx.has_normal;
5120 facet->has_fin_timeout = ctx.has_fin_timeout;
5121 facet->nf_flow.output_iface = ctx.nf_output_iface;
5122 facet->mirrors = ctx.mirrors;
5124 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
5125 if (subfacet->actions_len != odp_actions->size
5126 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
5127 free(subfacet->actions);
5128 subfacet->actions_len = odp_actions->size;
5129 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
5133 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5134 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5135 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5136 * since 'subfacet' was last updated.
5138 * Returns 0 if successful, otherwise a positive errno value. */
5140 subfacet_install(struct subfacet *subfacet,
5141 const struct nlattr *actions, size_t actions_len,
5142 struct dpif_flow_stats *stats,
5143 enum slow_path_reason slow)
5145 struct facet *facet = subfacet->facet;
5146 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5147 enum subfacet_path path = subfacet_want_path(slow);
5148 uint64_t slow_path_stub[128 / 8];
5149 enum dpif_flow_put_flags flags;
5152 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5154 flags |= DPIF_FP_ZERO_STATS;
5157 if (path == SF_SLOW_PATH) {
5158 compose_slow_path(ofproto, &facet->flow, slow,
5159 slow_path_stub, sizeof slow_path_stub,
5160 &actions, &actions_len);
5163 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5164 subfacet->key_len, actions, actions_len, stats);
5167 subfacet_reset_dp_stats(subfacet, stats);
5171 subfacet->path = path;
5177 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
5179 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
5180 stats, subfacet->slow);
5183 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5185 subfacet_uninstall(struct subfacet *subfacet)
5187 if (subfacet->path != SF_NOT_INSTALLED) {
5188 struct rule_dpif *rule = subfacet->facet->rule;
5189 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5190 struct dpif_flow_stats stats;
5193 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5194 subfacet->key_len, &stats);
5195 subfacet_reset_dp_stats(subfacet, &stats);
5197 subfacet_update_stats(subfacet, &stats);
5199 subfacet->path = SF_NOT_INSTALLED;
5201 ovs_assert(subfacet->dp_packet_count == 0);
5202 ovs_assert(subfacet->dp_byte_count == 0);
5206 /* Resets 'subfacet''s datapath statistics counters. This should be called
5207 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5208 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5209 * was reset in the datapath. 'stats' will be modified to include only
5210 * statistics new since 'subfacet' was last updated. */
5212 subfacet_reset_dp_stats(struct subfacet *subfacet,
5213 struct dpif_flow_stats *stats)
5216 && subfacet->dp_packet_count <= stats->n_packets
5217 && subfacet->dp_byte_count <= stats->n_bytes) {
5218 stats->n_packets -= subfacet->dp_packet_count;
5219 stats->n_bytes -= subfacet->dp_byte_count;
5222 subfacet->dp_packet_count = 0;
5223 subfacet->dp_byte_count = 0;
5226 /* Updates 'subfacet''s used time. The caller is responsible for calling
5227 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
5229 subfacet_update_time(struct subfacet *subfacet, long long int used)
5231 if (used > subfacet->used) {
5232 subfacet->used = used;
5233 facet_update_time(subfacet->facet, used);
5237 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5239 * Because of the meaning of a subfacet's counters, it only makes sense to do
5240 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5241 * represents a packet that was sent by hand or if it represents statistics
5242 * that have been cleared out of the datapath. */
5244 subfacet_update_stats(struct subfacet *subfacet,
5245 const struct dpif_flow_stats *stats)
5247 if (stats->n_packets || stats->used > subfacet->used) {
5248 struct facet *facet = subfacet->facet;
5250 subfacet_update_time(subfacet, stats->used);
5251 facet->packet_count += stats->n_packets;
5252 facet->byte_count += stats->n_bytes;
5253 facet->tcp_flags |= stats->tcp_flags;
5254 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
5260 static struct rule_dpif *
5261 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5263 struct rule_dpif *rule;
5265 rule = rule_dpif_lookup__(ofproto, flow, 0);
5270 return rule_dpif_miss_rule(ofproto, flow);
5273 static struct rule_dpif *
5274 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5277 struct cls_rule *cls_rule;
5278 struct classifier *cls;
5280 if (table_id >= N_TABLES) {
5284 cls = &ofproto->up.tables[table_id].cls;
5285 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5286 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5287 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5288 * are unavailable. */
5289 struct flow ofpc_normal_flow = *flow;
5290 ofpc_normal_flow.tp_src = htons(0);
5291 ofpc_normal_flow.tp_dst = htons(0);
5292 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5294 cls_rule = classifier_lookup(cls, flow);
5296 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5299 static struct rule_dpif *
5300 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5302 struct ofport_dpif *port;
5304 port = get_ofp_port(ofproto, flow->in_port);
5306 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5307 return ofproto->miss_rule;
5310 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5311 return ofproto->no_packet_in_rule;
5313 return ofproto->miss_rule;
5317 complete_operation(struct rule_dpif *rule)
5319 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5321 rule_invalidate(rule);
5323 struct dpif_completion *c = xmalloc(sizeof *c);
5324 c->op = rule->up.pending;
5325 list_push_back(&ofproto->completions, &c->list_node);
5327 ofoperation_complete(rule->up.pending, 0);
5331 static struct rule *
5334 struct rule_dpif *rule = xmalloc(sizeof *rule);
5339 rule_dealloc(struct rule *rule_)
5341 struct rule_dpif *rule = rule_dpif_cast(rule_);
5346 rule_construct(struct rule *rule_)
5348 struct rule_dpif *rule = rule_dpif_cast(rule_);
5349 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5350 struct rule_dpif *victim;
5353 rule->packet_count = 0;
5354 rule->byte_count = 0;
5356 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5357 if (victim && !list_is_empty(&victim->facets)) {
5358 struct facet *facet;
5360 rule->facets = victim->facets;
5361 list_moved(&rule->facets);
5362 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5363 /* XXX: We're only clearing our local counters here. It's possible
5364 * that quite a few packets are unaccounted for in the datapath
5365 * statistics. These will be accounted to the new rule instead of
5366 * cleared as required. This could be fixed by clearing out the
5367 * datapath statistics for this facet, but currently it doesn't
5369 facet_reset_counters(facet);
5373 /* Must avoid list_moved() in this case. */
5374 list_init(&rule->facets);
5377 table_id = rule->up.table_id;
5379 rule->tag = victim->tag;
5380 } else if (table_id == 0) {
5385 miniflow_expand(&rule->up.cr.match.flow, &flow);
5386 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5387 ofproto->tables[table_id].basis);
5390 complete_operation(rule);
5395 rule_destruct(struct rule *rule_)
5397 struct rule_dpif *rule = rule_dpif_cast(rule_);
5398 struct facet *facet, *next_facet;
5400 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5401 facet_revalidate(facet);
5404 complete_operation(rule);
5408 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5410 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule_->ofproto);
5411 struct rule_dpif *rule = rule_dpif_cast(rule_);
5412 struct facet *facet;
5414 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5415 facet_push_stats(facet);
5418 /* Start from historical data for 'rule' itself that are no longer tracked
5419 * in facets. This counts, for example, facets that have expired. */
5420 *packets = rule->packet_count;
5421 *bytes = rule->byte_count;
5423 /* Add any statistics that are tracked by facets. This includes
5424 * statistical data recently updated by ofproto_update_stats() as well as
5425 * stats for packets that were executed "by hand" via dpif_execute(). */
5426 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5427 *packets += facet->packet_count;
5428 *bytes += facet->byte_count;
5433 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5434 struct ofpbuf *packet)
5436 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5437 struct initial_vals initial_vals;
5438 struct dpif_flow_stats stats;
5439 struct action_xlate_ctx ctx;
5440 uint64_t odp_actions_stub[1024 / 8];
5441 struct ofpbuf odp_actions;
5443 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5444 rule_credit_stats(rule, &stats);
5446 initial_vals.vlan_tci = flow->vlan_tci;
5447 initial_vals.tunnel_ip_tos = flow->tunnel.ip_tos;
5448 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5449 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals,
5450 rule, stats.tcp_flags, packet);
5451 ctx.resubmit_stats = &stats;
5452 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
5454 execute_odp_actions(ofproto, flow, odp_actions.data,
5455 odp_actions.size, packet);
5457 ofpbuf_uninit(&odp_actions);
5461 rule_execute(struct rule *rule, const struct flow *flow,
5462 struct ofpbuf *packet)
5464 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5465 ofpbuf_delete(packet);
5470 rule_modify_actions(struct rule *rule_)
5472 struct rule_dpif *rule = rule_dpif_cast(rule_);
5474 complete_operation(rule);
5477 /* Sends 'packet' out 'ofport'.
5478 * May modify 'packet'.
5479 * Returns 0 if successful, otherwise a positive errno value. */
5481 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5483 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5484 uint64_t odp_actions_stub[1024 / 8];
5485 struct ofpbuf key, odp_actions;
5486 struct odputil_keybuf keybuf;
5491 flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow);
5492 if (netdev_vport_is_patch(ofport->up.netdev)) {
5493 struct ofproto_dpif *peer_ofproto;
5494 struct dpif_flow_stats stats;
5495 struct ofport_dpif *peer;
5496 struct rule_dpif *rule;
5498 peer = ofport_get_peer(ofport);
5503 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5504 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5505 netdev_vport_inc_rx(peer->up.netdev, &stats);
5507 flow.in_port = peer->up.ofp_port;
5508 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5509 rule = rule_dpif_lookup(peer_ofproto, &flow);
5510 rule_dpif_execute(rule, &flow, packet);
5515 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5517 if (ofport->tnl_port) {
5518 struct dpif_flow_stats stats;
5520 odp_port = tnl_port_send(ofport->tnl_port, &flow);
5521 if (odp_port == OVSP_NONE) {
5525 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5526 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5527 odp_put_tunnel_action(&flow.tunnel, &odp_actions);
5528 odp_put_skb_mark_action(flow.skb_mark, &odp_actions);
5530 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5532 if (odp_port != ofport->odp_port) {
5533 eth_pop_vlan(packet);
5534 flow.vlan_tci = htons(0);
5538 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5539 odp_flow_key_from_flow(&key, &flow,
5540 ofp_port_to_odp_port(ofproto, flow.in_port));
5542 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5544 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5545 error = dpif_execute(ofproto->backer->dpif,
5547 odp_actions.data, odp_actions.size,
5549 ofpbuf_uninit(&odp_actions);
5552 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5553 ofproto->up.name, odp_port, strerror(error));
5555 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5559 /* OpenFlow to datapath action translation. */
5561 static bool may_receive(const struct ofport_dpif *, struct action_xlate_ctx *);
5562 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5563 struct action_xlate_ctx *);
5564 static void xlate_normal(struct action_xlate_ctx *);
5566 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5567 * The action will state 'slow' as the reason that the action is in the slow
5568 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5569 * dump-flows" output to see why a flow is in the slow path.)
5571 * The 'stub_size' bytes in 'stub' will be used to store the action.
5572 * 'stub_size' must be large enough for the action.
5574 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5577 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5578 enum slow_path_reason slow,
5579 uint64_t *stub, size_t stub_size,
5580 const struct nlattr **actionsp, size_t *actions_lenp)
5582 union user_action_cookie cookie;
5585 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5586 cookie.slow_path.unused = 0;
5587 cookie.slow_path.reason = slow;
5589 ofpbuf_use_stack(&buf, stub, stub_size);
5590 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
5591 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5592 odp_put_userspace_action(pid, &cookie, &buf);
5594 put_userspace_action(ofproto, &buf, flow, &cookie);
5596 *actionsp = buf.data;
5597 *actions_lenp = buf.size;
5601 put_userspace_action(const struct ofproto_dpif *ofproto,
5602 struct ofpbuf *odp_actions,
5603 const struct flow *flow,
5604 const union user_action_cookie *cookie)
5608 pid = dpif_port_get_pid(ofproto->backer->dpif,
5609 ofp_port_to_odp_port(ofproto, flow->in_port));
5611 return odp_put_userspace_action(pid, cookie, odp_actions);
5615 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5616 ovs_be16 vlan_tci, uint32_t odp_port,
5617 unsigned int n_outputs, union user_action_cookie *cookie)
5621 cookie->type = USER_ACTION_COOKIE_SFLOW;
5622 cookie->sflow.vlan_tci = vlan_tci;
5624 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5625 * port information") for the interpretation of cookie->output. */
5626 switch (n_outputs) {
5628 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5629 cookie->sflow.output = 0x40000000 | 256;
5633 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5635 cookie->sflow.output = ifindex;
5640 /* 0x80000000 means "multiple output ports. */
5641 cookie->sflow.output = 0x80000000 | n_outputs;
5646 /* Compose SAMPLE action for sFlow. */
5648 compose_sflow_action(const struct ofproto_dpif *ofproto,
5649 struct ofpbuf *odp_actions,
5650 const struct flow *flow,
5653 uint32_t probability;
5654 union user_action_cookie cookie;
5655 size_t sample_offset, actions_offset;
5658 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5662 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5664 /* Number of packets out of UINT_MAX to sample. */
5665 probability = dpif_sflow_get_probability(ofproto->sflow);
5666 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5668 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5669 compose_sflow_cookie(ofproto, htons(0), odp_port,
5670 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5671 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
5673 nl_msg_end_nested(odp_actions, actions_offset);
5674 nl_msg_end_nested(odp_actions, sample_offset);
5675 return cookie_offset;
5678 /* SAMPLE action must be first action in any given list of actions.
5679 * At this point we do not have all information required to build it. So try to
5680 * build sample action as complete as possible. */
5682 add_sflow_action(struct action_xlate_ctx *ctx)
5684 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5686 &ctx->flow, OVSP_NONE);
5687 ctx->sflow_odp_port = 0;
5688 ctx->sflow_n_outputs = 0;
5691 /* Fix SAMPLE action according to data collected while composing ODP actions.
5692 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5693 * USERSPACE action's user-cookie which is required for sflow. */
5695 fix_sflow_action(struct action_xlate_ctx *ctx)
5697 const struct flow *base = &ctx->base_flow;
5698 union user_action_cookie *cookie;
5700 if (!ctx->user_cookie_offset) {
5704 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
5706 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5708 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5709 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5713 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
5716 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5717 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
5718 ovs_be64 flow_tun_id = ctx->flow.tunnel.tun_id;
5719 uint8_t flow_nw_tos = ctx->flow.nw_tos;
5720 struct priority_to_dscp *pdscp;
5721 uint32_t out_port, odp_port;
5723 /* If 'struct flow' gets additional metadata, we'll need to zero it out
5724 * before traversing a patch port. */
5725 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 18);
5728 xlate_report(ctx, "Nonexistent output port");
5730 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5731 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5733 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5734 xlate_report(ctx, "STP not in forwarding state, skipping output");
5738 if (netdev_vport_is_patch(ofport->up.netdev)) {
5739 struct ofport_dpif *peer = ofport_get_peer(ofport);
5740 struct flow old_flow = ctx->flow;
5741 const struct ofproto_dpif *peer_ofproto;
5742 enum slow_path_reason special;
5743 struct ofport_dpif *in_port;
5746 xlate_report(ctx, "Nonexistent patch port peer");
5750 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5751 if (peer_ofproto->backer != ctx->ofproto->backer) {
5752 xlate_report(ctx, "Patch port peer on a different datapath");
5756 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
5757 ctx->flow.in_port = peer->up.ofp_port;
5758 ctx->flow.metadata = htonll(0);
5759 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
5760 memset(ctx->flow.regs, 0, sizeof ctx->flow.regs);
5762 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5763 special = process_special(ctx->ofproto, &ctx->flow, in_port,
5766 ctx->slow |= special;
5767 } else if (!in_port || may_receive(in_port, ctx)) {
5768 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
5769 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5771 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
5772 * learning action look at the packet, then drop it. */
5773 struct flow old_base_flow = ctx->base_flow;
5774 size_t old_size = ctx->odp_actions->size;
5775 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5776 ctx->base_flow = old_base_flow;
5777 ctx->odp_actions->size = old_size;
5781 ctx->flow = old_flow;
5782 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5784 if (ctx->resubmit_stats) {
5785 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
5786 netdev_vport_inc_rx(peer->up.netdev, ctx->resubmit_stats);
5792 pdscp = get_priority(ofport, ctx->flow.skb_priority);
5794 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5795 ctx->flow.nw_tos |= pdscp->dscp;
5798 odp_port = ofp_port_to_odp_port(ctx->ofproto, ofp_port);
5799 if (ofport->tnl_port) {
5800 odp_port = tnl_port_send(ofport->tnl_port, &ctx->flow);
5801 if (odp_port == OVSP_NONE) {
5802 xlate_report(ctx, "Tunneling decided against output");
5806 if (ctx->resubmit_stats) {
5807 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
5809 out_port = odp_port;
5810 commit_odp_tunnel_action(&ctx->flow, &ctx->base_flow,
5813 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
5814 ctx->flow.vlan_tci);
5815 if (out_port != odp_port) {
5816 ctx->flow.vlan_tci = htons(0);
5818 ctx->flow.skb_mark &= ~IPSEC_MARK;
5820 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
5821 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
5823 ctx->sflow_odp_port = odp_port;
5824 ctx->sflow_n_outputs++;
5825 ctx->nf_output_iface = ofp_port;
5826 ctx->flow.tunnel.tun_id = flow_tun_id;
5827 ctx->flow.vlan_tci = flow_vlan_tci;
5828 ctx->flow.nw_tos = flow_nw_tos;
5832 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
5834 compose_output_action__(ctx, ofp_port, true);
5838 xlate_table_action(struct action_xlate_ctx *ctx,
5839 uint16_t in_port, uint8_t table_id, bool may_packet_in)
5841 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
5842 struct ofproto_dpif *ofproto = ctx->ofproto;
5843 struct rule_dpif *rule;
5844 uint16_t old_in_port;
5845 uint8_t old_table_id;
5847 old_table_id = ctx->table_id;
5848 ctx->table_id = table_id;
5850 /* Look up a flow with 'in_port' as the input port. */
5851 old_in_port = ctx->flow.in_port;
5852 ctx->flow.in_port = in_port;
5853 rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id);
5856 if (table_id > 0 && table_id < N_TABLES) {
5857 struct table_dpif *table = &ofproto->tables[table_id];
5858 if (table->other_table) {
5859 ctx->tags |= (rule && rule->tag
5861 : rule_calculate_tag(&ctx->flow,
5862 &table->other_table->mask,
5867 /* Restore the original input port. Otherwise OFPP_NORMAL and
5868 * OFPP_IN_PORT will have surprising behavior. */
5869 ctx->flow.in_port = old_in_port;
5871 if (ctx->resubmit_hook) {
5872 ctx->resubmit_hook(ctx, rule);
5875 if (rule == NULL && may_packet_in) {
5877 * check if table configuration flags
5878 * OFPTC_TABLE_MISS_CONTROLLER, default.
5879 * OFPTC_TABLE_MISS_CONTINUE,
5880 * OFPTC_TABLE_MISS_DROP
5881 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
5883 rule = rule_dpif_miss_rule(ofproto, &ctx->flow);
5887 struct rule_dpif *old_rule = ctx->rule;
5889 if (ctx->resubmit_stats) {
5890 rule_credit_stats(rule, ctx->resubmit_stats);
5895 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
5896 ctx->rule = old_rule;
5900 ctx->table_id = old_table_id;
5902 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5904 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
5905 MAX_RESUBMIT_RECURSION);
5906 ctx->max_resubmit_trigger = true;
5911 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
5912 const struct ofpact_resubmit *resubmit)
5917 in_port = resubmit->in_port;
5918 if (in_port == OFPP_IN_PORT) {
5919 in_port = ctx->flow.in_port;
5922 table_id = resubmit->table_id;
5923 if (table_id == 255) {
5924 table_id = ctx->table_id;
5927 xlate_table_action(ctx, in_port, table_id, false);
5931 flood_packets(struct action_xlate_ctx *ctx, bool all)
5933 struct ofport_dpif *ofport;
5935 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
5936 uint16_t ofp_port = ofport->up.ofp_port;
5938 if (ofp_port == ctx->flow.in_port) {
5943 compose_output_action__(ctx, ofp_port, false);
5944 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
5945 compose_output_action(ctx, ofp_port);
5949 ctx->nf_output_iface = NF_OUT_FLOOD;
5953 execute_controller_action(struct action_xlate_ctx *ctx, int len,
5954 enum ofp_packet_in_reason reason,
5955 uint16_t controller_id)
5957 struct ofputil_packet_in pin;
5958 struct ofpbuf *packet;
5960 ctx->slow |= SLOW_CONTROLLER;
5965 packet = ofpbuf_clone(ctx->packet);
5967 if (packet->l2 && packet->l3) {
5968 struct eth_header *eh;
5970 eth_pop_vlan(packet);
5973 /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
5974 * LLC frame. Calculating the Ethernet type of these frames is more
5975 * trouble than seems appropriate for a simple assertion. */
5976 ovs_assert(ntohs(eh->eth_type) < ETH_TYPE_MIN
5977 || eh->eth_type == ctx->flow.dl_type);
5979 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
5980 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
5982 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
5983 eth_push_vlan(packet, ctx->flow.vlan_tci);
5987 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5988 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
5989 ctx->flow.nw_tos, ctx->flow.nw_ttl);
5993 if (ctx->flow.nw_proto == IPPROTO_TCP) {
5994 packet_set_tcp_port(packet, ctx->flow.tp_src,
5996 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
5997 packet_set_udp_port(packet, ctx->flow.tp_src,
6004 pin.packet = packet->data;
6005 pin.packet_len = packet->size;
6006 pin.reason = reason;
6007 pin.controller_id = controller_id;
6008 pin.table_id = ctx->table_id;
6009 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
6012 flow_get_metadata(&ctx->flow, &pin.fmd);
6014 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6015 ofpbuf_delete(packet);
6019 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6021 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
6022 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6026 if (ctx->flow.nw_ttl > 1) {
6032 for (i = 0; i < ids->n_controllers; i++) {
6033 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6037 /* Stop processing for current table. */
6043 xlate_output_action(struct action_xlate_ctx *ctx,
6044 uint16_t port, uint16_t max_len, bool may_packet_in)
6046 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
6048 ctx->nf_output_iface = NF_OUT_DROP;
6052 compose_output_action(ctx, ctx->flow.in_port);
6055 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
6061 flood_packets(ctx, false);
6064 flood_packets(ctx, true);
6066 case OFPP_CONTROLLER:
6067 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6073 if (port != ctx->flow.in_port) {
6074 compose_output_action(ctx, port);
6076 xlate_report(ctx, "skipping output to input port");
6081 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6082 ctx->nf_output_iface = NF_OUT_FLOOD;
6083 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
6084 ctx->nf_output_iface = prev_nf_output_iface;
6085 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6086 ctx->nf_output_iface != NF_OUT_FLOOD) {
6087 ctx->nf_output_iface = NF_OUT_MULTI;
6092 xlate_output_reg_action(struct action_xlate_ctx *ctx,
6093 const struct ofpact_output_reg *or)
6095 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
6096 if (port <= UINT16_MAX) {
6097 xlate_output_action(ctx, port, or->max_len, false);
6102 xlate_enqueue_action(struct action_xlate_ctx *ctx,
6103 const struct ofpact_enqueue *enqueue)
6105 uint16_t ofp_port = enqueue->port;
6106 uint32_t queue_id = enqueue->queue;
6107 uint32_t flow_priority, priority;
6110 /* Translate queue to priority. */
6111 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6112 queue_id, &priority);
6114 /* Fall back to ordinary output action. */
6115 xlate_output_action(ctx, enqueue->port, 0, false);
6119 /* Check output port. */
6120 if (ofp_port == OFPP_IN_PORT) {
6121 ofp_port = ctx->flow.in_port;
6122 } else if (ofp_port == ctx->flow.in_port) {
6126 /* Add datapath actions. */
6127 flow_priority = ctx->flow.skb_priority;
6128 ctx->flow.skb_priority = priority;
6129 compose_output_action(ctx, ofp_port);
6130 ctx->flow.skb_priority = flow_priority;
6132 /* Update NetFlow output port. */
6133 if (ctx->nf_output_iface == NF_OUT_DROP) {
6134 ctx->nf_output_iface = ofp_port;
6135 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
6136 ctx->nf_output_iface = NF_OUT_MULTI;
6141 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
6143 uint32_t skb_priority;
6145 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6146 queue_id, &skb_priority)) {
6147 ctx->flow.skb_priority = skb_priority;
6149 /* Couldn't translate queue to a priority. Nothing to do. A warning
6150 * has already been logged. */
6154 struct xlate_reg_state {
6160 xlate_autopath(struct action_xlate_ctx *ctx,
6161 const struct ofpact_autopath *ap)
6163 uint16_t ofp_port = ap->port;
6164 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
6166 if (!port || !port->bundle) {
6167 ofp_port = OFPP_NONE;
6168 } else if (port->bundle->bond) {
6169 /* Autopath does not support VLAN hashing. */
6170 struct ofport_dpif *slave = bond_choose_output_slave(
6171 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
6173 ofp_port = slave->up.ofp_port;
6176 nxm_reg_load(&ap->dst, ofp_port, &ctx->flow);
6180 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6182 struct ofproto_dpif *ofproto = ofproto_;
6183 struct ofport_dpif *port;
6193 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6196 port = get_ofp_port(ofproto, ofp_port);
6197 return port ? port->may_enable : false;
6202 xlate_bundle_action(struct action_xlate_ctx *ctx,
6203 const struct ofpact_bundle *bundle)
6207 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
6208 if (bundle->dst.field) {
6209 nxm_reg_load(&bundle->dst, port, &ctx->flow);
6211 xlate_output_action(ctx, port, 0, false);
6216 xlate_learn_action(struct action_xlate_ctx *ctx,
6217 const struct ofpact_learn *learn)
6219 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6220 struct ofputil_flow_mod fm;
6221 uint64_t ofpacts_stub[1024 / 8];
6222 struct ofpbuf ofpacts;
6225 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6226 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
6228 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6229 if (error && !VLOG_DROP_WARN(&rl)) {
6230 VLOG_WARN("learning action failed to modify flow table (%s)",
6231 ofperr_get_name(error));
6234 ofpbuf_uninit(&ofpacts);
6237 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6238 * means "infinite". */
6240 reduce_timeout(uint16_t max, uint16_t *timeout)
6242 if (max && (!*timeout || *timeout > max)) {
6248 xlate_fin_timeout(struct action_xlate_ctx *ctx,
6249 const struct ofpact_fin_timeout *oft)
6251 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6252 struct rule_dpif *rule = ctx->rule;
6254 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6255 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6260 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
6262 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
6263 ? OFPUTIL_PC_NO_RECV_STP
6264 : OFPUTIL_PC_NO_RECV)) {
6268 /* Only drop packets here if both forwarding and learning are
6269 * disabled. If just learning is enabled, we need to have
6270 * OFPP_NORMAL and the learning action have a look at the packet
6271 * before we can drop it. */
6272 if (!stp_forward_in_state(port->stp_state)
6273 && !stp_learn_in_state(port->stp_state)) {
6281 tunnel_ecn_ok(struct action_xlate_ctx *ctx)
6283 if (is_ip_any(&ctx->base_flow)
6284 && (ctx->base_flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
6285 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
6286 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
6287 " but is not ECN capable");
6290 /* Set the ECN CE value in the tunneled packet. */
6291 ctx->flow.nw_tos |= IP_ECN_CE;
6299 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6300 struct action_xlate_ctx *ctx)
6302 bool was_evictable = true;
6303 const struct ofpact *a;
6306 /* Don't let the rule we're working on get evicted underneath us. */
6307 was_evictable = ctx->rule->up.evictable;
6308 ctx->rule->up.evictable = false;
6310 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6311 struct ofpact_controller *controller;
6312 const struct ofpact_metadata *metadata;
6320 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6321 ofpact_get_OUTPUT(a)->max_len, true);
6324 case OFPACT_CONTROLLER:
6325 controller = ofpact_get_CONTROLLER(a);
6326 execute_controller_action(ctx, controller->max_len,
6328 controller->controller_id);
6331 case OFPACT_ENQUEUE:
6332 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6335 case OFPACT_SET_VLAN_VID:
6336 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6337 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6341 case OFPACT_SET_VLAN_PCP:
6342 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6343 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
6348 case OFPACT_STRIP_VLAN:
6349 ctx->flow.vlan_tci = htons(0);
6352 case OFPACT_PUSH_VLAN:
6353 /* XXX 802.1AD(QinQ) */
6354 ctx->flow.vlan_tci = htons(VLAN_CFI);
6357 case OFPACT_SET_ETH_SRC:
6358 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6362 case OFPACT_SET_ETH_DST:
6363 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6367 case OFPACT_SET_IPV4_SRC:
6368 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6371 case OFPACT_SET_IPV4_DST:
6372 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6375 case OFPACT_SET_IPV4_DSCP:
6376 /* OpenFlow 1.0 only supports IPv4. */
6377 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6378 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
6379 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6383 case OFPACT_SET_L4_SRC_PORT:
6384 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6387 case OFPACT_SET_L4_DST_PORT:
6388 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6391 case OFPACT_RESUBMIT:
6392 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6395 case OFPACT_SET_TUNNEL:
6396 ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6399 case OFPACT_SET_QUEUE:
6400 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6403 case OFPACT_POP_QUEUE:
6404 ctx->flow.skb_priority = ctx->orig_skb_priority;
6407 case OFPACT_REG_MOVE:
6408 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
6411 case OFPACT_REG_LOAD:
6412 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
6415 case OFPACT_DEC_TTL:
6416 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6422 /* Nothing to do. */
6425 case OFPACT_MULTIPATH:
6426 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
6429 case OFPACT_AUTOPATH:
6430 xlate_autopath(ctx, ofpact_get_AUTOPATH(a));
6434 ctx->ofproto->has_bundle_action = true;
6435 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6438 case OFPACT_OUTPUT_REG:
6439 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6443 ctx->has_learn = true;
6444 if (ctx->may_learn) {
6445 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6453 case OFPACT_FIN_TIMEOUT:
6454 ctx->has_fin_timeout = true;
6455 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6458 case OFPACT_CLEAR_ACTIONS:
6460 * Nothing to do because writa-actions is not supported for now.
6461 * When writa-actions is supported, clear-actions also must
6462 * be supported at the same time.
6466 case OFPACT_WRITE_METADATA:
6467 metadata = ofpact_get_WRITE_METADATA(a);
6468 ctx->flow.metadata &= ~metadata->mask;
6469 ctx->flow.metadata |= metadata->metadata & metadata->mask;
6472 case OFPACT_GOTO_TABLE: {
6473 /* XXX remove recursion */
6474 /* It is assumed that goto-table is last action */
6475 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6476 ovs_assert(ctx->table_id < ogt->table_id);
6477 xlate_table_action(ctx, ctx->flow.in_port, ogt->table_id, true);
6485 ctx->rule->up.evictable = was_evictable;
6490 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
6491 struct ofproto_dpif *ofproto, const struct flow *flow,
6492 const struct initial_vals *initial_vals,
6493 struct rule_dpif *rule,
6494 uint8_t tcp_flags, const struct ofpbuf *packet)
6496 ovs_be64 initial_tun_id = flow->tunnel.tun_id;
6498 /* Flow initialization rules:
6499 * - 'base_flow' must match the kernel's view of the packet at the
6500 * time that action processing starts. 'flow' represents any
6501 * transformations we wish to make through actions.
6502 * - By default 'base_flow' and 'flow' are the same since the input
6503 * packet matches the output before any actions are applied.
6504 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
6505 * of the received packet as seen by the kernel. If we later output
6506 * to another device without any modifications this will cause us to
6507 * insert a new tag since the original one was stripped off by the
6509 * - Tunnel 'flow' is largely cleared when transitioning between
6510 * the input and output stages since it does not make sense to output
6511 * a packet with the exact headers that it was received with (i.e.
6512 * the destination IP is us). The one exception is the tun_id, which
6513 * is preserved to allow use in later resubmit lookups and loads into
6515 * - Tunnel 'base_flow' is completely cleared since that is what the
6516 * kernel does. If we wish to maintain the original values an action
6517 * needs to be generated. */
6519 ctx->ofproto = ofproto;
6521 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
6522 ctx->base_flow = ctx->flow;
6523 ctx->base_flow.vlan_tci = initial_vals->vlan_tci;
6524 ctx->base_flow.tunnel.ip_tos = initial_vals->tunnel_ip_tos;
6525 ctx->flow.tunnel.tun_id = initial_tun_id;
6527 ctx->packet = packet;
6528 ctx->may_learn = packet != NULL;
6529 ctx->tcp_flags = tcp_flags;
6530 ctx->resubmit_hook = NULL;
6531 ctx->report_hook = NULL;
6532 ctx->resubmit_stats = NULL;
6535 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6536 * into datapath actions in 'odp_actions', using 'ctx'. */
6538 xlate_actions(struct action_xlate_ctx *ctx,
6539 const struct ofpact *ofpacts, size_t ofpacts_len,
6540 struct ofpbuf *odp_actions)
6542 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6543 * that in the future we always keep a copy of the original flow for
6544 * tracing purposes. */
6545 static bool hit_resubmit_limit;
6547 enum slow_path_reason special;
6548 struct ofport_dpif *in_port;
6550 COVERAGE_INC(ofproto_dpif_xlate);
6552 ofpbuf_clear(odp_actions);
6553 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
6555 ctx->odp_actions = odp_actions;
6558 ctx->has_learn = false;
6559 ctx->has_normal = false;
6560 ctx->has_fin_timeout = false;
6561 ctx->nf_output_iface = NF_OUT_DROP;
6564 ctx->max_resubmit_trigger = false;
6565 ctx->orig_skb_priority = ctx->flow.skb_priority;
6569 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
6570 /* Do this conditionally because the copy is expensive enough that it
6571 * shows up in profiles.
6573 * We keep orig_flow in 'ctx' only because I couldn't make GCC 4.4
6574 * believe that I wasn't using it without initializing it if I kept it
6575 * in a local variable. */
6576 ctx->orig_flow = ctx->flow;
6579 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6580 switch (ctx->ofproto->up.frag_handling) {
6581 case OFPC_FRAG_NORMAL:
6582 /* We must pretend that transport ports are unavailable. */
6583 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
6584 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
6587 case OFPC_FRAG_DROP:
6590 case OFPC_FRAG_REASM:
6593 case OFPC_FRAG_NX_MATCH:
6594 /* Nothing to do. */
6597 case OFPC_INVALID_TTL_TO_CONTROLLER:
6602 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
6603 special = process_special(ctx->ofproto, &ctx->flow, in_port, ctx->packet);
6605 ctx->slow |= special;
6607 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6608 struct initial_vals initial_vals;
6609 uint32_t local_odp_port;
6611 initial_vals.vlan_tci = ctx->base_flow.vlan_tci;
6612 initial_vals.tunnel_ip_tos = ctx->base_flow.tunnel.ip_tos;
6614 add_sflow_action(ctx);
6616 if (tunnel_ecn_ok(ctx) && (!in_port || may_receive(in_port, ctx))) {
6617 do_xlate_actions(ofpacts, ofpacts_len, ctx);
6619 /* We've let OFPP_NORMAL and the learning action look at the
6620 * packet, so drop it now if forwarding is disabled. */
6621 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
6622 ofpbuf_clear(ctx->odp_actions);
6623 add_sflow_action(ctx);
6627 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
6628 if (!hit_resubmit_limit) {
6629 /* We didn't record the original flow. Make sure we do from
6631 hit_resubmit_limit = true;
6632 } else if (!VLOG_DROP_ERR(&trace_rl)) {
6633 struct ds ds = DS_EMPTY_INITIALIZER;
6635 ofproto_trace(ctx->ofproto, &ctx->orig_flow, ctx->packet,
6636 &initial_vals, &ds);
6637 VLOG_ERR("Trace triggered by excessive resubmit "
6638 "recursion:\n%s", ds_cstr(&ds));
6643 local_odp_port = ofp_port_to_odp_port(ctx->ofproto, OFPP_LOCAL);
6644 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
6646 ctx->odp_actions->data,
6647 ctx->odp_actions->size)) {
6648 ctx->slow |= SLOW_IN_BAND;
6650 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
6652 compose_output_action(ctx, OFPP_LOCAL);
6655 if (ctx->ofproto->has_mirrors) {
6656 add_mirror_actions(ctx, &ctx->orig_flow);
6658 fix_sflow_action(ctx);
6662 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
6663 * into datapath actions, using 'ctx', and discards the datapath actions. */
6665 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
6666 const struct ofpact *ofpacts,
6669 uint64_t odp_actions_stub[1024 / 8];
6670 struct ofpbuf odp_actions;
6672 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
6673 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
6674 ofpbuf_uninit(&odp_actions);
6678 xlate_report(struct action_xlate_ctx *ctx, const char *s)
6680 if (ctx->report_hook) {
6681 ctx->report_hook(ctx, s);
6685 /* OFPP_NORMAL implementation. */
6687 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
6689 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
6690 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
6691 * the bundle on which the packet was received, returns the VLAN to which the
6694 * Both 'vid' and the return value are in the range 0...4095. */
6696 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
6698 switch (in_bundle->vlan_mode) {
6699 case PORT_VLAN_ACCESS:
6700 return in_bundle->vlan;
6703 case PORT_VLAN_TRUNK:
6706 case PORT_VLAN_NATIVE_UNTAGGED:
6707 case PORT_VLAN_NATIVE_TAGGED:
6708 return vid ? vid : in_bundle->vlan;
6715 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
6716 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
6719 * 'vid' should be the VID obtained from the 802.1Q header that was received as
6720 * part of a packet (specify 0 if there was no 802.1Q header), in the range
6723 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
6725 /* Allow any VID on the OFPP_NONE port. */
6726 if (in_bundle == &ofpp_none_bundle) {
6730 switch (in_bundle->vlan_mode) {
6731 case PORT_VLAN_ACCESS:
6734 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6735 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
6736 "packet received on port %s configured as VLAN "
6737 "%"PRIu16" access port",
6738 in_bundle->ofproto->up.name, vid,
6739 in_bundle->name, in_bundle->vlan);
6745 case PORT_VLAN_NATIVE_UNTAGGED:
6746 case PORT_VLAN_NATIVE_TAGGED:
6748 /* Port must always carry its native VLAN. */
6752 case PORT_VLAN_TRUNK:
6753 if (!ofbundle_includes_vlan(in_bundle, vid)) {
6755 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6756 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
6757 "received on port %s not configured for trunking "
6759 in_bundle->ofproto->up.name, vid,
6760 in_bundle->name, vid);
6772 /* Given 'vlan', the VLAN that a packet belongs to, and
6773 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
6774 * that should be included in the 802.1Q header. (If the return value is 0,
6775 * then the 802.1Q header should only be included in the packet if there is a
6778 * Both 'vlan' and the return value are in the range 0...4095. */
6780 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
6782 switch (out_bundle->vlan_mode) {
6783 case PORT_VLAN_ACCESS:
6786 case PORT_VLAN_TRUNK:
6787 case PORT_VLAN_NATIVE_TAGGED:
6790 case PORT_VLAN_NATIVE_UNTAGGED:
6791 return vlan == out_bundle->vlan ? 0 : vlan;
6799 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
6802 struct ofport_dpif *port;
6804 ovs_be16 tci, old_tci;
6806 vid = output_vlan_to_vid(out_bundle, vlan);
6807 if (!out_bundle->bond) {
6808 port = ofbundle_get_a_port(out_bundle);
6810 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
6813 /* No slaves enabled, so drop packet. */
6818 old_tci = ctx->flow.vlan_tci;
6820 if (tci || out_bundle->use_priority_tags) {
6821 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
6823 tci |= htons(VLAN_CFI);
6826 ctx->flow.vlan_tci = tci;
6828 compose_output_action(ctx, port->up.ofp_port);
6829 ctx->flow.vlan_tci = old_tci;
6833 mirror_mask_ffs(mirror_mask_t mask)
6835 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
6840 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
6842 return (bundle->vlan_mode != PORT_VLAN_ACCESS
6843 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
6847 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
6849 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
6852 /* Returns an arbitrary interface within 'bundle'. */
6853 static struct ofport_dpif *
6854 ofbundle_get_a_port(const struct ofbundle *bundle)
6856 return CONTAINER_OF(list_front(&bundle->ports),
6857 struct ofport_dpif, bundle_node);
6861 vlan_is_mirrored(const struct ofmirror *m, int vlan)
6863 return !m->vlans || bitmap_is_set(m->vlans, vlan);
6867 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
6869 struct ofproto_dpif *ofproto = ctx->ofproto;
6870 mirror_mask_t mirrors;
6871 struct ofbundle *in_bundle;
6874 const struct nlattr *a;
6877 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
6878 ctx->packet != NULL, NULL);
6882 mirrors = in_bundle->src_mirrors;
6884 /* Drop frames on bundles reserved for mirroring. */
6885 if (in_bundle->mirror_out) {
6886 if (ctx->packet != NULL) {
6887 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6888 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6889 "%s, which is reserved exclusively for mirroring",
6890 ctx->ofproto->up.name, in_bundle->name);
6896 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
6897 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6900 vlan = input_vid_to_vlan(in_bundle, vid);
6902 /* Look at the output ports to check for destination selections. */
6904 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
6905 ctx->odp_actions->size) {
6906 enum ovs_action_attr type = nl_attr_type(a);
6907 struct ofport_dpif *ofport;
6909 if (type != OVS_ACTION_ATTR_OUTPUT) {
6913 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
6914 if (ofport && ofport->bundle) {
6915 mirrors |= ofport->bundle->dst_mirrors;
6923 /* Restore the original packet before adding the mirror actions. */
6924 ctx->flow = *orig_flow;
6929 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6931 if (!vlan_is_mirrored(m, vlan)) {
6932 mirrors = zero_rightmost_1bit(mirrors);
6936 mirrors &= ~m->dup_mirrors;
6937 ctx->mirrors |= m->dup_mirrors;
6939 output_normal(ctx, m->out, vlan);
6940 } else if (vlan != m->out_vlan
6941 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
6942 struct ofbundle *bundle;
6944 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
6945 if (ofbundle_includes_vlan(bundle, m->out_vlan)
6946 && !bundle->mirror_out) {
6947 output_normal(ctx, bundle, m->out_vlan);
6955 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
6956 uint64_t packets, uint64_t bytes)
6962 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
6965 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6968 /* In normal circumstances 'm' will not be NULL. However,
6969 * if mirrors are reconfigured, we can temporarily get out
6970 * of sync in facet_revalidate(). We could "correct" the
6971 * mirror list before reaching here, but doing that would
6972 * not properly account the traffic stats we've currently
6973 * accumulated for previous mirror configuration. */
6977 m->packet_count += packets;
6978 m->byte_count += bytes;
6982 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
6983 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
6984 * indicate this; newer upstream kernels use gratuitous ARP requests. */
6986 is_gratuitous_arp(const struct flow *flow)
6988 return (flow->dl_type == htons(ETH_TYPE_ARP)
6989 && eth_addr_is_broadcast(flow->dl_dst)
6990 && (flow->nw_proto == ARP_OP_REPLY
6991 || (flow->nw_proto == ARP_OP_REQUEST
6992 && flow->nw_src == flow->nw_dst)));
6996 update_learning_table(struct ofproto_dpif *ofproto,
6997 const struct flow *flow, int vlan,
6998 struct ofbundle *in_bundle)
7000 struct mac_entry *mac;
7002 /* Don't learn the OFPP_NONE port. */
7003 if (in_bundle == &ofpp_none_bundle) {
7007 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7011 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7012 if (is_gratuitous_arp(flow)) {
7013 /* We don't want to learn from gratuitous ARP packets that are
7014 * reflected back over bond slaves so we lock the learning table. */
7015 if (!in_bundle->bond) {
7016 mac_entry_set_grat_arp_lock(mac);
7017 } else if (mac_entry_is_grat_arp_locked(mac)) {
7022 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7023 /* The log messages here could actually be useful in debugging,
7024 * so keep the rate limit relatively high. */
7025 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7026 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7027 "on port %s in VLAN %d",
7028 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7029 in_bundle->name, vlan);
7031 mac->port.p = in_bundle;
7032 tag_set_add(&ofproto->backer->revalidate_set,
7033 mac_learning_changed(ofproto->ml, mac));
7037 static struct ofbundle *
7038 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7039 bool warn, struct ofport_dpif **in_ofportp)
7041 struct ofport_dpif *ofport;
7043 /* Find the port and bundle for the received packet. */
7044 ofport = get_ofp_port(ofproto, in_port);
7046 *in_ofportp = ofport;
7048 if (ofport && ofport->bundle) {
7049 return ofport->bundle;
7052 /* Special-case OFPP_NONE, which a controller may use as the ingress
7053 * port for traffic that it is sourcing. */
7054 if (in_port == OFPP_NONE) {
7055 return &ofpp_none_bundle;
7058 /* Odd. A few possible reasons here:
7060 * - We deleted a port but there are still a few packets queued up
7063 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7064 * we don't know about.
7066 * - The ofproto client didn't configure the port as part of a bundle.
7067 * This is particularly likely to happen if a packet was received on the
7068 * port after it was created, but before the client had a chance to
7069 * configure its bundle.
7072 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7074 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7075 "port %"PRIu16, ofproto->up.name, in_port);
7080 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7081 * dropped. Returns true if they may be forwarded, false if they should be
7084 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7085 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7087 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7088 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7089 * checked by input_vid_is_valid().
7091 * May also add tags to '*tags', although the current implementation only does
7092 * so in one special case.
7095 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
7098 struct ofproto_dpif *ofproto = ctx->ofproto;
7099 struct flow *flow = &ctx->flow;
7100 struct ofbundle *in_bundle = in_port->bundle;
7102 /* Drop frames for reserved multicast addresses
7103 * only if forward_bpdu option is absent. */
7104 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7105 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7109 if (in_bundle->bond) {
7110 struct mac_entry *mac;
7112 switch (bond_check_admissibility(in_bundle->bond, in_port,
7113 flow->dl_dst, &ctx->tags)) {
7118 xlate_report(ctx, "bonding refused admissibility, dropping");
7121 case BV_DROP_IF_MOVED:
7122 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7123 if (mac && mac->port.p != in_bundle &&
7124 (!is_gratuitous_arp(flow)
7125 || mac_entry_is_grat_arp_locked(mac))) {
7126 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7138 xlate_normal(struct action_xlate_ctx *ctx)
7140 struct ofport_dpif *in_port;
7141 struct ofbundle *in_bundle;
7142 struct mac_entry *mac;
7146 ctx->has_normal = true;
7148 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
7149 ctx->packet != NULL, &in_port);
7151 xlate_report(ctx, "no input bundle, dropping");
7155 /* Drop malformed frames. */
7156 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7157 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
7158 if (ctx->packet != NULL) {
7159 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7160 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7161 "VLAN tag received on port %s",
7162 ctx->ofproto->up.name, in_bundle->name);
7164 xlate_report(ctx, "partial VLAN tag, dropping");
7168 /* Drop frames on bundles reserved for mirroring. */
7169 if (in_bundle->mirror_out) {
7170 if (ctx->packet != NULL) {
7171 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7172 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7173 "%s, which is reserved exclusively for mirroring",
7174 ctx->ofproto->up.name, in_bundle->name);
7176 xlate_report(ctx, "input port is mirror output port, dropping");
7181 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
7182 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7183 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7186 vlan = input_vid_to_vlan(in_bundle, vid);
7188 /* Check other admissibility requirements. */
7189 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7193 /* Learn source MAC. */
7194 if (ctx->may_learn) {
7195 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
7198 /* Determine output bundle. */
7199 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
7202 if (mac->port.p != in_bundle) {
7203 xlate_report(ctx, "forwarding to learned port");
7204 output_normal(ctx, mac->port.p, vlan);
7206 xlate_report(ctx, "learned port is input port, dropping");
7209 struct ofbundle *bundle;
7211 xlate_report(ctx, "no learned MAC for destination, flooding");
7212 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7213 if (bundle != in_bundle
7214 && ofbundle_includes_vlan(bundle, vlan)
7215 && bundle->floodable
7216 && !bundle->mirror_out) {
7217 output_normal(ctx, bundle, vlan);
7220 ctx->nf_output_iface = NF_OUT_FLOOD;
7224 /* Optimized flow revalidation.
7226 * It's a difficult problem, in general, to tell which facets need to have
7227 * their actions recalculated whenever the OpenFlow flow table changes. We
7228 * don't try to solve that general problem: for most kinds of OpenFlow flow
7229 * table changes, we recalculate the actions for every facet. This is
7230 * relatively expensive, but it's good enough if the OpenFlow flow table
7231 * doesn't change very often.
7233 * However, we can expect one particular kind of OpenFlow flow table change to
7234 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7235 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7236 * table, we add a special case that applies to flow tables in which every rule
7237 * has the same form (that is, the same wildcards), except that the table is
7238 * also allowed to have a single "catch-all" flow that matches all packets. We
7239 * optimize this case by tagging all of the facets that resubmit into the table
7240 * and invalidating the same tag whenever a flow changes in that table. The
7241 * end result is that we revalidate just the facets that need it (and sometimes
7242 * a few more, but not all of the facets or even all of the facets that
7243 * resubmit to the table modified by MAC learning). */
7245 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7246 * into an OpenFlow table with the given 'basis'. */
7248 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7251 if (minimask_is_catchall(mask)) {
7254 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7255 return tag_create_deterministic(hash);
7259 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7260 * taggability of that table.
7262 * This function must be called after *each* change to a flow table. If you
7263 * skip calling it on some changes then the pointer comparisons at the end can
7264 * be invalid if you get unlucky. For example, if a flow removal causes a
7265 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7266 * different wildcards to be created with the same address, then this function
7267 * will incorrectly skip revalidation. */
7269 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7271 struct table_dpif *table = &ofproto->tables[table_id];
7272 const struct oftable *oftable = &ofproto->up.tables[table_id];
7273 struct cls_table *catchall, *other;
7274 struct cls_table *t;
7276 catchall = other = NULL;
7278 switch (hmap_count(&oftable->cls.tables)) {
7280 /* We could tag this OpenFlow table but it would make the logic a
7281 * little harder and it's a corner case that doesn't seem worth it
7287 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7288 if (cls_table_is_catchall(t)) {
7290 } else if (!other) {
7293 /* Indicate that we can't tag this by setting both tables to
7294 * NULL. (We know that 'catchall' is already NULL.) */
7301 /* Can't tag this table. */
7305 if (table->catchall_table != catchall || table->other_table != other) {
7306 table->catchall_table = catchall;
7307 table->other_table = other;
7308 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7312 /* Given 'rule' that has changed in some way (either it is a rule being
7313 * inserted, a rule being deleted, or a rule whose actions are being
7314 * modified), marks facets for revalidation to ensure that packets will be
7315 * forwarded correctly according to the new state of the flow table.
7317 * This function must be called after *each* change to a flow table. See
7318 * the comment on table_update_taggable() for more information. */
7320 rule_invalidate(const struct rule_dpif *rule)
7322 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7324 table_update_taggable(ofproto, rule->up.table_id);
7326 if (!ofproto->backer->need_revalidate) {
7327 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7329 if (table->other_table && rule->tag) {
7330 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7332 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7338 set_frag_handling(struct ofproto *ofproto_,
7339 enum ofp_config_flags frag_handling)
7341 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7342 if (frag_handling != OFPC_FRAG_REASM) {
7343 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7351 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7352 const struct flow *flow,
7353 const struct ofpact *ofpacts, size_t ofpacts_len)
7355 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7356 struct initial_vals initial_vals;
7357 struct odputil_keybuf keybuf;
7358 struct dpif_flow_stats stats;
7362 struct action_xlate_ctx ctx;
7363 uint64_t odp_actions_stub[1024 / 8];
7364 struct ofpbuf odp_actions;
7366 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7367 odp_flow_key_from_flow(&key, flow,
7368 ofp_port_to_odp_port(ofproto, flow->in_port));
7370 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7372 initial_vals.vlan_tci = flow->vlan_tci;
7373 initial_vals.tunnel_ip_tos = 0;
7374 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals, NULL,
7375 packet_get_tcp_flags(packet, flow), packet);
7376 ctx.resubmit_stats = &stats;
7378 ofpbuf_use_stub(&odp_actions,
7379 odp_actions_stub, sizeof odp_actions_stub);
7380 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
7381 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7382 odp_actions.data, odp_actions.size, packet);
7383 ofpbuf_uninit(&odp_actions);
7391 set_netflow(struct ofproto *ofproto_,
7392 const struct netflow_options *netflow_options)
7394 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7396 if (netflow_options) {
7397 if (!ofproto->netflow) {
7398 ofproto->netflow = netflow_create();
7400 return netflow_set_options(ofproto->netflow, netflow_options);
7402 netflow_destroy(ofproto->netflow);
7403 ofproto->netflow = NULL;
7409 get_netflow_ids(const struct ofproto *ofproto_,
7410 uint8_t *engine_type, uint8_t *engine_id)
7412 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7414 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7418 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7420 if (!facet_is_controller_flow(facet) &&
7421 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7422 struct subfacet *subfacet;
7423 struct ofexpired expired;
7425 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7426 if (subfacet->path == SF_FAST_PATH) {
7427 struct dpif_flow_stats stats;
7429 subfacet_reinstall(subfacet, &stats);
7430 subfacet_update_stats(subfacet, &stats);
7434 expired.flow = facet->flow;
7435 expired.packet_count = facet->packet_count;
7436 expired.byte_count = facet->byte_count;
7437 expired.used = facet->used;
7438 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
7443 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7445 struct facet *facet;
7447 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7448 send_active_timeout(ofproto, facet);
7452 static struct ofproto_dpif *
7453 ofproto_dpif_lookup(const char *name)
7455 struct ofproto_dpif *ofproto;
7457 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7458 hash_string(name, 0), &all_ofproto_dpifs) {
7459 if (!strcmp(ofproto->up.name, name)) {
7467 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7468 const char *argv[], void *aux OVS_UNUSED)
7470 struct ofproto_dpif *ofproto;
7473 ofproto = ofproto_dpif_lookup(argv[1]);
7475 unixctl_command_reply_error(conn, "no such bridge");
7478 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7480 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7481 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7485 unixctl_command_reply(conn, "table successfully flushed");
7489 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7490 const char *argv[], void *aux OVS_UNUSED)
7492 struct ds ds = DS_EMPTY_INITIALIZER;
7493 const struct ofproto_dpif *ofproto;
7494 const struct mac_entry *e;
7496 ofproto = ofproto_dpif_lookup(argv[1]);
7498 unixctl_command_reply_error(conn, "no such bridge");
7502 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7503 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7504 struct ofbundle *bundle = e->port.p;
7505 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7506 ofbundle_get_a_port(bundle)->odp_port,
7507 e->vlan, ETH_ADDR_ARGS(e->mac),
7508 mac_entry_age(ofproto->ml, e));
7510 unixctl_command_reply(conn, ds_cstr(&ds));
7515 struct action_xlate_ctx ctx;
7521 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7522 const struct rule_dpif *rule)
7524 ds_put_char_multiple(result, '\t', level);
7526 ds_put_cstr(result, "No match\n");
7530 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7531 table_id, ntohll(rule->up.flow_cookie));
7532 cls_rule_format(&rule->up.cr, result);
7533 ds_put_char(result, '\n');
7535 ds_put_char_multiple(result, '\t', level);
7536 ds_put_cstr(result, "OpenFlow ");
7537 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7538 ds_put_char(result, '\n');
7542 trace_format_flow(struct ds *result, int level, const char *title,
7543 struct trace_ctx *trace)
7545 ds_put_char_multiple(result, '\t', level);
7546 ds_put_format(result, "%s: ", title);
7547 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
7548 ds_put_cstr(result, "unchanged");
7550 flow_format(result, &trace->ctx.flow);
7551 trace->flow = trace->ctx.flow;
7553 ds_put_char(result, '\n');
7557 trace_format_regs(struct ds *result, int level, const char *title,
7558 struct trace_ctx *trace)
7562 ds_put_char_multiple(result, '\t', level);
7563 ds_put_format(result, "%s:", title);
7564 for (i = 0; i < FLOW_N_REGS; i++) {
7565 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7567 ds_put_char(result, '\n');
7571 trace_format_odp(struct ds *result, int level, const char *title,
7572 struct trace_ctx *trace)
7574 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
7576 ds_put_char_multiple(result, '\t', level);
7577 ds_put_format(result, "%s: ", title);
7578 format_odp_actions(result, odp_actions->data, odp_actions->size);
7579 ds_put_char(result, '\n');
7583 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
7585 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7586 struct ds *result = trace->result;
7588 ds_put_char(result, '\n');
7589 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7590 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
7591 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
7592 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
7596 trace_report(struct action_xlate_ctx *ctx, const char *s)
7598 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7599 struct ds *result = trace->result;
7601 ds_put_char_multiple(result, '\t', ctx->recurse);
7602 ds_put_cstr(result, s);
7603 ds_put_char(result, '\n');
7607 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
7608 void *aux OVS_UNUSED)
7610 const char *dpname = argv[1];
7611 struct ofproto_dpif *ofproto;
7612 struct ofpbuf odp_key;
7613 struct ofpbuf *packet;
7614 struct initial_vals initial_vals;
7620 ofpbuf_init(&odp_key, 0);
7623 ofproto = ofproto_dpif_lookup(dpname);
7625 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
7629 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
7630 /* ofproto/trace dpname flow [-generate] */
7631 const char *flow_s = argv[2];
7632 const char *generate_s = argv[3];
7634 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
7635 * flow. We guess which type it is based on whether 'flow_s' contains
7636 * an '(', since a datapath flow always contains '(') but an
7637 * OpenFlow-like flow should not (in fact it's allowed but I believe
7638 * that's not documented anywhere).
7640 * An alternative would be to try to parse 'flow_s' both ways, but then
7641 * it would be tricky giving a sensible error message. After all, do
7642 * you just say "syntax error" or do you present both error messages?
7643 * Both choices seem lousy. */
7644 if (strchr(flow_s, '(')) {
7647 /* Convert string to datapath key. */
7648 ofpbuf_init(&odp_key, 0);
7649 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
7651 unixctl_command_reply_error(conn, "Bad flow syntax");
7655 /* The user might have specified the wrong ofproto but within the
7656 * same backer. That's OK, ofproto_receive() can find the right
7658 if (ofproto_receive(ofproto->backer, NULL, odp_key.data,
7659 odp_key.size, &flow, NULL, &ofproto, NULL,
7661 unixctl_command_reply_error(conn, "Invalid flow");
7664 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
7668 error_s = parse_ofp_exact_flow(&flow, argv[2]);
7670 unixctl_command_reply_error(conn, error_s);
7675 initial_vals.vlan_tci = flow.vlan_tci;
7676 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
7679 /* Generate a packet, if requested. */
7681 packet = ofpbuf_new(0);
7682 flow_compose(packet, &flow);
7684 } else if (argc == 7) {
7685 /* ofproto/trace dpname priority tun_id in_port mark packet */
7686 const char *priority_s = argv[2];
7687 const char *tun_id_s = argv[3];
7688 const char *in_port_s = argv[4];
7689 const char *mark_s = argv[5];
7690 const char *packet_s = argv[6];
7691 uint32_t in_port = atoi(in_port_s);
7692 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
7693 uint32_t priority = atoi(priority_s);
7694 uint32_t mark = atoi(mark_s);
7697 msg = eth_from_hex(packet_s, &packet);
7699 unixctl_command_reply_error(conn, msg);
7703 ds_put_cstr(&result, "Packet: ");
7704 s = ofp_packet_to_string(packet->data, packet->size);
7705 ds_put_cstr(&result, s);
7708 flow_extract(packet, priority, mark, NULL, in_port, &flow);
7709 flow.tunnel.tun_id = tun_id;
7710 initial_vals.vlan_tci = flow.vlan_tci;
7711 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
7713 unixctl_command_reply_error(conn, "Bad command syntax");
7717 ofproto_trace(ofproto, &flow, packet, &initial_vals, &result);
7718 unixctl_command_reply(conn, ds_cstr(&result));
7721 ds_destroy(&result);
7722 ofpbuf_delete(packet);
7723 ofpbuf_uninit(&odp_key);
7727 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
7728 const struct ofpbuf *packet,
7729 const struct initial_vals *initial_vals, struct ds *ds)
7731 struct rule_dpif *rule;
7733 ds_put_cstr(ds, "Flow: ");
7734 flow_format(ds, flow);
7735 ds_put_char(ds, '\n');
7737 rule = rule_dpif_lookup(ofproto, flow);
7739 trace_format_rule(ds, 0, 0, rule);
7740 if (rule == ofproto->miss_rule) {
7741 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
7742 } else if (rule == ofproto->no_packet_in_rule) {
7743 ds_put_cstr(ds, "\nNo match, packets dropped because "
7744 "OFPPC_NO_PACKET_IN is set on in_port.\n");
7748 uint64_t odp_actions_stub[1024 / 8];
7749 struct ofpbuf odp_actions;
7751 struct trace_ctx trace;
7754 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
7757 ofpbuf_use_stub(&odp_actions,
7758 odp_actions_stub, sizeof odp_actions_stub);
7759 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_vals,
7760 rule, tcp_flags, packet);
7761 trace.ctx.resubmit_hook = trace_resubmit;
7762 trace.ctx.report_hook = trace_report;
7763 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
7766 ds_put_char(ds, '\n');
7767 trace_format_flow(ds, 0, "Final flow", &trace);
7768 ds_put_cstr(ds, "Datapath actions: ");
7769 format_odp_actions(ds, odp_actions.data, odp_actions.size);
7770 ofpbuf_uninit(&odp_actions);
7772 if (trace.ctx.slow) {
7773 enum slow_path_reason slow;
7775 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
7776 "slow path because it:");
7777 for (slow = trace.ctx.slow; slow; ) {
7778 enum slow_path_reason bit = rightmost_1bit(slow);
7782 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
7785 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
7788 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
7791 ds_put_cstr(ds, "\n\t- Needs in-band special case "
7794 ds_put_cstr(ds, "\n\t (The datapath actions are "
7795 "incomplete--for complete actions, "
7796 "please supply a packet.)");
7799 case SLOW_CONTROLLER:
7800 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
7801 "to the OpenFlow controller.");
7804 ds_put_cstr(ds, "\n\t- Needs more specific matching "
7805 "than the datapath supports.");
7812 if (slow & ~SLOW_MATCH) {
7813 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
7814 "the special slow-path processing.");
7821 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7822 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7825 unixctl_command_reply(conn, NULL);
7829 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7830 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7833 unixctl_command_reply(conn, NULL);
7836 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
7837 * 'reply' describing the results. */
7839 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
7841 struct facet *facet;
7845 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7846 if (!facet_check_consistency(facet)) {
7851 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
7855 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
7856 ofproto->up.name, errors);
7858 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
7863 ofproto_dpif_self_check(struct unixctl_conn *conn,
7864 int argc, const char *argv[], void *aux OVS_UNUSED)
7866 struct ds reply = DS_EMPTY_INITIALIZER;
7867 struct ofproto_dpif *ofproto;
7870 ofproto = ofproto_dpif_lookup(argv[1]);
7872 unixctl_command_reply_error(conn, "Unknown ofproto (use "
7873 "ofproto/list for help)");
7876 ofproto_dpif_self_check__(ofproto, &reply);
7878 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7879 ofproto_dpif_self_check__(ofproto, &reply);
7883 unixctl_command_reply(conn, ds_cstr(&reply));
7887 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
7888 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
7889 * to destroy 'ofproto_shash' and free the returned value. */
7890 static const struct shash_node **
7891 get_ofprotos(struct shash *ofproto_shash)
7893 const struct ofproto_dpif *ofproto;
7895 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7896 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
7897 shash_add_nocopy(ofproto_shash, name, ofproto);
7900 return shash_sort(ofproto_shash);
7904 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
7905 const char *argv[] OVS_UNUSED,
7906 void *aux OVS_UNUSED)
7908 struct ds ds = DS_EMPTY_INITIALIZER;
7909 struct shash ofproto_shash;
7910 const struct shash_node **sorted_ofprotos;
7913 shash_init(&ofproto_shash);
7914 sorted_ofprotos = get_ofprotos(&ofproto_shash);
7915 for (i = 0; i < shash_count(&ofproto_shash); i++) {
7916 const struct shash_node *node = sorted_ofprotos[i];
7917 ds_put_format(&ds, "%s\n", node->name);
7920 shash_destroy(&ofproto_shash);
7921 free(sorted_ofprotos);
7923 unixctl_command_reply(conn, ds_cstr(&ds));
7928 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
7930 struct dpif_dp_stats s;
7931 const struct shash_node **ports;
7934 dpif_get_dp_stats(ofproto->backer->dpif, &s);
7936 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
7937 dpif_name(ofproto->backer->dpif));
7938 /* xxx It would be better to show bridge-specific stats instead
7939 * xxx of dp ones. */
7941 "\tlookups: hit:%"PRIu64" missed:%"PRIu64" lost:%"PRIu64"\n",
7942 s.n_hit, s.n_missed, s.n_lost);
7943 ds_put_format(ds, "\tflows: %zu\n",
7944 hmap_count(&ofproto->subfacets));
7946 ports = shash_sort(&ofproto->up.port_by_name);
7947 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
7948 const struct shash_node *node = ports[i];
7949 struct ofport *ofport = node->data;
7950 const char *name = netdev_get_name(ofport->netdev);
7951 const char *type = netdev_get_type(ofport->netdev);
7954 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
7956 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
7957 if (odp_port != OVSP_NONE) {
7958 ds_put_format(ds, "%"PRIu32":", odp_port);
7960 ds_put_cstr(ds, "none:");
7963 if (strcmp(type, "system")) {
7964 struct netdev *netdev;
7967 ds_put_format(ds, " (%s", type);
7969 error = netdev_open(name, type, &netdev);
7974 error = netdev_get_config(netdev, &config);
7976 const struct smap_node **nodes;
7979 nodes = smap_sort(&config);
7980 for (i = 0; i < smap_count(&config); i++) {
7981 const struct smap_node *node = nodes[i];
7982 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
7983 node->key, node->value);
7987 smap_destroy(&config);
7989 netdev_close(netdev);
7991 ds_put_char(ds, ')');
7993 ds_put_char(ds, '\n');
7999 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
8000 const char *argv[], void *aux OVS_UNUSED)
8002 struct ds ds = DS_EMPTY_INITIALIZER;
8003 const struct ofproto_dpif *ofproto;
8007 for (i = 1; i < argc; i++) {
8008 ofproto = ofproto_dpif_lookup(argv[i]);
8010 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
8011 "for help)", argv[i]);
8012 unixctl_command_reply_error(conn, ds_cstr(&ds));
8015 show_dp_format(ofproto, &ds);
8018 struct shash ofproto_shash;
8019 const struct shash_node **sorted_ofprotos;
8022 shash_init(&ofproto_shash);
8023 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8024 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8025 const struct shash_node *node = sorted_ofprotos[i];
8026 show_dp_format(node->data, &ds);
8029 shash_destroy(&ofproto_shash);
8030 free(sorted_ofprotos);
8033 unixctl_command_reply(conn, ds_cstr(&ds));
8038 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8039 int argc OVS_UNUSED, const char *argv[],
8040 void *aux OVS_UNUSED)
8042 struct ds ds = DS_EMPTY_INITIALIZER;
8043 const struct ofproto_dpif *ofproto;
8044 struct subfacet *subfacet;
8046 ofproto = ofproto_dpif_lookup(argv[1]);
8048 unixctl_command_reply_error(conn, "no such bridge");
8052 update_stats(ofproto->backer);
8054 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8055 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
8057 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8058 subfacet->dp_packet_count, subfacet->dp_byte_count);
8059 if (subfacet->used) {
8060 ds_put_format(&ds, "%.3fs",
8061 (time_msec() - subfacet->used) / 1000.0);
8063 ds_put_format(&ds, "never");
8065 if (subfacet->facet->tcp_flags) {
8066 ds_put_cstr(&ds, ", flags:");
8067 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8070 ds_put_cstr(&ds, ", actions:");
8071 if (subfacet->slow) {
8072 uint64_t slow_path_stub[128 / 8];
8073 const struct nlattr *actions;
8076 compose_slow_path(ofproto, &subfacet->facet->flow, subfacet->slow,
8077 slow_path_stub, sizeof slow_path_stub,
8078 &actions, &actions_len);
8079 format_odp_actions(&ds, actions, actions_len);
8081 format_odp_actions(&ds, subfacet->actions, subfacet->actions_len);
8083 ds_put_char(&ds, '\n');
8086 unixctl_command_reply(conn, ds_cstr(&ds));
8091 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8092 int argc OVS_UNUSED, const char *argv[],
8093 void *aux OVS_UNUSED)
8095 struct ds ds = DS_EMPTY_INITIALIZER;
8096 struct ofproto_dpif *ofproto;
8098 ofproto = ofproto_dpif_lookup(argv[1]);
8100 unixctl_command_reply_error(conn, "no such bridge");
8104 flush(&ofproto->up);
8106 unixctl_command_reply(conn, ds_cstr(&ds));
8111 ofproto_dpif_unixctl_init(void)
8113 static bool registered;
8119 unixctl_command_register(
8121 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
8122 2, 6, ofproto_unixctl_trace, NULL);
8123 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8124 ofproto_unixctl_fdb_flush, NULL);
8125 unixctl_command_register("fdb/show", "bridge", 1, 1,
8126 ofproto_unixctl_fdb_show, NULL);
8127 unixctl_command_register("ofproto/clog", "", 0, 0,
8128 ofproto_dpif_clog, NULL);
8129 unixctl_command_register("ofproto/unclog", "", 0, 0,
8130 ofproto_dpif_unclog, NULL);
8131 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8132 ofproto_dpif_self_check, NULL);
8133 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8134 ofproto_unixctl_dpif_dump_dps, NULL);
8135 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8136 ofproto_unixctl_dpif_show, NULL);
8137 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8138 ofproto_unixctl_dpif_dump_flows, NULL);
8139 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8140 ofproto_unixctl_dpif_del_flows, NULL);
8143 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8145 * This is deprecated. It is only for compatibility with broken device drivers
8146 * in old versions of Linux that do not properly support VLANs when VLAN
8147 * devices are not used. When broken device drivers are no longer in
8148 * widespread use, we will delete these interfaces. */
8151 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8153 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8154 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8156 if (realdev_ofp_port == ofport->realdev_ofp_port
8157 && vid == ofport->vlandev_vid) {
8161 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8163 if (ofport->realdev_ofp_port) {
8166 if (realdev_ofp_port && ofport->bundle) {
8167 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8168 * themselves be part of a bundle. */
8169 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8172 ofport->realdev_ofp_port = realdev_ofp_port;
8173 ofport->vlandev_vid = vid;
8175 if (realdev_ofp_port) {
8176 vsp_add(ofport, realdev_ofp_port, vid);
8183 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8185 return hash_2words(realdev_ofp_port, vid);
8188 /* Returns the ODP port number of the Linux VLAN device that corresponds to
8189 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
8190 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
8191 * it would return the port number of eth0.9.
8193 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
8194 * function just returns its 'realdev_odp_port' argument. */
8196 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8197 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
8199 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8200 uint16_t realdev_ofp_port;
8201 int vid = vlan_tci_to_vid(vlan_tci);
8202 const struct vlan_splinter *vsp;
8204 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
8205 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8206 hash_realdev_vid(realdev_ofp_port, vid),
8207 &ofproto->realdev_vid_map) {
8208 if (vsp->realdev_ofp_port == realdev_ofp_port
8209 && vsp->vid == vid) {
8210 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
8214 return realdev_odp_port;
8217 static struct vlan_splinter *
8218 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8220 struct vlan_splinter *vsp;
8222 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8223 &ofproto->vlandev_map) {
8224 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8232 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8233 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8234 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8235 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8236 * eth0 and store 9 in '*vid'.
8238 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8239 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8242 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8243 uint16_t vlandev_ofp_port, int *vid)
8245 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8246 const struct vlan_splinter *vsp;
8248 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8253 return vsp->realdev_ofp_port;
8259 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8260 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8261 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8262 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8263 * always the case unless VLAN splinters are enabled), returns false without
8264 * making any changes. */
8266 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8271 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8276 /* Cause the flow to be processed as if it came in on the real device with
8277 * the VLAN device's VLAN ID. */
8278 flow->in_port = realdev;
8279 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8284 vsp_remove(struct ofport_dpif *port)
8286 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8287 struct vlan_splinter *vsp;
8289 vsp = vlandev_find(ofproto, port->up.ofp_port);
8291 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8292 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8295 port->realdev_ofp_port = 0;
8297 VLOG_ERR("missing vlan device record");
8302 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8304 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8306 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8307 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8308 == realdev_ofp_port)) {
8309 struct vlan_splinter *vsp;
8311 vsp = xmalloc(sizeof *vsp);
8312 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8313 hash_int(port->up.ofp_port, 0));
8314 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8315 hash_realdev_vid(realdev_ofp_port, vid));
8316 vsp->realdev_ofp_port = realdev_ofp_port;
8317 vsp->vlandev_ofp_port = port->up.ofp_port;
8320 port->realdev_ofp_port = realdev_ofp_port;
8322 VLOG_ERR("duplicate vlan device record");
8327 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8329 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8330 return ofport ? ofport->odp_port : OVSP_NONE;
8333 static struct ofport_dpif *
8334 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8336 struct ofport_dpif *port;
8338 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8339 hash_int(odp_port, 0),
8340 &backer->odp_to_ofport_map) {
8341 if (port->odp_port == odp_port) {
8350 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8352 struct ofport_dpif *port;
8354 port = odp_port_to_ofport(ofproto->backer, odp_port);
8355 if (port && &ofproto->up == port->up.ofproto) {
8356 return port->up.ofp_port;
8362 const struct ofproto_class ofproto_dpif_class = {
8397 port_is_lacp_current,
8398 NULL, /* rule_choose_table */
8405 rule_modify_actions,
8414 get_cfm_remote_mpids,
8419 get_stp_port_status,
8426 is_mirror_output_bundle,
8427 forward_bpdu_changed,
8428 set_mac_table_config,